Siobhan Sutcliffe,Kala Visvanathan,
Elizabeth A. Platz
SUMMARY OF KEY POINTS
|
Epidemiology is a design and analytical method used to investigate disease occurrence in human populations. Epidemiology encompasses several different design approaches, including randomized trial and observational approaches; this chapter focuses on observational approaches.
|
EPIDEMIOLOGY
Epidemiology is a design and analytical method used to investigate disease occurrence in human populations. Although traditionally used to test hypotheses about disease causation, epidemiologic methods also can be used to address a wider range of hypotheses in biomedical research, including translational and clinical hypotheses generated in oncology. Epidemiology encompasses several different design approaches, including experimental (randomized trial) and observational approaches, but this chapter focuses on observational approaches. To motivate the use of epidemiology in oncology, two hypothetical translational and clinical research questions are posed and used to illustrate concepts and approaches throughout this chapter:
|
1. |
Prognosis: Basic scientists recently have identified a protein expressed to a greater extent in high- than low-grade prostate tumors and not expressed in normal adjacent epithelium. This protein can be measured in tissue using a newly developed sensitive technique. To help identify which men with prostate cancer should receive adjuvant therapy following prostatectomy, investigators would like to know whether high expression of this protein predicts prostate cancer recurrence after prostatectomy independent of known prognostic characteristics. |
|
|
2. |
Therapeutic effectiveness: Researchers have determined that a gene encoding a particular P450 metabolic enzyme enhances the action of a chemotherapeutic agent currently being evaluated in combination with standard therapy for treatment of non-Hodgkin's lymphoma. To help determine which patients with non-Hodgkin's lymphoma may benefit most from this new chemotherapeutic agent, investigators would like to know whether a particular haplotype in the gene that encodes the P450 metabolic enzyme predicts overall survival after treatment with the chemotherapeutic agent in combination with standard therapy. |
STATING RESEARCH QUESTIONS AS HYPOTHESES TESTABLE USING EPIDEMIOLOGIC METHODS
To determine the optimal epidemiologic approach for addressing translational and clinical research questions in oncology, the research questions should be restated as testable hypotheses. Stating a research question as a testable hypothesis involves determining which groups should be compared. For instance, the prognosis and therapeutic effectiveness research questions can be restated as testable hypotheses as follows:
|
Prognosis: To test whether patients with high expression of the tissue marker are more likely to recur after prostatectomy than patients with low expression of the tissue marker. |
||
|
Therapeutic effectiveness: To test whether patients with non-Hodgkin's lymphoma with a particular haplotype are more likely to survive than patients without the haplotype after treatment with the new chemotherapeutic agent in combination with standard therapy. |
ELEMENTS IN THE DESIGN OF STUDIES USING EPIDEMIOLOGIC METHODS
Defining the Target, Source, and Study Populations
Once the research question is defined and stated as a testable hypothesis, the population (source population) from which study participants (study population) will be drawn can be identified. In selecting the source population, investigators should be cognizant of all patients who may benefit from the results of the study (target population), beyond those included in the study ( Fig. 25-1 ).
|
Figure 25-1 Relationships among study, source, and target populations. |
In the cancer center setting, several source populations from which the study population can be drawn typically are readily available (convenience sample): patients seeking screening and risk reduction options because of elevated risks of specific cancers; patients referred because of signs or symptoms suggestive of cancer; patients presenting with newly diagnosed cancer for treatment or a second opinion; patients under surveillance during and after completing treatment; and patients seeking salvage or palliative care for recurrence. For the two hypothetical research questions, possible source populations might be the following:
|
Prognosis: Patients newly diagnosed with prostate cancer eligible for prostatectomy. |
||
|
Therapeutic effectiveness: Patients with non-Hodgkin's lymphoma eligible to receive the new chemotherapeutic agent. |
Although they provide a good start, these definitions of the source population still are too broad. The source population should be narrowly defined with respect to person, place, and time. Doing so reduces the heterogeneity among individuals under study to help investigators draw sound inferences. For the two hypothetical translational and clinical research questions, the source populations can be more narrowly defined as follows:
|
Prognosis: Men newly diagnosed with clinical stage T1c or T2 prostate adenocarcinoma who underwent radical retropubic prostatectomy between 1993 and 2002 at one academic hospital and for whom resected prostate tissue was collected and stored. |
||
|
Therapeutic effectiveness: Adult men and women newly diagnosed with diffuse large β-cell lymphoma randomized to the new chemotherapeutic agent arm of a multicenter clinical trial evaluating the efficacy of this agent in combination with standard therapy. All trial participants were required to have a histologically confirmed diagnosis of diffuse large β-cell lymphoma according to the World Health Organization Lymphoma Classification, to be less than 60 years of age, to be CD20-positive, and to have at least two risk factors according to the Age-Adjusted International Prognostic Index. Participants were enrolled in 2000 and followed for 5 years. |
The study population typically is a subset of the source population. Reasons for not including all individuals from the source population in the study include feasibility issues, such as assay cost and throughput; patient willingness to participate; and availability or completeness of medical records and biologic specimens. For instance, in the prognosis example, the investigators may have intended to enroll all men with early-stage prostate cancer treated by radical prostatectomy from 1993–2002. However, some men may have chosen not to participate and, therefore, were not included in the study population, although they were included in the source population. Alternatively, the investigators may assemble the study population in 2007 by searching the medical records for men who underwent prostatectomy for early-stage prostate cancer between 1993 and 2002. Some of these men may no longer have archived prostatectomy specimens and, thus, cannot be included in the study population, despite being in the source population.
In the therapeutic effectiveness example, the source population is the group of participants consented and randomized to the new chemotherapeutic agent arm of the trial. At the outset, participants may have been asked to provide consent for both randomization to treatment and collection of blood specimens for genetic testing, including testing for the new clinical research question. In this case, the source and study population are one and the same. Alternatively, the new clinical research question may have been posed subsequent to the start of the trial. In this case, the investigators would have had to approach participants at a later trial visit to collect blood specimens for genetic testing. The study population might then differ from the source population because some participants might not have consented to this new correlative study or some might no longer have been participating in the trial because of death or other reasons for withdrawal. Investigators should always consider reasons for differences between the source and study population, as these may affect the results of the study (see section on Comparability).
As stated earlier, when selecting the source population, investigators always should be aware of the target population. For the two hypothetical translational and clinical research questions, possible target populations might be the following:
|
Prognosis: Men newly diagnosed with clinically organ-confined prostate cancer. |
||
|
Therapeutic effectiveness: Adult men and women less than 60 years of age newly diagnosed with CD20-positive diffuse large β-cell lymphoma with at least two risk factors according to the Age-Adjusted International Prognostic Index. |
Although study populations should be representative of the target population when estimating the occurrence of disease or factor in the target population (e.g., Americans living with cancer in 2007), the source and study populations do not necessarily have to be representative of all patients in the target population when addressing translational or clinical research questions that involve comparing the experiences of two or more groups.
For instance, in the therapeutic effectiveness example, members of the study population may be more likely to be white, younger, and healthier, on average, than members of the target population. This difference might arise because members of different groups may choose to participate to differing extents even when effort is made to recruit adequate proportions of each group. Given this difference, an important question to consider is whether findings from the study population will apply to the broader target population. For example, if the new chemotherapeutic agent trial is conducted primarily among white, younger, and healthier patients and it is known that white individuals have a higher frequency than non-white individuals of the haplotype that produces a version of the P450 enzyme that enhances the action of the new agent, would the finding that white patients with the haplotype are more likely to survive after treatment than white patients without the haplotype also apply to non-white patients? All else being equal, it might be expected that the findings would apply; what might differ is the percentage of patients in a particular racial or ethnic group who have the beneficial haplotype and who would thus benefit from treatment with the new agent. Of course, if all else is not equal, then the findings might not apply perfectly. For example, if non-white patients tend to have more comorbidities for which they are simultaneously prescribed other medications that compete for activation by the enzyme, then the new agent might be less effective in non-white patients with the haplotype than in white patients with the haplotype. Thus, the gain in survival with the haplotype would appear to be lower in non-whites than in whites. Note that in this example the new agent might still be equally as effective in white and non-white participants with the haplotype who do not have comorbidities, and less effective, but equally so, in white and non-white participants with the haplotype who have comorbidities and are taking the same array of drugs activated by the enzyme.
Therefore, the message is: Before embarking on a study, contemplate the likely generalizability of the findings to other groups of patients beyond those who are most likely to participate in a translational or clinical research study. Then, at the minimum, collect adequate information to characterize the study population so that differences between the participants and the target population are known, and, optimally, design the study and enroll members of the target population so that differences in findings among key segments of the target population may be determined directly.
Defining the Endpoint of Interest
Once the study population is defined, a detailed definition of the endpoint of interest (case definition) should be developed. Depending on the translational or clinical research question, a wide range of endpoint definitions is possible in oncologic research. These include cancer-specific endpoints, such as cancer diagnosis, which can be further divided by stage (e.g., localized versus advanced stage), histology, or expression of specific receptors; cancer recurrence or progression; or cancer-specific death. Other possible endpoints include diagnosis of second cancers, death from all causes, or short- and long-term complications of treatment among cancer survivors (e.g., cognition, peripheral neuropathy, or neutropenic sepsis). Optimally, the case definition should be narrow and specific to limit the endpoint spectrum to only those relevant or believed to be relevant to the factor of interest. For the prognosis and therapeutic effectiveness examples, the endpoint definitions might be as follows:
|
Prognosis: Prostate cancer recurrence as defined by a rise in serum prostate-specific antigen (PSA) concentration >0.2 ng/mL on two consecutive tests, metastasis, or death from prostate cancer as the underlying cause. |
||
|
Therapeutic effectiveness: Death from diffuse large β-cell lymphoma as the underlying cause. |
Although the endpoint definition ideally should be narrowly defined, investigators tend to use broader endpoint definitions to increase the number of cases. In doing so, however, they may include more cases not related to the factor of interest. In the prognosis example, if the tissue marker is expressed only in those cases that have the potential to metastasize and cause death, then the case definition—serum PSA concentration >0.2 ng/mL on two consecutive tests, metastasis, or death from prostate cancer as the underlying cause—is a sensitive case definition; the investigators are unlikely to miss cases that ultimately will metastasize and result in death. On the other hand, this case definition is not very specific; many of the men who only have PSA recurrence might never develop metastases or die from their prostate cancer.
The undesired effect of a heterogeneous population of cases, which increases the opportunity for including cases that are not related to the factor of interest, is that the association between the factor and the endpoint will be diluted. One way to investigate the possibility of a heterogeneous case mix is to perform separate analyses by characteristics of the endpoint, for example, by PSA recurrence only or by metastasis and death from prostate cancer, or, more generally, by tumor histology, tumor-node-metastasis stage, differentiation status, location in the organ, size, and known molecular markers (e.g., estrogen/progesterone receptor positivity in breast cancer, BCL/ABL translocation in chronic myelogenous leukemia).
Defining the Factor of Interest
As for the endpoint of interest, the factor of interest in oncologic research also can be wide-ranging, and does not need to be limited to a simple dichotomy of exposed and unexposed. Depending on the translational or clinical research question, the factor of interest can be the following:
|
• |
Exposures, whether measured in a participant's environment (e.g., air pollution, cigarette advertisements); as the participant's behaviors and their extent (e.g., pack-years of cigarette smoking, frequency of intake and dose of selenium from the diet and supplements, cancer screening uptake and frequency, frequency and dose of prescription medications, or reproductive history); or in the participant's blood or target organs (e.g., biomarkers of internal dose of selenium or DNA damage caused by ingestion of heterocyclic amines formed in charred meat) |
|
|
• |
Markers of susceptibility, whether measured as germline genetic variation, inherited epigenetic variation (e.g., imprinting), or, possibly, family history of cancer as a surrogate for susceptibility |
|
|
• |
Markers of biological pathways influenced by both genetic and nongenetic factors (e.g., age at menarche and menopause, circulating sex steroid hormones, or tissue cytokines) |
|
|
• |
Characteristics of premalignant lesions, cancer cells/tumors and surrounding tissues (e.g., altered protein/receptor expression, gain or loss of CpG methylation, chromosomal aberrations, or telomere shortening) among other attributes of the individuals, their tissues, their cancers, and their environments. |
In the two hypothetical translational and clinical questions, definitions of the factors of interest might be as follows:
|
Prognosis: Expression of the tissue marker in foci of prostate adenocarcinoma as detected by image analysis. A composite measure of immunohistochemical staining intensity and the proportion of cells staining positive is used. For possible future use in the clinic, the distribution of the composite measure is dichotomized as high versus low expression. |
||
|
Therapeutic effectiveness: Carrying no copies versus at least one copy of the specified haplotype that is known to encode a catalytically more active version of the P450 enzyme. Germline DNA extracted from peripheral blood lymphocytes is used to determine haplotypes. |
Note that for these particular examples, the factors of interest have been dichotomized for greater ease of description in later analytic sections, although more complicated expressions of the factors could be used (e.g., tissue marker—continuous expression or none, low, medium, high expression; haplotype—0, 1, or 2 copies).
In defining the factor of interest, investigators must consider the most relevant and feasible measure of the factor (factor definitions) in relation to the endpoint. For the prognosis and therapeutic effectiveness examples, the definitions of the factors of interest are relatively straightforward because the investigators can measure the most relevant factors directly at the most relevant times relative to the endpoint. In the prognosis example, the investigators are interested in testing whether patients with high expression of the tissue marker in the tumor are more likely than patients with low expression to have recurrences after prostatectomy. The most relevant factor of interest in this case is expression of the tissue marker in the cancer at the time of prostatectomy. In the therapeutic effectiveness example, the investigators are interested in testing whether patients with diffuse large β-cell non-Hodgkin's lymphoma with a particular haplotype are more likely to survive than patients without the haplotype after treatment with the new chemotherapeutic agent in combination with standard therapy. In this case, the most relevant factor of interest is a particular haplotype in the gene encoding one of the P450 enzymes. The time at which haplotypes are determined is less important because germline haplotypes remain the same throughout life.
Although the factor definitions for the hypothetical examples were relatively straightforward, there may be other situations in which the factor of interest is much more difficult to define. As an example, in a prognostic study of prostate cancer, the investigators may be interested in testing whether obesity is related to prostate cancer recurrence. However, deciding which is the most relevant measure of obesity (e.g., self-reported or measured/calculated weight, body mass index, waist-to-hip ratio, or percent body fat) and the most relevant and feasible time to measure obesity (e.g., at the time of prostatectomy, in the 10 years before prostatectomy, or in adolescence) may be difficult.
For instance, the most feasible factor for the investigators to measure might be body mass index calculated from self-reported weight and height at the time of prostatectomy, but the most relevant factor for prostate cancer prognosis might be the amount of adipose tissue at the time of prostatectomy. If the investigators choose to use the most feasible factor, then they may observe a weaker association between self-reported body mass index and prostate cancer recurrence than if they had used percent body fat depending on how well self-reported body mass index correlates with the amount of adipose tissue at the time of prostatectomy.
As a more extreme example, if the most relevant factor is the amount of adipose tissue during adolescence (at the time of prostate maturation), but the investigators choose to use the most feasible factor, then they may not observe an association between self-reported body mass index at the time of prostatectomy and prostate cancer recurrence if self-reported body mass index is not correlated (or not sufficiently correlated) with the amount of adipose tissue during adolescence. The investigators might then conclude that obesity is not associated with prostate cancer recurrence, even though obesity during adolescence might be important for prostate cancer prognosis. Therefore, the investigators must carefully consider the factor definition in relation to the proposed hypothesis. Defining and assessing the factor is an art, and may involve trade-offs in feasibility, accuracy, and relevance.
Design Considerations for Causal Inference: The Cohort Study
The typical goal of studies using epidemiologic methods, including translational and clinical studies, is to evaluate whether a factor is causally related to an endpoint of interest. Important considerations in drawing causal inferences include assessment of temporality and comparability.
Temporality
For a factor to be causally related to an endpoint of interest it must necessarily precede the occurrence of the endpoint; that is, temporality must be established. This is accomplished in experimental designs, whether in the laboratory or in a clinical trial, by randomly assigning a factor to laboratory animals or human participants, respectively, who do not currently have the endpoint of interest, and then observing these animals or participants over time for endpoint development. Therefore, in these designs it is clear that the factor preceded the endpoint. By contrast, in observational studies, investigators do not assign the factor to study participants. Rather, the investigators measure the existing factor status. Optimally, to establish temporality, the investigators measure the factor in individuals who have not yet experienced the endpoint of interest, but who are at risk for the endpoint, and then they follow those individuals over time for endpoint development. This design is called the prospective cohort study ( Fig. 25-2 ). Prospective indicates that the factor status and its assessment predate the development of the endpoint. Cohort study denotes follow-up of a specified group of initially endpoint-free individuals over time for endpoint development. In a simple analysis of the data from a prospective cohort study, the investigators compare the extent of occurrence of the endpoint in those with and without the factor. The prospective cohort study is the gold standard observational design. Returning to the two hypothetical translational and clinical questions, the prospective cohort design is illustrated as follows:
|
Prognosis: Cohort members are enrolled at the time of prostatectomy. Expression of the marker is assessed in foci of prostate adenocarcinoma from tissue resected at prostatectomy. Cohort members are followed through 2006 (maximum follow-up of 14 years) for recurrence. The frequency of recurrence is compared between those with high and low expression of the tissue marker ( Fig. 25-3A ). |
||
|
Therapeutic effectiveness: Cohort members are trial participants randomized to the new chemotherapeutic agent arm. Presence or absence of the haplotype is determined from peripheral blood lymphocytes collected at the start of the trial. Patients are followed through to the end of the trial for death from diffuse large β-cell lymphoma. Survival is compared between those with and without the haplotype. Note that, although the parent study is a randomized clinical trial (of the efficacy of a new chemotherapeutic agent in combination with standard therapy versus standard therapy alone), this example question is not itself investigated using the randomized trial design because participants are not randomized to the factor of interest—haplotype for the gene encoding the enzyme ( Fig. 25-3B ). |
|
Figure 25-2 Schematic of the prospective cohort study. |
|
Figure 25-3 A, Schematic of the hypothetical prognostic research question as a prospective cohort study. B, Schematic of the hypothetical therapeutic effectiveness research question as a prospective cohort study. |
As an alternative to the prospective cohort study design, the investigators may instead identify a cohort of individuals and reconstruct their past factor status using pre-existing records or specimens. This design is called the retrospective cohort study ( Fig. 25-4 ). In this case, retrospective denotes that the factor status predates endpoint development, but the assessment of the factor postdates endpoint development. Otherwise the approach is the same as in a prospective cohort study. Returning to the hypothetical translational and clinical questions, the retrospective cohort design is illustrated as follows:
|
Prognosis: The investigators assemble the study population in 2007 by searching the medical records for men who underwent prostatectomy for clinically organ-confined prostate cancer between 1993 and 2002. |
||
|
Therapeutic effectiveness: After the trial has finished, the investigators identify all participants in the new chemotherapeutic agent arm of the trial, and search each site of the multicenter clinical trial for blood specimens archived at the first visit of the trial. |
|
Figure 25-4 Schematic of the retrospective cohort study. |
Although the distinction may seem subtle, the major difference between the prospective and retrospective designs is the position of the investigators relative to factor assessment and follow-up for the endpoint of interest—whether the investigators (1) assemble the cohort and collect information on the factor or samples in which the factor status will be assessed at the start of follow-up, or (2) assemble the cohort sometime after the start of the follow-up and attempt to reconstruct the factor status at the time of the start of follow-up using existing records or samples. The retrospective cohort design may not be optimal because of uncertainties in factor assessment, including incomplete or missing medical or other records, and missing or degraded samples in which the factor status would have been assessed. For example, in the prognosis study, prostate tissue from some cohort members might no longer be available because it has already been used for other tissue studies.
A further problem with factor assessment in retrospective cohort studies is that the extent of uncertainty in the factor assessment may differ according to whether or not a participant experienced the endpoint. Say, for instance, in the therapeutic effectiveness example that blood specimens were not saved from the first visit of the new chemotherapeutic agent trial. In this case, investigators would have to collect blood samples from participants at a later date in the trial to determine the haplotype status of each participant. Although not consistent with the typical retrospective cohort study design, this approach still maintains temporality because germline haplotypes are present from the time of conception and are invariant throughout life and, therefore, must have preceded the endpoint of interest. However, by collecting blood at a later date in the trial, the investigators may still have introduced a distortion (bias) into the results of the study.
For example, suppose that absence of the haplotype is associated with shorter duration of survival after recurrence compared with presence of the haplotype. At the time of blood draw partway through the trial, proportionally more participants without the haplotype (a fact not yet known to the investigators) who had recurrences would not be able to provide a blood specimen because they had already died as opposed to those with the haplotype who had recurred, whereas there would be no difference in the ability to obtain a blood specimen from those who had not yet recurred among those with and without the haplotype. Thus, this differential inclusion of participants by both haplotype and death could then make it appear as though the haplotype were not as protective for death from lymphoma after treatment, not associated with death from lymphoma at all, or even associated with an increased risk of death from lymphoma, depending on the extent to which participants without the haplotype who died were excluded from the study because haplotype status could not be determined. For this and other reasons, investigators should think very carefully about the implications of reconstructing a cohort and its factor status to avoid biased results. Differential inclusion of participants by both the factor of interest and the endpoint of interest, as illustrated in this example, is called selection bias, a topic that will be discussed in the next section of the chapter.
Comparability
Another major consideration when drawing inferences from translational and clinical studies that use epidemiologic methods is comparability of the two (or more) groups being contrasted. To help think about comparability, consider laboratory animal experiments: the researcher has complete control over the groups being contrasted and, therefore, can make the groups as similar as possible on all aspects except for the factor of interest. The laboratory researcher typically selects genetically identical animals, assigns the factor (or treatment) to half of the animals but otherwise feeds and handles the animals in exactly the same manner, and then euthanizes the animals after a certain length of time to assess the endpoint of interest. The only difference between the two groups being contrasted is the factor of interest; that is, the two groups are comparable save the factor. When coupled with temporality (i.e., the factor/treatment was given before the endpoint occurred), comparability allows the laboratory researcher to conclude that the factor was the cause of the endpoint.
Trialists attempt to accomplish a similar objective in randomized clinical trials by assigning participants to the treatment (or other intervention) group or placebo (or standard of care) group through a random process. This process helps to ensure that participants in the treatment and placebo groups have similar distributions of known and unknown inherent and modifiable characteristics, including those related to disease incidence or prognosis. Although randomization cannot ensure comparability, the larger the number of participants randomized, the more likely it is that the two groups will have similar distributions of known and unknown characteristics. This level of comparability allows the investigators to conclude that the treatment was the likely cause of any observed differences in the occurrence of the endpoint between the two groups.
Had the investigators not used a randomized design, they would not have been able to reach such a strong conclusion about the relation between the treatment and the endpoint. If the investigators had not randomized participants, but instead had allowed patients or their oncologists to choose whether or not to take the new treatment, sicker patients (i.e., those already more likely to die of their cancers) might have been more likely to select the new therapy over the standard of care in the hope that cure might be more likely. In this case, patients receiving the new therapy would not have been comparable to patients receiving the standard of care with respect to disease prognosis. If these two groups had been compared with respect to the occurrence of the endpoint (e.g., death from cancer), as might be the case in a prospective cohort study, then the new treatment would have appeared to be less efficacious than it should. This bias is called selection bias or confounding by indication. Selection bias occurs when enrollment or inclusion of individuals with and without the factor (e.g., those who selected the new treatment and those who selected the standard of care) is differentially dependent on their likelihood of developing the endpoint of interest (e.g., on their likelihood of dying from cancer).
Just as randomization helps to equalize the two groups being contrasted on disease incidence or prognosis, it also helps to equalize the two groups on all other known and unknown characteristics that also could alter the relation between a factor and an endpoint. For instance, in the preceding example, if the investigators had not randomized participants to the new treatment or standard of care, but instead had allowed them to select their treatment, patients or their oncologists might have selected treatment based on cost, rather than (or in addition to) disease prognosis. If oncologists only prescribed the new treatment to patients with higher socioeconomic status whose health insurance would be more likely to cover the cost of the expensive new treatment, and if patients with higher socioeconomic status were less likely to die from their cancer because of better diet, less adverse health behaviors, and greater access to adjuvant therapies, then the new treatment would have appeared to be more efficacious than it should simply because of differences in socioeconomic status between the two groups. In this case, socioeconomic status is called a confounder. Simply defined, confounders are factors that are (1) correlated with the factor of interest; (2) associated with the endpoint; and (3) not steps in the pathway in how the factor of interest influences the endpoint. In the preceding example, socioeconomic status is a confounder because it is (1) correlated with treatment; (2) associated with death from cancer; and (3) not a step in the pathway between treatment and death from cancer because it is unlikely that treatment influenced socioeconomic status, which then influenced death from cancer.
For reasons of both selection bias and confounding, randomized trials are the gold standard design for testing interventions, whether medications, devices, screening and diagnostic tests, other medical procedures and surgeries, or behavioral interventions. However, there are, of course, many circumstances in which a randomized trial would not be ethical, such as when a screening test is already in routine clinical practice, but was never formally tested for efficacy, or when the factor of interest is believed to be harmful. In addition, not all factors of interest in oncology are amenable to randomization because they are inherent characteristics of individuals or are otherwise nonmodifiable, such as family history of breast cancer, or age at menarche. In these cases, the cohort study may be an ethical alternative, and with proper design and analytic consideration, should provide sound inferences. Selection bias can be reduced in cohort studies by restricting the extent of disease among participants (e.g., by limiting the source population to a single stage at diagnosis) or by collecting detailed information on extent of disease at the start of the study, so that it may be taken into account in the statistical analysis (e.g., adjustingon stage at diagnosis, see the section on Comparing Two Groups: Measures of Association). Selection bias also is much less likely in cohort studies in which the baseline study population is free of a diagnosis of the endpoint of interest (e.g., a cohort of men and women who do not currently have a cancer diagnosis) as is typically the case in etiologic studies (i.e., studies seeking to identify the causes of an endpoint of interest). Selection bias also is less likely in prospective cohort studies than in retrospective cohort studies (see the illustration of selection bias in the retrospective cohort example of therapeutic effectiveness under the section on Design Considerations for Causal Inference: The Cohort Study).
Confounding can be minimized in cohort studies by collecting or measuring information on potential confounders, such as socioeconomic status, at the start of the study. These confounders can then be taken into account in either the design phase (e.g., by restricting to one level of the confounder—high socioeconomic status) and/or the statistical analysis of the study (by adjustment; see the section on Comparing Two Groups: Measures of Association). However, even with forethought, investigators may not always be able to anticipate all confounders or may not measure them perfectly. In such cases, residual confounding may be present and may still explain any observed associations.
Returning to the two hypothetical translational and clinical questions, randomized trials addressing the prognostic ability of the tissue biomarker or the difference in therapeutic effectiveness of the new agent by the enzyme haplotype are not feasible. Both the tissue biomarker and the haplotype are nonmodifiable characteristics of the patients’ tumors or the patients themselves. Therefore, these questions are optimally addressed in cohort studies. Comparability in these example cohort studies is discussed here:
|
Prognosis: Patients whose prostate tumors have high and low expression of the tissue marker are made comparable with respect to a known prognostic factor, clinical stage at diagnosis, in the design phase of the study by restricting patients to only those with clinical stage T1c or T2 disease. In the conduct of the study, information is collected on pathologic stage and grade and other key prognostic factors so that they may be taken into consideration in the analysis to make those patients with high and low expression of the marker even more comparable on likelihood of recurrence. At the time of prostatectomy, demographic, anthropometric, and behavioral characteristics also are collected. These potential confounders can be taken into consideration in the statistical analysis. This level of comparability will allow investigators to determine the independent prognostic value of the tissue marker. |
||
|
Therapeutic effectiveness: Although this study is embedded in a randomized trial of a new chemotherapeutic agent, the research question about the metabolic enzyme haplotype does not take advantage of the randomization. Only one arm of the trial is used: patients randomized to the new agent in combination with standard therapy. Therefore, comparability of those with and without at least one copy of the haplotype must be carefully considered. Patients with and without the haplotype are comparable with respect to histologic type of non-Hodgkin's lymphoma by design because only patients with diffuse large β-cell lymphoma were included in the trial. Participants also were required to be less than 60 years of age and to have at least two risk factors according to the Prognostic Index, making them more comparable on age and prognosis. Age, specific prognostic factors, and potentially confounding factors collected as part of the parent trial can be taken into consideration in the analysis to make those with and without at least one copy of the haplotype even more comparable on likelihood of survival with non-Hodgkin's lymphoma. This level of comparability will allow the investigators to determine the independent predictive capability of the haplotype for survival after treatment with the new agent. |
Another possible source of lack of comparability common to both experimental and observational designs is observation bias or information bias. This bias occurs when the accuracy of observing the endpoint differs between those with and without the factor. One possible source of observation bias in both experimental (including randomized trials) and observational designs is differential loss-to-follow-up (i.e., loss-to-follow-up that is differential with respect to the likelihood of the endpoint of interest) between those with and those without the factor.
For example, a new complementary cancer treatment is being tested against placebo for improvement in quality of life after cancer treatment. Quality of life is assessed at the start and then again at the end of the trial. Suppose that the new treatment improves quality of life in some patients, but in many other patients quality of life dramatically declines because of an adverse response to the treatment. Those in the complementary treatment arm who experience the adverse response are much less likely to continue participating in the trial than those whose quality of life improves. In the placebo arm in which many fewer patients have dramatic declines in quality of life, the likelihood of continued participation to the end of the trial is only slightly lower among those whose quality of life does not improve compared with those whose quality of life does improve. This bias would make the complementary treatment appear more efficacious than it is simply because patients receiving the treatment who experienced a decline in quality of life are under-represented in the analysis.
Another possible source of observation bias is detection bias or medical surveillance bias. This bias occurs when those with and without the factor of interest are surveyed for the endpoint to different degrees. For example, in a prospective study of obesity and prostate cancer recurrence, if urologists suspect that obesity may indeed increase the risk of recurrence, they might monitor obese men for a postprostatectomy rise in PSA more often than they monitor nonobese men. Thus, it may appear as though obesity is associated with earlier prostate cancer recurrence simply because obese men are monitored more frequently for recurrence.
In summary, comparability between those with and without the factor of interest is a major consideration in drawing causal inferences in studies using epidemiologic methods. To enhance comparability, investigators should think carefully about selection bias, confounding, and observation bias in the design, conduct, and analysis of their study.
Other Cohort Design Considerations
So far, the cohort design has been described as a single study population divided into those with and without the factor and then followed for the endpoint of interest. This approach of a cohort formed from a single study population is desirable because it helps to ensure that the two groups being contrasted are from a relatively homogeneous population with similar methods of factor, confounder, and endpoint assessment to enhance comparability between the two groups.
However, sometimes investigators cannot identify groups of participants with and without the factor of interest from the same population, and must resort to comparing groups from different populations. The classic example comes from cancer etiology studies in the occupational setting, in which all members of the study population have the factor of interest, such as employment at a large factory that manufactures a particular chemical. To get a sense of the risk experience the employees would have had if they had not been exposed to the chemical, an external unexposed comparison group is needed. One possible comparison group is employees at a factory that does not manufacture the particular chemical, but is otherwise similar to the factory that does manufacture the chemical. Another comparison group could be the entire population of the country in which the factory is located, if the risk of the endpoint (e.g., cancer) is known for that particular country. In using a national comparison group, investigators assume that the general population in that country is not exposed to the factor of interest and is otherwise comparable to the exposed group. This latter assumption should be carefully considered, for reasons such as the fact that workers tend to be healthier on average than the general population because they are able to work (healthy worker effect).
In the context of oncology, one setting in which investigators might need to use an external comparison group is in the introduction of a new cancer treatment before it is ready to be tested in a clinical trial. In this case, investigators might use a historical comparison group—a group of patients who were treated using the standard of care earlier in calendar time—to contrast with the current group of patients treated with the new surgical or pharmacologic cancer therapy. However, again care should be taken when comparing these two groups because of differences in the case mix and other patient characteristics over time (secular trends).
Statistical Analysis of Cohort Studies
Measures of the Occurrence of an Endpoint
The fundamental measure of occurrence of an endpoint (e.g., development of disease or recurrence of disease) in a study using the cohort design is the probability of the endpoint, called the cumulative incidence or risk. Most simply defined, the cumulative incidence is the number of new endpoints that occur during a particular time interval divided by the number of individuals at risk at the start of follow-up in a cohort study (or clinical trial). The numerator typically only includes new or incident endpoints, not existing or prevalent cases. In oncology, investigators usually are most interested in the first occurrence of an endpoint (e.g., cancer diagnosis or cancer recurrence). The denominator only includes those able to experience the endpoint (e.g., women who have had a hysterectomy are not at risk for endometrial cancer and would not be included in the denominator when estimating the cumulative incidence of endometrial cancer), and all of the endpoints must have arisen from among individuals at risk in the denominator. As an example of a cumulative incidence calculation, if 100 men and women were enrolled in a cohort study and 25 died by the end of 5 years of follow-up, then the cumulative incidence of death would be 25/100 or 0.25 over 5 years. The range of values for the cumulative incidence is 0 if none of the at-risk individuals experience the endpoint and 1 if all of the at-risk individuals experience the endpoint. The time interval over which the cumulative incidence is estimated must be specified: a 0.25 (or 25%) risk over 1 month is substantially larger than a 0.25 risk over the lifetime.
This simple cumulative incidence calculation assumes that all participants enter follow-up at the same time; follow-up time can be expressed as calendar time, age, or time on study (e.g., time since mastectomy irrespective of calendar year or age). It also assumes that follow-up only ends because of occurrence of the endpoint or end of the study (administrative censoring). In other words, participants who do not experience the endpoint of interest are assumed to remain at risk for the endpoint of interest (e.g., not die from another disease) and to remain under follow-up (e.g., not stop participating in the study) until the end of the study.
Rarely do all participants remain under follow-up until the end of the study. Some individuals may die of something other than the endpoint of interest. Others may choose not to continue participating or are withdrawn by the investigators. All of these individuals are considered lost to follow-up. In this case, the simple cumulative incidence calculation is not appropriate; the estimate will be too small because the at-risk denominator presumes that all participants were followed for the full study period.
One appropriate method to estimate cumulative incidence when there is loss-to-follow-up is the Kaplan-Meier method. This method allows all of the time during which an individual is at risk and under follow-up to be used, even for participants not observed for the full study period. The first step of the Kaplan-Meier method is to align all participants with respect to follow-up time (e.g., time since mastectomy). Then, at each time that an endpoint occurs, the probability of the endpoint is calculated among only those individuals who survived until the time of that particular endpoint (conditional probability of the endpoint). Lack of survival until the time that a particular endpoint occurs may be due to having already experienced the endpoint, no longer being at risk for other reasons (e.g., death from another cause). The group of individuals still at risk at the time each endpoint occurs is called the risk set. The next step in the Kaplan-Meier method is to calculate the probability of surviving to the time that each endpoint occurs for each of the risk sets (1 – conditional probability of the endpoint; conditional probability of survival). Then, the cumulative probability of survival up to the time of each endpoint is calculated as the product of the conditional probability of survival at the time of each endpoint and the cumulative probability of survival to the time of the prior endpoint.
The cumulative probability of survival over the entire follow-up period is the cumulative probability at the time of the last endpoint. The cumulative incidence over the entire follow-up period can then be calculated as the complement of the cumulative probability of survival over the entire follow-up period (1 – cumulative probability of survival; Figs. 25-5A and B ). The cumulative probabilities of survival at each time an endpoint occurs can be plotted against follow-up time to obtain the Kaplan-Meier survival curve ( Fig. 25-5C ). Note that survival curves may be generated for endpoints other than death; survival merely means not having had the endpoint of interest and still being at risk and under observation. Survival is assumed to be constant (flat line on the graph) until an endpoint occurs, at which time the cumulative survival drops to the value for the next interval. Direct examination of the Kaplan-Meier curve can show whether survival is constant or changes over follow-up time. In Figure 25-5C , the change in cumulative survival over the first 5 years is greater than the change in cumulative survival in subsequent years as shown by the difference in the steepness of the curve in the first 5 years compared with later years. The Kaplan-Meier curve also can be used to estimate such measures as median survival time by determining the time at which half of the population has experienced the endpoint of interest.
|
Figure 25-5 Use of the Kaplan-Meier method to calculate the cumulative incidence of the endpoint and to plot the survival curve. A, Time at risk for 10 hypothetical individuals. B, Calculation of the cumulative probabilities of survival and incidence for the 10 hypothetical individuals. C, Kaplan-Meier survival curve for the 10 hypothetical individuals. |
Two assumptions are made when using the Kaplan-Meier approach. The first is that individuals who do not experience the endpoint and do not survive until the next endpoint time have a similar risk of the endpoint as individuals who remain under follow-up. If this assumption is violated, then the estimated cumulative incidence will be incorrect. The second assumption is that no change in the background risk occurs over the accrual period (e.g., secular changes in the risk of cancer death might occur if there is a stage shift due to implementation of a new screening test and if this early detection followed by appropriate treatment produces longer survival than if not detected early). If this problem is present, then separate Kaplan-Meier estimates and curves should be presented for different intervals of accrual time.
Other methods of estimating cumulative incidence are available. For example, if the time that the endpoint occurs cannot be pinpointed, as may happen when cancer recurrence is detected on a blood test performed every 6 months, then the actuarial or life-table method may be used. The actuarial method is similar to the Kaplan-Meier method, except that the times of detection of the endpoints are binned into time intervals. For instance, in this cancer recurrence example, the endpoints would be binned into 6-month intervals (see Gordis[1] for further details).
Another measure of endpoint occurrence more typically used by epidemiologists is the incidence rate. When death is the endpoint, the incidence rate of death is usually called the mortality rate. The incidence rate conveys how fast the endpoint is occurring in a population at risk. It is used when follow-up for some participants begins later than for others or when some participants are not followed for the full study period. The incidence rate is calculated as the number of new endpoints divided by the person-time at risk. Like the cumulative incidence, the numerator is the number of incident endpoints. The denominator is the sum of the amount of time that each individual is at risk for the endpoint and is under follow-up. In Figure 25-6 , individual 1 contributes 1 person-year at risk before he develops the endpoint of interest, individual 3 contributes 2.5 person-years at risk before she stops participating in the study, and individual 10 contributes 14 person-years at risk by the end of the study. Together, these 10 individuals contribute 58 person-years at risk to the denominator of the incidence rate. Whereas the cumulative incidence requires a separate specification of the relevant time period over which it applies, time is inherent in the calculation of the incidence rate. Person-time at risk may be expressed in any unit of time—years, months, weeks, seconds—although person-years are most often used for cancer incidence, recurrence, and death. The incidence rate may range from 0 per person-time if no endpoints occur during follow-up to infinity per person-time if one or more endpoints occur during the briefest instant of follow-up.
|
Figure 25-6 Time at risk for 10 hypothetical individuals. Individual 1 contributes 1 person-year at risk before he develops the endpoint of interest; individual 3 contributes 2.5 person-years at risk before she stops participating in the study; and individual 10 contributes 14 person-years at risk by the end of the study. |
As for the calculation of cumulative incidence, several assumptions are made in calculating the incidence rate. The first two assumptions are the same as for the cumulative incidence: (1) individuals who do not experience the endpoint and do not survive until the next endpoint time have a similar risk of the endpoint as individuals who remain under follow-up; and (2) no change in the background risk occurs over the accrual period. A third assumption in calculating the incidence rate is the equivalence of person-time at risk over time and across individuals; that is, 1 person-year contributed by individual 10 during the first year of follow-up is the same as 1 person-year contributed by individual 10 during the last year of follow-up. Likewise, any year contributed by individual 5 is the same as any year contributed by individual 7. When this equivalence does not hold, for example when the incidence rate differs by age or calendar time and/or by characteristics of the participants, then separate incidence rates should be calculated by categories of age, time, or participant characteristics.
A third measure of endpoint occurrence is the hazard, which is defined as the instantaneous occurrence of the endpoint among those still at risk. Although commonly calculated as part of a ratio (hazard ratio), the hazard itself is calculated very rarely because of complexities involved with the calculation.
When should the cumulative incidence versus the incidence rate be presented as the measure of endpoint occurrence? It depends on the goal of the research. If investigators would like to generate information useful for communicating to patients (e.g., What is the likelihood that my cancer will recur after I am treated?), then the cumulative incidence might be the most useful measure. When communicating this measure to patients, oncologists also should make it clear that the cumulative incidence is not specific to any one particular patient, but instead refers to the average cumulative incidence in a group of patients with similar characteristics. The incidence rate typically is calculated in the context of etiologic studies. The risk and the rate are mathematically related, and with certain assumptions, the cumulative incidence can be estimated from the observed incidence rates (see Rothman[2] for further details).
Comparing Two Groups: Measures of Association
Testing a hypothesis usually involves comparing the occurrence of the endpoint in two or more groups that differ on a factor of interest. One approach is to compare the survival curves generated from the Kaplan-Meier method between individuals with and without the factor of interest. The statistical test used to determine whether the survival curves are likely different from each other is called the log-rank test.
Sometimes investigators would like to know more than merely whether or not the survival experiences between the two groups differ. Instead, they may want to quantify the magnitude of this difference. If the endpoint experiences of the two groups are expressed as cumulative incidences (calculated from the Kaplan-Meier or other methods) or rates, then either the ratio or the difference of these measures may be calculated. Ratio measures include the cumulative incidence ratio, rate ratio, or hazard ratio, collectively known as the relative risk (RR) of the endpoint. The RR is calculated as follows:
Cumulative incidence ratio = (cumulative incidence in group 1)/(cumulative incidence in group 2)
Rate ratio = (incidence rate in group 1)/(incidence rate in group 2)
The hazard ratio rarely is calculated directly because, as stated in the previous section, individual hazards are difficult to calculate.
The order of the two groups being compared in the calculation of the RR has direct bearing on its interpretation. Typically, in studies using epidemiologic methods, investigators are particularly interested in the endpoint experience in one group, the group with the factor of interest, as compared with a second group, the group without the factor of interest. In this case, the investigators would calculate the RR as follows:
RR = (occurrence of the endpoint in the group with the factor)/(occurrence of the endpoint in the group without the factor)
Alternatively, investigators might be interested in comparing those without the factor of interest to those with the factor. The choice of which group belongs in the numerator of the ratio and which belongs in the denominator depends on the statement of the research question. For instance, in the prognosis example, the investigators asked whether high expression of the tissue marker is associated with greater prostate cancer recurrence; thus, they would calculate the RR as the cumulative incidence of prostate cancer recurrence in patients with high expression of the tissue marker divided by the cumulative incidence of prostate cancer recurrence in patients with low expression of the tissue marker. If the cumulative incidence were higher among patients with high expression than among patients with low expression, then the RR would be greater than 1 and the magnitude of the RR would indicate how many times higher the cumulative incidence of recurrence was in those with high expression than in those with low expression of the tissue marker (i.e., it would indicate the strength of the association). If contrary to the hypothesis, the cumulative incidence were lower among patients with high expression than among patients with low expression, then the RR would be less than 1 and the magnitude of the RR would indicate how many times lower the cumulative incidence of recurrence was in those with high compared with low expression. The RR is unitless and can range from 0 to infinity.
Measures of the difference in the occurrence of the endpoint include the cumulative incidence difference and rate difference, collectively known as the risk difference. The risk difference is calculated as follows:
Cumulative incidence difference = (cumulative incidence in group 1) - (cumulative incidence in group 2)
Rate difference = (incidence rate in group 1) - (incidence rate in group 2)
As for the RR, the order of the two groups also is important in the interpretation of the risk difference. Typically, investigators calculate the risk difference as the occurrence of the endpoint in the group with the factor of interest minus the occurrence of the endpoint in the group without the factor of interest. The cumulative incidence difference and rate difference have the same units as the cumulative incidence and rate, respectively. The cumulative incidence difference can range from -1 to 1, while the rate difference can range from minus infinity to positive infinity.
Whereas the RR often is calculated in studies using epidemiologic methods, the risk difference is calculated less often. It usually is calculated only when the factor has been identified as a likely cause of the endpoint and when the absolute size or burden of the endpoint attributable to this factor is of interest. For this reason, the risk difference also is known as the attributable risk. Another way to interpret the attributable risk is the extent of occurrence of the endpoint that would have been avoided in individuals with the factor if they had never experienced the factor. Note that a high RR does not necessarily imply a high attributable risk. If, for instance, the cumulative incidence in those with the factor is 1 in 10,000,000 people over the lifetime and the cumulative incidence in those without the factor is 1 in 100,000,000 people over the lifetime, then the RR is 10 (large), but the attributable risk is only 9 in 100,000,000 over the lifetime (small). Sometimes the population attributable risk percent is reported as a measure of a population's excess burden due to some members of the population having the factor of interest. Another way to view this measure is the proportion of the occurrence of the endpoint that would have been avoided in a population if individuals with the factor had never experienced the factor. The population attributable risk percent is calculated as the cumulative incidence (rate) in the total population minus the cumulative incidence (rate) in those without the factor all divided by the cumulative incidence (rate) in the total population.
As alluded to in the discussion of comparability, it may be necessary to take potentially confounding factors into account in the estimation of the RR or the risk difference. Two main statistical approaches may be used: stratification or adjustment. Stratification involves estimating the RR or risk difference in categories of the potential confounder known as strata. Within each stratum of the confounder, all individuals have the same confounder status, and thus the stratum-specific association between the factor of interest and the endpoint is independent of the confounder. If an unconfounded summary estimate of the RR or risk difference is desired, then the summary estimate can be calculated as the weighted average (often using Mantel-Haenszel weights) of these independent stratum-specific estimates. Note that age-standardization, which is used routinely to adjust cancer incidence and mortality rates for confounding by age (because different populations may have different age structures), is merely a form of stratification with weights derived from a specified standard population, such as the 2000 U.S. Census Population as used by the National Cancer Institute's Surveillance, Epidemiology, and End Results Program (http://seer.cancer.gov/) or the World Standard Population as used by the World Health Organization (Globocan: http://www-dep.iarc.fr/).
Another statistical method that takes confounding into account is adjustment using regression techniques. Detailed descriptions of regression models can be found in biostatistics textbooks. Briefly, choice of the appropriate regression model is determined by the nature of the endpoint of interest. For instance, is the endpoint binary? This often is the type of endpoint studied in oncology (e.g., cancer diagnosis or cancer recurrence). Currently, there are no straightforward methods for modeling the cumulative incidence ratio when some participants are lost to follow-up, and thus it is difficult to calculate an adjusted estimate of association. If, instead, the rate ratio or the hazard ratio is the desired measure of association, then either Poisson regression or Cox proportional hazards regression, respectively, can be used to calculate adjusted estimates. The Cox model circumvents difficulties in directly estimating individual hazards by allowing the unmeasured hazard in those without the factor to vary from moment to moment, and then assuming that the ratio of the hazards in those with and without the factor is constant, allowing it to only estimate the hazard ratio.
Design Considerations for Causal Inference: Alternative Prospective Designs
Nested Case-Control Study
So far, this chapter has discussed only the cohort approach to addressing research questions. However, sometimes it is not feasible to evaluate the factor status of each participant in the cohort study, for example, when the factor is assessed using an expensive or labor-intensive method.
In this case, alternative prospective epidemiologic designs can be used that maintain the correct temporal sequence between the factor and the endpoint, but use a smaller sample size—the nested case-control study and the case-cohort study. In these designs, all individuals in the cohort must still be followed for the endpoint of interest, and material—either biological (e.g., blood specimen) or physical (e.g., paper medical records)—from which to ascertain the factor must be collected before the endpoint occurs. In the nested case-control study ( Fig. 25-7 ), all individuals who develop the endpoint of interest (or a random sample of these individuals) are selected as cases. Then, at each time that a case occurs, one or more individuals are randomly selected from the risk set, which consists of those who have not yet developed the endpoint and who are still at risk and under follow-up. These individuals are called controls. Only these cases and controls are then tested for the factor of interest.
|
Figure 25-7 Schematic of a nested case-control study within a prospective cohort of 10 hypothetical individuals. All cases are sampled. A subset of all possible controls is sampled using incidence density sampling. |
Looking at Figure 25-7 , at the time that individual 1 recurs, individuals 2 through 10 are still at risk of recurrence and under follow-up. Individual 9 is randomly selected to be the control. At the time that individual 4 recurs, individuals 5 through 10 are still at risk of recurrence and under follow-up. Individual 6 is randomly selected to be control. This method of control sampling is called incidence density sampling. Note that an individual who later becomes a case may be sampled earlier as a control for another case (individual 9). Additionally, the same individual may be sampled as a control for more than one case. Individual 9 could have been sampled as a control for both case 1 and case 4.
To help illustrate why this sampling approach is correct, consider how in the Kaplan-Meier method the conditional probability of the endpoint is estimated at each endpoint time as the number of cases divided by the number of individuals still at risk and under follow-up; that is, each individual contributes to the estimate of the conditional probability at each time that an endpoint occurs, including individuals who later become cases. In the nested case-control study, each control is sampled from the risk set to reflect the risk of the members of the risk set. As in the Kaplan-Meier method, each individual still at risk and under follow-up has the opportunity to be sampled each time an endpoint occurs.
To enhance the statistical ability to adjust for confounding, control subjects often are sampled such that they have characteristics similar to the cases, a process called matching. In a nested case-control study with incidence density sampling of controls, cases and controls are necessarily matched on follow-up time by design. Note that statistical analysis of matched cases and controls should take into account the matched design (see section on Statistical Analysis of Case-Control Studies). Despite the incorrect common belief that matching itself enhances comparability and reduces confounding, not considering the matching in the analysis may produce a biased estimate of the association.
Rather than selecting controls from the risk set, it might be tempting to select controls from among individuals who never experience the endpoint of interest by the end of cohort follow-up (i.e., not matched on follow-up time). For two reasons, this approach is not optimal. First, because of the requirement that individuals sampled as controls remain free of the endpoint until the end of follow-up, they are less likely to include individuals who later become cases than incidence density sampled controls, who are required to be free of the endpoint only until the time of case diagnosis. Thus, if the factor of interest is associated with the endpoint, then the prevalence of the factor will be even lower (if the factor increases the risk or even higher if the factor decreases the risk) in controls selected from the end of follow-up than in controls selected at the time of case diagnosis. This lower prevalence of the factor results in an overestimation of the true rate/hazard ratio (see section on Statistical Analysis of Case-Control Studies).
The second reason that selecting controls at the end of study follow-up is not optimal is that, because controls are required to remain free of the endpoint until the end of follow-up, they also typically are required to remain alive for a longer period of time than cases who may die after their diagnosis. Therefore, if the factor of interest is associated with longevity, then investigators may observe an association between the factor of interest and the endpoint simply because controls are required to remain alive for a longer period of time than cases. This bias is called survival bias.
Case-Cohort Study
Another alternative to the prospective cohort design is the case-cohort design. It is used in the same settings as the nested case-control study, albeit less frequently, because in the past software to analyze data from this design was not readily available. In the case-cohort study ( Fig. 25-8 ), the entire cohort is followed for the endpoint of interest, and material from which to ascertain the factor is collected before the endpoint occurs for all individuals. Then, individuals who develop the endpoint of interest (or a random sample of these individuals) are selected as cases. Next, a random sample (or random samples within categories of characteristics) of the entire cohort at the start of the study is selected and called the subcohort. All of the cases and all of the members of the subcohort are tested for the factor of interest. Looking at Figure 25-8 , at the time that individual 6 recurs, individuals 7 through 15 in the subcohort are still at risk and under follow-up. Thus, individual 6 is compared to individuals 7 through 15, but not to individuals 1 through 5, who are not members of the subcohort. At the time that individual 1 recurs, individuals 7 through 15 in the subcohort are still at risk and under follow-up. In this case, individual 1 is compared with individuals 7 through 15. Thus, similar to controls in nested case-control studies, members of the subcohort in any given risk set are used to reflect the risk of all members in the entire cohort who would have been in that risk set. However, in contrast to the nested case-control setting, a greater number of individuals typically are used to reflect the risk set in the case-cohort setting than in the nested case-control setting.
|
Figure 25-8 Schematic of a case-cohort study within a prospective cohort of 10 hypothetical individuals. All cases are sampled. A subset of the entire cohort at baseline is sampled. |
When should a case-cohort design be used instead of a nested case-control design? If investigators have several endpoints of interest, it might be more cost-effective to use a case-cohort design in which the subcohort can be used to reflect the risk experience of the entire cohort for multiple case groups rather than selecting a separate control group for each case group as would be done in a nested case-control study.
Statistical Analysis of Alternative Prospective Designs
Data from both the nested case-control and case-cohort approaches require special analytic techniques, conditional logistic regression, and Cox proportional hazards regression with a variance correction, respectively. The measures of association estimated from these data approximate the hazard ratio. Adjustment for potentially confounding factors in these models is done as usual.
Design Considerations for Causal Inference: The Case-Control Study
Thus far, all of the epidemiologic study approaches described have been variations on the cohort design. However, the cohort design often is not practical or feasible in oncologic research, for example when the source population is not enumerable, when the endpoint of interest is very rare, or when investigators do not have time to wait until enough endpoints have accrued in a cohort. For these reasons, investigators may choose to use the case-control design ( Fig. 25-9 ). In this design, cases are individuals with the endpoint of interest, and controls are individuals who do not have the endpoint, but are at risk for its occurrence. Once cases and controls have been identified, the factor of interest is assessed at the time that the case or control is ascertained for the study. Thus, in contrast to prospective designs in which assessment of current or prior factor status is made before the case is diagnosed, in case-control designs, assessment of current or prior factor status is made after case diagnosis. Controls are used to obtain an estimate of the prevalence of the factor in the source population at risk for the endpoint.
|
Figure 25-9 Schematic of a case-control study. |
In case-control studies, cases may be identified from several different sources: hospitals and clinics, community outreach screenings, and cancer and other state and national registries. Controls may be identified from those same sources, as well as from other types of registries, such as drivers, voters, population (e.g., in Scandinavia), and the Centers for Medicare and Medicaid Services (United States), or from a general population using mechanisms such as random-digit dialing. Two classic types of case-control studies have been defined: hospital (clinic)-based and population-based. In a hospital-based case-control study, cases and controls are sampled from the same practice, clinic, hospital, or medical care system. Typically, controls are individuals diagnosed with conditions other than the endpoint of interest that are thought not to be related to the factor of interest (i.e., the prevalence of the factor in the controls is expected to be the same as in the source population). This assumption can be difficult to prove, and, if incorrect, can lead to a biased estimate of the association between the factor and the endpoint. In a population-based case-control study, cases and controls are sampled from the same catchment area or the general population. For instance, cases may be sampled from a community hospital, and controls may be randomly sampled from the residents of the hospital's catchment area. Alternatively, cases may be sampled from a state cancer registry and controls may be sampled from residents of the state by random-digit dialing.
The factor of interest and its extent may be assessed using several possible methods. These include review of medical records, interview or administration of a questionnaire, or measurement of biological specimens, again at the time of ascertainment of the cases and controls.
As a simple illustration of the case-control design, consider a new hypothetical research question addressing second cancer risk:
To appropriately counsel women diagnosed with premenopausal breast cancer, oncologists would like to know whether these women have a higher risk of developing ovarian cancer than women not diagnosed with premenopausal breast cancer. If oncologists do not have access to a large cohort of women with information on breast and ovarian cancer diagnoses, then they may consider conducting a case-control study to address their research question in a faster and less expensive way than might be done if they developed a new prospective cohort study. In this case, the investigators would identify cases of ovarian cancer and controls without ovarian cancer and then ascertain the women's prior history of premenopausal breast cancer. Incident cases of ovarian cancer could be identified from the state cancer registry during a specified number of years, and controls free of a diagnosis of ovarian cancer could be identified from random-digit dialing of state residents who must have been residing in the state during the same time interval as the cases. Both cases and controls could then be sent a questionnaire asking them about their prior history of premenopausal breast cancer, and medical record information could be requested to confirm any reported diagnoses.
At first glance, the case-control design might appear simple and thus, straightforward to conduct. Oncologists have access to cases and “just need to find some controls.” In fact, studies of this design require substantial forethought in sampling cases and controls and in assessing the factor of interest to avoid inaccurate results. Thinking back to the cohort design, two considerations for causal inference were described—temporality and comparability.
Temporality
In the case-control design, temporality cannot be ensured because the factor is determined at or after the occurrence of the endpoint of interest, rather than before the occurrence of the endpoint. For example, if investigators collect information about cases’ and controls’ recent history of the factor, they may not capture the relevant factor status because cases may have changed their factor status as a result of having experienced the endpoint (e.g., cases may change their diet near the time of recurrence due to prodromal symptoms). Likewise, high-grade tumors (cases) may have higher expression of a biomarker (factor of interest) in tissue removed at surgery than low-grade tumors (controls) because the tumor influenced the production of the factor, rather than the factor influencing the development of high-grade disease.
Comparability
In contrast to cohort studies, where comparability refers to similarity of those with and without the factor of interest, in case-control studies it refers to similarity of those with and without the endpoint of interest. Comparability encompasses similarity in (1) recruitment from the source population; (2) assessment of the factor of interest; and (3) assessment and handling of confounders. Pertinent to recruitment from the same source population, a classic problem in case-control studies is selection bias, which occurs when inclusion of individuals with and without the endpoint is differentially dependent on whether or not they have the factor of interest. Recall that selection bias in a cohort study occurs when inclusion of individuals with and without the factor is differentially dependent on their likelihood of developing the endpoint of interest. In a case-control study, selection bias can result from differences in the mechanisms and timing of selection of cases and controls. Selection bias can be induced unwittingly by the investigators or by mechanisms beyond the control of the investigators.
Ideally, controls should be sampled at the same time (or during the same interval of time) as cases ( Fig. 25-10 ) to avoid differences in the factor experience of the source population over time. Time refers to the most relevant time axis, for instance, calendar time, age or time since cancer treatment. Looking at Figure 25-10 , the source population can be visualized as an at-risk cohort, although the cohort is not enumerated and the cohort members are not observable during their entire time at risk. At the time that individual 5 develops the endpoint, individuals 2 and 8 are eligible to serve as controls because they have not yet developed the endpoint. Individual 8 might be selected as a control. Note how similar this diagram is to earlier diagrams of the Kaplan-Meier method and the nested case-control study design. To illustrate this sampling scheme, consider the second cancer example:
|
Figure 25-10 Schematic of a case-control study in which controls are sampled at the same time as the cases. |
Cases are women recently diagnosed with ovarian cancer identified from a state cancer registry. At the time (i.e., calendar time) that each case is diagnosed, investigators select a woman who is the same age as the case (±2 years) and who has not been diagnosed with ovarian cancer to serve as the control from the female population of the state obtained through random-digit dialing. In this way, cases and controls are matched on both age and calendar time and, thus, should have a similar opportunity to develop premenopausal breast cancer and ovarian cancer.
Sometimes in oncologic case-control studies, controls are selected after ascertaining cases for practical reasons ( Fig. 25-11 ). For instance, in the second cancer example, it may take investigators 10 years to identify a sufficient number of cases because ovarian cancer is a relatively rare cancer. Investigators may decide that it is too time-consuming and expensive to ascertain controls at the same time as cases. Instead, they may decide to sample controls after all of the cases have been ascertained. However, by sampling controls later in calendar time than cases, controls may not necessarily be sampled from the same source population as cases because time has elapsed. This sampling scheme may lead to at least two possible biases. The first is selection bias due to secular trends in factors or confounders not recognized, and therefore not measured, by investigators. For instance, in the second cancer example, if the investigators were not aware that a mammography campaign had been introduced into the state in the last 3 years of the study, they might observe a higher prevalence of prior premenopausal breast cancer among controls than among cases simply because controls would have been more likely than cases to be screened during the mammography campaign and to have occult breast cancer diagnosed before enrollment in the study.
|
Figure 25-11 Schematic of a case-control study in which controls are sampled after the cases. |
A second example of selection bias related to sampling controls at a later date than cases is survival bias. This bias is the same as the one described earlier (see discussion of Nested Case-Control Study under Design Considerations for Causal Inference: Alternative Prospective Designs) and occurs when the factor of interest is associated with longevity, but not necessarily with the endpoint of interest. In this case, investigators may observe an association between the factor and endpoint merely because controls are required to remain alive for a longer period of time (i.e., to an older age) than cases.
Even if cases and controls are sampled at the same time, selection bias still may be introduced if cases and controls are recruited from different source populations. Consider the second cancer example redesigned as a hospital-based case-control study. Instead of selecting ovarian cancer cases from a state cancer registry and controls from the general population, investigators might decide to select cases and controls from an academic medical center with a large high-risk ovarian and breast cancer clinic. Cases might be defined as women recently diagnosed with ovarian cancer at the high-risk ovarian and breast cancer clinic because it is a convenient source of ovarian cancer cases. Controls might be defined as similarly aged women attending the medical center for minor foot surgery. Cases and controls appear to be sampled from the same source population, namely, the same academic medical center. However, they actually are sampled from two different source populations: cases are sampled from a source population enriched with women with a prior history of premenopausal breast cancer because they are visiting the high-risk clinic, whereas controls are not enriched with women with a prior history of premenopausal breast cancer because they are not visiting the high-risk clinic. Therefore, even if premenopausal breast cancer truly is associated with subsequent development of ovarian cancer, the association will be incorrectly too large because cases and controls were sampled from two different source populations. A better choice of controls for these cases would be other similarly aged women attending the same high-risk ovarian and breast cancer clinic who have not been diagnosed with ovarian cancer.
Another example of selection bias is detection bias or medical surveillance bias. This bias occurs when those subjects with the factor of interest are more or less likely to be screened or investigated or otherwise more or less likely to have the endpoint detected than those without the factor of interest. In the second cancer example, physicians may screen women with premenopausal breast cancer more often for other types of cancer, such as ovarian cancer, than women without a diagnosis of breast cancer. If investigators do not take differences in ovarian cancer screening into account, then they may observe a higher prevalence of prior premenopausal breast cancer among cases than among controls simply because women with a diagnosis of premenopausal breast cancer are more heavily screened for ovarian cancer than women without a diagnosis of breast cancer.
A further example of selection bias is another form of survival bias. This bias occurs when those with the factor of interest have a different likelihood of being ascertained for the study than those without the factor, usually when the factor affects the likelihood of surviving long enough to be enrolled in the study (e.g., to complete a questionnaire or to collect biological samples). In the second cancer example, some women may be too sick or may have died from their ovarian cancer before investigators have the chance to mail them a questionnaire. In this case, only women who live a sufficiently long time after their diagnosis will be able to complete the questionnaire. Therefore, investigators may falsely attribute factors associated with shorter or longer survival following the endpoint to the development of the endpoint itself. Optimally, cases should be ascertained soon after they experience the endpoint of interest; enrolling incident rather than prevalent cases avoids this bias.
In addition to bias introduced in the process of case and control recruitment, bias also can be introduced at the time of factor assessment. This bias is called observation bias, and it occurs when the accuracy of assessing the factor of interest differs by case and control status. Sources of observation bias in case-control studies are plentiful. Two examples are given here. Cases may report their factor history more accurately than controls because they are seeking to explain their disease or are more willing to admit their factor history; this form of observation bias is called recall bias. One way to reduce the potential for recall bias is to blind cases and controls to the study hypotheses or to ascertain the factor status very soon after diagnosis to reduce the amount of time that the subject can contemplate his or her diagnosis. Recall bias is not the inability to accurately remember one's past factor history; this is merely poor recall. The key to recall bias is that the accuracy (or inaccuracy) of the recollection differs between those with and without the endpoint. Another example of observation bias is when an interviewer prompts cases more than controls, resulting in a more complete factor history for cases. This form of observation bias is called interviewer bias, and can be minimized by blinding the interviewer to case-control status.
As in cohort studies, confounders also may contribute to lack of comparability, although, in this case, it is comparability between cases and controls. In case-control studies, confounders are handled in the same way as in cohort studies. To address confounders in a case-control study, investigators should collect information on potential confounders from cases and controls at the time of enrollment and then take these variables into consideration in the analyses (see the section on Statistical Analysis of Case-Control Studies). However, sometimes the distribution of these potentially confounding variables is very different between cases and controls, making it difficult to take these variables into account using statistical methods. One way to make the distributions of these potentially confounding variables more similar between cases and controls is to match controls to cases on these confounders and then analyze the data, taking into consideration the matching. Matching can be done at the individual level (individual matching), where each control is matched to one particular case on potential confounders (including time axes such as age), or at the group level (frequency matching), where controls are selected to have the same distribution of potential confounders as the entire group of cases. Matching enhances the ability to take these variables into account in the analysis, but, as noted previously, matching does not eliminate confounding. If cases and controls were matched on a true confounder and the data were not analyzed taking into consideration the matching, the results would be attenuated. The reason for this attenuation is that matching cases and controls on a third factor that is also correlated with the factor of interest (i.e., the confounder) makes cases and controls more similar on the factor of interest. Therefore, an analytic approach that focuses on the remaining differences between cases and controls is necessary.
Other Design Considerations In Case-Control Studies
Stated colloquially, the general principle guiding the design of case-control studies is: whatever investigators do to cases they must also do to controls. Stated more formally, selection of cases and observation of their factor and confounder status should be the same as selection of controls and observation of their factor and confounder status to reduce the possibility for selection bias, observation bias, and confounding.
There are very occasional exceptions to this statement. One is sampling of cases and controls from different source populations in studies of genetic variation and cancer. For example, in a study of polymorphisms (on chromosomes other than X and Y) in relation to testicular cancer, women could potentially be sampled as controls if they were sampled from individuals with the same racial and ethnic heritage as testicular cancer cases (to minimize confounding by race/ethnicity, i.e., population stratification), and if they were matched to cases on age (to minimize survival bias). Even though women are not at risk for testicular cancer, and thus are not in the same source population as cases, they could potentially still provide a valid estimate of allele frequencies in the source population that gave rise to the cases. However, investigators should still think very carefully about other potential differences between testicular cancer cases and female controls that might distort the relationship between polymorphisms and testicular cancer.
In contrast to the preceding example, blood bank donors, a convenient source of controls, may not provide a valid estimate of allele frequencies in the population that gave rise to the cases. Blood donors must meet very restrictive eligibility criteria, which may be related to their factor experience, such as susceptibility to infection, risk-taking behaviors, altruism, and so on. Therefore, in general, extreme caution must be taken in using controls that, from the outset, are clearly not from the source population that gave rise to the cases.
Statistical Analysis of Case-Control Studies
Due to differences in the design of case-control and prospective studies, data from case-control studies must be analyzed differently than data from prospective studies. In a case-control study, investigators decide how many cases and controls to sample. Typically, investigators do not select all possible cases and controls from the source population. More often, they select a sample of possible cases and controls, often with a greater sampling fraction for cases than controls. Therefore, even if lack of temporality were not of concern, cumulative incidences (or rates) could not be calculated from case-control data, because investigators set the number of cases relative to controls in the study. For example, in a source population that consists of 1000 people followed for 1 year, of whom 100 develop the endpoint and 900 do not (assuming no losses to follow-up), investigators might sample all of the cases (n = 100, 100% sampling fraction) and an equal number of controls (n = 100, 11% sampling fraction) for their case-control study. If investigators were to calculate a cumulative incidence based on these case-control data, the incorrectly calculated measure would be 100/(100 + 100) = 0.5 over 1 year, whereas the true cumulative incidence of the endpoint is actually 100/1000 = 0.1 over 1 year.
Instead of the cumulative incidence (or rate), the odds of the factor should be calculated in case-control studies in which cases and controls are not matched. The odds is the ratio of the probability of having the factor of interest divided by the probability of not having the factor:
where PF means the probability of having the factor and 1 - PF means the probability of not having the factor.
The odds can range from 0 when the probability of the factor is 0 to infinity when the probability of the factor is 1. One odds is calculated for cases and another one for controls. Then, to determine the association between the factor of interest and the endpoint, the ratio of the two odds—or the odds ratio (OR)—is calculated as follows:
To illustrate using the second cancer example, the probability of a prior history of premenopausal breast cancer is 15% or 0.15 in ovarian cancer cases and 0.08 in controls without ovarian cancer. The odds of prior premenopausal breast cancer in cases is thus 0.15/0.85 = 0.1765, and the odds of prior premenopausal breast cancer in controls is 0.08/0.92 = 0.08700. The OR then equals 0.1765/0.08700 = 2.03, suggesting a positive association between premenopausal breast cancer and ovarian cancer. A simpler way of calculating the OR is to multiply the number of exposed cases by the number of unexposed controls and then to divide by the product of the number of unexposed cases and exposed controls ( Fig. 25-12 ). The OR is unitless and can range from 0 to infinity.
|
Figure 25-12 Calculation of the odds ratio as the measure of association in the hypothetical Second Cancer case-control study. |
If cases and controls are individually matched, then the OR is calculated differently. Only pairs in which the case and its control differ on their factor status contribute information. These discordant pairs are used to calculate the matched OR, defined as the number of pairs in which the case has the factor and the control does not, divided by the number of pairs in which the case does not have the factor and the control does have the factor ( Fig. 25-13 ).
|
Figure 25-13 Calculation of the matched odds ratio as the measure of association in a matched case-control study. |
Recall that controls should be sampled during the same calendar time as cases (see Fig. 25-10 ) but sometimes are sampled after the cases have been ascertained (see Fig. 25-11 ). If the former is done, then the OR is a direct estimate of the true rate/hazard ratio for the same reason as in the nested case-control study, assuming no selection bias, observation bias, or confounding. Note, though, that from an inferential perspective, the case-control study nested within a cohort is preferable to the case-control study in which the source population is not enumerated at baseline, because the factor is assessed before the endpoint occurs, and the likelihood of selection bias is reduced.
When controls are sampled after cases in calendar time, the OR estimates the cumulative incidence ratio (assuming no selection bias, observation bias, or confounding), but only when the endpoint is uncommon in the source population (e.g., ovarian cancer in the second cancer example). Here is the logic:
The OR just described (i.e., the factor OR) is mathematically equivalent to the endpoint OR, which is defined as the odds of the endpoint among those with the factor divided by the odds of the endpoint in those without the factor:
where PD|F+ means the probability of the endpoint among those with the factor and PD|F- means the probability of the endpoint among those without the factor.
The endpoint OR can be written as a function of the cumulative incidence ratio:
or
When the endpoint is uncommon—for example, less than 10%—among those with and without the factor in the source population, then
Thus,
Except when the OR equals 1, the value of the OR is always further away from 1 than the value of the cumulative incidence ratio. For example, when the cumulative incidence ratio is greater than 1
Therefore, Endpoint OR is greater than the cumulative incidence ratio.
In this case, the OR suggests a stronger strength of association than the cumulative incidence ratio. If the cumulative incidence ratio had been less than 1, the OR would have been less than the cumulative incidence ratio, suggesting a stronger protective association. The more common the endpoint is in the source population (e.g., death from diffuse large β-cell lymphoma in the therapeutic effectivenessexample), the greater is the overestimation of the cumulative incidence ratio by the OR.
If investigators wish to take potentially confounding factors into account, then the OR can be estimated by logistic regression for studies in which cases and controls are not matched, and by conditional logistic regression for studies in which cases and controls are matched. Detailed descriptions of these regression models can be found in biostatistics textbooks.
DRAWING INFERENCES FROM STUDIES USING EPIDEMIOLOGIC METHODS
Once the results are in hand, investigators must consider the ability to draw causal inferences—in other words, they must consider all possible explanations for their finding beyond cause. These alternative explanations may include selection bias, observation bias, and confounding, as well as chance variability.
Even if temporality and comparability were considered in the design phase of the study, error still may arise during the conduct of observational studies. For example, even if a rigorous definition of the source population and a rigorous approach to sampling from the source population were used, selection bias still is possible because participation is voluntary. The investigators always must be cognizant of these issues and should try to investigate how likely they were to have occurred and the extent to which they could have affected the results.
Chance variability pertains to the notion that only one sample of all possible members of the underlying population is studied. Assuming no selection bias, this sample should differ from the underlying population only in a random way. Thus, estimates of the measure of endpoint occurrence or association from this sample should differ from the true value in the full underlying population only in a random or nonsystematic way. Chance variability is reduced by studying a larger sample. Therefore, in the design phase of the study, investigators should consider the most appropriate sample size to optimize the statistical ability to detect an association if one truly exists. Sample size determination is based, in part, on the expected cumulative incidence or rate of the endpoint in those without the factor (cohort study), the expected prevalence of the factor among controls (case-control study), and/or the expected size of the association. In the analysis phase of the study, the extent of variability in the estimate of the measure of endpoint occurrence or association should be quantified, usually with a confidence interval or P value. Investigators should then consider this variability when drawing inferences from the study. For instance, if the RR is > > 1, but the P is greater than 0.05 (the typical cut-point for statistical significance), is there an association or not? It depends: if the hypothesis were motivated by existing scientific evidence, then the conclusion from the study might be that the findings suggest an association, rather than that no association exists. Interpretation of association magnitudes and significance is an art; it is not dictated by simple rules.
After considering alternative explanations for an epidemiologic finding, investigators still cannot conclude that the observed association reflects causation. In epidemiology, any given research question must be evaluated many times using different designs and study populations, and usually by independent investigators.
Each epidemiologic study design has different potentials for each type of bias. If findings from studies that used different designs are similar, then this may provide evidence in favor of a causal association. If findings differ, then more weight should be given to studies that used designs that maintained temporality and were the least susceptible to bias.
In general, prospective cohort studies and their derivative designs provide stronger evidence for a causal association than retrospective cohort studies and case-control studies, because they maintain the correct temporal sequence between the factor and the endpoint and because they are less susceptible to selection and observation biases. Although both prospective cohort studies and randomized trials are likely to have the correct temporal sequence, evidence from randomized trials usually trumps evidence from cohort studies because randomization reduces the likelihood of selection bias and confounding.
Different study populations (defined, for example, by country, race/ethnicity, or socioeconomic status) may have different characteristics that influence the likelihood or extent of confounding. If findings are the same across study populations, then this may provide evidence in favor of a causal association. If findings differ, reasons may include differences in unmeasured confounders or differences between study populations in the presence and prevalence of factors that might modify the association. At this point, investigators should think about possible reasons for differences across study populations and investigate these reasons in subsequent studies.
Chance variability is possible in any study. When the results from independent epidemiologic studies are not too heterogeneous, meta-analyses can be performed to generate a more stable (i.e., less variable) summary estimate of the association.
When considering whether an observed association between a factor and endpoint is causal, some investigators find it useful to think about the nine “aspects” of an association that Sir Austin Bradford Hill described in 1965[3]:
|
1. |
Strength: How big is the measure of association? |
|
|
2. |
Consistency: Are findings from studies conducted using different designs in different populations similar? |
|
|
3. |
Specificity: Is the factor only associated with the endpoint of interest, and is this the only factor that has been found to be associated with the endpoint of interest? |
|
|
4. |
Temporality: Was the factor experienced before the endpoint occurred? |
|
|
5. |
Biological gradient: Does the magnitude of the measure of association increase (or decrease) with increasing extent of the factor? |
|
|
6. |
Plausibility and 7. Coherence: Is the association supported or refuted by the contemporary biologic literature? |
|
|
8. |
Experiment: Does the magnitude of the measure of association decrease (or increase) after changing the factor status? |
|
|
9. |
Analogy: Have associations between similar factors and similar endpoints been observed? |
Temporality has been highlighted in this chapter because it is the only aspect of an association necessary for causality. The other aspects neither rule in nor rule out the potential for an association to be causal.
In light of all these considerations, the original therapeutic effectiveness example is now reexamined. The investigators observed that patients with 2 copies of the haplotype had 0.5 times the risk of death, and patients with 1 copy had 0.7 times the risk of death compared with patients with no copies of the haplotype. These results were statistically significant (P < 0.05).
What should investigators infer? They conducted a prospective cohort study in which haplotyping was performed using samples collected at the start of the study; hence, the correct temporal sequence was obtained. All individuals from one arm of the trial were included in the study; therefore, selection bias is unlikely to have been introduced into the study. All trial participants were followed actively and in the same manner for survival. Therefore, it is unlikely that the ability to observe the endpoint differed by haplotype (i.e., no observation bias). The investigators observed that the prevalence of the haplotype did not vary by any of the participant characteristics measured at the start of the study, including race or ethnicity; therefore, confounding also is unlikely. The strength of the association between the haplotype and survival is relatively strong, suggesting that confounding alone is unlikely to explain the association. Although the findings were statistically significant, chance cannot be ruled out as an explanation for the findings. This is the first study of this hypothesis, so there are no other studies to which to compare the findings. The RR of death decreased with increasing number of copies of the haplotype, demonstrating a biological gradient. The hypothesis was well motivated by scientific evidence that the P450 enzyme enhances the action of the chemotherapeutic agent, and that the haplotype of interest encodes a more active form of the enzyme. Findings from this study are analogous to those from a study that investigated genetic variation in a different metabolic enzyme with cancer survival after treatment with a different chemotherapeutic agent.
After contemplating all of these aspects of the study and its findings, the investigators might conclude that the findings are promising, but because the study is observational and the first to investigate this hypothesis, additional studies should be conducted before the findings can be translated into clinical practice. As already mentioned, a randomized trial of this research question is not possible because haplotype is an inherent characteristic of an individual. However, if, in the future, findings from this study are upheld in other studies, the next step might be to conduct a trial in which patients without the beneficial haplotype are randomized to either a drug that induces the P450 enzyme or placebo, concurrent with administration of the new chemotherapeutic agent.
FINAL THOUGHTS ON THE USE OF EPIDEMIOLOGIC METHODS IN ONCOLOGIC RESEARCH
This chapter has described epidemiologic methods for formally testing hypotheses pertinent to oncology and discussed their rationale. Epidemiology is a flexible method that can be applied to many different types of translational and clinical studies that do not necessarily have an etiologic or population-level focus. For instance, in the hypothetical prognosis example, the focus of the study appears to be on the measurement of a newly discovered tissue marker using an innovative technique. However, use of this innovative technique does not ensure correct inferences about whether or not the tissue marker predicts prognosis in the study population, nor its applicability to other populations, even if the technique is highly accurate. Obtaining a valid answer also hinges on choosing the appropriate study population, design, and analysis, all of which require epidemiologic thinking. For this reason, a multidisciplinary research team that includes collaborators from clinical science, basic science, biostatistics, and epidemiology is optimal for addressing the complex translational and clinical questions posed in modern oncology.
REFERENCES
Copyright © 2008 Elsevier Inc. All rights reserved. - www.mdconsult.com
Abeloff: Abeloff's Clinical Oncology, 4th ed.
Copyright © 2008 Churchill Livingstone, An Imprint of Elsevier
Chapter 26 – Cancer Prevention, Screening, and Early Detection
Jason A. Zell,Frank L. Meyskens
SUMMARY OF KEY POINTS
|
INTRODUCTION
The guiding principles of this chapter and oncology should be that the best treatment of malignant disease is its prevention, and that the disease to be prevented is carcinogenesis, not cancer ( Fig. 26-1 ).[1]
|
Figure 26-1 Integration of the biology of carcinogenesis and prevention. *Hereditary alteration(s) or baseline polygenic representation provides the constitutive “set point” on which postzygote changes occur. † Precancer, premalignant lesion in general clinical parlance. |
By the time a cancer is diagnosed, even with the advanced techniques now available, more than 90% of the biologic life of the tumor is over, and the best chance to control the malignant process has been missed. The extensive advances in our understanding of carcinogenesis at the molecular level in the past decade, the rediscovery of intraepithelial neoplasia as an early, recognizable precursor of many solid tumors that can be managed simply, and the well-defined successes of screening and early detection in reducing the morbidity and mortality from several major cancers, need to be brought to bear on the problem of malignancy in a concerted and widespread fashion, with clinical oncologists working closely with primary care physicians and subspecialists. Because fewer than 50% of cancers are cured, once established, and because gains in treatment effectiveness have been increasingly incremental and expensive, early detection and prevention of cancer should be pursued aggressively as a means to reduce morbidity and mortality.
Many major diseases of humankind have been controlled by the systematic application of prevention strategies, including morbidity and mortality from nutritional and infectious diseases and vehicular trauma.[1] Among chronic diseases, the incidence of cardiovascular disease has decreased markedly as smoking has declined, cholesterol and blood pressure lowered, and exercise encouraged. It is likely that these simple approaches have led to a greater overall benefit to health for the population than the effect of all intensive care units, but such direct comparisons are difficult to make. In general, appreciation of the role of prevention strategies in the overall management of cancer has been neglected by clinical oncologists, although health care planners and society as a whole are intensely interested in this topic.[2] Cancer prevention strategies can be considered at three different major levels: primary, secondary, and tertiary. This chapter will deal primarily with primary and secondary prevention and with tertiary prevention as represented by chemoprevention of second malignancies.
Normal, asymptomatic individuals are the population at which primary prevention is addressed. Major strategies for risk reduction include changes in diet, increased physical activity, tobacco awareness, decreased exposure to the sun, and reduced intake of alcohol. With the increasing identification of constitutive genetic alterations that predispose individuals to cancer, this group has been targeted for primary interventions such as prophylactic surgery.[3] Annual screening mammography in women older than 50 years of age and smoking cessation or chemoprevention in a group of asymptomatic smoking individuals are also examples of targeted primary prevention.
Secondary prevention is directed toward individuals with evidence of preneoplastic, clinically identifiable progression, but without frank malignancy. The phenomenon of intraepithelial neoplasia, also called preneoplasia or precancer, has become of widespread interest, and management of these lesions has the potential to abrogate the disease process early. Many organ sites have preneoplastic counterparts that should be amenable to early intervention ( Table 26-1 ). Representative examples of this type of secondary prevention include suppression or reversal of oral leukoplakia, cervical intraepithelial neoplasia (CIN) or Barrett's esophagus and inhibition of polyp formation or progression ( Fig. 26-2 ).
Table 26-1 -- Common Clinical Precursors (Intraepithelial Neoplasia) of Cancer
|
Organ Site |
Precursor |
Method of Detection[*] |
|
Oropharynx |
Leukoplakia |
Visual[†] |
|
Skin |
Actinic keratoses/moles |
Visual[†] |
|
Esophagus |
Barrett's esophagus |
Endoscopy |
|
Colon |
Adenoma (polyp) |
Sigmoidoscopy, colonoscopy |
|
Breast |
LCIS, DCIS[‡] |
Mammography, ultrasound, MRI |
|
Cervix |
Intraepithelial neoplasia |
Colposcopy |
|
* |
Cytology and/or biopsy is required in almost all cases before definitive therapy can be initiated. |
|
† |
Elegant in situ optical spectroscopic methods are being developed to detect early preneoplastic changes, including enhancing the signals with fluorescent molecules. |
|
‡ |
Lobular and ductal carcinoma in situ. |
|
Figure 26-2 Examples of premalignant lesions. A, Leukoplakia. Whitish lesion on side of tongue. B, Erythroplakia. Reddish (dark area) lesion in otherwise normal-looking buccal cavity. C, Barrett's esophagus (pale area) with high-grade lesion (dark portion). D,Adenomatous polyp in proximal sigmoid colon. (Courtesy of Bill Armstrong and Ken Chang.) |
Tertiary prevention involves decreasing the morbidity of established disease. Chemoprevention of second malignancies is a good example of tertiary prevention. The distinction between primary, secondary, and tertiary prevention can sometimes become blurred. Further, tertiary prevention and adjuvant therapies can share many of the same goals. From the viewpoint of the clinical oncologist, probably the best way to look at prevention is as one more therapeutic modality for the management of cancer, directed at its control in the earliest stages. The observation that the addition of a retinoid after bone marrow transplantation markedly enhances the survival of children with refractory neuroblastoma represents an informative synthesis of a quaternary treatment approach and a tertiary prevention modality.[4]
Avoidable Causes
An extensive analysis of the topic of avoidable causes of cancer was performed over 2 decades ago by Doll and Peto[5]; these investigators concluded that 50% to 70% of all human cancers were preventable. No new data have emerged that would alter that overall estimate, although some of the specifics have changed.[6] The major avoidable risk factors can be broadly separated into four areas: tobacco, infectious, chemical (including hormonal), and diet.
Tobacco smoke is far and away the most important carcinogen to which humans are exposed on a routine basis. The morbidity and mortality from tobacco smoke is huge and represents the major preventable cause of all diseases, not just cancer, in modern and many undeveloped societies. It is estimated that more than 500 million smokers now living will die of tobacco-related illnesses. [7] [8] What is generally not appreciated is the wide carcinogenic range of molecular damage and the numerous organ sites affected by cigarette smoke. [9] [10] In addition to the lungs, cigarette smoking contributes significant attributable risk to the development of cancers of the oropharynx (75%), bladder (50%), esophagus (50%), pancreas (25%), cervix (20%), kidney (15%), and bone marrow (10%). Cigarette smoke facilitates chromosomal instability and enhances transformation at all levels of cancer formation (initiation, promotion, progression), including adversely affecting the natural history of successfully resected early-stage lung cancer.[11] Although the incidence of cigarette smoking has fallen among males, in 1994 lung cancer surpassed breast cancer as the most common cause of death from cancer in females. Lung cancer incidence among females has increased from 1975 through 2003, although the rate of increase has declined since 1991.[12] Well-tested modules have been developed to assist health care workers, including physicians, in applying smoking prevention and cessation strategies.[7] Various forms of nicotine (gum, patch, inhalers, nasal sprays) and behavioral modulators (bupropion) have been effective in increasing the quit rate significantly at very low cost.[8]
The evidence for infectious involvement in human cancer has increased dramatically in the last 15 years ( Table 26-2 ). Hepatitis and human papillomavirus (HPV) clearly play major roles in the development and evolution of hepatocellular and cervical carcinoma, respectively. Vaccines using various viral components as the target have recently been tested; results from these trials strongly suggest that liver cancer and cervical cancer are preventable. [13] [14] In addition to the classical and long-recognized associations of the parasites C. sinensis to cholangiocarcinoma and S. haematobium to squamous cell carcinoma of the bladder, the bacterium H. pylori has now been accepted as an etiologic agent associated with gastric dysplasia, stomach cancer, and a rare type of lymphoma.[15] Early on, these diseases can be treated with antibiotics and the process reversed. Finally, it seems that viruses have a role in the evolution of some lymphomas (human T-lymphotropic virus-1, Epstein-Barr virus [EBV]). These findings all offer new approaches for primary and secondary prevention using standard and new microbiologic and immunologic approaches.
Table 26-2 -- Cancers with an Infectious Etiology
|
Cancer |
Agent |
Major Mode of Transmission |
Intervention |
|
Hepatocellular carcinoma |
Hepatitis virus |
Maternal, oral |
Vaccine |
|
Gastric |
Helicobacter pylori |
Oral |
Antibiotics |
|
Cervix |
Papillomavirus |
Sexual |
Vaccine |
|
Several major cancers are of infectious etiology and can be eradicated by preventive intervention. |
A number of chemicals are known to play a role in cancer causation, perhaps the most widespread being aniline dyes (bladder cancer), asbestos (lung, mesothelioma), and hormones (breast, prostate). The role of endogenous and exogenous hormones in cancer causation is complex and is considered in detail in the sections on breast, prostate, and gynecologic organ sites.[16] Some of the most intensely debated issues in medicine relate to this area. For example, is the overall health benefit of hormone replacement therapy (HRT) worthwhile? Results from the Women's Health Initiative (WHI) trial suggest not,[17] and long-term use of HRT is not recommended.
The question of the specific role of dietary components in cancer prevention remains largely unanswered. Many comprehensive reviews on the topic are available, and the overall recommendation to eat an abundant amount of fruits and vegetables has not changed in 20 years.[18] Ambitious campaigns such as the well-known “5-a-day for better health” campaign to encourage large-scale dietary changes continue.[19] The specific components responsible for the protective effects against cardiovascular disease and cancer remain unclear. The roles of macronutrients, fat, and fiber in prevention have been topics of much discussion. The general recommendation to reduce total calories and fat consumption and to increase fiber is a good one with regard to cardiovascular disease prophylaxis, but whether such a strategy affects cancer outcome remains unproven. Increasing epidemiologic data suggests that physical activity and basal metabolic index play critical roles, and several trials are underway to address these issues.[20]
There is growing evidence that changes in the insulin-like growth factor pathways play an important role in many aspects of lifestyle changes represented by a high basal metabolic index and its control.[20]With increasingly positive protective effects of physical exercise on cardiovascular disease being shown, there has been a renewed interest in the influence of physical exercise on malignant transformation and progression.[21] There has been a great deal of interest in micronutrients as preventive agents, but the results emanating from supplementation in well-done randomized clinical trials thus far have been disappointing.[22] Notable exceptions have included the report that supplementation with a modest dose of vitamin A (25,000 IU per day) can decrease the appearance of cutaneous squamous cell cancer of the skin in individuals with prior actinic keratoses and several trials showing that supplementation with a modest dose of calcium can decrease the subsequent prevalence of cancer polyps by 20%. [23] [24] [25]
Probably the most exciting diet-related development has been the identification of a wide range of potentially new and active chemoprevention compounds in food, such as protease inhibitors (soybeans), monoterpenes (citrus fruit oils), polyphenols (nuts), dithiolethiones (cruciferous vegetables), alliums (onion/garlic family), and many others.[26] The opportunity to genetically engineer foods to reduce the risk of heart disease and cancer is a topic of much scientific interest and commercial activity.[27] Nature created these molecules to deal with a hostile toxic environment, and figuring out how to use them for the prevention of cancer should be both scientifically interesting and clinically rewarding.
Screening and Early Detection
Strictly speaking, screening is limited to normal individuals. The science of screening identifies many pitfalls in the design, analysis, and interpretation of such trials, including length and lead-time biases and many others. [28] [29] Beyond the technical issues involved in study design, three other requirements addressing implementation, analysis, and interpretation, must be met to demonstrate that a screening test is useful:
|
1. |
A test must be available that will detect cancer earlier than routine methods (e.g., clinical or self-examination). |
|
|
2. |
There must be evidence that treatment at an earlier stage of disease will result in an improved outcome (decreased cause-specific morbidity or mortality). |
|
|
3. |
There must be evidence of a total health benefit. For example, the benefits of early detection via screening that meets the aforementioned criteria must also outweigh the adverse risks of subsequent diagnostic and therapeutic interventions. In current screening trials, disease-specific vs. all-cause mortality, and risk-benefit have become increasingly important issues—particularly among older individuals. [30] [31] |
Fulfilling these requirements is difficult, and the issues specifically related to screening of different organ sites for precancers or cancers are discussed in those sections. Some generic comments are worthwhile. Enough evidence exists for a specific test for some organ sites that has been proven effective to recommend the routine adoption of screening ( Table 26-3 ). Although the availability of cancer screening is generally increasing, usage is relatively low for some organ sites (e.g., colon) and among groups that lack health insurance or a usual source of care.[32] Many screening tests, however, are ineffective (e.g., routine chest roentgenograms in smokers being the most notable).
Table 26-3 -- Effectiveness of Major Screening Approaches for Cancer[*]
|
Organ Site |
Test |
Positive Level of Evidence[†] |
Recommended |
|
Breast |
|||
|
Over age 50 |
Mammography |
Strong |
Yes |
|
Age 40–50 |
Mammography |
Fairly strong |
Yes |
|
Cervix |
Papanicolaou[‡] |
Strong |
Yes |
|
Colorectal |
|||
|
Over age 50 |
Occult fecal blood |
Strong |
Yes |
|
Sigmoidoscopy |
Strong |
Yes |
|
|
Colonoscopy[§] |
Fairly strong |
Yes |
|
|
Lung |
Chest roentgenogram |
None[‖] |
No |
|
Melanoma |
Skin examination |
Moderate |
Yes |
|
Prostate |
Prostate-specific antigen |
Moderate |
Yes[¶] |
|
* |
Listed here are organ sites for which sufficient data exist to make a judgment. Although no specific trial evidence exists, routine physical examination of the skin, oral cavity, testicles, and ovary/uterus is worthwhile, because treatment success is closely related to stage at diagnosis and effective treatment is available in most cases. |
|
† |
The concept of level of evidence is a valuable approach—a quantitative approach that is presented in detail in the PDQ section of the NCI Web site (http://cancernet.nci:nih.gov/clinpolq/screening). A randomized trial with survival as an endpoint is at the top of the hierarchy, whereas anecdotal evidence by experts is at the lowest. |
|
‡ |
Screening should begin with the onset of sexual activity. |
|
§ |
Evidence also exists that colonoscopy with excisional biopsy is an effective therapeutic maneuver, but the cost of the procedure has precluded its general usage. |
|
‖ |
Several randomized trials of screening chest roentgenograms showed no effect on outcome. Spiral CT is currently being tested in a large national trial. |
|
¶ |
With careful follow-up and appropriate testing. |
Also, a positive screening test may lead to aggressive intervention that could allow “cure” of the organ site disease but result in an overall increased morbidity or mortality that is not efficacious for a person's general health (e.g., radical prostatectomy for older individuals with a minimally increased PSA). Finally, screening for currently incurable malignancies (e.g., pancreatic cancer) offers new ethical dilemmas. If we have little to offer therapeutically, do we want to know the risk? Maybe, maybe not. Perhaps earlier prophylactic surgery might be able to affect the outcome in a few patients—for example, in families with early-onset pancreatic cancer.
Over the past few years, there have been rapid developments applying new imaging modalities for the purpose of cancer screenings. The United States Preventive Services Task Force now recommends screening with imaging techniques for breast cancer (mammography for women aged 40 to 70 years) and colorectal cancer (barium enema, fecal occult blood, endoscopy).[33] Cancer screening trials are ongoing for breast cancer (ultrasound, magnetic resonance imaging [MRI]), colorectal cancer (computed tomography [CT] colonography), liver cancer (ultrasound, CT), and pancreas (endoscopic ultrasound, CT, endoscopic retrograde cholangiopancreatography).[34] Major trials have completed accrual, including the Prostate, Lung, Colorectal and Ovarian (PLCO) trial and the National Lung Screening Trial. Mortality results from the PLCO are expected in 2015 and will assess the role of chest radiography (i.e., x-ray) in lung cancer screening, and transvaginal ultrasound for ovarian cancer screening. Mortality data from the National Lung Screening Trial are expected in 2009, to evaluate low-dose CT for lung cancer screening among high-risk adults.
The age of molecular diagnosis in screening is upon us and holds both promise and peril.[35] Novel functional and molecular imaging techniques for cancer diagnosis currently include measurements of tumor angiogenesis via dynamic contrast-enhanced (DCE)-MRI, DCE-CT with ultrasound, positron emission tomography (PET), and combinations of these techniques.[36] Diffuse optical imaging is another functional imaging technique that has emerged as an adjunct for cancer diagnosis. This noninvasive imaging method uses near-infrared light intensity to address tissue-specific changes between normal and tumor tissue (i.e., tissue hemoglobin concentration, tissue oxygen saturation) in vivo.[37] DCE-MRI, DCE-CT with ultrasound, and PET imaging have varying capabilities to profile microvessel density—a hallmark of angiogenesis—which is essential for tumor growth and invasion. However, even in the setting of cancer diagnosis, these methods are not yet standardized.[36] Appropriate methods for incorporating these functional and molecular imaging techniques into screening trials must be considered, particularly in light of their associated high costs. Thus, the potential impact of such molecular and functional imaging techniques on cancer screening is great, though not yet realized.
Although early detection is, formally speaking, the evaluation of a symptomatic individual for cancer and is therefore different from screening, many of the caveats regarding evaluation of this approach are the same. The increasing ability to identify high-risk populations, either by phenotypic criteria or by genetic analysis, also tends to lead to a blurring of the classic division between screening and early detection. With the rapid advances in molecular diagnostics, routine genetic typing of individuals at risk for major tumor types should not be too distant in the future, and quantitation of that risk (a concept we proposed quite some time ago)[38] and its evaluation at the time of detection leads to real-world angst. Identification and referral of families at high risk for cancer susceptibility should be an increasing emphasis of clinical oncologists, but thus far, participation by oncologists has been low. [2] [39] At the very least, the ability to downshift the stage of a disease at the time of detection should eventually lead to improved survival, because new treatment approaches emanate from causative understanding of a particular cancer. Proving this point, though, has been difficult for cancers of many organ sites.
The effectiveness of screening for the major types of cancer is summarized in Table 26-3 and is discussed in detail in the individual sections of this chapter. There are effective screening modalities for breast cancer, colorectal cancers, melanoma, and cervical cancers, whereas no compelling evidence exists for the value of screening for lung cancer. The effectiveness of PSA in screening for prostate cancer has been a subject of intense debate; we feel that the tide of evidence has turned and that the evidence supports the routine use of PSA screening over age 50 with thoughtful management and appropriate follow-up of abnormal values.
Carcinogenesis and Chemoprevention
Carcinogenesis
Advances in our understanding of the biology of carcinogenesis (cancer formation) have sharpened our thoughts about screening and early detection and provided a guide to thinking about risk assessment and chemoprevention.[40] Figure 26-1 serves as a useful general roadmap to reflect on these issues for all tumor types (see Shureiqi and colleagues).[41] The classical model of carcinogenesis divides cancer evolution into three major epochs: initiation, promotion, and progression. This classification has served as a useful heuristic model for which considerable experimental evidence has been developed. In the past 15 years, the genetic paradigm for the development of cancer has been elegantly articulated and experimentally confirmed for some organ sites. A series of steps in response to separate molecular events at the genetic level is a useful platform from which to understand carcinogenesis in human epithelial tumors. Almost all human cancers examined in any detail have shown evidence of several acquired molecular abnormalities. Although the “pathway” is different for each cancer, the tumor suppressor genes p53 and p16 seem to play a central role in many malignancies. [42] [43] The expression of these abnormalities has allowed the development of markers that could serve as indicators of cancer risk or disease progression or possibly as surrogate endpoints for chemoprevention agent testing.[44]
Use of markers of carcinogenesis to assess the status of the disease relative to diagnosis and treatment and for assessment of chemoprevention effect is an important and complex issue.[45] The continued development and validation of markers will be critical to the intelligent management of early-stage cancer.[46] What also has become clear is that environmental phenomena (e.g., hormones, diet, carcinogens) can influence the expression of genetic changes. At one extreme of the paradigm is retinoblastoma, in which loss of a single gene inevitably results in an ocular tumor at a young age; however, most common solid tumors in adults seem to have underlying polygenic contributions, which can be affected by a large range of exogenous factors, even when a deleterious mutation such as BRCA1 orBRCA2 is present.
Chemoprevention
The idea of the chemoprevention of human cancer has been with us for nearly 3 decades, but only in the recent years have positive clinical trials validated the results of preclinical data and their potential for use in human beings. [47] [48] [49] [50] Retinoids are a major group of compounds that have provided convincing “proof of principle” of chemoprevention in humans, but in general they have been too toxic for widespread use and have not been adopted widely.[51] The overall results of some of the key randomized chemoprevention trials are summarized in Table 26-4 and are discussed in more detail in the individual organ site sections. Major problems in developing chemoprevention as a modality for cancer management have been the length and size of the trials required to show changes in a definitive endpoint. [52] [53] Consequently, only the National Cancer Institute and a few large research groups have been able to marshal the resources to develop broad-based chemoprevention efforts. Another significant issue in the design of early trials was that many large studies evolved primarily from epidemiologic observations, with little experimental data available. Because the implementation of large phase III or IV chemoprevention trials is a 10- to 100-million-dollar exercise, political influences on the funding process have also been substantial.
Table 26-4 -- Current Overall Status of Chemoprevention in Preventing Human Cancers[*]
|
Organ Site |
Pretrial Level of Evidence |
Agent |
Status of Chemoprevention |
Comment |
|
Breast |
Strong |
Tamoxifen |
Effective |
One large trial, very positive; two smaller trials showed no effect. Overall long-term health benefits must be determined. |
|
Cervix[†] |
Strong |
Multiple |
Ineffective to marginal |
Numerous phase III trials of several compounds have not substantiated phase II trials except for topical all-trans-retinoic acid. |
|
Colon |
Strong |
Multiple |
Mixed |
Slight decrease (25%) in colon polyp recurrence by calcium or aspirin |
|
Head and neck (secondary) |
Moderately strong |
13-cis retinoic acid |
Effective but toxic |
Follow-up trial at lower dose ineffective |
|
Leukoplakia |
Moderately strong |
β-carotene |
Promising |
Single randomized trial needs confirmation. |
|
Tertiary |
Strong |
13-cis-retinoic acid |
Effective but toxic |
Follow-up studies at lower doses in progress |
|
Lung |
||||
|
Primary |
Strong |
β-carotene |
Ineffective |
More lung cancers and increased higher overall mortality |
|
Secondary (metaplasia) |
Strong |
Retinoids |
Ineffective |
Impressively negative |
|
Prostate |
Moderately strong |
Finasteride |
Accrual complete |
Results indicate positive effects but are preliminary. |
|
SELECT[‡] |
Ongoing |
Results in 2010 |
||
|
Skin |
Strong |
Retinoids |
Effective in some cases |
Seems to depend on stage of cancer development and strength of agent |
|
Diclofenac |
Effective |
Causes regression of actinic keratoses |
|
* |
Details are discussed in individual sections. |
|
† |
Cervix, effects on regression of cervical intraepithelial neoplasia. |
|
‡ |
Selenium and vitamin E. Epidemiologic data from secondary analysis; experimental data moderate. |
An attempt to develop chemoprevention agents logically has been outlined, and systematic preclinical testing and evolution of sequential clinical trials are likely to avoid some of the mistakes of the past.[53] [54] Several key elements are featured in this decision analysis process:
|
• |
Preclinical in vitro and in vivo testing against a battery of molecular targets and cellular and animal models |
|
|
• |
Accurate identification of side effects and assessment of their importance |
|
|
• |
Evidence of modulation of anticipated biochemical or molecular markers in the relevant tissue in short-term human trials |
|
|
• |
A randomized 6- to 12-month study of multiple low doses of the candidate agent in a relevant patient/participant population, with careful identification of side effects and assessment of their importance, and evaluation of their biochemical, molecular, and/or histologic effects |
In this regard we have performed a particularly informative series of studies in assessing topical all-trans-retinoic acid in cervical cancer prevention and difluoromethylornithine (DFMO) in colon cancer prevention, [55] [56] [57] whereas the M.D. Anderson group[58] has performed a series of important trials in aerodigestive cancers and the Arizona group has done the same for skin cancers.[59] Following this logical pathway of chemoprevention agent development assures that the probability of conducting a definitive phase III or IV study will be high and maximizes the chance for a successful outcome. Although the term “chemoprevention” has been widely used to describe chemical or dietary intervention to prevent or reverse cancer, it is a misnomer that is out of sync with the nosology used in other areas of medicine. Therapeutic prevention might be a better term—for example, cholesterol-lowering agents and antihypertensives to prevent cardiovascular diseases, prophylactic antibiotics to prevent infections, and many more. Risk-benefit of a drug in a prevention setting has always been a key issue, particularly of individuals at low risk. The unexpected increase of lung cancers and adverse cardiovascular events in two large trials in which β-carotene was administered called this issue into sharp focus in the mid-1990s. [60] [61] Recently, a different level of risk has been appreciated, designated risk-risk. In several large trials, selective/specific COX-2 inhibitors markedly reduced the incidence of colonic polyps but at the price of increased serious cardiovascular events. [62] [63] A careful analysis indicated that overall morbidity would be increased by using selective COX-2 inhibitors to reduce adenomatous colon polyps in individuals at low risk.[64] These important observations do not necessarily preclude using other nonsteroidal anti-inflammatory drugs (NSAIDs) that have more complex effects on the prostaglandin pathways (e.g., aspirin, sulindac); the results of other trials will help in this decision making, but the unexpected outcomes in the COX-2 inhibitor trials has clouded the field of chemoprevention for reduction of cancer incidence, particularly in individuals at low risk. It is important to recall that the development of cardiovascular risk reduction trials met similar problems in the early days until effective and nontoxic agents were developed, and 20 to 30 years ago the regulatory burdens and the ethical bars were considerably lower.[65]
Although thus far most agents have been developed based on epidemiologic observations or carcinogen models in animals, increasing knowledge about the molecular basis of cancer progression in human tumors should result in the discovery and synthesis of highly specific drugs based on altered biochemical and signaling pathways. [66] [67] The number of specific tumors for which prevention strategies could be reviewed is large. In this chapter we review the major organ sites (aerodigestive, colon, breast, prostate) and those sites (skin, ovary, cervix) in which sufficient evidence exists to suggest that preventive strategies currently have a role in clinical oncology. We also offer a few comments on less studied cancers (stomach, liver) that are extremely common outside the United States and Europe. With the rapidly increasing scientific understanding of the biologic basis for many tumor types and the recognition that screening, early detection, and chemoprevention should play a large role in the management of the carcinogenic process, we can anticipate that the list of therapeutic prevention strategies for intraepithelial neoplasia and possibly earlier manifestations of cancer will grow rapidly over the next few years.
AERODIGESTIVE MALIGNANCIES
Risk Reduction
Aerodigestive malignancies encompass a subset of cancers including those that arise from the oral cavity, pharynx, esophagus, and lung.[58] These organ sites have been grouped together, because they share a mucosal epithelial field that is directly subject to malignant transformation by the common toxin (tobacco), the major underlying etiologic agent for these malignancies. In addition to cigarette smoke, smokeless tobacco has also played an increasing contributory role in oral carcinogenesis, and oral cancer in the young adult has become increasingly common.[68] Alcohol also clearly plays a synergistic role with tobacco carcinogens in the development of oral and esophageal cancers, including second cancers. [69] [70] Polymorphisms in the alcohol dehydrogenase gene involved with tobacco carcinogen metabolism may also play an important role in determining risk for head and neck and tobacco-related malignancies.[71] The identification of HPV in more than 50% of oropharyngeal and nearly 100% of laryngeal tumors has led to acceptance of the role of HPV in head and neck squamous cell carcinomas (HNSCCs) pathogenesis. [72] [73] Compared with HPV-negative HNSCC tumors, which harbor tobacco-induced p53 mutations, HPV-positive HNSCC tumors have wild-type p53, which is inactivated by the viral oncoprotein. [73] [74] Interestingly, HPV-positive HNSCC patients have a different clinical profile compared with HPV-negative patients: they are younger, less likely to ingest alcohol or use tobacco, and show equal gender distribution; the tumors show poorly differentiated and basaloid histology, but such patients have improved survival characteristics. [73] [75] Progress in understanding the biology of tobacco-associated carcinogenesis in the past few years has been rapid, and molecular models of head and neck and lung cancers have been characterized to a substantial degree.[76] What is clear from cytogenetic genomic hybridization and other studies is that, although aerodigestive cancers share many similar changes early on (e.g., loss and gain of 3p and cyclin alterations), discrete subsets exist.[77] Because different genes are involved, this information should have a practical effect on the development of chemoprevention and other interventions.
In 2006 more than 230,000 cancers are expected to develop in aerodigestive sites, and 188,000 related deaths are anticipated in the United States.[78] Because it is estimated that the etiology of more than 80% of aerodigestive cancers is tobacco related, the cost to society of this legal carcinogen is extraordinarily high. The application of prevention strategies should have a favorable impact on decreasing morbidity and mortality from aerodigestive cancers; particularly important for the medical profession should be the adoption of proven, physician-facilitated smoking cessation methods. [7] [8]
Screening and Early Detection
Oropharyngeal cancer occurs in a region of the body that is easily accessible to examination by a health care worker. The morbidity and mortality from oropharyngeal cancer is directly related to the stage at diagnosis, so effective screening and early detection should be worthwhile.[79] However, no definitive trial has demonstrated that an early detection program can downshift stage at diagnosis of oropharyngeal cancer or reduce mortality in a screened population. Nevertheless, an inspection of the oral cavity should be part of every examination in high-risk patients (smokers) and can be made efficacious by careful inspection of the soft palate, tongue, and floor of the mouth, where 90% of all squamous cell cancers occur.[80] The preneoplastic lesions, leukoplakia and particularly erythroplakia, should be identified and biopsy specimens obtained if necessary, because they represent early observable signs of squamous cell carcinomas with different prognoses (see Fig. 26-2A and B ). In appropriately screened populations of high-risk smokers and drinkers over age 40, a detection rate of oral cancers as high as 1 cancer in every 200 individuals examined has been achieved.[81] A successful screening program can be mounted using health caseworkers; for example, in Sri Lanka, where oral cancer is a common malignancy, a sensitivity of 58% was obtained for 660 patients with suspected cancers.[82]
Recent advances in optical biology also suggest that screening using autofluorescent techniques may allow detection of premalignant changes before the clinical appearance of disease.[83] The strong association of HPV with oral cancers in young, nonsmoking individuals[72] and the success of screening techniques in detecting cervical intraepithelial neoplasia (see the discussion that follows) suggest that oral screening for HPV should be adopted for those who are sexually active. Established risk factors for HNSCC related to sexual behavior (i.e., history of genital warts, young age at onset of sexual activity, and a high number of sexual partners) indicate that this group is the same population at risk for genital HPV.
Routine screening of esophageal cancer in the United States has not been attempted to a significant degree, because it is relatively uncommon and therapeutic options are poor. In China and other Asian countries, however, the disease is much more common and found at high frequency in certain geographic locales. In these areas, screening and early detection using esophageal cytology are widely used, although the efficacy of these approaches has yet to be firmly established.[84]
Although lung cancer incidence among men has declined since the 1980s, the incidence for females has increased from 1975 through 2003 in the United States.[12] This trend may be changing, in that recent data indicate that the rate of increase among women has declined since 1991.[12] The issue of screening for lung cancer has been a long and complicated one, without a strong supporting evidence base. In the Mayo Lung Project (a large controlled trial designed to assess lung cancer screening with chest roentgenograms and sputum cytology among high-risk adults conducted in the 1970s and 1980s), chest x-ray imaging was noted to detect early-stage lung cancers, without leading to any difference in lung cancer mortality. [85] [86] Furthermore, excess cases were noted in the intervention arm, leading to “overdiagnosis.”[87] Four large randomized trials of screening chest x-rays in smokers have demonstrated no difference in survival between the randomized groups.[88] In subsequent chest CT screening trials at The Mayo Clinic, CT was shown to detect early-stage tumors, but with a high rate of false-positives (i.e., benign nodule detection). Recent support for lung cancer CT screening among high-risk individuals comes from the singular outcome analysis of a large trial in which 85% of the screen-detected lung tumors were stage I at diagnosis, and 10-year overall survival was 88% for these stage I patients.[89] Although these observational findings are impressive, we must await the results of confirmatory randomized controlled prospective screening trials before making broad screening recommendations for lung cancer. Thus, the National Lung Screening Trial (CT imaging) and PLCO (chest x-ray) trials for lung cancer screening will serve as important validation studies to evaluate the efficacy of lung cancer screening among high-risk individuals.
Chemoprevention
Epithelial cancers of the upper aerodigestive tract and lungs are the most extensively studied system for chemoprevention in humans, and the results are the most negative. The natural history of the disease process has been studied extensively and provides a rich platform from which to conduct chemoprevention trials. Field carcinogenesis by tobacco carcinogens with its associated epidemiologic risk and characterized molecular changes is a straightforward concept that has guided the development of chemoprevention studies in this area.[46] The recent identification of molecular (e.g., DNA repair, telomerase), metabolic (e.g., cytochrome P450, alcohol dehydrogenase), and mutagen sensitivity profiles that predispose to aerodigestive cancers, acquired chromosomal abnormalities in the field and in the cancers, and alterations of several molecular parameters that predict responsiveness and unresponsiveness, have recently provided useful detail from which to consider the next generation of rational chemoprevention trials.[90] The identification of a variety of molecular changes during head and neck cancer progression, in addition to readily identifiable histologic precursors, has provided a biologic base for understanding the interaction of carcinogenesis and chemoprevention of this disease. Recent studies of the molecular changes that accompany the progression of lung cancer also provide a useful paradigm and platform from which to develop well-considered chemoprevention approaches.
Thus far, however, the results of primary, secondary, and tertiary chemoprevention trials of the lung have been disappointing (see Goodman[91]). Two large placebo-controlled, multiagent randomized trials in heavy smokers (more than 47,000 participants) have been negative and showed no beneficial effect of retinol (vitamin A) or α-tocopherol (vitamin E). [60] [61] More disturbingly, these two large, randomized trials indicate that current smokers supplemented with oral β-carotene developed lung cancers at a rate 25% greater than the placebo group and also showed an increased overall mortality. Although these findings remain unexplained, possibilities that might help explain the adverse effect of β-carotene in these studies include the following:
|
• |
The formation of cyclic epoxides in the setting of tobacco carcinogens, inflammation, and high β-carotene concentrations |
|
|
• |
The suppression of RAR-b, a major transcription factor important for differentiation in epithelial tissues |
|
|
• |
Lowering of the concentration of other micronutrients that might be protective |
|
|
• |
Stimulation of preneoplastic clones by enhancement of growth factor production |
|
|
• |
Complex genetic polymorphisms that lead to alteration of tobacco carcinogen metabolism |
Definitive secondary (metaplasia, atypia) and tertiary (second malignancy) chemoprevention trials that use a number of different retinoids and other compounds (folic acid, N-acetylcysteine) have also yielded negative results. [91] [92] Treatment with anethole dithiolethione (an organo-sulfur compound) was shown in a randomized trial to reduce development of new bronchial dysplasia lesions and to slow progression of pre-existing disease in current or former smokers.[93] However, a randomized trial of inhaled budesonide in smokers with bronchial dysplasia indicated no effect of the active agent in the regression of bronchial dysplastic lesions or prevention of new lesions.[94] A modest decrease in p53 and BCL-2 protein expression in bronchial biopsy specimens was seen, as was a slightly higher rate of resolution of CT-detected lung nodules. Whether budesonide or anethole dithiolthione will be useful in managing preneoplastic lesions of the lung will require assessment in larger and longer trials.
Studies of secondary and tertiary chemoprevention of head and neck cancers have led to somewhat more encouraging results. Randomized trials have shown that isotretinoin causes regression of oral leukoplakia, though accompanied by substantial side effects.[95] Two randomized trials confirmed activity of β-carotene, although results from the later study were less convincing. [96] [97] Several other, less toxic agents (retinol, 4-[hydroxyphenyl]-retinamide, and α-tocopherol) and selenium have also produced responses of premalignant lesions in phase II trials.[98] A randomized trial of the cyclooxygenase inhibitor ketorolac as an oral rinse was negative in patients with oropharyngeal leukoplakia.[99] A recently reported phase II trial of the complex retinoid fenretinide suggested activity for this compound in patients with retinoic acid–resistant oral leukoplakia.[100] We have demonstrated substantial potential activity of Bowman-Birk inhibitor (a soybean-derived compound) against oral leukoplakia.[101] The results of randomized studies for these latter two compounds have not yet been reported.
In a randomized phase III adjuvant trial of patients treated for head and neck cancer by local therapy, the synthetic retinoid 13-cis-retinoic acid (isotretinoin, Accutane) at a high daily oral dose (50 to 100 mg/m2) led to a reduction in the incidence of second primary tumors, a difference that was maintained for more than 5 years.[102] The rate of second primary tumors was affected greatly by tobacco smoking status, with the efficacy of chemoprevention decreasing sequentially in current and former smokers as compared with nonsmokers.[103] The side effects in the 13-cis-retinoic acid trial were substantial at this dosage level, however. These results with 13-cis-retinoic acid were particularly significant in that a similarly designed randomized trial using another retinoid (etretinate) at a high dose showed no reduction of second primary tumors.[104] Therefore, the efficacy of low-dose isotretinoin (30 mg/day) to prevent second primary tumors after treatment of early-stage (I and II) head and neck cancer was tested in a randomized trial, with the hope that side effects could be decreased without losing efficacy. This strategy was unsuccessful, and no difference in the appearance of second malignancies in the placebo and treatment groups could be demonstrated; it is noteworthy that patients who continued to smoke had an increased rate of second primary tumors and death.[105]
The incidence of adenocarcinoma of the esophagus has been increasing over the past 2 decades. Gastroesophageal reflux has been identified as a risk factor leading to the development of a columnar-lined esophagus, called Barrett's esophagus, which progresses to adenocarcinoma via a metaplasia-dysplasia sequence ( Fig. 26-2C ). Both endoscopic resection and thermal or photodynamic ablation have been used to treat this condition,[106] but long-term benefit is unknown. Epidemiologic and clinical observations also suggest that NSAIDs and aspirin, by inhibiting COX enzymes, and proton pump inhibitors by decreasing gastric reflux[107] should be effective as chemopreventive agents, but randomized controlled trials proving this supposition have not yet been reported. Attempts to reverse or suppress these lesions with 13-cis-retinoic acid have been unsuccessful.[108]
Overall, these trials suggest that oral leukoplakia, but not bronchial metaplasia, can be reversed or suppressed by currently available chemopreventive agents. Thus far, “proof of principle” of chemoprevention in head and neck cancers has been achieved. However, large-scale phase III trials will have to show efficacy with a favorable risk-benefit profile before the strategy of chemoprevention can be adopted into standard medical practice for the management of aerodigestive cancers.
COLORECTAL CANCER
Screening for and early detection of colorectal cancers results in 5-year survival rates of 90% for colon cancer and 80% for rectal cancer, providing that diagnosis and treatment occur before the lesions have spread beyond the bowel to regional lymph nodes or distant metastatic sites. Unfortunately, more than 60% of patients still present with higher staged disease, leading to a lower overall 5-year survival rate of 65%.[78] In the United States, colorectal cancer affects 148,000 individuals annually and is responsible for more than 55,000 deaths per year, which is surpassed only by cancer deaths secondary to lung cancer.[78] These statistics highlight the critical importance of identifying individuals at risk for the development of colorectal cancer and of screening and early detection in its management. This section will discuss various risk factors, both modifiable (i.e., extrinsic risk factors) and not modifiable (i.e., intrinsic risk factors), that increase susceptibility for the development of colorectal cancer and will review current screening guidelines. This information can be used to design more effective preventive strategies, which could make use of genetic testing for individuals at risk and apply behavioral modification and chemoprevention.
Pathogenesis
Numerous epidemiologic, international, and experimental studies have evaluated various hereditary and environmental factors that lead directly or indirectly to the development of colorectal cancers. It is believed that colon cancer is the result of a complex series of genetic and epigenetic events that occur when environmental factors interact with an individual's inherited or acquired susceptibility. [42] [109]This interaction produces somatic mutations that accumulate over time and lead to neoplastic transformation of normal colonic epithelium into premalignant adenomatous polyps (see Fig. 26-2D ) and ultimately into invasive disease. The natural history preceding the development of cancer can progress through several decades. Adenomatous polyps, especially the villous subtype, are the premalignant lesions in more than 90% of colorectal cancers. The risk of malignant degeneration depends on the size of the polyp (increasing greatly in those greater than 1 to 2 cm in size), duration of its presence, number present at the time of the initial examination, and the histologic type.[110] Only adenomatous polyps seem to carry a premalignant risk; however, hyperplastic polyps are diagnosed more commonly in individuals with a smoking or drinking history, two predisposing factors for adenomatous polyp development.[111] The presence of hyperplastic polyps may warrant increased screening and prevention counseling, although further study of this issue is needed before definitive conclusions can be made.
Etiology
Heredity
Our understanding of the genetic and molecular alterations that precede the development of colorectal cancers has broadened and deepened over the past 10 to 15 years ( Fig. 26-3 ).[112] This information has facilitated the identification of individuals who might benefit from early interventions with more vigilant screening, chemoprevention, or treatment. Based on studies of family histories, it is estimated that 20% to 30% of colorectal cancers have a significant hereditary component.[112] Thus far, however, genes associated with only two major syndromes—familial adenomatous polyposis (FAP) and hereditary nonpolyposis colon cancer (HNPCC)—have been identified clearly. Allelic deletions have been identified in patients diagnosed with these two autosomal dominant syndromes, FAP and HNPCC.[113] [114] [115] FAP accounts for only 1% of colon cancer cases per year and is associated with a deletion of the APC gene on chromosome 5 (band q21). These patients develop thousands of adenomatous polyps that tend to be evenly distributed throughout the colon and rectum by the second or third decades of life. If surgical treatment by complete colectomy is not done, affected individuals are at high risk to develop colon cancer by the age of 40. A highly specific mutation (T to A at nucleotide 3920) has been found in 6% of Ashkenazi Jews, and about 28% of Ashkenazim have a family history of colorectal cancer.[116] This mutation created a small hypermutable region of the gene, thereby indirectly causing predisposition.
|
Figure 26-3 Colorectal cancer results from inherited genetic predisposition and acquired molecular alterations interacting with environmental and endogenous toxins that are themselves modified by gene products. TSG, tumor suppressor gene; ONC, oncogene; ROS/RNS, reactive oxygen and reactive nitrogen species form from normal metabolic processes (endogenous) and from exposure to carcinogens and toxins (external). |
The incidence of HNPCC was determined to be less than 1% of annual colon cancer cases in a large population-based study, although estimates of 5% are typically reported from nonpopulation-based studies.[117] Diagnosis requires that three or more relatives be diagnosed with colorectal cancer, representing at least two successive generations, and at least one relative must have been diagnosed before the age of 50.[115] One relative must be a first-degree relative of the proband patient. Patients tend to have cancers that arise in the proximal colon, and they also develop ovarian and endometrial cancers at a higher rate than the population at large. This syndrome is associated with defective DNA repair mechanisms, which lead to aberrant cell growth and tumor formation. These mutations occur on chromosomes 3 (hMLH1, 3p21) and 2 (hMSH2, 2p). [118] [119] Based on extensive experimental and clinical data, Vogelstein and colleagues [42] [113] proposed that it is the progressive accumulation of mutations that ultimately lead to invasive disease. This proposal has been substantiated extensively, and mutations associated with colorectal cancers have been identified involving proto-oncogenes, tumor suppressor genes, and certain key regulatory enzymes, such as cytochrome P450 and acetyltransferase. [120] [121] [122] It has been suggested that colorectal cancers in adults develop through one of three different pathways (chromosomal instability, microsatellite instability, and CpG island methylator phenotype) and have different biologic behaviors.[109] The role of polymorphisms in metabolizing key molecules (including those present in the diet) is being examined closely and should provide a platform from which to understand gene-environment interactions. For example, polymorphisms in hepatic cytochrome P450 and acetyltransferase enzymes lead to rapid oxidation and acetylation of genotoxic compounds such as heterocyclic amines, which are present in processed foods.[122] Accelerated metabolism of these compounds increases an individual's risk of developing colorectal cancers threefold.
Diet
Although diet seems to play a significant role in colon carcinogenesis, the degree to which individual macronutrients and micronutrients contribute to the development of colorectal cancer has been elusive. In part, this difficulty stems from differences in design and methodology in studies that have been performed to evaluate this subject, including the type of dietary questionnaire administered, differences in cohorts such as age and ethnicity, confounding effects of other dietary components, selection and recall biases, sample size, and length of follow-up.
The majority of past evidence has demonstrated an increase in incidence and mortality rates from colorectal cancers in groups of people who consume a more “westernized” diet that is high in animal fat, total calories, and red meat but low in fiber and fruit and vegetable intake.[123] International and migrant studies have supported this observation. Recent studies, however, have indicated that the older evidence should be reconsidered. Large prospective cohort studies and several large randomized trials indicate that fiber does not seem to be protective nor fat contributory to colon cancer development.[124] [125] [126] In contrast, mechanistic considerations, metabolic studies, and epidemiologic studies suggest a strong protective effect of folate and an important role of insulin and insulin-like growth factors in colon cancer pathogenesis. Potter[109] and other researchers have done a particularly nice job in attempting to relate genetic changes, risk factors (including diet), and downstream molecules and pathogenesis.
Other Factors (Alcohol, Smoking, Exercise, Body Mass Index)
Primary prevention of colorectal cancer also requires that we understand factors other than diet that increase risk for colorectal carcinoma by initiating or promoting carcinogenesis. These include use of alcohol and tobacco, sedentary lifestyle, and the metabolic changes that proceed from these. [18] [127]
Many studies have demonstrated a relationship between alcohol use and colorectal cancer and adenoma formation. [128] [129] It is still uncertain whether alcohol directly initiates DNA damage or acts as a promoter on cells that already have undergone preneoplastic changes. A low-methionine or low-folate diet might contribute to a situation leading to adenomas and colorectal cancer, in that both methionine and folate are cofactors for DNA synthesis; lowered concentrations of these compounds leads to hypomethylation of DNA, which is a precursor to aneuploidy and loss of heterozygosity.[130] Various forms of a key enzyme, 5,10-methylenetetrahydrofolate reductase, which catalyzes the conversion of 5,10-methylenetetrahydrofolate to 5-methyltetrahydrofolate, have also been identified.[131] Some mutations of this enzyme increase its activity, whereas others decrease it. Low activity leads to decreased methionine synthesis and antagonizes methyl group metabolism in DNA synthesis.[131] This theory gained some support in the U.S. Male Health Professionals Study.[111] An association was found between high alcohol intake and methionine-deficient diets, after controlling for intakes of fat, red meat, fiber, level of physical activity, body mass index (BMI), and multivitamin and aspirin supplementation. Alcohol might be particularly important for progression of large adenomas to tumors.[128] Avoiding excess alcohol, while increasing dietary folate and methionine, seems like a reasonable approach to decreasing risk for colon cancer. Several large studies are underway to test whether supplementation with folate can reduce adenomata development.
Cigarette smoking has been consistently associated with adenoma formation but less so with colon tumors, an observation that could be explained by the molecular nature of a subset of colon tumors in which microsatellite instability and/or p53-negative status is a prominent feature.[132] The Health Professionals Follow-up Cohort Study and the Nurse's Health Study observed more adenomas in individuals with a history of smoking than in those who did not smoke.[133] Analysis of the very large American Cancer Society Cancer Prevention II study indicates that 20% of colorectal cancers and 12% of deaths are associated with long-term cigarette smoking[134] ( Fig. 26-4 ).
|
Figure 26-4 The potential effect of diet on risk of colon cancer. NEFA, nonesterified fatty acids; ROS, reactive oxygen species. |
Physical inactivity and high BMI also increase one's risk for colorectal cancers. A prospective study found a significant inverse association between leisure-time physical activity and incidence of colon cancer in participants of the Nurse's Health Study.[135] An inverse association was also observed between physical activity and the development of large (>1 cm) adenomas in the distal colon. In this same study, more adenomas were observed in individuals with a high BMI. Obesity, and in particular abdominal adiposity, has also been associated with an elevated risk for adenomatous polyps and colon cancer.[136] Recently, two separate cohort studies have demonstrated that physical activity is associated with improved outcomes among resected colon cancer patients. [137] [138] Increasing physical activity and maintaining lean body weight for the prevention of colorectal cancer probably has considerable merit for decreasing the incidence of polyps and colon cancer, as well as of other chronic diseases.
The mechanisms underlying these proposed effects are not clear, but a unifying hypothesis has been proposed recently and involves the sequential steps of consumption of excess dietary energy, development of insulin resistance, and increased circulating levels of insulin, triglycerides, and nonesterified fatty acids, which results in secondary colonic epithelial damage.[139] Thus, the beneficial effects of physical activity on colon cancer risk and outcomes represents the sum total of numerous cellular and molecular events. A goal of prevention research is to determine which molecular pathways are being affected by physical activity and to evaluate potential biomarkers in this process. However, validated mechanisms have been lacking. Epidemiologic evidence associating insulin, insulin-like growth factor-1, and insulin-like growth factor–binding protein levels with colorectal cancer incidence provides considerable support for the proposal that changes in diet and physical activity may affect key molecular events.[140]
Screening and Early Detection
Population-Based Data
A definitive amount of data has accumulated indicating that colorectal cancer screening for persons at average risk is effective and reduces colorectal morbidity; agreement on the best screening modality remains unsettled, however.[141] Cost-saving analysis even supports the use of universal colonoscopy over the long term, notwithstanding the considerable practical challenges and total cost in achieving this goal. Probably everyone over 50 years of age should be screened. Yet surprisingly, fewer than 50% of individuals who should be screened are evaluated by fecal occult blood tests (FOBT), digital rectal exams (DRE), and/or sigmoidoscopy.[32]
Conflicting recommendations from government and private agencies contribute to the confusion. In general, all groups advocate screening in men and women beginning at the age of 50, because the incidence of colon cancer rises sharply between the ages of 50 and 55 and continues to double with each succeeding decade, reaching a peak by the age of 75. The American Cancer Society and the American College of Obstetricians and Gynecologists support yearly DREs beginning at age 40 and FOBTs performed yearly, once a person reaches 50 years of age. Sigmoidoscopy should occur every 3 to 5 years beginning at 50 years of age. Flexible sigmoidoscopy is superior to rigid sigmoidoscopy, because it allows the examiner to visualize up to 60 cm of bowel mucosa and is easier on the patient.
There are numerous studies validating the use of DRE, FOBT, and sigmoidoscopy as effective screening tools, providing regular screening is performed (to detect lesions in this disease with a long preinvasive phase). [142] [143] [144] This is particularly important with the FOBT, because reported sensitivities are low and range between 22% and 92%. Sensitivity is higher when at least three tests are performed on different days and the samples are rehydrated with hydrogen peroxide. To decrease false-negative results, which also increases the sensitivity of the test, patients should be instructed to avoid vitamin C and to eat a high-residue diet for several days before the test. The incidence of false positives is lowered when gastrointestinal irritants (e.g., aspirin, oral iron, and meat products) are not consumed a few days before the FOBT. In general, screened, asymptomatic patients have a positive test 4% to 6% of the time. Only 5% to 10% of these individuals have colorectal cancers, and an additional 30% have benign polyps. [145] [146] The positive predictive value is only about 20%, so a positive test can be costly, because follow-up requires evaluation by sigmoidoscopy or colonoscopy. Randomized clinical trials have demonstrated a decrease in mortality by 15% to 33% with the regular use of serial FOBT, however.[143] Patients who received FOBT in conjunction with sigmoidoscopy in a Memorial Sloan-Kettering colon cancer trial had a significantly higher survival probability when compared with those who received sigmoidoscopy alone (70% vs. 48%, respectively).[143] This is an impressive result.
Regular screening with sigmoidoscopy among patients over 50 years of age both reduces mortality from colorectal cancer and prolongs survival. [147] [148] Studies must be performed to clarify the optimal interval between screening and to develop recommendations for individuals at higher risk for adenomas or colorectal cancers. The early results of a “once-only” sigmoidoscopy at age 60, in which a high yield of adenomas was obtained, suggests that a less intense approach to screening could be a cost-effective strategy to prevent colon cancers.[149] Follow-up colonoscopy suggested that a significant number of proximal adenomas were present. The efficacy of screening colonoscopy for adenoma detection was directly related to prolonged endoscope withdrawal time in a recent study, a finding that may influence patient management in the near future.[150] There is little doubt that screening colonoscopy is effective at identifying silent, large adenomas and colon cancers, but its cost effectiveness must be demonstrated; so must the optimal usage of fecal occult blood, sigmoidoscopy, and colonoscopy with regard to efficacy and cost. A particularly exciting approach is the identification of colon cancer-specific mutations in fecal DNA.[151] A study of the adenomatous polyposis coli (APC) gene, the initiating abnormality in most sporadic colon cancers, in the fecal DNA of normal individuals and patients with polyps shows considerable promise.[152] Although the specificity of this test was high (100%), the sensitivity was only 57%. However, integration of the test into the screening paradigm (perhaps with FOBT) as a low-cost, general screen is an important goal. The DRE should be included in all examinations, because it permits evaluation of the distal rectum and prostate. Its sensitivity has decreased, however, with the temporal shift to more proximal lesions in the colon.[149] Other tests, including double-contrast barium enema, have not been evaluated completely as screening tests and are more expensive than DRE, FOBT, and sigmoidoscopy. We feel that these modalities should be reserved for those patients with positive screening tests.
High-Risk Individuals
More frequent screening is recommended for certain postpoly-pectomy individuals and other individuals at increased risk for the development of adenomatous polyps or colorectal cancers.[153] One particularly important group includes those who have had previous treatment for a primary colorectal cancer. Recent analysis indicates that evaluation of this high-risk group has been inadequate. Proctosigmoidoscopy or colonoscopy should begin at age 10 for first-degree relatives of individuals with FAP and at age 25 for individuals with a strong family history suggestive of HNPCC. [154] [155]Genetic testing for the mutated APC gene in peripheral mononuclear cells of patients with early onset of adenomatous polyps or in first-degree relatives of FAP-affected individuals will help to determine who needs closer surveillance.[156] Counseling about diet, chemoprevention, and lifestyle issues can be provided for individuals who test positive for the mutated APC gene. In addition, a colectomy should be considered if polyps are found.
Genetic testing for mutations in DNA repair genes should be performed on anyone whose family history is suggestive of HNPCC; counseling can then be provided for individuals who test positive.[157] A positive family history of colorectal cancer that does not meet criteria for FAP or HNPCC probably also warrants earlier screening, but definitive guidelines have not been established. A prospective study of approximately 120,000 men and women who underwent colonoscopy or sigmoidoscopy surveillance concluded that the age-adjusted relative risk for cancer was 1.72 with one first-degree relative with the disease, 2.75 with two first-degree relatives, and 5.37 with one first-degree relative who was under age 45 at the time of diagnosis.[158] It is estimated that as many as 25% of individuals with colon cancer have positive family histories. Baseline colonoscopy before age 50 seems to be a reasonable consideration among this group of patients with a positive family history.
Other high-risk conditions requiring close surveillance include individuals with a long-standing history of inflammatory bowel disease, a prior history of polypectomy or ureterosigmoidostomy, a personal history of ovarian, endometrial, or colon cancer, and finally anyone who has been treated for Streptococcus bovis bacteremia. [154] [159] [160] [161]
Other Health Factors
In considering health care resources, other issues that affect the development of colon cancer should be considered. The role of estrogen replacement in the development of colon cancer and in other health parameters in postmenopausal women is of great importance. The results of the Nurse's Health Study suggested that current estrogen replacement in postmenopausal women can reduce the risk of colon cancer (relative risk = 0.65); this effect disappeared 5 years after discontinuation of estrogen replacement.[162] However, a detailed analysis of the group of postmenopausal women in the WHI trial who received estrogen plus progestin indicated that the colorectal cancers were diagnosed in a more advanced stage even though the number of colorectal cancers was reduced by 45%.[163] When reproductive factors were examined among women who were diagnosed with colon cancer in this study, oral contraceptive pill use and later age of menarche were also associated with a decreased risk. In another study, women who delivered more than five children, especially if they had a positive family history of adenomas, were at increased risk for the development of adenomas.[164] In this study, however, no association was found with age at menarche, menopause, first birth, or oral contraceptive pill use. Although estrogen plus progestin lowered the risk for colorectal cancer in the WHI trial,[163] these favorable results will have to be balanced against the increased risk for breast cancer and cardiovascular disease (see later discussion).
Chemoprevention
Despite the enormous amount of epidemiologic observations and experimental data that support a protective role of many dietary constituents against the development of adenomatous polyps and colorectal polyps, the results from definitive randomized trials have been modest, at best. Many micronutrients and dietary constituents have been studied. Trials using vitamins C, D, and E, and β-carotene as well as supplementation with fiber or lowering of dietary fat have been uniformly negative. Trials in which calcium has been supplemented have been modestly positive with a 25% to 35% reduction of adenomatous polyps demonstrated. [24] [25] However, daily supplementation of calcium with vitamin D for 7 years had no effect on the incidence of colorectal cancer among postmenopausal women in the WHI.[165] These results suggest that calcium inhibits colon carcinogenesis at early but not late stages.
Additionally, in a double-blind, randomized, placebo-controlled trial of selenium supplementation for skin cancer prevention, a secondary analysis showed that the numbers of colon, prostate, and lung cancers were found to be reduced by 50% in the treatment arm.[166] The results of additional randomized trials are pending and should become available in the next 3 to 5 years.
Study of the effects of micronutrient supplementation on the formation of polyps and colorectal cancers is difficult, given the inherent complexity of carcinogenesis. Understanding how various dietary components inhibit carcinogenesis will be instrumental to the development of novel dietary-derived chemopreventive agents in the future. Factors such as type of micronutrient, dose, and duration of treatment, as well as cohort demographics (age and geographic location), and endpoints (polyp formation, or changes in the incidence and mortality of invasive cancer) are all important variables that will affect trial results. However, despite these difficulties in study design, analysis, and interpretation, we should not dismiss the large amount of epidemiologic evidence and supportive experimental data demonstrating that the consumption of diets rich in fruits and vegetables, but low in fat, have a lower incidence of bowel cancer and cancer in general.[167] Observational and animal studies suggest that reduction of caloric intake and moderate physical exercise should also decrease the risk of colorectal cancer.[168] Meat consumption—particularly red meat that has been processed, has been associated with increased risk of colorectal cancer in numerous epidemiologic studies. Substances in meat that have been implicated in colon carcinogenesis include the known carcinogens: heterocyclic amines (formed in meat during cooking), polycyclic aromatic hydrocarbons and N-nitroso compounds (both found in processed meats), and the amino acid arginine (i.e., a key substrate of the polyamine synthetic pathway).[169] [170] [171] [172] Research from clinical trials should provide us with invaluable data on which to base dietary and lifestyle recommendations for the prevention of colorectal cancer.
There are also abundant scientific opportunities to explore the role of nondietary chemoprevention compounds in controlling colorectal cancer based on substantial studies of colon carcinogenesis (see review[173]). Some of the more active nondietary compounds being studied include NSAIDs and DFMO. Mechanistic studies indicate that epithelial regeneration and focal inflammation may be important early changes in the pathogenic process, [139] [173] so that the use of nontoxic antiproliferative and anti-inflammatory agents as chemoprevention agents has a strong rationale. These biologic features of colon carcinogenesis are also used to support a role for probiotics and prebiotics for the prevention of colorectal cancer.[174]
A considerable amount of experimental and epidemiologic evidence exists to support the use of NSAIDs to decrease the risk of colon cancer.[173] These compounds exert their antiproliferative effects on colonic cells through inhibiting prostaglandin synthesis by reversibly binding to cyclooxgenase as well as through several other newly discovered mechanisms. Laboratory studies have consistently demonstrated that NSAIDs can inhibit chemically induced and transplanted tumors in rodents. The interpretation of epidemiologic trials involving NSAIDs is challenging because of differences in design and methodology, including the particular agent chosen, the dose, frequency, and duration of use, and variable follow-up periods. Nevertheless, most case-control and cohort studies have demonstrated an association of a reduced risk of colon cancer with increased consumption of NSAIDs.
Analyses of subgroups of patients who routinely take these drugs—such as patients with rheumatoid arthritis (aspirin), inflammatory bowel disease (sulfasalazine), and FAP (sulindac)—have reported a decrease in either adenomatous polyp formation or the development of colorectal cancer. A particularly important study was the Nurse's Health Study, which used three consecutive questionnaires to determine the rate of colorectal cancer among women who consumed aspirin and compared these rates to women who reported no aspirin use.[175] After at least a decade of regular aspirin use, at doses similar to those recommended for the prevention of cardiovascular disease, aspirin consumption was found to reduce the risk of colorectal cancer substantially. Both celecoxib and sulindac have been shown to cause regression of polyps in patients with FAP. [176] [177] Sulindac was not effective in preventing the development of new polyps in these patients, however.[178]
The results of several large randomized trials using NSAIDs have been reported. [168] [179] [180] Aspirin has uniformly reduced the recurrence of adenomatous polyps by 25% to 35% in patients at moderate risk with acceptable toxicity. Studies with COX-2 selective inhibitors have produced an unusual conundrum. Several randomized trials have demonstrated substantial benefit with a 40% to 50% reduction in recurrence of adenomas; however, cardiovascular events were markedly increased in the treatment arm. [62] [63] These results have put a significant damper on the development of chemoprevention agents for cancer. Whether similar adverse results will be seen with less selective NSAIDs (e.g., sulindac) is unknown and awaits the results of ongoing trials.
DFMO has been found to be a potent inhibitor of carcinogenesis in experimental animal models by reducing the number and size of adenomas and carcinomas. This drug exerts its effects by irreversibly inhibiting ornithine decarboxylase, the first enzyme in the polyamine synthesis pathway. Suppressing intracellular pools of polyamines decreases cell growth and interferes with the process of carcinogenesis in essentially all animal models. We have reported the results of a long-term clinical trial, which serially measured the effects of different doses of DMFO on rectal mucosal polyamines over a 12-month time period and demonstrated consistent suppression without side effects.[57] Demonstration of a dose of a chemopreventive agent that has a substantial biochemical effect without producing clinical side effects is an important goal. Currently, we are studying the prevention of adenomas in patients that have undergone polypectomies for adenomatous polyps. DMFO will be given in conjunction with sulindac over a 3-year period, and its effect in reducing polyp recurrence is being studied in a randomized, placebo-controlled trial.
The role of tertiary prevention in colorectal cancer has been little explored. As effective chemoprevention agents are developed, the group of patients who have been “cured” by standard therapy should become a focus of investigation, because the incidence of second primary colon cancers is high (about 25%).[181]
Integration of Prevention Activities
Several approaches exist to prevent the development of colorectal cancers. Successful primary prevention depends on public education and counseling about behavioral and dietary modifications that can be made to decrease an individual's risk, including increased physical activity and reduction of total calorie intake. In patients with adenomatous polyps, polypectomy is a useful preventive measure. Patients with FAP or HNPCC might require colectomies at a younger age to prevent development of cancer. New advances in genetic testing will help to select people who not only need closer surveillance but also might benefit from surgical treatment before cancer or premalignant polyps develop.
More vigilant screening in individuals at increased risk for the development of colorectal cancer is reasonable, but the appropriate screening tests, and the optimal interval between tests, require further clarification. More important, given that the majority of cancers occur in patients without a family history, everyone probably should be screened beginning at 50 years of age. Clinical applications of chemoprevention with DFMO, NSAIDs, and various micronutrients are still under development but may offer important alternatives for the future or as part of an overall approach to prevention ( Fig. 26-5).
|
Figure 26-5 Routine screening and early detection for colorectal cancer and its prevention. *There is a growing consensus that a colonoscopy should be performed at age 50 and all polyps removed. If negative, repeat about every 5 years. If positive, repeat in 3 years. If there is a familial tendency, start 5 years before age of youngest family member with disease. |
BREAST CANCER
The morbidity and mortality from breast cancer remains high despite significant advances in our understanding and management over the last several decades. Therefore, prevention and early detection have become important challenges for the medical community. In addition to an enormous health benefit, several billion dollars would be saved annually if breast cancer were prevented and/or the disease were detected at an earlier stage. The widespread use of screening mammography, the increasing recognition that breast density is a major risk factor, the identification of high-risk individuals based on family history, the detection of deleterious mutations, and the “proof of principle” that tamoxifen and raloxifene can reduce the risk for breast cancer all anticipate more effective early management of this disease.
Etiology
Heredity
Primary prevention depends on our ability to identify individuals who are at increased risk for the development of breast cancer.[182] Although many risk factors cannot be changed, knowledge of their presence can be used to identify high-risk individuals. Age, socioeconomic class, geographic location, race, and ages of menopause, menarche, and first birth are examples of risk factors that are difficult to change but important to recognize. The incidence of breast cancer, like that of most cancers, increases with age. The majority of cases are diagnosed in women older than 40 years of age, with only 10% to 15% occurring in women younger than 40 years old and fewer than 5% occurring among women younger than 35 years of age. Affluent women and individuals born in colder climates or in the Western hemisphere also tend to have a higher incidence of breast cancer. White women have more breast cancers than black, Asian, Hispanic, or Native American women. It is of considerable interest that Hispanic and Native American women have many of the same demographic variables (obesity, high fat, and low vegetable diet) that are associated with a high incidence of breast cancer in whites, yet their incidence of breast cancer is less than half that of whites.[183] Determining whether this difference is genetically based or if it reflects some other protective dietary or environmental factor is an important issue to address. By identifying who is at higher risk for breast cancer, health professionals can then counsel this subgroup of women and their families about the risks for breast disease and various ways to modify these risks, and they can encourage enrollment into clinical trials aimed at studying novel approaches for breast cancer risk reduction.
The most important step in trying to discern who is at risk is to take a detailed personal and family history extending back at least three generations.[184] Nearly 25% of women diagnosed with breast cancer have a family history of the disease.[184] Recent advances in our understanding of the molecular biology of breast cancer have led to the identification of specific mutations that might help identify women with a hereditary predisposition to developing breast cancer and might help predict who will respond to adjuvant therapy. Medical records, including pathology reports, should be obtained whenever possible to help complete an accurate pedigree. Recall bias is a significant problem when constructing pedigrees and can profoundly influence how we counsel patients; therefore, it is important to collect documentation whenever possible. Family histories must be gathered from the maternal and paternal sides of the family. This latter step is often neglected and makes it impossible to counsel anyone in a meaningful way.
Although most cases of breast cancer are sporadic and a product of many genetic insults, approximately 5% are due to specific inherited germline mutations in the BRCA1 and BRCA2 tumor suppressor genes.[185] The estimated lifetime risk for breast cancer in BRCA1 and BRCA2 mutation carriers ranges from 55% to 85%, in comparison with the 13% lifetime risk for the general population. [185] [186] [187]Women with these mutations are also at increased risk for the development of a second breast cancer; BRCA1 mutation carriers carry up to a 65% lifetime risk, and BRCA2 carriers might share a similar risk.[187] Therefore, carriers of mutations in the BRCA genes and women with a personal or family history might benefit from prevention strategies and genetic counseling.
Genetic testing should be offered to individuals with a strong family history (breast or ovarian cancer in two or more generations), a history of multiple primaries (ovarian or breast, colon, endometrial), early age of onset of breast cancer (<35 years), or individuals of Ashkenazi Jewish descent, who carry a frequency of mutations in the BRCA genes estimated to be 2.2%. [185] [186] Approximately one out of every 300 to 800 American women carries a BRCA1 mutation; however, not every one of these women will develop breast cancer. This variable penetrance seen in BRCA mutation carriers highlights one of the dilemmas of genetic testing for BRCA gene abnormalities. Furthermore, a negative test might offer false reassurance, because it could represent a false negative. In addition, although a number of mutations in the BRCA genes have been sequenced, there are probably many more that we have not identified and for which we therefore cannot test. Practitioners should be aware of the uncertainty inherent in genetic testing for BRCA mutations and be prepared to counsel their patients accordingly.
Family history without a clearly defined genetic syndrome is also important in counseling women about their risks for developing breast cancer. [188] [189] A woman with a first-degree relative with premenopausal breast cancer carries anywhere between a 1.8- and an 8.8-fold increased risk of developing a breast cancer in the future and is at high risk for harboring a deleterious mutation. This risk decreases to 1.2- to 4.0-fold in a woman who has a first-degree relative who developed breast cancer after menopause. Having a second-degree relative with breast cancer increases a woman's risk by approximately 1.5-fold.
Hormonal Factors
Women with a long lifetime exposure to estrogen are also more likely to develop breast cancer. The risk for breast cancer increases by 20% if menarche occurs before the age of 12.[190] Furthermore, women who experience a late menopause, are nulliparous, or deliver their first child after 30 are also at increased risk. An induced abortion does not result in an increased risk of breast cancer ( Box 26-1 ).[191]
Box 26-1
|
INDICATIONS FOR GENETIC TESTING IN BREAST CANCER[*] |
|||||||||
|
|||||||||
|
* The indications for genetic testing in breast cancer and in other cancers are in rapid evolution as the true risks become better defi ned and as prevention (e.g., tamoxifen) and early detection (e.g., mammography, MRI) strategies mature. |
The subject of HRT and its role in the etiology and progression of breast cancer has been among the most intensely studied in medicine. Studies evaluating the role of hormone replacement in postmenopausal women have reported contradictory results—not surprising given that many of the studies evaluated different doses, preparations, follow-up times, and different age cohorts. [192] [193] [194]Results from the WHI study (i.e., the largest prospective randomized clinical trial to address the issue of HRT on breast cancer development), however, conclusively demonstrated an increased risk of invasive breast cancer (relative risk = 1.26) among those women taking estrogen plus progestin compared with placebo control subjects.[17] Additionally, the supplementation of postmenopausal women with estrogen plus progestin in the WHI trial led to increased mammographic density and breast cancer, [195] [196] These findings have sparked an intense debate, with proponents of HRT advocating no change in prescribing habits and opponents vociferously advocating the opposite.
Compelling data in support of the aforementioned WHI results have emerged from recent epidemiologic reports indicating a sharp decrease in breast cancer incidence in 2003, particularly in estrogen receptor (ER)–positive tumors. [197] [198] This time period represents the first year after results of the WHI trial were released, with a resultant immediate 50% decrease in HRT use. The rapidity of this observed decreased breast cancer incidence supports an effect of hormone therapy on progression of subclinical lesions. These reports from the California Cancer Registry and in the national U.S. Surveillance, Epidemiology, and End Results (SEER) registry have sparked a renewed debate about the role of HRT in early breast cancer development, because it represents the first significant decrease in breast cancer incidence in the United States in more than half a century.
In contrast, the preponderance of evidence suggests that the risk for breast cancer from oral contraceptive use is very low or nonexistent. An overview of 54 epidemiologic studies evaluating the role of oral contraceptives and breast cancer found the RR of current users to be 1.24 vs. no risk for women who had not taken them for 10 or more years.[199] Thomas[200] summarized the results of five cohort studies and found that the overall risk for ever-users to be 1.06. On the other hand, in a large, matched case-control study of BRCA mutation carriers use of oral contraceptives was associated with increased risk of breast cancer in BRCA1 (overall risk = 1.20) but not in BRCA2.[201]
Oral contraceptives are used by more than 150 million women worldwide and offer a number of health benefits, including reduction in dysmenorrhea, fibrocystic breast changes, iron deficiency anemia, pelvic pain secondary to endometriosis, ectopic pregnancy, pelvic inflammatory disease, functional ovarian cyst formation, and the incidence of endometrial and ovarian cancers.[189] Women should be counseled that there could be a slightly increased risk of developing breast cancer, although some believe that the apparent increase is due to surveillance bias, because physicians examine women taking oral contraceptive pills more frequently. Supporting this viewpoint is the observation that women with breast cancer who have taken oral contraceptives in the past do not experience lower survival rates when compared with women who have not taken them. Based on these data, current prescribing practices for oral contraceptives should not be changed, although the special case of BCRA mutation carriers must be assessed carefully on an individual basis.[201]
Other risk factors associated with breast cancer, such as proliferative fibrocystic changes in the breast, sedentary lifestyle, and diet, may be used to design preventive strategies for the high-risk woman. Overall, typical or atypical proliferative fibrocystic changes of the breast are associated with a two- to fourfold increased risk for the development of breast cancer.[202] Because the clinical significance varies depending upon the degree of hyperplasia and atypia present, the Cancer Committee of the College of American Pathologists has replaced the term fibrocystic disease with fibrocystic changes.Possible treatment options for relieving symptoms related to fibrocystic changes include decreasing the consumption of foods rich in methylxanthines (e.g., chocolate, coffee, tea, and cola) or taking a supplement such as vitamin E, or medications such as danocrine, bromocriptine, or tamoxifen. Although consumption of these drugs has been shown to decrease fibrocystic changes in the breast, studies have not been done to evaluate whether these changes actually decrease the increased risk of breast cancer.
Diet
Numerous observational studies have reported on the role of diet in breast cancer development with disparate findings, whereas experimental studies are more definitive. [203] [204] [205] Many public health agencies advocate a low-fat, high-fiber diet to prevent breast cancer. Data linking diet to breast cancer come largely from international and migration studies. First-generation Japanese-American women and women who have recently migrated from Japan have risks for the development of breast cancer that approach those of Native American women.[206] Differences are seen among women from various countries as well; for example, women in Great Britain have age-standardized mortality rates of approximately 28 per 100,000 versus those of Japanese women of 6 per 100,000.[207]
The WHI study is the only randomized prospective study performed to address the role of diet in the development of breast cancer, in which women were randomized to low-fat diets or no dietary intervention.[208] In this randomized trial of a low-fat dietary intervention among postmenopausal women aged 50 through 79 years, the low-fat diet resulted in a nonsignificant 8% reduction in breast cancer incidence over an 8-year study period. However, secondary analysis revealed a benefit from the dietary intervention among women with baseline high-fat diets, and women who were strictly adherent to the intervention. Therefore, we consider these results to be promising, if not definitive data related to dietary modification of breast cancer risk. Initial data from the WHI study related to calcium–vitamin D supplementation compared to placebo among postmenopausal women revealed no reduction in breast cancer risk, although the incident breast cancers in the intervention group were smaller in size.[209]
Women of tall stature or who have high body fat and mass have higher rates of breast cancer than other women. [203] [204] [205] An increase in estrone and estradiol as BMI increases has also been documented.[210] The recognition that a hormone (leptin) produced by fat cells vigorously stimulates the growth of normal and malignant breast cells may provide an important biologic link to the phenomenologic observation.[211] Some animal studies, however, have shown that caloric restriction in general, rather than a low-fat diet per se, decreases risk of breast cancer. This effect of caloric restriction could underlie the observation that women who exercise have a lower risk of breast cancer. Another consideration is that women who exercise ovulate less frequently and therefore are not exposed to the higher levels of estrogen that normally occur in women who ovulate regularly.
Numerous epidemiologic studies suggest that alcohol has an effect on the development of invasive breast cancer. An extensive and detailed meta-analysis of the six largest prospective cohort studies addressing this issue showed that alcohol consumption was associated with a linear increase in breast cancer incidence for intakes less than 60 g/day (about two to five drinks).[212] The association was not modified by other factors, and higher alcoholic intakes (>60 g/day) were not associated with additional increased risk. Low dietary levels of selenium and antioxidants such as vitamins C and E, and β-carotene have been associated with breast cancer development and differences in survival.[213] β-carotene is the major provitamin A carotenoid and has differentiating and antiproliferative effects on a variety of cells, including mammary carcinomas. Levels of β-carotene have been analyzed in numerous studies and are lower in women with higher staged breast cancer and breast cancer in general. A case-control study from Europe, however, observed no differences in vitamin A or β-carotene levels between cases and control subjects.[214] Studies of soy intake in Singapore Chinese women are of great interest. [210] [215] Soy intake was significantly associated with lowered plasma estrone and with more favorable mammographic patterns. Recently, there has been a great interest in gene-environment interactions and studies of micronutrients in relationship to metabolic pathways, and genetic polymorphisms are likely to be informative. Among women at high risk for breast cancer for other reasons (e.g., heredity), reducing alcohol consumption should be a straightforward way to reduce breast cancer risk.
Screening and Early Detection
Secondary prevention is aimed at detecting preinvasive lesions such as ductal carcinoma in situ, lobular carcinoma in situ, or early-staged breast cancers that have the potential to be cured with limited treatment. Screening tests include the breast self-examination, the clinical breast examination administered by health care professionals, and mammography. Successful implementation of wide-scale screening programs that incorporate these techniques, followed by treatment of detected lesions, is probably responsible for most of the decline in the overall death rate from breast cancer that occurred among American women from 1989 through 1993.[216] This decline continues and probably represents both the increased use of mammography and the effectiveness of systemic adjuvant therapy. Currently, 5-year survival rates for localized breast cancers have increased to more than 98%.[78]
Although we encourage women to perform monthly breast examinations, a randomized trial indicated that this practice does not decrease overall mortality rates.[217] The most effective combination for decreasing the incidence of invasive disease is the clinical breast examination and mammography. The Breast Cancer Detection Demonstration Project showed that the sensitivity of the clinical breast examination and mammogram together was 70% to 80%, with sensitivity increased for older patients.[218] Although it is standard practice for a clinical breast examination to be performed annually, there has been a great deal of controversy surrounding the appropriate time to begin routine screening with mammograms. Randomized controlled trials of a large number of women on trials from several countries have unequivocally demonstrated a 40% decrease in mortality from breast cancer in women who have annual mammograms beginning at the age of 50.[219] Throughout the years, controversies have erupted regarding the magnitude of benefit from screening mammography. We continue to recommend the practice to women beginning at the age of 50 but encourage discussion of the risks and benefits associated with this procedure.
The opinion on routine screening in women between the ages of 40 and 49, however, is mixed.[220] Eight randomized controlled trials performed between 1963 and 1982 do not demonstrate a statistically significant difference in breast cancer mortality within 7 years after screening was initiated in women randomized to receive or not receive screening mammograms. A majority in a recent consensus panel used this information to state that there currently was not sufficient evidence to advocate routine screening mammography in women ages 40 to 49. Five of these trials, however, demonstrated a 16% decrease in mortality if follow-up continued for 10 years, prompting release of a minority report advocating routine screening in the 40 to 49 age group.[220] As the minority report highlighted, the goal of mammography is to detect preinvasive lesions, and 15% to 20% of breast cancers are now diagnosed as ductal carcinoma in situ or lobular carcinoma in situ in younger women. The minority report correctly pointed out that the risk from radiation during mammography screening was overemphasized. The U.S. Preventive Task Force has considered the issues carefully and recommends a screening mammography every 1 or 2 years for women aged 40 and older,[221] a position we favor. An important recent observation is that mammographic density is strongly associated with risk and is heritable.[222]This information should further target the premenopausal woman at higher risk for breast cancer; determination of the underlying genetic basis for breast density has now become of great interest, particularly in that the use of a postmenopausal hormone therapy was strongly associated with an increase in mammographic density in the Postmenopausal Estrogen/Progestin Interventions and WHI trials.[195] [196]
New areas that are currently being evaluated for the enhancement of primary and secondary prevention strategies include digital mammography, DCE-MRI, DCE-CT with ultrasound, PET, optical scanning, ductal lavage for cytologies and molecular testing, nipple aspirates, and blood and urine assays for growth factors and autoantibodies to oncoproteins and to tumor DNA. Validated biologic markers of breast cancer risk and/or more sophisticated screening modalities might well increase our ability to detect lesions earlier in high-risk populations.
Chemoprevention
Successful therapeutic prevention for breast cancer has progressed faster than for any other malignancy.[223] Two compounds, tamoxifen and raloxifene, which are estrogen receptor antagonists, have been shown to reduce the incidence of primary breast cancer in women at high risk. [224] [225] In a head-to-head randomized trial of tamoxifen and raloxifene, efficacy was comparable with a 50% reduction of breast cancers; the toxicity of raloxifene was considerably less than that of tamoxifen.[226] Based on the positive activity of anastrozole (an inhibitor of estrogen synthesis) in the adjuvant setting and its minimal toxicity,[227] a randomized trial of this compound versus raloxifene in women at high risk for breast cancer is now underway. Preclinical studies at our institution using a transgenic mouse model suggest that antiprogestational agents may be of particular benefit in BCRA1-positive breast cancers.[228] The pros and cons of chemoprevention for breast cancer were recently reviewed, and the overall conclusion was that it cannot yet be recommended for general usage but should be useful for reduction of risk among high-risk individuals.[229]
There are several other options for a woman who is at very high risk for breast cancer, including bilateral mastectomy or oophorectomy and lifestyle modification. Prophylactic bilateral mastectomies have been performed on some mutation carriers, but cases of breast cancer developing in the remaining breast tissue after subcutaneous and total mastectomies have been reported. In addition, such surgeries are dramatic procedures for a woman who has only a “probability” of developing breast cancer, and a decision analysis paradigm for these interventions is available.[230] No long-term data exist on the effect of these surgeries in increasing the life expectancy of mutation carriers, yet one meta-analysis reveals considerable global variation in the utility of prophylactic surgery among unaffected BRCA1 and BRCA2mutation carriers.[231] Bilateral oophorectomy has been proposed as another option for premenopausal women who have completed their childbearing. Although castration has been shown to decrease the risk of breast cancer in young, nulliparous women, especially when it is performed before the age of 35, this remains a very controversial area in the management of breast disease. Secondary prevention can be accomplished by instructing these high-risk women about the importance of clinical breast examinations by a physician and screening mammograms, which should begin at a younger age, preferably at least 5 years earlier than the age at which the relative developed breast cancer. Observational studies and initial clinical trials suggest that moderate physical exercise and control of obesity may decrease the risk for breast cancer, [232] [233] [234] as has been suggested for colorectal cancer. Recently, several randomized interventional trials have been initiated to definitively address these issues.
PROSTATE CANCER
Etiology
The age-adjusted incidence of prostate cancer rose slowly from 1965 to 1985 for unclear reasons. Parallel with aging of the baby-boomer population, the prevalence has also markedly increased since the general recommendation in the late 1980s by the American Cancer Society and the American Urological Association of yearly screening for PSA after age 50. This recommendation led to widespread screening and a rapid increase in the incidence of prostate cancer that peaked in the early 1990s.[235] In the most recent national estimates using SEER data from 1999 through 2003, the age-adjusted incidence of prostate cancer was 165 cases per 100,000 individuals—the highest rate of any cancer type among men.[12] Major differences in the incidence of prostate cancer are observed across the major U.S. ethnic groups, with blacks having the highest incidence rates (243 per 100,000) and Asian-Pacific Islanders having the lowest incidence rates (104 per 100,000). Overall, 234,460 cases of prostate cancer are expected in the United States in 2006, along with 27,350 deaths—making prostate cancer the third deadliest cancer in men after cancer of the lung and colorectal cancer.[78]
Several factors—age, familial/genetic, environmental, and hormonal—seem to contribute to the development of prostate cancer.[236] Prostate cancer shows a familial tendency that is currently not well defined, but at least one study suggests that 10% to 15% of cases could have a strong genetic component.[237] The loss of heterozygosity in certain chromosomes in prostate cancer suggests that a gene related to some prostate cancers will be found.[238] The existence of a locus in chromosome 1 (band q 24) that predisposes men to develop early-onset prostate cancer has been verified, but a gene has not yet been isolated.[239]
The androgen dependence of prostate cancer led to the interesting hypothesis that variations in transcriptional activity by the androgen receptor regulated by CAG repeats could determine risk. Subsequent findings, however, argue against such an association, although there may be specific underlying situations in which other genotype influences lead to such an effect.[240] Because these studies concentrate on identifying prostate cancer risk and most prostate cancers are not clinically significant, others have argued for identifying genotypes that are associated with clinically aggressive cancers that predict outcome (mortality). In this regard, a report documenting extensive mitochondrial mutations in primary prostate cancers might provide new insights into progression.[241]
Both epidemiologic and experimental data suggest that hormones, particularly testosterone, play a definitive role in the development of prostate cancer. In the rat model, testosterone induces prostate cancer, and in humans prostate cancer rarely occurs in castrated men. [242] [243] Also, black men have a higher incidence of prostate cancer at all ages than white men, and Japanese men have the lowest incidence.[244] [245] Whether this racial-ethnic variation in prostate cancer risk has a hormonal basis is still unclear, but a substantive amount of data supports this viewpoint.[245]
A high-fat diet and obesity may be associated with an increased risk of prostate cancer, but the studies to date have yielded inconsistent results.[246] One investigation suggests that the preadult hormonal milieu, as reflected in attained height and childhood obesity, could have a strong influence on prostate carcinogenesis.[247] Epidemiologic, animal model, and in vitro studies indicate that n-3 polyunsaturated fatty acids, lycopene, and selenium might also be important in the pathogenesis of prostate cancer.[248] Additionally, GSTP1 has been proposed as a caretaker gene that serves to detoxify carcinogens associated with various lifestyle habits.[249]
Screening and Early Detection
The relative benefits and costs of screening for prostate cancer are currently among the most contentious issues in the medical community. [250] [251] [252] There are several major reasons why this controversy continues:
|
• |
All available first-line techniques (DRE and serum PSA) have high rates of false-positive results. This leads to a relatively low positive predictive value and the unnecessary workup of many normal individuals. |
|
|
• |
The natural history of prostatic intraepithelial neoplasia (PIN), the probable precursor of prostate cancer, is highly variable, and the natural history of the disease cannot currently be predicted reliably in any one particular case or by any specific biologic or pathologic marker. |
|
|
• |
The workup of abnormal screening tests is invasive, requiring several biopsies of the prostate. |
|
|
• |
The treatment of prostate cancer produces significant morbidity and measurable mortality. |
|
|
• |
The rate of false-negative results is also high, which can produce a false level of assurance about the reliability of the screening tests ( Fig. 26-6 ). |
|
Figure 26-6 Screening and early detection for prostate cancer. |
These same five concerns regarding the use of PSA and DRE for screening also exist for their use for early detection purposes, but the consequences are mitigated somewhat, because individuals are by definition symptomatic on presentation.
The two most commonly used screening tests for prostate cancer are PSA and DRE, with transrectal ultrasound (TRUSP) reserved for patients with a positive PSA and/or DRE. Before the 1990s, yearly DRE after age 50 was the standard test used both for detection of prostate cancer and for screening. Although many primary care physicians use the DRE as part of a routine physical examination, assessment of its routine use indicates that DRE is performed in less than 50% of primary care encounters in which one would expect it to be done.[253]
A summary of the data indicates that the positive predictive value of DRE is relatively low (11% to 26%), whereas the negative predictive value (85% to 96%) is relatively high.[254] The most complete assessment thus far evaluated 811 unselected serial patients from 50 to 80 years of age who underwent DREs; 43 patients had a palpable nodule, and the positive predictive value of the 38 patients who underwent biopsy was 25%.[254] It is of great interest that 68% of the detected tumors were clinically localized, but only 30% were pathologically localized after radical prostatectomy. These data and other studies suggest that only about 20% to 25% of cases are localized at the time of a positive DRE; on the other hand, more than 25% of cases of prostate cancer are metastatic by the time a detectable palpable lump is detected on DRE. [254] [255] Although the effectiveness of DRE is probably also significantly influenced by the skill of the examiner, the proper technique can easily be taught to health workers, is inexpensive, and is relatively noninvasive. Its usage as a primary screening tool, however, has not been widely adopted, probably because of its inconvenience. Whether routine screening by DRE alone can reduce mortality from prostate cancer is unknown. With the emergence of serum PSA as the screening test of choice, it is unlikely that the specific value of DRE in reducing prostate cancer morbidity and mortality per se will ever be demonstrated conclusively.
Annual measurement of serum PSA as a screening test for prostate cancer has been adopted widely following the initial 1993 recommendation of the American Cancer Society. At the current time, the professional community remains split over the question whether serum PSA should be recommended for routine screening in men older than age 50. The issues have been presented and analyzed extensively, and the same arguments that are used to discourage routine screening are used by others to recommend its widespread use. Estimates of overdiagnosis of clinically insignificant lesions have ranged from 15% to 84% in well-done investigations. [251] [252]
Several studies have demonstrated that PSA screening results in a stage downshift and increases the detection rate of early-stage cancers. [256] [257] The false-positive rate (25% to 50%) is high, however, resulting in a positive predictive value of PSA in screening studies of about 30%. Because most of the studies have been done on symptomatic men, the positive predictive values in a true screening effort are likely to be lower (i.e., due to an expected lower prevalence of disease). In practical terms, this observation means that less than one-third of men with an elevated PSA will have biopsy-proven prostate cancer, and two-thirds will have a biopsy result that is negative for prostate cancer. Even if biologically aggressive tumors (and this is unlikely in most cases) were being identified (see the discussion later in this chapter), a large number of men would undergo unnecessary prostate biopsies with the attendant fiscal cost and morbidity.
Because the value of a “normal” level of PSA (<4 ng/mL) is influenced by several physiologic parameters, there has been a great interest in enhancing the specificity of the test. Techniques to improve the positive predictive value of PSA have included using age-adjusted reference ranges, PSA density (PSA level/size of prostate as measured by ultrasound), PSA velocity (change in PSA per unit of time), and ratio of bound to free PSA. [258] [259] Although the validity of these strategies is currently unconfirmed, modeling has been important in determining the most effective use of tumor markers in other diseases, and it is likely that with enough time and information, the rate of false-positive results with PSA screening can be reduced. One way to reduce the false-positive rate is to combine PSA screening with DRE, and when appropriate, with TRUSP. A positive DRE increases the likelihood that a positive (i.e., abnormal) determination PSA is a true positive and therefore enhances the positive predictive value of the test. Additional evaluations with TRUSP should further increase the positive predictive value. The prostate component of the large, multicenter, randomized, controlled PLCO Cancer Screening Trial is expected to contribute significantly to our understanding of the value of PSA and DRE screening on mortality among men aged 55 to 74 years. In this study, the intervention arm receives screening examinations at scheduled intervals, but clinical management of these results are left to the individual study clinicians. However, such mortality data are not expected until the year 2015.
Initial reports addressing the rate of diagnostic procedures performed from the PLCO have emerged. At baseline through the year 2000, 4801 men with positive PSA (i.e., >4 ng/mL) or DRE were identified.[260] Among the 2717 men with elevated PSA, 64% received biopsy within 3 years, and among men with positive DRE and negative PSA only 27% underwent biopsy within 3 years. As might be expected, higher biopsy rates were noted for those with PSA above 7 ng/mL as compared with those whose PSA level was 4 to 7 ng/mL. Follow-up data regarding mortality, and also the positive predictive value for PSA, DRE, or both among different demographic strata are not yet available but are expected to assist greatly in our understanding of the utility of these commonly used screening tests.
The second major obstacle to the successful use of any screening modality for prostate cancer is that the biologic aggressiveness of PIN and early prostate cancer is not identified with high reliability by serum PSA. Progressive prostate cancer is a serious disease with high morbidity and mortality; however, not all prostate cancers are serious, and indolent behavior is more common than not. For example, about 30% of men over age 50 have histologic evidence of prostate cancer at routine autopsy, suggesting a prevalence of prostate cancer of about 9 million.[261] About 1.2 to 1.5 million of these 9 million men, or about 15%, will eventually die of their disease. Thus, most prostate cancers in the population are latent and do not progress to clinical adversity; therefore, an aggressive workup of an elevated PSA should not be a reflex action.
A third major consideration in evaluating PSA as a useful screening test is that the subsequent workup and treatment have a significant complication rate. Follow-up testing of an elevated PSA requires a repeat PSA, DRE, TRUSP, and biopsy. These are relatively safe procedures, but about 0.1% to 0.4% of the 20% of screened men who undergo biopsy experience infection or bleeding, and almost all experience considerable anxiety while waiting for the results. [262] [263] The potential complications of treatment can be quite serious and include impotence, incontinence, and death from radical prostatectomy. Adverse outcomes of radical prostatectomy have been compared with watchful waiting in a randomized clinical trial. In this study, erectile dysfunction (80% vs. 45%) and urinary leakage (49% vs. 21%) were more common among patients who underwent radical prostatectomy; however, symptoms of urinary obstruction were improved (28% vs. 44%).[264] Radiation therapy is no less benign, but it carries a lower incidence of incontinence, a higher incidence of acute gastrointestinal complications, and a similar incidence of impotence.
A fourth major issue in PSA screening is the high rate of false-negative results. Numerous studies have demonstrated that many individuals (25%) with a normal PSA level have disease beyond the prostate.[265] Such results can lead to false reassurances and decreased follow-up when other factors suggest that a more aggressive workup might be reasonable. Several studies suggest that a rising PSA, even in the normal range, is a cause for concern and reason enough to biopsy. [266] [267]
Although serum measurement of PSA has been widely adopted in men over age 50 as a primary screening tool for prostate cancer, its value in improving the overall health of men has not been shown to date. The equally important issue of whether screening does more harm than good also remains unanswered, because the natural history of prostate cancer is so variable.[265] Decision analysis has been used to determine the benefits and risks of age- and quality-adjusted survival, but the results remain inconclusive. [268] [269] Other studies suggest that screening might have the potential to decrease survival, particularly in the older individual.[269] There is, of course, no lack of critics of this viewpoint.[270] What information is needed to resolve this difficult and important issue? Perhaps only a series of randomized trials can lay this question to rest, and to this aim, the prostate component of the PLCO Cancer Screening Trial may provide answers to these important questions. The results of a randomized trial involving more than 40,000 men in the city of Quebec have been reported, and those subjects with a regular PSA screening had a 60% decrease in mortality from prostate cancer after 7 years.[271] The conduct and interpretation of this type of trial is complex, and the results of several other large screening studies will have to be available before definitive recommendations about the value of routine screening PSA can be made.[272] These targeted trials, however, will provide only a general guide regarding population-based screening using serum PSA in men over age 50 as an approach to identify potential prostate cancers. The current consensus by the U.S. Preventive Services Task Force is that the evidence is insufficient to recommend for or against routine screening for prostate cancer using PSA.[272]
An equally important issue is this: How do we identify and distinguish a biologically aggressive tumor in any one individual from those that will remain latent for the life of the individual? This is a very difficult problem to study. Although the earliest features of prostate cancer pathogenesis remain obscure, recent studies of the biologic features of intraepithelial hyperplasia of the prostate and of the “normal” prostate in individuals with a strong family history could shed some light on this issue. [272] [273] [274] Recent studies of the cytogenetic and molecular alterations in high-grade PIN have indicated that loss of heterozygosity is prominent and that certain oncogenes are expressed.[274] Defining the biologic features of the preclinical phase of prostate cancer is critical to answer for innumerable reasons, not the least of which is to increase the effectiveness of PSA screening. In this regard, a large, randomized trial of men with T1b, T1c, or T2 prostate cancer demonstrated that radical prostatectomy was superior to “watchful waiting” in terms of disease-free survival but not in terms of overall survival.[275]
Chemoprevention
Although the development of rat prostate tumors has been studied for some time, this model system has been regarded as a poor one for carcinogenesis of the human prostate. The development of transgenic models that simulate the human disease represents an improvement in this regard.[276] Epithelial changes, including PIN, were identified in the human prostate long ago, although only recently have the biologic (and clinical) implications of these changes been recognized. The importance of these alterations, the recognition of the analogous evolution of the process to other epithelial cancers (e.g., cervical, oral), and its association with a wide spectrum of biologic abnormalities has moved PIN into the forefront as the probable, but clinically uncommon, preneoplastic precursor of prostate cancer.[276] An impressive array of studies measuring various biologic and molecular parameters in PIN have been done, and various biologic changes associated with the progression of prostate cancer have been identified.[277] What should be done now is to relate these biologic findings to the clinical aggressiveness of PIN and/or the eventual outcome of clinically relevant (nonindolent) prostate cancer. To be able to do so will help guide the difficult decisions after detection of an elevated serum PSA in biopsy samples during the screening and/or identification of PIN and during the early detection process. Three major categories of chemoprevention agents are being currently considered: inhibitors of proliferation, hormonal modulators, and stimulators of differentiation.
Thus far, two definitive randomized trials have been launched. The results of a phase III study using the 5a-reductase inhibitor finasteride (thereby lowering levels of dihydrotestosterone, the active metabolite of testosterone) in a high-risk population has been reported. The incidence of prostate cancer was decreased about 25%, and the side effects were minimal.[278] However, conclusions about the overall benefit were compromised by the finding that the risk for higher grade tumors was increased in the treatment arm. Although a large number of explanations have been offered for these contradictory findings, the future of finasteride as a therapeutic agent for prostate prevention is uncertain.
A second trial uses selenium, vitamin E, or both. The rationale for the study was based on secondary analyses of several large intervention trials in which prostate cancer was not the target, although recently some supportive experimental data also have become available. [279] [280] Accrual to this 2 × 2 factorial randomized trial (vitamin E, selenium) is ongoing, and results are anticipated in 2013. A large number of compounds are being investigated at the preclinical level, and a few have advanced to the phase I/II clinical level. Studies of fenretinide failed to show an effect on relevant surrogate markers, whereas DFMO was more successful. [273] [281]
Two other relatively unexplored areas of chemoprevention research in prostate cancer should also be mentioned: PIN and familial risk. Just as understanding the biology of PIN will affect our screening and early detection decisions, PIN should also serve as a useful marker in chemoprevention studies. Although the heterogeneity of lesions will make interpretation of effect of an intervention a challenge, PIN represents an important parameter for advancing our knowledge of early prostate cancer carcinogenesis and its modulation by candidate chemoprevention agents. Several studies are in progress to use PIN as a screening tool for new chemoprevention agents including studies of biologic markers to determine risk.[282] The roles of family studies and genetics in identifying individuals at high risk for prostate cancer are in their infancy, but epidemiologic studies support the notion that genetic risk plays a role, and clinical studies support the observation that early prostate cancer in some individuals is highly aggressive, whereas in others it is indolent. Linking these two parameters should identify a population of individuals in whom screening, early detection, and chemoprevention agents should be intensively directed.
Advances in the systemic therapy of advanced prostate cancer have been slow in coming. In a real sense, advances in the management of prostate cancer have been minimal since the introduction of hormonal therapy more than 50 years ago. It is likely that the widespread use of screening and early detection with an appropriate follow-up will reduce the morbidity and mortality from prostate cancer in a substantial way and that effective chemoprevention will be developed, because the major biologic enhancer (androgens) of prostate cancer carcinogenesis is known.
SKIN CANCERS
Each year, more than $2 billion is spent to treat patients diagnosed with skin cancers, the majority of which are malignant melanoma and basal and squamous cell carcinomas. These figures underestimate the true cost, because many of these cancers are treated in physicians’ offices, and nonmelanoma skin cancers are not routinely reported to tumor registries. An aging population, depletion of the stratospheric ozone layer, and increased recreational exposure to ultraviolet radiation (UVR) represent some of the factors that contribute to the development of over 1 million cases of nonmelanoma (basal and squamous cell carcinomas), 49,000 cases of melanoma in situ, and 62,000 cases of invasive melanoma diagnosed annually in the United States.[78] Understanding how these and other risk factors lead to alterations in key cellular processes like DNA synthesis and repair, oncogene activation, cell-cycle control, and apoptosis is the focus of intense research efforts aimed at designing novel preventive, diagnostic, and treatment strategies.[59] This section details the various primary, secondary, and tertiary preventive approaches for melanoma and nonmelanoma skin cancers. Incorporating these strategies into medical practice should decrease the incidence and mortality from skin cancer and should also decrease health care costs.
Etiology and Primary Prevention
Environmental
Successful primary prevention depends on the ability to identify individuals at risk for skin cancer and to use this information to educate both high-risk groups and the general population about various ways to reduce risk ( Table 26-5 ). Many factors have been identified that increase an individual's risk for the development of melanoma and nonmelanoma skin cancers. Exposure to UVR is a major risk factor.[283] Not only is cumulative UVR exposure important in the development of skin cancers, but it is apparent that acute, intermittent exposure to UVR is carcinogenic.
Table 26-5 -- Predisposition and Risk Factors for Skin Cancer
|
The electromagnetic spectrum is composed of infrared, visible, and UV light; the latter is responsible for causing the cellular and architectural changes in the epidermis and dermis that lead to photoaging and skin cancer.[283] Although the UVR spectrum is broad, UVR-β (290–320 nm) and UVR-A (320–400 nm) are the only wavelengths that routinely reach the earth's surface, in that shorter wavelengths (UVR-C) are absorbed by the ozone layer. UVR-β is more potent than UVR-A in inducing neoplastic transformation in epidermal keratinocytes and melanocytes, which give rise to basal and squamous cell cancers, and melanoma, respectively. UVR-A, however, has been found to penetrate the skin more deeply and is the predominant wavelength emitted from artificial lamps found in tanning salons.[283] More than a million adolescent and young women frequent these facilities daily and expose themselves to up to five times the amount of UVR that is emitted from the sun at any given time. The role of UVR-A radiation in the development of skin cancer will increase as this industry continues to grow.
The mechanism of action of UVR on the skin has been studied extensively. Once photons penetrate through the stratum corneum, they are absorbed by cellular DNA and produce base substitutions in pyrimidines.[284] The substitution of thymidine for cytosine is pathognomonic for UVB-induced skin damage and is found in the tumor suppressor gene p53 in more than 90% of squamous cell skin cancers. [285] [286] Basal cell cancers also contain p53 mutations.[287] Although UVR is regarded as contributing to the pathogenesis of melanoma, these types of mutations are uncommon, therefore raising the likelihood that the role of UVR is associative or complementary to the process. Normally, p53 acts to protect damaged cells by either inducing cell-cycle arrest (so that mutated DNA can be repaired or excised) or by inducing apoptosis.[288] UVR-induced p53 mutations disturb the cell cycle by inhibiting cyclin-dependent kinases, leading to uncontrolled cell proliferation. Cells with one mutated p53 allele can undergo clonal expansion, and, if the other p53 allele is mutated, neoplastic transformation occurs. Therefore, UVR could both initiate and promote carcinogenesis.[288] UV light might also have immunosuppressive effects by interfering with the ability of Langerhans cells to process antigens.[289]
Other risk factors that increase susceptibility for the development of skin cancers include skin complexion and response to sunlight, degree of freckling, ethnicity, gender, age, geographic location, presence of premalignant skin lesions, medical history of exposure to ionizing radiation or psoralen and UV therapy, chronic skin irritation (ulcers, inflammation, or trauma), or a personal history of a germatodermatoses (xeroderma pigmentosum, nevoid basal cell carcinoma syndrome, and familial dysplastic nevus syndrome), lymphoreticular malignancy, granulomatous diseases, or other immunosuppressed states such as organ transplantation in which development of multiple cutaneous squamous cell carcinomas is a major problem.
Health care providers should be aware of several premalignant dermatoses for the purpose of identifying individuals who are at increased risk for the development of skin cancers. The most common lesion, actinic keratoses (solar keratoses), has been reported to undergo malignant transformation to squamous cell cancer in 12% of patients.[290] Histologic evaluation of white patches occurring on mucous membranes, known as leukoplakia, is also important, because up to 20% could be dysplastic, with 3% to 6% becoming invasive cancers. Atypical and dysplastic nevi, large congenital nevi (>9 cm), and an increased number of moles are common precursor lesions of melanoma.[291] Chronic skin irritation from radiation (radiation dermatitis), chemicals (tar and arsenical keratoses), infrared light (thermal keratoses), and scars (scar keratoses) might also lead to malignant transformation. Any patient who is immunosuppressed (human immunodeficiency virus [HIV] diagnosis or transplant recipient) or who has a history of epidermodysplasia veruciformis or Bowen's disease (an intradermal carcinoma that often occurs on sun-exposed areas) should be considered for prevention protocols. Anyone who has a prior history of skin cancer is also at risk for a second primary cutaneous malignancy. [292] [293] Patients diagnosed with thin melanomas (<75 mm in thickness; Breslow staging) were found to have a 4% chance of developing a second primary melanoma.[294] First-degree relatives of skin cancer patients also carry an increased risk.[292]
Recently, a risk model for melanoma risk has been proposed using data from a case-control study, based on similar methodologies developed by Gail for the breast cancer risk model developed nearly 2 decades ago. [295] [296] This melanoma risk model incorporates subject responses to simple questions about complexion, history of blistering sunburn, and ability to tan, with findings on examination of the back describing the number of small and large nevi. This risk model captures attributable risk for melanoma development of 86% for men and 89% for women. Though promising, further risk models are still needed for individuals with a history of melanoma or nonmelanoma skin cancer, or persons with a first-degree relative with melanoma. Such models could identify high-risk individuals who could be referred for chemoprevention trials.[297]
Heredity
Several molecular abnormalities have been identified that could be responsible for the genetic instability preceding the development of invasive nonmelanoma skin cancers ( Table 26-6 ). Defects in DNA repair genes, oncogenes, and tumor suppressor genes, as well as allelic losses in a number of chromosomes, have been described.[287] Mutations in the Ras oncogene have been shown to initiate epidermal skin cancers.[298] Mutations in DNA repair genes (xeroderma pigmentosum, for example) bring about an inability to repair UVR damage to DNA efficiently, leading to the development of melanoma and epidermal skin cancers.[299] DNA repair capacity might be particularly important for individuals with other strong risk factors, such as low tanning ability and the presence of dysplastic nevi.[300] There are numerous examples of specific chromosomal abnormalities in nonmelanoma skin cancers. Allelic losses have been found in 9p, 13q, 17p, 17q, and 3p in squamous cell cancers and in 9q in basal cell cancers.[298] Studies of keratinocyte transformation have also focused on the UVR-mediated phosphatidylinositol 3-kinase and p38 mitogen-activated protein kinase pathways.[301]
Table 26-6 -- Molecular Determinants of Carcinogenesis in Skin Cancer
|
The genes and genetics of melanoma have been summarized.[302] One autosomal dominant syndrome that markedly increases an individual's lifetime risk for melanoma is familial dysplastic nevus syndrome. Melanoma in one or more first-degree relatives and the presence of a large number of moles (between 10 and 100) are required to diagnose this syndrome.[303] Rearrangements or deletions of genes have been found to occur in chromosomes 9 and 10 in patients with familial melanoma, atypical nevi, or early melanoma lesions. Two cyclin-dependent, kinase, tumor suppressor genes have been isolated on chromosome 9 (p15, p16). Mutations in this region lead to uncontrolled cell proliferation, because the transition from G1 to S of the cell cycle is no longer inhibited.[304] The finding that the penetrance of one of these genes to frank melanoma is dependent on geographic variation emphasizes the role of environment in genetic expression[305] and the potential interaction of UVR and sunburn genotype.[306] Certain polymorphisms in the melanocortin-1 receptor gene have been shown to correlate with a reduced response to melanotrophin, the major natural hormone that regulates cutaneous pigmentation.[307] Other studies suggest that loss-of-function mutations in the MCIR gene sensitize human melanocytes to the DNA-damaging effects of UVR, which could increase melanoma cancer risk.[308]
The identification of frequent RafB mutations in both melanomas and benign moles is also of great interest and suggests that this alteration may be the initial molecular change leading to immortalization.[309] Once the significance of the genetic abnormalities in causing nonmelanoma and melanoma skin cancers is understood, we might be able to incorporate this information into new risk models to identify high-risk individuals who would benefit from prevention protocols and increased surveillance.
Preventive Measures
Numerous examples of primary preventive strategies exist. Protective clothing (hats, long sleeves, special fabrics approved by the U.S. Food and Drug Administration), behavioral modification (avoidance of peak sun from 10:00 am to 3:00 pm, avoidance of sun tanning salons, and use of appropriate shading), and liberal application of sunscreens are three such examples.[310] There are two types of sunscreens: blockers and reflectors. Blockers such as paraminobenzoic acid absorb UVR-β only. Nonparaminobenzoic acid blocker sunscreens absorb both UVR-A and UVR-B. Reflectors such as zinc oxide completely reflect UVR light. The SPF, or sun protective factor, is a measure of the comparison of the minimal erythema dose with and without sunscreen and should be at least 15. Finally, sunscreens should not wash off easily when bathing or sweating.
The current role of sunscreens in preventing skin cancer is the subject of considerable controversy.[311] Randomized trials have demonstrated that sunscreen use encourages prolonged sun exposure.[312]Another randomized study, however, has shown a protective effect of sunscreen use against the development of squamous cell cancer and nevi.[313] Based on these findings, we probably can conclude that regular sunscreen application should be combined with protective clothing to reduce the long-term risk for the development of melanoma and nonmelanoma skin cancers.[283]
In Australia, which has the highest incidence of skin cancers, other approaches have been adopted, including the passage of laws for employers to provide sun protection for employees, distribution of free sunscreens, tree planting campaigns, and public shade structures. In the United States, the Federal Trade Commission requires that protective eyewear be worn in tanning salons. In addition, signed informed consents and signs about health risks of UVR exposure in tanning salons are required. Furthermore, in Texas an adult must accompany children attending tanning salons, and parental permission is mandatory for all minors. Attitude and behavioral modification of children (and their parents) informed by education and knowledge about sun exposure and skin damage/cancer is a reasonable goal and is most effective when started at a young age ( Box 26-2 ). [314] [315]
Box 26-2
|
USEFUL THINGS TO TELL YOUR PATIENTS ABOUT THE PREVENTION OF SKIN CANCER |
|||||||||||||||||||||
|
|||||||||||||||||||||
|
* A wet T-shirt has an SPF of 0; several companies now make clothes that are specifically treated to give a high SPF. |
Screening and Early Detection
The following are essential for effective screening:
|
• |
An understanding of the four characteristics of skin lesions that are suggestive of premalignant or malignant changes (i.e., the ABCDs): asymmetry, border irregularity, color variegation, and diameter (6 mm or greater) |
|
|
• |
Knowledge of important prognostic factors that should be recorded: anatomic location, ulceration, number of atypical lesions, presence of lymph nodes |
|
|
• |
Familiarity with the types and indications for the particular type of biopsy: shave biopsy, incisional biopsy, excisional biopsy, punch biopsy |
|
|
• |
Access to skilled pathologists who can comment on key histologic criteria such as thickness (Breslow staging), margins, ulceration, regression, satellitosis, angiolymphatic invasion, mitotic activity, precursor lesions, host response, and growth phase (radial vs. horizontal) |
Screening for skin cancer, and in particular for melanoma, is supported by several criteria. Skin cancer is the most common cancer worldwide and is an important public health problem. Melanoma is second only to leukemia in terms of years of potential life lost, because it often affects younger people during the most productive periods of their lives.[315] Although basal and squamous cell cancers have a much better prognosis than melanoma, they cause considerable local disfigurement if not diagnosed and treated early. In addition, screening skin examinations are acceptable to both patients and health care providers. Premalignant cutaneous lesions tend to have a long latent phase, making early diagnosis and treatment possible with evidence that supports subsequent decreases in both incidence and mortality rates.[290]
Although no randomized, prospective studies evaluating the efficacy of screening for skin cancer have been conducted, nonrandomized studies support its practice. Thinner melanoma lesions (stages I and II) are diagnosed more frequently with routine screening and intensive education programs. The Sydney, Australia Melanoma Project demonstrated that widespread screening and education led to a decrease in the thickness of lesions from 2.5 mm to 0.8 mm, a decrease in the number of ulcerated lesions, an increase in the number of melanomas diagnosed in the radial rather than the vertical growth phase, and an increase in the 5-year survival rate to 94%.[316] A decrease in lesion thickness and an increase in the number of melanomas diagnosed were also confirmed in studies in Scotland and the United States.[316] [317] Interestingly, a large skin cancer education and screening demonstration project found that 80% of study participants did not have a regular dermatologist, 50% saw their physician only because of the free skin examination, and 80% were receiving their first skin examination. Another interesting population-based, case-control study investigated whether skin self-examination would reduce the incidence of melanoma.[318] Although only 15% of participants in the study cohort practiced skin self-examination, they found that skin self-examination was associated with a reduced risk of melanoma in general and a reduced incidence of more advanced disease in melanoma patients. The authors concluded that mortality from melanoma might be reduced by as much as 63% if regular skin self-examinations were performed.
New imaging technologies may play increasingly important roles in screening and early detection for skin cancers in the future. To this aim, diffuse optical spectroscopy represents an exciting new technology that is emerging. Recently, diffuse optical spectroscopy technology has been used via a hand-held probe with direct skin contact to define the optical properties of skin in vivo, including quantification of melanin, hemoglobin, and water concentrations.[319] Thus, the potential for noninvasive identification of malignant versus benign skin lesions is an anticipated application for this technology. However, the utility in accurately identifying malignant cutaneous lesions has not been described analytically, and the appropriate validation studies on melanoma and nonmelanoma skin cancers must be performed before widespread usage of this novel screening device can be recommended.
Challenges to effective screening examinations include the following:
|
• |
Lack of standardization of the examination and training that health care providers receive |
|
|
• |
Lack of accurate reporting to tumor registries (for melanoma; nonmelanoma skin cancers are not reported at all) |
|
|
• |
Inability to motivate certain high-risk groups (e.g., white males) to come in for examination |
|
|
• |
An inability to screen adequately for the 1% to 2% of amelanotic cases of melanoma |
One step toward overcoming these challenges has been to designate the months of May and June as free skin cancer screening months in the United States in an attempt to attract more people for evaluation.
Chemoprevention
Nonmelanoma Skin Cancers
Clinical trials investigating the use of β-carotene, 13-cis-retinoic acid, selenium, and NSAIDs have been performed on individuals with a history of nonmelanoma skin cancers. Results using vitamin A derivatives have been mixed, which most likely reflects the complex biochemical and molecular mechanisms underlying the prodifferentiation or promaturation changes that these compounds produce. Three randomized clinical trials did not find a decrease in the recurrence rates of basal or squamous cell cancers in individuals at high risk for new skin cancers. Levine and coworkers[320] found no beneficial effect when using isotretinoin or retinol in high-risk subjects with at least four prior basal cell or squamous cell cancers. The Isotretinoin–Basal Cell Carcinoma Study Group also demonstrated that isotretinoin did not prevent the recurrence of basal cell cancer in patients previously treated for basal cell cancer.[321] Finally β-carotene was not shown to prevent nonmelanoma cancers.[322] In contrast, in 2297 moderate-risk subjects with a history of actinic keratoses and at most two squamous or basal cell carcinomas, supplementation with a moderate dose of retinol (25,000 IU daily) reduced the incidence of squamous cell (but not basal cell) cancers by 25%.[23] In addition, patients with xeroderma pigmentosum develop fewer new skin cancers after receiving high-dose isotretinoin.[323] However, interest in the retinoids as preventive agents has waned, and it is unlikely that their usage will be adopted unless a new generation of nontoxic compounds can be developed.
Negative results have been produced in a large multicenter study involving selenium supplementation.[166] In this double-blind, randomized, placebo-controlled study, 200 mg of selenium daily did nor seem to prevent the development of future nonmelanoma skin cancers in patients with a history of basal or squamous cell skin cancer. A secondary analysis of the effect of NSAIDs in this trial on nonmelanoma skin cancer also suggested no effect for this class of agents.[324] Finally, NSAIDs have been found to prevent the erythema that occurs 6 to 12 hours after acute sun exposure. In a large, randomized trial, topical application of the COX-1 inhibitor diclofenac was superior to placebo in clearing actinic keratosis lesions, and this formulation has now been approved for clinical use, although recent concern about the cardiovascular toxicity of selective COX inhibitors leave its usage uncertain.[325] A small placebo-controlled study of 2-DFMO suggests that this compound might have comparable activity.[326] A comprehensive strategy for the development of chemoprevention drugs for nonmelanoma skin cancers has recently been described.[59]
Melanoma and Dysplastic Nevi
Topical β-all-trans-retinoic acid has been demonstrated to cause regression of dysplastic nevi [327] [328]; however, the inconvenience of topical application has limited its usage.
The development of chemoprevention agents for melanoma has progressed slowly, although the development of a risk model for melanoma is an important first step, as is the development of an in vitro model for assessing the potential value of chemoprevention agents. [297] [324] A great deal of new preclinical work has advanced our understanding of melanoma pathogenesis as well,[329] including a unique redox-based paradigm.[330] The recent development of a transgenic model that simulates the human disease is also an important advance and should lead to a more systematic development of chemoprevention drugs.[331]
Tertiary Prevention
Lifetime surveillance of patients who have been diagnosed and treated for skin cancer is an important component of prevention. Many studies have shown that laboratory and radiologic surveillance do not detect second primaries or recurrences beyond what is found on skin examination in patients with a history of melanoma. Patients with a history of melanoma should be evaluated every 6 months with skin and regional lymph node examinations. Particular attention should be paid to the scar of the previous excision site during these examinations. After 2 years of normal examinations, the interval between visits can be extended to 1 year. If an individual has any atypical moles, a positive family history for melanoma, or other poor prognostic factors, an evaluation (including cutaneous photography) every 3 to 6 months for the first 2 years (lengthening the interval between examinations only if the atypical moles are stable) is recommended. No recent findings have affected these general recommendations. Newer biomarkers are being explored to detect melanoma earlier, but thus far no validated successes have been reported.
Prevention of skin cancer depends on increasing the awareness of health care professionals and the public about the importance of early diagnosis and skin self-examinations. More research needs to be devoted to ways of motivating high-risk individuals to receive screening examinations. Public policy measures should be expanded and consideration given to other approaches, such as mandatory use of sunscreen and adequate clothing in daycare centers and public schools. The true measure of the various prevention strategies will come from studies of their ability to decrease both the incidence of and the mortality from skin cancers. Recent evaluations of the value of screening and early detection for cutaneous melanoma suggest that survival has been increased from 50% 30 years ago to over 85% today.
OVARIAN CANCER
Ovarian cancer will affect an estimated 20,180 women in the United States during 2006 and is the fifth leading cause of cancer mortality among women, after cancer of the lung, breast, colorectum, and pancreas.[78] Approximately 10% of ovarian cancers are associated with a BRCA mutation, whereas the remainder are sporadic.[332] The lifetime risk of developing ovarian cancer for the general population is approximately 1.5%.[333] Advanced-stage disease (spread beyond the pelvis or FIGO stage III/IV) is present in 70% of these women; this has led to a 40% overall 5-year survival rate with low 5-year survival rates for advanced-stage disease (stage III 5-year survival, 25.1%; stage IV 5-year survival, 11.1%).[334] Hence, most women who present with advanced disease die from their disease. Therefore, prevention of ovarian cancer is an important goal. Given that 90% of all ovarian cancers arise from the epithelial lining of the ovary rather than from the germ cells or sex-cord derivatives, preventive strategies outlined in the following paragraphs pertain to the prevention of epithelial cancers.
Primary Prevention and Risk Reduction
Epigenetic Factors
Several reproductive, environmental, and genetic factors could influence an individual's risk of developing ovarian cancer ( Table 26-7 ). Advancing age is an important risk factor, in that data from SEER and the National Center for Health Statistics reveal that more than 48% of all cases of ovarian cancer occurred among women older than 65 years of age.[335]
Table 26-7 -- Risk Factors for the Development of Ovarian Cancer
|
Important protective reproductive factors include increased number of pregnancies, oral contraceptive use, and breastfeeding. Reanalysis of six large case-control studies involving 2768 incident cases demonstrated an overall risk of 0.66 in oral contraceptive users that continued to fall with duration of use; the effect was projected to persist for a lifetime after stopping.[336] Only a limited number of studies have examined the effect of HRT on ovarian cancer development. In the largest investigation reported thus far, a cohort study of 44,241 participants in the Breast Cancer Detection and Demonstration Project (329 cases of ovarian cancer), estrogen-only use (but not combined estrogen-progesterone) was associated with a modest increased relative risk of ovarian cancer.[337] No relationship between ovarian cancer risk and age of menarche, age of menopause, or duration of HRT has been observed in one large meta-analysis.[338] Conversely, increased risk was observed in nulliparous women who had a history of infertility. Critical analysis suggests that women with a history of infertility alone are at increased risk, and fertility drugs might not alter this risk further.[339] Tubal ligation, and to a lesser degree hysterectomy, are associated with decreased risk of ovarian cancer.[340]
A large prospective cohort study of 300,537 women (1511 deaths from ovarian cancer) showed that mortality rates from ovarian cancer were significantly increased in overweight and obese women who had never used HRT.[341] Other dietary factors, such as consumption of milk and selenium, might also modify an individual's risk for ovarian cancer, but the data are not convincing.
Finally, talc has been suggested as a possible factor leading to the development of ovarian cancer, in that epithelial ovarian malignancy shares similarities with mesotheliomas, and talc is structurally similar to asbestos, a proven cause of mesotheliomas. Cramer and associates[342] first reported this association in 1982 when they assessed genital exposure to talc in 215 white females with epithelial ovarian cancer and 215 matched controls. They reported that 92 of the patients (48%) regularly used talc, either as a dusting powder on the perineum or on sanitary napkins, compared with 61 controls (28.4%). A relative risk of 1.92 was associated with these practices in the women with ovarian cancer. In a follow-up study, researchers determined that the risk for ovarian cancer was highest in women who applied talc directly to the perineum or undergarments on a daily basis for more than 10 years.[343]
Genetic Factors
Approximately 10% of all epithelial ovarian cancers are related to genetic mutations (hereditary, familial, other rare syndromes). Women considered to have a hereditary ovarian cancer syndrome must have at least two first-degree relatives with histologically confirmed ovarian cancer. These individuals carry an overall lifetime probability of developing ovarian cancer of approximately 50%. There are three “hereditary” autosomal dominant syndromes, which include the site-specific ovarian cancer syndrome, hereditary breast and ovarian cancer syndrome, and HNPCC, Lynch II.[344] Together, these cancer syndromes account for 1% to 5% of the ovarian cancers that are diagnosed and generally occur 1 to 2 decades earlier than nonhereditary ovarian cancer.[345] The genetic linkage for the majority of the hereditary breast and ovarian cancer and the site-specific ovarian cancer syndromes has been found on the BRCA1 and BRCA2 loci on chromosome 17 (band q21). Specifically, the lifetime risk of developing ovarian cancers among BRCA1 and BRCA2 carriers is 15% to 60% and 15% to 28%, respectively.[346]
Women who report a family history of either a single first-degree relative and/or one or more non-first-degree relatives with ovarian cancer meet the definition of familial ovarian cancer syndrome. About 5% of diagnosed ovarian cancers fall into this category. Ovarian cancers can also occur as part of other rare, inherited, genetic syndromes such as Cowden's disease or Li-Fraumeni syndrome. Once individuals are identified with a familial or hereditary ovarian cancer syndrome, or a germline BRCA mutation carrier, primary preventive strategies can be implemented, including oral contraceptive pill use, risk-reducing salpingo-oophorectomy, and possibly dietary modification.
Prophylactic oophorectomy is a subject of substantial controversy. It has been estimated that 700 prophylactic oophorectomies might need to be performed to prevent one case of ovarian cancer.[347] A person's age, medical history, reproductive plans, and proximity to menopause must be considered when trying to make an informed decision. In general, postmenopausal women or women older than 50 years are encouraged to undergo this procedure, because the risk of ovarian cancer increases with age. Women who are younger than 50 but older than 40 years of age should be counseled on the risks and benefits associated with surgical menopause and subsequent HRT, and on the risks of developing ovarian cancer based on their pedigree. There are two exceptions to these rules. One is the performance of prophylactic oophorectomies on women who are being operated on for a bowel cancer, because metastases are found in 25% of cases when the ovaries are carefully examined.[347] In addition, women with one of the three hereditary ovarian cancer syndromes, and also BRCA1 or BRCA2 carriers should strongly consider risk-reducing salpingo-oophorectomy at age 35 or at the completion of childbearing. [348] [349] These women, however, should be counseled that they are still at increased risk for the development of intra-abdominal carcinomatosis, which is histologically similar to epithelial ovarian cancer but arises from the peritoneal lining. As mentioned, carriers of germline mutations in BRCA1 and BRCA2 genes should be counseled about the option of prophylactic risk-reducing salpingo-oophorectomy.[350]The recommendation is problematic, however, because there is heterogeneity in the type and number of mutations at the BRCA loci, which thus far has made it difficult to quantify an individual's risk of developing ovarian cancer. An interesting decision analysis study has been performed[351] to determine the effect of prophylactic oophorectomy on life expectancy in women with BRCA1 and BRCA2mutations. On average, a 30-year-old woman would increase her life expectancy from 0.3 to 1.7 years after undergoing bilateral oophorectomies. The analysis suggested that a woman could delay oophorectomy until she was 40 years of age with minimal loss of life expectancy.
Screening and Early Diagnosis
Early detection of ovarian cancer has not contributed to an overall improvement in survival of gynecologic malignancies.[352] Effective screening for a disease requires that the disease of interest has a premalignant phase, a long preclinical phase, stage-dependent outcome, curative treatments that are acceptable and readily available, and a cost-effective screening test that demonstrates good sensitivity, specificity, and positive predictive values. Unfortunately, the premalignant condition that progresses to invasive ovarian cancer has not been well defined. In addition, once invasive ovarian cancers develop, the preclinical phase is asymptomatic, with the result that most women present with advanced disease. Survival is not only stage dependent but also grade dependent. Primary treatment is surgical and in general is curative only if disease is confined to the ovaries. The available screening tests (pelvic ultrasound, serum cancer antigen [CA-125] levels, and pelvic examinations) are acceptable to patients and physicians but suffer from decreased specificity secondary to unacceptably high false-positive rates. Furthermore, the likelihood that an individual has ovarian cancer after a positive test (positive predictive value) is too low for available tests to be used for mass screening. In addition, screening might actually increase morbidity and mortality, because surgical exploration is the only way to diagnose ovarian cancer definitively. A decrease in mortality from ovarian cancer will occur only if we can detect early-stage disease reliably, which is not possible given available screening tests. Currently, there is insufficient evidence to support routine screening with transvaginal ultrasound and serial CA-125 serum measurements.
Specific screening tests have been evaluated both individually and together in a variety of populations. Although the pelvic examination should be included in every annual physical examination, studies have shown that it is not extremely effective in diagnosing adnexal masses.[353] CA-125 is an antigenic determinant on a high-molecular-weight glycoprotein that is recognized by a monoclonal antibody (OC125).[354] It is expressed by 82% of ovarian carcinomas and in a percentage of other normal or pathologic conditions including pelvic infections, endometriosis, pregnancy, menstruation, pancreatitis, renal failure, hepatitis, peritonitis, and congestive heart failure. Even in early-stage disease, however, CA-125 can detect only 50% of stage I disease and 60% of stage II disease.[354] Therefore, a normal CA-125 in the setting of an abnormal examination should not keep the physician from performing other diagnostic studies or surgical exploration. The real value of this tumor marker is in a premenopausal or postmenopausal woman who has a pelvic mass, particularly in the postoperative period, to measure response to treatment and progression of disease ( Box 26-3 ).
Box 26-3
|
PHILOSOPHY OF SCREENING AND EARLY DETECTION FOR OVARIAN CANCER |
|||||||||
|
Ultrasonography by itself is easy to perform, but like CA-125 measurements, it is not an acceptable enough screening test to warrant its general use. Currently, color flow imaging and measurements of pulsatility indexes are being coupled to transvaginal ultrasound to try to decrease the rate of false-positive results. Improvements in the ability to visualize both ovaries during a given examination might improve the accuracy of transvaginal ultrasound as a screening test. So might more accurate measurements of the morphologic variations that exist between ovaries, among patients, and during different periods of the reproductive life cycle of a woman.[354] Numerous studies have combined pelvic examinations, CA-125 serum measurements, and transvaginal ultrasound. Results from an ongoing National Cancer Institute–supported multicenter trial evaluating the utility of transvaginal ultrasound and serial CA-125 measurements in 74,000 women who have been randomized to either annual pelvic examinations with CA-125 measurements and pelvic ultrasonography or annual pelvic examinations without CA-125 or ultrasound evaluation will help direct future screening practices. Until improvements in test characteristics of the various screening techniques can be accomplished, multimodality screening for ovarian cancer is recommended only for women with a hereditary ovarian cancer syndrome orBRCA1 or BRCA2 mutation, although no specific studies are available to support this practice. Individuals with a strong family history that does not demonstrate an autosomal dominant inheritance pattern should be evaluated on a case-by-case basis.
Chemoprevention
Chemoprevention studies for epithelial ovarian cancer have not been performed, although the data for oral contraceptive pills suggest strongly that these agents would be effective. In addition, DePalo and colleagues[355] completed a preliminary analysis of women with T1-T2 breast cancer who were randomized to oral 4-hydroxy-fenretinamide or placebo. Six cases of ovarian cancer were diagnosed in the placebo group vs. none in the treated group (P = 0.02).[355] Long-term follow-up after the intervention phase has shown that no significant difference between the two groups was maintained once the intervention was stopped. Fenretinide and its derivatives continue to have considerable appeal, however, and further studies should clarify the role that this vitamin A derivative might play in the chemoprevention of ovarian cancer.[356] Studies in primates suggest a strong synergistic effect of oral contraceptives plus fenretinide on markers that should be associated with a reduction in ovarian cancer[357]; however, proving via a classical randomized trial whether ovarian cancer can be prevented will be difficult.
In conclusion, a great deal of work remains to be done both in establishing the etiology of ovarian cancer and in developing strategies to prevent its occurrence and reoccurrence, but the opportunities seem great.
CERVICAL CANCER
Prevention of cervical cancer is one of medicine's greatest accomplishments. The recent demonstration of an effective vaccine against HPV, the major etiologic agent of cervical cancer, suggests that this disease eventually could become a concern of the past.[358] The U.S. annual mortality rate for cervical cancer has decreased from 26,000 deaths in 1941 to an estimated 3700 deaths in 2006, which is impressive given the twofold increase in population in the United States during the second half of the 20th century.[78] Widespread usage of Papanicolaou smear screening is largely responsible for reduction of cervical cancer from the leading cause of cancer death in American women to an uncommon one. In fact, the incidence of cervical cancer has decreased by 70% since widespread screening began in the early 1940s.
Unfortunately, these favorable statistics cannot be generalized to other countries or to certain subgroups within the United States. Cervical cancer is still the number one killer of women in underdeveloped nations, with 500,000 cases still being diagnosed worldwide every year. Despite the tremendous resources dedicated to improved screening methodologies and identification of high-risk individuals, simple visual inspection of the cervix after acetic acid application is the most effective approach in underdeveloped countries, in terms of both cost and lives saved.[359]
Primary Prevention and Risk Reduction
Numerous risk factors for preinvasive and invasive cervical disease have been identified, some of which are modifiable and others not ( Table 26-8 ). Age and race are nonmodifiable risk factors, whereas socioeconomic status and degree of immunosuppression (HIV-positive individuals, transplant recipients) are risk factors that are difficult to change. Other factors, such as sexual and behavioral risk factors—including a high number of sexual partners for a woman or her partner; early age at first coitus, pregnancy, or marriage; history of sexually transmitted diseases; HPV infection; contraceptive choice; nutritional status; tobacco smoking; and frequency of Papanicolaou smear screening—are modifiable risk factors, and areas for strategic prevention efforts.
Table 26-8 -- Risk Factors for the Development of Cervical Cancer
|
|
* |
Sexual behavior is probably largely a surrogate for the risk of exposure to HPV. |
Cervical cancer is a sexually transmitted disease. Many of the epidemiologic and behavioral risk factors are direct and indirect surrogate markers for infection with HPV. Recently the National Institutes of Health has released a consensus statement that cervical cancer is largely preventable if young people modify their sexual behavior and decrease their exposure to HPV, which is the most important risk factor for the development of preinvasive and invasive cervical disease.[360] More than 100 types of this double-stranded DNA virus exist, with approximately 50 types found in the epithelial cells of the genital tract. HPV is further subclassified into high, intermediate, and low-risk types. For example, HPV 16 and 18 are high-risk types and have been found in many high-grade, preneoplastic, and invasive cervical lesions. In contrast, HPV 6 and 11 are low-risk types and are generally found in condylomata accuminata (genital warts) or low-grade cervical lesions. Almost 90% of invasive cervical cancer specimens and more than 75% of high-grade cervical lesions have measurable titers of HPV detected by available molecular biologic techniques (hybridization or the polymerase chain reaction).[361]
The mechanism of action of HPV oncogenesis is related to viral production of the E6 and E7 proteins that bind and inactivate tumor suppressor genes such as p53 and Rb (E7) in cervical cells. [362] [363]This, in turn, could lead to neoplastic transformation. HPV DNA has been found to integrate into host chromosomal DNA in many high-grade dysplastic and invasive cancer cervical specimens but not in the majority of low-grade cervical cancer precursor lesions. In the latter, the HPV DNA tends to exist in an unintegrated, circular form known as an episome. Molecular epidemiologic studies suggesting that different HPV 16 variants have different oncogenic potential could partly explain the wide geographic variation of CIN and cervical cancer in similarly infected populations.[364]
In general, women with cytologic or histologic evidence of HPV DNA should be counseled that they are at increased risk for the development of cervical disease, even though it is difficult to provide definitive risk assessments. In addition, Koutsky and associates[13] performed two large prospective studies of women who had a normal Pap smear and found that those who tested HPV-positive had a 10- to 15-fold chance of developing a cytologic lesion compared with those who were HPV-negative.[13]
Clearly, not everyone with cytologic evidence of HPV will progress to developing a preneoplastic cervical lesion, and even fewer will progress to invasive disease. Investigators have detected HPV DNA in 6% of women of reproductive age with healthy cervixes.[365] The presence of HPV DNA in women with normal Pap smears might be as high as 40%, however, and it is estimated that approximately 10 to 20 million women in the United States have detectable HPV DNA. [366] [367] Although numerous experimental, clinical, and epidemiologic studies have established a definitive role for HPV in the development of cervical cancer, other factors such as age, contraceptive method, smoking history, degree of immunosuppression, and nutritional status probably play a role in the progression and neoplastic transformation of HPV-infected cervical cells.
The grade of the cervical lesion also influences the risk of a cervical lesion progressing to invasive cancer. Many prospective studies have documented the percentage of the different grades of squamous intraepithelial lesions that will progress to higher grade lesions.[368] Estimates indicate that it could take approximately 10 years, on average, for a preinvasive lesion to progress to an invasive cancer. Sixty percent of low-grade lesions spontaneously regress on their own without treatment, in comparison with 30% of high-grade cervical lesions. It is estimated that anywhere from 35% (CIN 2) to 56% (CIN 3) of high-grade lesions will persist.[368] Therefore, the higher the grade of the lesion, the more likely it will either persist or progress. These wide ranges suggest that other factors play a significant role in determining the eventual outcome of premalignant cervical lesions.
Differences in incidence and mortality rates of carcinoma in situ and cervical cancer have been reported across different ethnic groups. Although socioeconomic factors might be substantially responsible for the discrepancies, African Americans and Mexican Americans, especially if they do not speak English, are more commonly diagnosed with cervical cancer and ultimately will die more frequently than white women.[369] The age-adjusted incidence rates for blacks are 14 per 100,000, compared with 7.8 per 100,000 for white Americans.[370] Subsequent data from the National Institutes of Health, however, revealed that the difference in incidence of cervical cancer between blacks and whites is disappearing, which suggests that efforts to target this high-risk group are succeeding.[360] This is not the case for Hispanics, because approximately 1.6 million Hispanic women remained unscreened in the United States.
Different methods of contraception might have different effects on the development of cervical disease. Barrier methods of contraception have been found to significantly decrease an individual's risk of developing cervical cancer (relative risk ≈ 0.4).[371] Conversely, a weakly positive association has been observed between oral contraceptive use and the development of cervical disease.[372] It is difficult to conclude definitively that oral contraceptive use increases a person's risk for cervical disease because of important confounders such as sexual behavior and HPV infection.[373]
Evidence strongly implicates smoking as an independent risk factor in the development of cervical disease.[374] It has been estimated that smokers have a 4.5-fold increased risk of carcinoma in situ as compared with matched controls.[375] In a 10-year prospective study of cervical dysplasia, tobacco smoking was associated with a two- to fourfold increased risk of CIN 3 and invasive cervical cancer in those women also infected with oncogenic HPV.[376] In addition, a significant dose-response relationship has been observed in women who smoked more or for longer periods of time.[374] Elevated levels of nicotine and cotinine have been found in the cervical mucus of smokers, which could alter local defense mechanisms (such as Langerhans cells) and/or be mutagenic themselves, leading to transformation of HPV-infected cervical cells.[377]
The relationship between immunosuppression and neoplasia is strong and has been documented in several ways. Studies of renal allograft patients have demonstrated a 4- to 16-fold increase in cervical dysplasia, a 2- to 9-fold increase in HPV infection, and a marked increase in synchronous cervical, vulvar, vaginal, and anal lesions.[378] The HIV epidemic has created a large population of immunosuppressed patients at risk for the development of cancer in general. In 1993, cervical cancer was added to Kaposi's sarcoma and non-Hodgkin's lymphoma as an acquired immunodeficiency syndrome–defining malignancy based on published reports describing the development of more aggressive cervical cancers in HIV-infected women.[379] It was observed that HIV-infected women, though still asymptomatic from their disease, presented with more advanced cancers, of higher grade, and at a younger age than what was expected in the immunocompetent population.
Screening and Early Detection
The Pap smear is considered to be medicine's most successful screening test. Although it has never been subject to a randomized controlled clinical trial, this diagnostic test has been in widespread use since Drs. Traut and Papanicolaou published their findings in 1941.[380] International and regional surveys have documented decreases in both the incidence and mortality rates from cervical cancer among clinics that have widely adopted Pap smear screening. Scandinavian countries have reported reductions in mortality rates from cervical cancer by 30% to 80%, depending on the country, the duration of the screening program, and frequency of screening (Iceland, 80% reduction; Finland, 50%; Sweden, 34%).[381] Canada and the United States also have experienced similar reductions in both incidence and mortality rates. The Pap smear is an effective screening test because it can detect disease early—at a stage when it can subsequently be managed by available and effective treatments ( Fig. 26-7 ). In addition, the Pap smear is cost effective and acceptable to both physicians and patients. Currently, 50 million Pap smears are performed annually in the United States.
|
Figure 26-7 Routine screening and early detection for cervical cancer and its prevention. |
Despite the profound impact that the Pap test has had on decreasing the incidence of cervical cancer by facilitating early detection of preinvasive lesions, its validity has been questioned periodically. General misperceptions by the media, public regulatory agencies, plaintiffs, physicians, and the community at large exist regarding the role of this screening test.[382]
Several problems have arisen regarding the Pap smear as a screening tool. A great deal of media attention has focused on screening errors that have occurred in “Pap mills,” where technicians were paid according to the number of slides they could screen. Studies have reported that the false-negative rate ranges between 5% and 20% in good laboratories, suggesting that the overall sensitivity is approximately 80%. Many of the false-negative results are actually due to sampling errors rather than to screening or interpretation errors. In sampling errors, abnormal cells are either absent or unidentifiable because of inappropriate technique, unsatisfactory equipment, or difficult patient examination, making it hard to obtain a satisfactory sample. Interpretation errors occur when the health care provider fails to fill out the patient history on the cytology requisition or cytotechnologists are supervised inadequately.
New technologies to improve the accuracy of the Pap smear are emerging, but there are some concerns that these tests might be too costly for high-risk patients who need them the most (low-income, minority, or elderly women). If this new technology improves the sensitivity of testing sufficiently, however, there ultimately might be a positive cost-benefit effect through the prevention of cervical cancer. A detailed analysis of the costs and benefits of different strategies to screen for cervical cancer in less-developed countries has been presented.[383] Compared with no well-organized screening, all strategies saved lives, at costs ranging from $121 to $6720 per life year saved, and they reduced mortality by as much as 58%. The simple approach of visual inspection of the cervix after applying acetic acid with appropriate follow-up was highly cost effective ($524 per life year saved and 83% reduction in mortality) and could be a more reasonable approach in less well-developed countries or certain hard-to-reach populations in the United States.
Current screening guidelines of the American College of Gynecologists and the American Cancer Society include annual Pap smear and pelvic examinations for any woman who is sexually active and/or 18 or more years of age. After three consecutive normal Pap smears and examinations, a “low-risk” patient may increase the interval between screenings from 1 year to 3 years. A high-risk individual is anyone who has had two or more sexual partners in her lifetime, intercourse before 20 years of age, a relationship with a male with multiple sexual partners, history of an abnormal Pap smear or gynecologic cancer, or anyone who is immunosuppressed. Most American women should probably be considered to be at high risk and therefore should undergo yearly Pap smear screening until sexually inactive.
Conventional Pap smear screening has proven to be extremely effective. The success of this test and future screening modalities depends on realistic expectations about screening tests, continued quality assurance, improved sampling of cervical tissue, and the incor poration of newer technologies with a view to improving the accuracy of screening and interpretation of Pap smears. Successful implementation, however, must be cost effective and lead to further declines in the incidence of and mortality from cervical cancer. Follow-up rates for abnormal Pap smears, particularly among underserved populations, remain another barrier for successful implementation of this important test. An attempt to improve the follow-up rates for severely abnormal Pap test results in underserved populations has been performed in the setting of a randomized controlled trial.[384] Patients in the single-visit intervention arm were required to wait in the clinic until the Pap smear results were available, and then large-loop electrosurgical excision procedure was performed immediately for severely abnormal results. This single-visit intervention resulted in a higher proportion of patients with abnormal Pap results receiving a full course of therapy within 6 months as compared with those in the usual-care study arm (88% vs. 53%), with a higher rate of follow-up at 1 year (63% vs. 21%). These results demonstrate that a single-visit cervical screening program was feasible among patients in an underserved population. Adoption of such a program could help narrow the gap in cervical cancer rates for underserved and minority populations. The challenge of the next decade will be for cervical cancer screening to optimize the medical outcomes and economic costs in general and in special high-risk populations.[385]
Chemoprevention
The goal of chemoprevention of cervical disease is to prevent or delay the development of cervical cancer and its precursor lesions by interrupting or preventing the process of carcinogenesis at the cellular level. The cervix is ideally suited for clinical studies that evaluate the process of carcinogenesis, because it is easily accessible for evaluation by Pap smear and colposcopy, and in general, abnormal cervical cells progress from low-grade lesions (HPV-positive cells and CIN 1) to high-grade lesions (CIN 2, CIN 3, carcinoma in situ) over an extended period of time.
Numerous experimental and epidemiologic trials have implicated HPV in the etiology of cervical cancer, and several large randomized trials have demonstrated spectacular efficacy (100% effectiveness) of an HPV vaccine in individuals who are sexually naive and/or HPV unaffected,[13] an effect that has been sustained over the long term.[14] Based on these results, a vaccine effective against a range of common subtypes has been approved for administration to young girls. [386] [387] Epidemiologic studies of diet have suggested that low intake of several nutrients is associated with cervical cancer; however, the results of supplementation trials thus far have been almost uniformly negative, notwithstanding the experiences with topical retinoic acid and positive results from a recent randomized trial of celecoxib in which a favorable effect on high-grade dysplasia was demonstrated. [388] [389] [390] The eventual role of supplementation in late stages of the disease remains unconfirmed; clearly, continuation of smoking has an adverse effect,[391] an observation that remains important for overall cancer control.
INTERNATIONALLY IMPORTANT CANCERS
Hepatocellular carcinoma (HCC) and stomach cancer are the number one and two causes of death from cancer worldwide.[392] With the increasing influx of Asian immigrants, these diseases have become more common in the United States as well. Over the past several years, important advances related to their control have become evident, and therefore a brief review of these findings is presented. An excellent review recently has become available.[393] The major risk factor for HCC is exposure to hepatitis B or C virus, which is typically transmitted vertically (i.e., mother to child) in Asian countries and horizontally (i.e., intravenous injection, sexual contact, blood transfusion) in the United States. [393] [394] Contributory risk factors include hepatotoxins such as aflatoxin, excess alcohol exposure, and smoking, as well as male gender. [393] [395] Numerous studies have demonstrated that vaccination can prevent the development of HCC caused by hepatitis B. [395] [396] Likewise, interferon treatment seems effective in interrupting or slowing the processes that lead to HCC initiated by hepatitis C.[397] A large phase III randomized controlled trial addressing HCC screening is currently underway to evaluate the optimal screening interval (3 months versus 6 months) for liver ultrasonography or CT among cirrhotic patients.[398]
Chemoprevention agents also seem to be effective in populations at risk for HCC due to hepatitis C infection. A randomized trial of a cyclic retinoid has shown that this compound reduced development of a second primary HCC by one-third, but this study has not been repeated or confirmed.[399] A phase III randomized trial of an anti-hepatitis C vaccine is underway, and a unique trial of supplementation withS-adenosylmethionine to prevent progression has recently been initiated at the University of California, Irvine. Overall, advances in microbial oncology have led to a great deal of progress in understanding the etiologic basis of HCC and in developing strategies for its eradication and prevention.
Our basic understanding of the etiologic origin of stomach cancer has been affected greatly by a paradigm shift, and the recognition that H. pylori infection is central to gastric carcinogenesis in humans has been critical to the development of new prevention strategies.[400] Although individual susceptibility, lifestyle issues, and diet all play a role in the pathogenesis of gastric carcinoma, attributable risks for H. pylori range as high as 73%.[401] There now exists a full-fledged attempt to prevent gastric cancer by appropriate screening in high-risk populations followed by H. pylori eradication. [401] [402]Epidemiologic data are supportive of a protective role for fruit and vegetable consumption. Two large randomized trials suggest that H. pylori treatment interferes with the progression of gastric preneoplasia and gastric cancer [403] [404] although a definitive trial is needed to prove the latter issue. Supplementation with antioxidants in both trials failed to add benefits to eradication of the bacterium.
Rational approaches to the early detection and prevention of the microbial origin of HCC and stomach cancer are now established. Concerted public health efforts should lead to screening and their eradication.
Nasopharyngeal carcinoma (NPC) is uncommon in the United States but has a high incidence in China and Southeast Asia, due to increased susceptibility to EBV.[405] In the endemic regions NPC is typically undifferentiated carcinoma (World Health Organization [WHO] type 3), whereas in the United States the majority of patients have the keratinizing squamous cell carcinoma (WHO type 1) NPC subtype. The WHO type 1 NPC subtype histologically resembles typical HNSCC and is more commonly associated with tobacco exposure and alcohol consumption. Nearly all type III NPC tumors involve latent EBV infection, whereas EBV infection is absent from type I NPC tumors in nonendemic regions; furthermore, survival is improved for type III NPC as compared with type I NPC.[406] Although the majority of type I NPC patients in the United States are Caucasian, Asian ethnicity within type I NPC has been shown to be an independent prognostic factor for improved survival.[407] This may be related to a higher proportion of EBV-positive type I tumors among U.S. Asians, or genotypic differences such as overexpression of epidermal growth factor receptor 1, which has been associated with poor survival in NPC and HNSCCs. [408] [409] [410] Recent advances in molecular targeted therapeutics of HNSCC involving epidermal growth factor receptor small molecule inhibitors and monoclonal antibodies have emerged.[411] Based on the findings and toxicity profile of agents in such therapeutic trials, opportunities may present for a targeted approach to prevention strategies for NPC and HNSCC among selected patients.
USEFUL RESOURCES
For those interested in cancer prevention, particularly useful resources include the Journal of the National Cancer Institute and the Cancer Epidemiology Biomarkers and Prevention journal, and publications from the International Agency for Research on Cancer (http://www.iarc.fr/). This latter entity publishes a continuing series on early detection and prevention that represents an invaluable and critical analysis of current controversies. A particularly important Task Force Report from the American Association for Cancer Research on chemoprevention drug development recently has been published.[412] Useful Internet sites include
|
The Division of Cancer Prevention, National Cancer Institute, National Institutes of Health (http://www.prevention.cancer.gov). |
||
|
Chemoprevention—The Answer to Cancer? (http://ohioline.ag.ohio-state.edu/hyg-fact/5000/5051.html). |
||
|
Harvard Center for Cancer Prevention, Harvard School of Public Health (http://www.hsph.harvard.edu/cancer/). |
||
|
American Institute of Cancer Research (http://www.aicr.org/index). |
||
|
International Society of Cancer Chemoprevention (http://www.iscac.org). |
||
|
Cancer Research Foundation of America (http://www.preventcancer.org). |
||
|
National Foundation for Cancer Research (http://www.nfcr.org). |
||
|
ClinicalTrials.gov, search words “Cancer Prevention,” “Screening,” and “Chemoprevention” (http://www.clinicaltrials.gov). |
||
|
United States Department of Health and Human Services, Agency for Healthcare Research and Quality, United States Preventive Services Task Force (USPSTF) (http://www.ahrq.gov/clinic/prevenix.htm). |
ACKNOWLEDGMENTS
The authors thank Sandy Schroeder for excellent administrative assistance in the preparation of this chapter. We thank Bill Armstrong and Ken Chang for the photographs in Figure 26-2 . This work was supported in part by P30CA62203 from the National Institutes of Health.
REFERENCES