Kaare Christensen
At a first glance, oral clefts (OCs) seem an ideal group of congenital malformations to be studied. They are among the most common congenital malformations and, compared to most other anomalies, easily diagnosed and described. However, even such a seemingly straightforward task as estimating the frequency of OC imposes its own problems. The first part of this chapter addresses methodological issues in the assessment of OC frequency, including frequency measurements and the completeness and biases in various ascertainment sources. The second part covers some methodological issues in connection with the two major design options for assessing environmental risk factors OC: case-control studies and follow-up studies.
Assessing the frequency of Oral Clefts
Although OCs are among the most frequent congenital malformations and relatively easily assessed, the statistical power of most etiological studies is low, especially because most studies subdivide OCs into cleft lip (CL) with or without cleft palate (CL/P) and isolated cleft palate (CP) and, furthermore, into syndromic and nonsyndromic cases. These subdivisions are based on observations that nonsyndromic CL/P and CP rarely segregate in the same family (Fogh-Andersen, 1942) and that a substantial number of syndromic cases have a strong association with specific genetic mutations. Additionally, it may be of interest to study OCs stratified by familial occurrence and severity [bilateral vs. unilateral CL/P, cleft lip and palate (CLP) vs. CL] (Mitchell, 1997). Finally, some studies have suggested that the nonsyndromic CP group may consist of two etiologically different subtypes: CP affecting the soft palate only and CP affecting both the hard and soft palate (Christensen and Fogh-Andersen, 1994; Clementi et al., 1997). A number of etiological factors are likely shared by these subgroups, while other factors may be unique to a subgroup. Therefore, the emerging international collaborative efforts are essential to obtain reasonable statistical power in analyses of the subgroups.
Incidence or Prevalence?
In epidemiology, the three major measures of disease frequency are incidence, prevalence, and risk. Incidence is the number of disease onsets per risk time (observation time), prevalence is the proportion of a population that has a disease at a specific point in time, and risk is a cumulative incidence proportion, i.e., the proportion of a given population that becomes cases within a given time period (Rothman, 1986). The frequency of OCs is one usually reported as 1 to 2/1000 births. There has been a long semantic and conceptual discussion about this frequency measurement: is it a measure of incidence, prevalence, or risk?
The primary palate and the palatine shelves fuse at the end of the first trimester of gestation. Hence, the population at risk includes all conceptions surviving to the time of normal fusion. Since it is not possible to ascertain all conceptions that occur and survive until this point and not feasible to assess whether fusion has occurred in all cases, neither the incidence nor the risk measure of disease frequency is obtainable. What could be measured is the proportion of OCs among newborns, stillborns, and abortions occurring after the first trimester. Therefore, only OC prevalence may be measured (Rothman and Greenland, 1998).
Prevalence data have severe limitations when used for etiological research because prevalence depends on both the incidence of new cases and the survival of cases. If first-trimester exposure to an environmental risk factor increases OC risk when the exposure intensity is moderate and the same exposure leads to spontaneous abortion when the exposure is intense (the so-called two-threshold model), then paradoxical results may occur (Dronamraju et al., 1982; Rothman and Greenland, 1998). A reduction in the exposure could lead to a reduction in the number of spontaneous abortions and an increase in OC prevalence. Therefore, in principle, it would not be possible to know whether an association between an exposure (environmental or genetic) and OC occurrence is due either to better survival of an exposed OC fetus compared to nonexposed OC fetuses or to an increased susceptibility of OC formation in exposed fetuses. Formally, additional evidence is needed to disentangle these effects, e.g., through studies of spontaneous abortion. In reality, however, estimating prevalence is the basis of OC epidemiology research; therefore, high-quality methods of collecting prevalence data are essential.
Change in Oral Cleft Frequency over Time
Most countries have undergone dramatic changes during the last century in terms of living conditions, work environment, health care, and lifestyle. Therefore, it is of considerable interest to test whether the frequency of OCs is associated with both gradual changes, such as improved living conditions, and more delimited exposures, such as war, changes in legislation, and new health programs. Such studies, however, rely heavily on the assumption of the same level of completeness of registration. False results may be obtained, e.g., if registration is less complete during war or a famine. However, spurious results may also occur even in settings with very good registration systems. The following example will illustrate this point.
A uniform and standardized registration system has been used in Denmark since the 1930s. These data have been entered into the Danish Facial Cleft Register, which includes 7290 oral cleft cases born between 1936 and 1987. Figure 9.1 is based on the Danish Facial Cleft Register and summarizes the estimated yearly prevalence of CL/P and CP in the period 1936–1987. From Figure 9.1, it is clear that in the second half of the period (1962-1987) the prevalence at birth was fairly constant for both CL/P (around 1.4 to 1.5/1000) and CP (around 0.7 to 0.9/1000). In contrast, there was apparently a steady increase in the prevalence of both CL/P and CP during the first half of the period (1936–1961).
However, the most likely explanation of this increase in the middle of the century is an improvement in the survival of newborn CL/P cases (especially those with associated anomalies) and a better ascertainment during these 26 years (especially of milder CP forms and OC cases with associated anomalies). Figure 9.1 shows that a substantial part of the increase in prevalence in that period was due to an increased proportion of cases with associated anomalies and milder forms of OC. The nearly constant prevalence of the most severe OC forms (nonsyndromic bilateral CL/P and CP including the hard palate) suggests that no major changes in the prevalence of OC have occurred in the 52-year period. If the overall figures were taken at face value, it might be concluded that OCs became increasingly common during the middle of the twentieth century. Detailed description of the cases, however, suggests the presence of ascertainment bias.
Change in Oral Cleft Frequency with Season
Another example of why trends in rates should be interpreted with caution is seasonal variation in the dates of birth of children with congenital malformations. Seasonality has been studied to gain insight into the possible role of diet, infections, and other factors that may vary with season. Seasonal variation in the dates of birth of children with nonsyndromic CL/P has been studied in nearly 600 cases born in Montreal between 1950 and 1996 (Fraser andGwyn, 1998). A significant tendency for children with CL/P to be born more often in the summer than in the winter was found. In another report, which used the Danish CL/P data set from Fogh-Andersen's (1942) thesis, no variation was found in month of birth among females, but a peak in April-May was observed for males. These analyses were based on the distribution of month of birth for CL/P cases. Applying this method to the entire Danish 1936-1987 cohort yielded a small seasonal variation similar to that observed in the data from Canada for both nonsyndromic CL/P and CP.
A critical point to take into account when studying OC seasonally is that pregnancy-timing preferences (i.e., when parents prefer to have their children) may vary between settings and time periods. For example, in the Scandinavian countries, the frequency of births is 10% to 20% higher in spring and summer compared to winter (NOMESKO, 1993). Hence, the Danish CL/P and CP seasonality pattern can be explained by the variation in the overall seasonality pattern of births; i.e., the larger number of OC births in the spring and summer can simply be explained by the larger number of total births at that time of year.
|
|
|
FIG. 9.1. Change in the frequency of oral clefts over time. Smoothed prevalence (5-year moving average) of subgroups of cleft lip with or without cleft palate (CL/P) and cleft (CP) per 1000 live births in Denmark in the period 1936–1987. CL, cleft lip; CLP, lip and palate; CPS, nonsyndromic CP including only the soft palate (submucous CP included here); CPH, nonsyndromic CP including the hard palate. (Modified from Christensen, 1999, with permission.) |
Oral Clefts and Associated Anomalies
A longstanding debate in OC research is whether cases with associated anomalies should be excluded (“splitting”) or included (“lumping”) in etiological studies. The rationale for splitting is to reduce etiological heterogeneity, i.e., to eliminate all OC cases with known etiology (e.g., monogenic diseases such as Van der Woude's syndrome and chromosomal abnormalities such as trisomy 18). The rationale for lumping is that there may be common etiological factors in nonsyndromic and syndromic OC cases. For example, although OC occurs frequently in trisomy 18, far from all trisomy 18 cases have OC.
The factors that might increase susceptibility to OC in trisomy 18 may be the same as those that increase risk in chromosomally normal individuals.
The splitting vs. lumping problem also applies to minor and more subtle associated anomalies, for which it is considerably harder to make a uniform ascertainment and classification. The wide range in the reported frequency of associated anomalies for OC (CL/P 2%-15%, CP 10%-50%) is partially due to differences in the definition of associated anomalies, how long after birth and how carefully the individuals are examined, and the selection of patients.
The 22qll deletion syndrome illustrates this problem. Patients with this syndrome often display only CP and a minor heart defect and/or learning disability, and many such cases are undiagnosed (Goldberg et al., 1993; Brøndum-Nielsen and Christensen, 1996). The inclusion of cases with 22qll deletion syndrome in a study of nonsyndromic CP could reduce the power of the study if the risk factors for CP in this syndrome are different from the risk factors for nonsyndromic CP.
As mentioned above, the rationale behind splitting is to reduce etiological heterogeneity. However, splitting may introduce bias. If the definition of minor anomalies includes conditions that occur at a rather high frequency in the general population (e.g., learning disabilities), OC cases without any such conditions might represent an otherwise very healthy group. Furthermore, as mentioned above, a number of environmental and genetic factors likely affect both syndromic and nonsyndromic OC cases; i.e., syndromic and nonsyndromic cases may have some common etiological factors.
Therefore, as long as the delineation of strictly nonsyndromic OC cases depends strongly on definitions of associated anomalies as well as the length and intensity of follow-up, the most reasonable approach is to obtain as much information on associated anomalies as possible and, later, to perform analyses with and without the OC cases with associated anomalies.
Sources of Ascertainment
The baseline epidemiological characteristics of OC are important both for scientific purposes and for public health planning. These data are usually obtained through population-based clinical records, cleft treatment centers, surveillance systems, or birth certificates. The completeness of such files highly depends on the practice and organization of the country, state, or facility providing them.
Population-Based Clinical Records
Population-based records of treated or hospitalized cases introduce a bias in favor of surviving cleft cases, which leads to underreporting of OC (Aabyholm, 1978). Surgical files have played a major role in OC research; e.g., the Danish surgical files have been the basis of numerous, often quoted studies (Fogh-Andersen, 1942, 1971; Bixler et al., 1971; Shields et 1979, 1981; Melnick 1980; Marazita et al., 1984; Chung 1986).
The completeness of various ascertainment sources was studied in Denmark for the period 1983-1987 using three nationwide ascertainment sources and an autopsy study in a 10% sample of the Danish population. The nationwide ascertainment sources were (1) a centralized surgical facility, which had treated all OC cases since the 1930s; (2) the National Institute for Defects of Speech, which has coordinated the registration and follow-up treatment of all OC cases, also since the 1930s; and (3) the National Register of Congenital Malformations, which was based on doctors’ notifications and reports (Christensen et al., 1992).
Based on these comparisons, it was estimated that more than 95% ascertainment was obtained by means of surgical files for CL/P without associated malformations/syndromes. However, the comparison showed that surgical files should not be used for studying CP. Surgical files included only 60% of these cases. Surgical files were found to be unsuitable for studying the prevalence of associated malformations or syndromes in cases with OC. Figure 9.2 illustrates the completeness of the centralized surgical files in Denmark and the reasons for nonregistration.
Cleft Treatment Centers
Cleft treatment centers often have high-quality information on OC cases in terms of associated malformations and other characteristics. However, these cases often cannot be used for estimating the prevalence of OC or other epidemiological characteristics because the population from which they come is not well defined. Furthermore, cleft treatment centers often have a very high frequency of associated anomalies among OC cases. This is partially due to differences in the definition of associated anomalies, how long after birth and how carefully the individuals are examined, as well as the referral practice. A bias in ascertainment arises in highly specialized centers that attract severe or complicated cases (Fraser, 1970; Shprintzen et al., 1985).
|
|
|
FIG. 9.2. Sources of ascertainment of oral cleft cases. Classification of 663 known facial cleft cases born in Denmark during 1983–1987, identified through three nationwide ascertainment sources. Small slices represent non-operated cases. When a non-operated case could be classified as belonging to more than one slice, the vertically uppermost was chosen. Parentheses show the number of cases with cleft lip with or without cleft palate, isolated cleft palate, and atypical facial clefts within each slice. (From Christensen et al., 1992, with permission.) |
Surveillance Systems
Registries of congenital malformations were established in various countries after the thalidomide epidemic. The primary aim of these registries was surveillance to avoid similar epidemics. While some of the registries are limited to surveillance, others include etiological substudies within the register (Dolk et al., 1998). A number of these registries have also been used in OC research. The efficiency of these record systems is highly dependent on the local practice and organization, but even the most well-established registries may have significant underreporting of cases with OC, particularly CP, which showed an under-reporting of 12% in one study (Ericson et al., 1977).
Birth Certificates
Birth certificates are a poor source for OC data, particularly for milder forms of CL/P and CP. Green et al. (1979) found that only 65% of known OC cases in a register of congenital malformation in Arkansas were recorded as having a cleft on the birth certificate and that only 48% were correctly classified regarding cleft type. The same pattern was observed in Denmark by Olsen (1982a,b), who found that only approximately 75% of OC cases had any indication of OC on the birth certificate.
Death Certificates and Autopsy Series
Death certificates and autopsies of fetuses can supplement other ascertainment sources. However, OCs might not be recorded on death certificates when multiple defects coexist because other, more severe malformations are most likely to be registered as the causes of death. Underreporting in autopsies of fetuses is also likely because of the difficulty of recognizing and diagnosing OCs in them (Kraus et al., 1963; Christensen 1992).
Design Options for Assessing Environmental Risk Factors for Oral Clefts: Methodological Issues
As pointed out in Chapter 17, there is convincing evidence that both genetic and environmental factors play a role in the etiology of OCs. However, what environmental and genetic factors these are is still largely an enigma. A number of design options for assessing OC risk factors are available. This section covers some methodological issues in connection with the two major design options for assessing environmental risk factors for OC: case-control studies and follow-up studies.
Case-Control Studies
The case-control study design is by far the most commonly used to evaluate environmental risk factors for OC. During the last decade, case-control studies have been used increasingly for genetic association studies as well. The basic concept is to compare the first-trimester exposures of mothers OC cases with those of mothers of controls. Ideally, a case-control study uses incident cases from a well-defined population and has four to five controls, who would have been cases in the study if they had had OC, for each case (Rothman and Greenland, 1998). As mentioned above, it is not possible to obtain incident cases in studies of OC; the best available option is to include cases and controls in the study as shortly after birth (or abortion) as possible.
If one uses older cases with OC, the window of exposure of interest (the first trimester) will be further back in time, increasing the likelihood of recall bias (Werler et al., 1989). This bias would arise if case mothers reported differently from control mothers (e.g., if case mothers were more inclined to remember all exposures during the first trimester). Therefore, a study would be substantially strengthened if exposure information were obtained during pregnancy (e.g., from general practitioners’ records), before the mother knew whether the fetus had OC or not. For self-reported exposures, it is often useful to collect the date of recognition of the pregnancy as well the exposure information before and after this date. This is because many women change their lifestyle to some degree when they recognize that they are pregnant. For example, in a Danish case-control study of OC, both case mothers and control mothers reported changes in their lifestyle after recognition of pregnancy: cigarette consumption decreased by one-third and alcohol consumption decreased by 60% among all mothers (who were interviewed on average 1-2 weeks after birth). If the date of recognition of the pregnancy is known as well as the exposure before and after the recognition, a first-trimester dose can be calculated for each mother (Christensen et al., 1999).
The number of controls necessary for each case is a compromise between statistical power and costs. However, it can be shown that including more than four or five controls per case increases the statistical power only slightly (Rothman, 1986). Selection of controls needs special attention. Matching on time and place of birth is a commonly used method that is logistically appealing. However, a disadvantage is that it does not allow the study of time and place of birth as potential risk factors. It has been argued that the control group should be selected among newborns with congenital malformations other than OC. The rationale for this is to avoid recall bias since both cases and controls would be equally motivated to recall exposures. However, this approach presents several disadvantages. First, risk factors will be overlooked if they are associated with both OCs and the congenital malformations in the control group. Second, it might be difficult to obtain a large control group if it consists of children with other congenital malformations. Third, it may be difficult to inform the parents of children with congenital malformations in the control group that the study is aimed at identifying causes of OC but not of the problem their children have. If the inclusion criterion in the case group is isolated OC, then the control group should have an isolated anomaly as well. If, however, OC cases with associated anomalies/syndromes are included, then controls with other anomalies/syndromes should not be excluded.
A special situation arises when genetic factors are included in case-control studies of ethnically heterogeneous populations. If the ethnic composition is different in the case and control groups, then any genetic factor which is unequally distributed among ethnic groups might show an association with OC (without having any etiological role); this is called population stratification. This potential bias can be reduced by matching on ethnic background or by conducting studies in ethnically homogeneous populations. Recent developments in methodology extend the case-control design to overcome such and other problems embedded in case-control studies of genetic factors. One option is to sample case triads (i.e., an OC case and his or her parents) and similar control-triads in the study. The use of such triads makes it possible to assess or avoid bias, such as population stratification (Umbach and Weinberg, 2000).
Follow-Up Studies (Cohort Studies)
From a strictly methodological point of view, prospective follow-up or cohort study is preferable over a case-control study. If pregnant women are followed-up from recognition of pregnancy, exposure information can be obtained with no recall bias. This is because these pregnant women do not yet know who will eventually be a case mother or a control mother. Despite OCs being among the most common congenital malformations, the drawback of this approach is that 1000 women would have to be followed in order to identify two cases. Such studies are enormous undertakings. In the period 1959-1966, such a study was conducted in the United States, the Collaborative Perinatal Project. In this study, about 56,000 pregnant women were followed from the first prenatal visit through labor and delivery, and their offspring were later followed-up to assess, among other things, the frequency of congenital malformations (Myrianthopoulos and Chung, 1974).
An even larger study is emerging from Denmark. During the years 1998-2002, approximately 100,000 pregnant women and their offspring will be included in a longitudinal study through a large collaborative effort. Women are typically enrolled in the study at the first visit to their general practitioner (usually in the first trimester). At this time, women who decide to participate, provide a blood sample and are interviewed about a broad range of health-related issues, including potential fetal risk factors and lifestyle information. Around the beginning of the third trimester, women who wish to participate again provide a blood sample and are interviewed. At birth, an umbilical blood sample is drawn, and later, punches are obtained from the newborn's screening card. Finally, 6 and 18 months after delivery, mothers participate in an interview. At this time, data including information on hospitalizations and surgeries are obtained from national health registries. This study allows for the assessment of risk factors even before the woman knows the outcome of her pregnancy, thereby avoiding recall bias. Also, it makes it feasible to include both maternal and infant genotypes in the analysis and to measure changes in maternal plasma antibodies during pregnancy. The latter will allow evaluation of the influence of various maternal infections in the etiology of OC. It should be emphasized, however, that even such a large study will not be able to obtain incident data since women experiencing spontaneous abortions prior to their first prenatal visit will not be included.
Summary
Oral clefts are among the most common congenital malformations and, compared to most other anomalies, relatively easily diagnosed and described. Delineation of nonsyndromic OC cases depends on the definition of associated anomalies as well as on the length and intensity of follow-up. The most reasonable approach in epidemiological studies is to collect as much information as possible on associated anomalies and to perform analyses with and without OC cases with associated anomalies. A fundamental problem of all studies is that they use only cases with OC who have survived at least two trimesters. Hence, it is not possible to distinguish factors which increase the risk of OC in fetuses and factors which enhance the survival chances of OC fetuses. In reality, however, prevalence measures of OC are the basis of OC epidemiology research; therefore, high-quality methods of collecting prevalence data are essential.
The validity of comparing OC frequency between settings and at different time periods depends heavily on the degree of ascertainment. Many different ascertainment sources are used, and all have limitations. Population-based clinical records (surgical files) can provide high ascertainment of cases of CL/P without associated malformations/syndromes. However, they are not useful for CP cases or cases with associated malformations or syndromes. For assessing environmental risk factors, the case-control study is logistically the most appealing design. Ideally, a case-control study uses cases ascertained shortly after birth from a well-defined population and has four or five controls for each case. However, the case-control study is subject to recall bias and selection of appropriate controls might be difficult. Follow-up studies of pregnant women and their offspring yield excellent exposure information, but such studies are logistically enormous.
Acknowledgements
This work was supported by the Egmont Foundation and the National Institute of Dental and Craniofacial Research (grant R01 DE 11948).
References
Aabyholm, FR (1978). Cleft lip and palate in Norway I. Registration, incidence and early mortality of infants with CLP. Scand J Plast Reconstr Surg 12: 29–34.
Bixler, D, Fogh-Andersen, P, Conneally, PM (1971). Incidence of cleft lip and palate in the offspring of cleft parents. Clin Genet 2: 155–159.
Brøndum-Nielsen, K, Christensen, K (1996). Chromosome 22qll deletion and other chromosome aberrations in cases with cleft palate, congenital heart defects and/or mental disabilities. A survey based on the Danish facial cleft register. Clin Genet 50: 116–120.
Christensen, K (1999) The 20th century Danish facial cleft population—epidemiological and genetic-epidemiological studies. Cleft Palate Craniofac J 36: 96–104.
Christensen, K, Fogh-Andersen, P (1994). Etiological subgroups in non-syndromic isolated cleft palate? A genetic-epidemiological study of 52 Danish birth cohorts. Clin Genet 46: 329–335.
Christensen, K, Holm, NV, Olsen, J, et al. (1992). Selection bias in genetic-epidemiological studies of cleft lip and palate. Am J Hum Genet 51: 654–659.
Christensen, K, Olsen, J, N0rgaard-Pedersen, B, et al. (1999). Oral clefts, transforming-growth-factor-alpha gene variants, and maternal smoking. A population based case-control study in Denmark 1991–1994. Am J Epidemiol 149: 248–255.
Chung, CS, Bixler, D, Watanabe, T, et al. (1986). Segregation analysis of cleft lip with or without palate: a comparison Danish and Japanese data. Am J Hum Genet 39: 603–611.
Clementi, M, Tenconi, R, Forabosco, P, et al. (1997). Inheritance of cleft palate in Italy. Evidence for a major autosomal recessive locus. Hum Genet 100: 204–207.
Dolk, H, Vrijheid, M, Armstrong, B, et al. (1998). Risk of congenital anomalies near hazardous-waste landfill sites in Europe: the EUROHAZCON study. Lancet 352: 423–427.
Dronamraju, KR, Bixler, D, Majumder, PP (1982). Fetal mortality associated with cleft lip and cleft palate. Johns Hopkins Med J 151: 287–289.
Ericson, A, Kallen, B, Winberg, J (1977). Surveillance of malformations at birth: a comparison of two record systems run in parallel. Int J Epidemiol 6: 35–41.
Fogh-Andersen, P (1942). Inheritance of Harelip and Cleft Palate. Copenhagen: Arnold Busck.
Fogh-Andersen, P (1971) Epidemiology and etiology of clefts. Birth Defects 7: 51–53.
Fraser, FC (1970). Counseling in genetics: its intent and scope. Birth Defects 6: 7–12.
Fraser, FC, Gwyn, A (1998). Seasonal variation in birth date of children with cleft lip. Teratology 57: 93–95.
Goldberg, R, Motzkin, B, Marion, et al. (1993). Velo-cardiofacial syndrome: a review of 120 patients. Am J Med Genet 45: 313–319.
Green, HG, Nelson, CJ, Gaylor, DW, Holson, JF (1979). Accuracy of birth certificate data for detecting facial cleft defects in Arkansas children. Cleft Palate J 16(2): 167–170.
Kraus, BS, Kitamura, H, Ooe, T (1963). Malformations associated with cleft lip and palate in human embryos and fetuses. Am J Obstet Gynecol 86: 321–328.
Marazita, ML, Spence, MA, Melnick, M (1984). Genetic analysis of cleft lip with or without cleft palate in Danish kindreds. Am J Med Genet 19: 9–18.
Melnick, M, Bixler, D, Fogh-Andersen, P, Conneally, PM (1980). Cleft lip +/- cleft palate: an overview of the literature and an analysis of Danish cases born between 1941 and 1968. Am J Med Genet 6: 83–97.
Mitchell, LE, Christensen, K (1997). Evaluation of family history data for Danish twins with nonsyndromic cleft lip with or without cleft palate. Am J Med Genet 72: 120–121.
Myrianthopoulos, NC, Chung, CS (1974). Congenital malformations in singletons: epidemiologic survey. Report from the Collaborative Perinatal project. Birth Defects 10: 1–58.
NOMESKO (1993). Birth and infant mortality in the Nordic countries. Copenhagen: Nordic Medico-Statistical Committee.
Olsen, J (1982a). The register of congenital anomalies in Funen County. I. The number of notifications and their sources [in Danish]. Ugeskr Laeger 144: 1333–1337.
Olsen, JH (1982b). Forplantning, miljo og kemiske staffer. Copenhagen: Milj0styrelsen.
Rothman, KJ (1986). Modern Epidemiology. Boston: Little, Brown.
Rothman, KJ, Greenland, S (1998). Modern Epidemiology, 2nd ed. Philadelphia: Lippincott-Raven.
Shields, ED, Bixler, D, Fogh-Andersen, P (1979). Facial clefts in Danish twins. Cleft Palate J 16: 1–6.
Shields, ED, Bixler, D, Fogh-Andersen, P (1981). Cleft palate: a genetic and epidemiological investigation. Clin Genet 20: 13–24.
Shprintzen, RJ, Siegel-Sadewitz, VL, Armato, J, Goldberg, RB (1985). Anomalies associated with cleft lip, cleft palate or both. Am J Med Genet 20: 585–595.
Umbach, DM, Weinberg, CR (2000). The use of case-parent triads to study joint effects of genotype and exposure. Am J Hum Genet 66: 251–261.
Werler, MM, Pober, BR, Nelson, K, Holmes, LB (1989). Reporting accuracy among mothers of malformed and non-malformed infants. Am J Epidemiol 129: 415–421.
