Keith F. Woeltje
Introduction
Computers have become ubiquitous in modern society. Children's toys contain more raw computing power than was present on the Apollo spacecraft that carried men to the moon. Not surprising, computers have become essential tools in healthcare epidemiology. Nevertheless, their use in this setting varies widely; their uses will be reviewed in this chapter.
Personal Computers
Hardware
Personal computers (PCs) started as machines built by geeky enthusiasts in the 1970s but became widely commercially available in the 1980s. In the first years of the 21st century, computers have become commodity items. PC prices have fallen steadily while computing power has increased exponentially. Until a few years ago, considerable thought had to be given to buying a personal computer in balancing price, capability, and upgradeability. Now, even the most inexpensive PCs available at large electronics stores can readily handle basic computer needs.
Hospitals or universities may have recommended (or required) minimum standards for PCs purchased for work use and may even have specified that computers be purchased from certain vendors. As a general rule, spending money on more memory (RAM) rather than a slightly faster processor is a better investment. Because more and more training materials (and software) are now being distributed on DVDs, a DVD drive should be considered essential (and it is now uncommon to find a computer sold without one). Also essential is the ability to connect to a local network. This is typically via a direct Ethernet cable connection but may, in some settings, be done wirelessly. Unless the user will be storing many multimedia files (e.g., photos and video), most data and reports generated by an infection control (IC) program does not use a lot of hard drive space, so a massive hard drive is typically not high on the priority list. Being able to write to optical media (such as a CD-R or DVD-R) is a nice feature for sharing and archiving data.
The most important decision in choosing computer hardware is based on determining what operating system one wants to use. The operating system (OS) is computer software that controls the actual workings of the hardware (e.g., displaying the information on a screen, taking input from a keyboard). Other software is written to work with the OS; software written to run on one OS will not run on another OS. However, some software may have different versions that run on different OSs. If someone needs a particular program and that program is only available for one OS, this will dictate what OS to choose.
Currently, different versions of Microsoft Windows taken together are the most commonly used OS. The newest version, Windows Vista, should be available by the time
P.122
this text is printed. Microsoft Windows runs on computer hardware from a wide variety of vendors. Because it is the most commonly used OS, many computer programs are available only in versions that run on Microsoft Windows. Another OS is Macintosh OS X; it will run only on computer hardware made by Apple Computer. Some computer programs that are made for Microsoft Windows also have different versions that run on Macintosh OS X. Recent changes in Apple Macintosh hardware allows users in certain circumstances also to run the Microsoft Windows OS, potentially allowing users to run programs written for both OSs on their machines.
Linux (www.linux.org) is another OS; it originally was used primarily by computer enthusiasts and information technology (IT) professionals. Initially, it was used particularly for servers (computers attached to a network that provides services to other computers, e.g., storing files or sending out Web pages). However, Linux has made some inroads as a desktop OS in some large companies, especially outside of North America. Linux is an example of “open-source software,” which is developed as a collaborative effort of programmers around the world. Open-source software is typically available free of charge with any associated cost being for the cost of the media or for technical support, not the software itself. Another key characteristic of open-source software is that the source code (the text of the program written by the programmers) also is available, so that end-users can make or commission modifications in the program if they choose. Linux comes in a bewildering variety of versions (termed “distributions” or, more commonly, “distros”) that vary primarily as to the software that comes along with the OS (www.distrowatch.com). Linux runs on a wide variety of computer hardware.
Basic Software
A computer and its OS alone would not be able to accomplish very much. Additional software is needed to perform basic tasks, such as word processing. Although stand-alone programs of productivity software are available, they typically are obtained in suites with all components somewhat integrated, allowing consistency of use across programs. Microsoft Office has become the most commonly used office suite, but a variety of competitors exist, including the free, open-source, OpenOffice.org (www.openoffice.org). Most institutions prefer to have all employees use the same program for ease of interchanging information, but most of these programs can convert between the most common formats.
Word processing software (e.g., Microsoft Word or Corel's Word Perfect) is the cornerstone of computer productivity software. It allows letters, reports, and so on to be easily edited and formatted for distribution. Presentation software (e.g., Microsoft PowerPoint, Apple Keynote) assists with the development of attractive “slide shows” (although actual physical slides are rarely used now with the increasing use of digital projectors). The most common error with this software is the temptation to include excessive special effects, to have distracting backgrounds, and to include too much information on one slide. Presentation software also can be easily used to lay out posters for display at scientific meetings.
Spreadsheet software (e.g., Microsoft Excel, OpenOffice.org Math) is designed primarily to manipulate numbers, not text. Spreadsheets are extremely useful to healthcare epidemiologists for calculating rates and doing basic statistics. The column and row structure of spreadsheets also allows them to be used as simple databases, such as a line-listing for an outbreak investigation. Spreadsheet software also can make graphs for visualization of data. These graphs can then be imported into a word processing document or slide presentation for distribution to others.
Desktop relational database software (e.g., Microsoft Access, FileMaker) is not included in many basic editions of office suite software but may be available in more extended or “professional” editions. Although in some respects more difficult to use than other desktop computer productivity software, database software can offer healthcare epidemiologists significant advantages over other methods of storing data, such as using spreadsheet software. For example, if blood culture data are being stored as part of a study, if a given blood culture has >1 organism, then in a spreadsheet either a number of columns need to be included (e.g., “organism_1,” “organisms_2”) if data from a given culture are all to be included on one row, or a number of separate rows need to be used to record the information from a single culture. Both alternatives may introduce difficulties in subsequent data retrieval and analysis. A relational database, however, would allow the information to be structured in a manner that avoids these issues.
Because most office productivity software can be used at a basic level without any special training, the benefits of formal training are frequently underappreciated. Various levels of training classes may be offered by larger organizations or may be available at local community colleges. On-line training or structured textbooks provide additional alternatives. The time and expense spent on training will be returned many times in productivity gains. In particular, database software requires some training to use it appropriately. An excellent introduction to databases at a conceptual level (and not tied to any particular program) is Database Design for Mere Mortals [1].
Networked Computers
Although a stand-alone PC can bring a huge productivity boost to a healthcare epidemiology/IC office, the potential utility of the computer increases significantly when it is networked to other computers. The PC then is no longer limited to data that have been entered by hand or files brought to it on some form of solid media. Instead, the PC
P.123
can now share information with other computers at high speeds. The more computers that are networked together, the more valuable the network, termed “Metcalf's Law” [2].
Internet
The Internet is really a network of computer networks [3]. Precursors to the Internet began as a research project in the late 1960s sponsored by the U.S. Department of Defense's Advanced Research Project Agency. The Internet as we know it started on January 1, 1983 [4]. The Internet has grown exponentially in the intervening decades. Research is underway for the next generation Internet, termed “Internet2” (www.internet2.edu/).
World Wide Web
For many people, the World Wide Web (“the Web”) is synonymous with the Internet. The World Wide Web started in 1990 as a research project at the European Organization for Nuclear Research (CERN) [5]. The purpose of the Web is to provide access to on-line documents (Web pages), including an easy mechanism for one page to refer to another. These pages can be viewed with software called Web browsers. Since its inception, the Web has grown beyond simple text to include images and multimedia presentations and allow downloads of files of various types. For healthcare epidemiologists, the Web provides a wealth of resources. An example is the Supercourse, a Web-based set of lectures in epidemiology (available at www.pitt.edu/~super1/). Many professional organizations have Web sites that provide news, guidelines, and links to other resources for both members and nonmembers. Examples include the Society for Healthcare Epidemiology of America (SHEA, www.shea-online.org), the Association of Professionals in Infection Control and Epidemiology (APIC, www.apic.org), the Community and Hospital Infection Control Association—Canada (CHICA—Canada, www.chica.org), the Hospital Infection Society (HIS, www.his.org.uk), and the International Federation of Infection Control (IFIC, www.theific.org). Government organizations such as the Centers for Disease Control and Prevention (CDC, www.cdc.gov) provide a wealth of resources including guidelines, information on specific diseases and outbreaks, and reference materials. State and local health departments also may have Web sites that provide valuable information on local issues. The Web also provides easy access to information from companies regarding their products. Literature searches of the U.S. National Library of Medicine's MEDLINE database also can be conducted on the Web using the PubMed system (www.pubmed.gov). Some hospitals and other organizations block or severely limit access of employees to the Web. Healthcare epidemiology/IC programs can make a strong argument for having relatively unfettered Web access to do their jobs correctly.
Electronic mail, or e-mail, is probably even more popular than the World Wide Web in terms of total numbers of users. Sending files as attachments in an e-mail has become a preferred method of sending data from one person to another. In addition, e-mail can serve as a means to alert healthcare epidemiologists to significant issues in a timely fashion. Organizations (e.g., SHEA and APIC) send e-mail alerts to their members when warranted. Healthcare professionals also can sign up for alerts and updates on terrorism and emergency response from the CDC at www.bt.cdc.gov/clinregistry/index.asp In addition, the CDC's Division of Healthcare Quality Promotion has a Rapid Notification System for Healthcare Professionals that sends out e-mail alerts related to outbreaks and product recalls. The sign-up page is at www2a.cdc.gov/ncidod/hip/rns/hip_rns_subscribe.html Alternatively, one can send an e-mail message to LISTSERV@CDC.GOV with a blank subject line and the text “subscribe HIP-RNS” (without the quotes) in the message.
E-mail “list-serv” software allows e-mail to be sent to many persons by sending an e-mail to a particular e-mail address. This facilitates group discussions via e-mail. Popular e-mail groups for healthcare epidemiologists include ProMED-mail (sign up at www.promedmail.org/), which is sponsored by the International Infectious Diseases Society, and the Emerging Infections Network (request sign-up information from ein@uiowa.edu), which is sponsored by the Infectious Diseases Society of America and the CDC. Because of the rise of unsolicited commercial e-mails (termed “spam”), many institutions have “spam filters” in place to reduce the influx of these nuisance messages. Unfortunately, these filters may filter out legitimate e-mail. List-serv e-mails, in particular, may be filtered out as “bulk e-mail.” Local IT personnel should be consulted to determine what steps are needed to ensure that desired e-mail reaches the recipient.
Specialized Software
An IC program could be quite productive and organized using only the software contained in an office suite. To go beyond basic data analysis requires substantial effort to set up formulas on a spreadsheet. Even with such effort, some advanced data analysis would simply not be possible. Advanced needs require specialized software.
General Statistical Software
For more extensive analysis than can be done with a spreadsheet, a healthcare epidemiologist could use general-purpose statistical software. A wide range of programs exists, from basic statistical packages included with some statistics texts to expensive, very complex programs that
P.124
can perform even the most esoteric analysis. Widely used statistics packages include SAS, SPSS, and Stata. Many other very capable commercial programs are available. R (www.r-project.org) is an open-source general-purpose statistical package. EpiTools (available from www.epitools.net/) is a set of tools for use with the R statistical package designed to add functions of use to epidemiologists. Although some of these programs allow for some form of direct data entry, they really are designed to import data that was entered using another program (e.g., a database or spreadsheet). Thus, using these programs for routine healthcare epidemiology use takes some effort. A user who is facile with a given program may choose to do basic analyses in that program. For the most part, these statistical programs are overkill for most IC programs and are used primarily in research settings or by facilities that have dedicated statistical analysts.
Healthcare Epidemiology–Specific Software
A number of software packages have been designed specifically for healthcare epidemiology/IC programs. They allow the user to enter surveillance data (both denominator and numerator data) and will then generate reports and graphs on rates. Popular programs include AICE Millenium and EpiQuest. Some of these programs include the ability to compare a facility's rates with benchmarks from the CDC's National Nosocomial Infections Surveillance (NNIS) System (now known as the National Healthcare Safety Network, or NHSN). Some of these programs also have provisions for importing data directly from available electronic sources. Using these sources will likely require working with hospital IT support services to ensure that data are sent in an appropriate format.
Some healthcare epidemiology–specific software is designed to import essentially all information from available electronic sources. Such programs include Infection Control Assistant, MedMined, QC Pathfinder, and SETNET. Although each of these programs has different features, they tend toward more real-time analysis of electronically available data, such as microbiology results. These programs may detect clusters of antimicrobial-resistant organisms relatively quickly but may be less able to generate surgical site infection (SSI) rate graphs, for example.
The advantage of these specialized IC/healthcare epidemiology software packages is that they are designed with the needs of an infection control practitioner (ICP)/healthcare epidemiologist in mind. Although they are somewhat configurable, they require minimal set up to generate useful information. This narrow specialization also is their downside; if a user needs a specific functionality that is not included in the package, the user must use other software. Vendors are eager to hear what features their users would like to see, but a feature is typically added only if there is substantial interest. A wide variety of options are available, and the features offered by each program continually increases. Each program may have features not offered by any of the others. Facilities interested in such software should determine what features they require, request information from each vendor, and do a careful cost-benefit analysis.
General Epidemiology Software
Some programs occupy a middle ground between the general statistical software packages and the healthcare epidemiology–specific software. They provide more support for data entry and management than general-purpose statistical programs while providing more flexible statistical analysis than most specialty IC/healthcare–epidemiology software packages. The primary downside is the considerable work that would be necessary to set up specific functionality that comes “ready-made” with IC specific software. The benefit is the ability to design in the specific functionality the user desires.
Epi Info
Epi Info (www.cdc.gov/epiinfo/) is a program designed and distributed free by the CDC. Early versions of the program were designed to run on Microsoft MS-DOS to assist CDC Epidemiologic Intelligence Service Officers (EISOs) in investigating outbreaks. It was steadily upgraded, and in the late 1990s a version for Microsoft Windows was finally released. Initially named “Epi Info 2000,” to distinguish it from the MS-DOS version, the name has subsequently been simplified to “Epi Info” again. Versions for other OSs are not available.
Epi Info does not generate specific reports or graphs designed for healthcare epidemiologists. Rather, it is a collection of tools that can be used in a wide variety of ways, including collecting and analyzing epidemiologic data. Functions are available to design data entry screens. Once designed, data can be entered, stored, and retrieved using the program. Internal data are stored in a relational database, which allows for sophisticated data storage if that is needed. Double data entry (see later Data Entry section) for ensuring data integrity is supported.
Epi Info provides an extensive range of statistical analysis tools. Analysis is not limited to data entered and stored using the software. The program can import and analyze data that have been stored in a number of database and spreadsheet formats. Results also can be displayed in a variety of graphical formats. Advanced statistical analysis, including logistic regression and Kaplan-Meier survival analysis, is available.
In addition to statistical analysis, Epi Info contains modules for nutritional anthropometry, mapping of data using geographic information system (GIS) standards, and even a simple word processor for producing reports. Additional modules can be developed by programmers outside the CDC to extend the program's capabilities. The software allows for data entry on different computers with
P.125
later merging of the data into one dataset. Overall, Epi Info provides a framework for developing quite sophisticated systems for healthcare epidemiology–data gathering and analysis.
EpiData
EpiData (www.epidata.dk/) started as a Windows-friendly data entry program for the older MS-DOS versions of Epi Info. This program has now become EpiData Entry. It enables the user to design data entry screens and then enter and manage data. EpiData Entry supports double data entry. Once entered, the data can be exported in a variety of formats for additional analysis. This includes export into SAS, SPSS, and Stata formats. As such, EpiData Entry is a good companion program to a general statistical package to provide the data entry and management functions the statistical programs lack.
EpiData Analysis is a newer program released in the fall of 2005. It provides additional data management functions beyond those offered by EpiData Entry. It also provides some basic descriptive statistical analysis and graphical functions, including statistical process control (SPC) charts. Although not presently as full featured as Epi Info, additional analysis functions are apparently planned.
Other Software and Resources
Sometimes an epidemiologist just needs to do a quick calculation, such as a 2 × 2 table. Such a simple task can actually be tedious to do in a full-fledged statistical package. Epi Info has a StatCalc module that is designed for quick calculations; however, it is based on the old MS-DOS version and is not as user friendly as more modern programs. A similar and free program that is designed specifically for Microsoft Windows is EpiCalc 2000, written by Mark Myatt (www.brixtonhealth.com; this Web site also contains links to a wide variety of other programs that may be of use to healthcare epidemiologists). For those with ready access to the Internet, the OpenEpi project (www.openepi.com) provides epidemiology analysis tools available in any Web browser.
Many more software packages that are of use to healthcare epidemiologists exist than can be described here. Many of the Web pages mentioned provide links to additional resources. A source of many links for statistical analysis is statpages.org/. A quick search using an Internet search engine can yield a wide variety of additional options.
Data Entry and Integrity
In order to use computers for any form of data analysis, the data must first be entered in a form that the computer can access. For a stand-alone computer, this is typically done through keyboard entry. If entering a large amount of data (e.g., transcribing information from paper data collection forms), double data entry can provide a higher level of assurance that the data have been entered correctly. This entails entering every form into two separate copies of the same database (often done by two different people) and then comparing the results to ensure that they are identical. Any discrepancies are resolved by referring back to the original data form. Not all software is capable of allowing for double data entry, but it is a nice feature to use to provide the highest level of data integrity. The EpiData Web site (www.epidata.dk/documentation.php) has a variety of publications that describe additional good practices for data management that can be downloaded.
With the increasing use of computers in healthcare facilities, more and more data needed by the healthcare epidemiologist already is available in electronic form. Patient demographics may be available from the admissions office, information on surgical procedures may be available from the operating room scheduling system, and so on. As noted previously, depending on what software is being used, data may need to be converted into a specific format to be imported. A variety of de facto standards exist (e.g., many programs can export and import data in Microsoft Excel format). If data are not already available in a format that can be used directly, frequently the hospital's IT department can write a program to convert the information into a useable format. As electronic medical records [6] become more commonly used, even more information may be available to the healthcare epidemiologist. It will be essential to work with the appropriate committees as an electronic medical record is being implemented at an institution to ensure that important data (e.g., presence of a central venous catheter on a given day) are being captured in a manner that allows for easy data querying and aggregation [7].
For items that are not available electronically, a variety of methods beyond simple paper forms have been used to collect the information. These include using forms that can be scanned in rather than requiring manual keyboarding of the information [8]. Small handheld computers (personal digital assistants, or PDAs) also have been used to collect data at the point of care [9]. Some software for PDAs allows for selecting choices from a variety of options, thus making data entry easier than entering words character by character. Such a system has been used at Barnes-Jewish Corporation (BJC) HealthCare member hospitals to report information to the system IC and Healthcare Epidemiology Consortium [10].
Once data have been entered by whatever means, provisions must be made to prevent data loss. An external means of backing up data should be considered essential. A variety of options are available, including external hard drives, external tape drives (less common as hard drive prices have fallen), and recordable CDs or DVDs. If using a second hard drive, an external drive is preferred to a second internal drive so that an internal component failure (e.g., a short circuit in the power supply) does not damage both drives simultaneously. For networked computers, storage space may be available on their local network; this provides a good alternative to an external hard drive. All back-up
P.126
systems are useless if not utilized. Users should remember to make backups regularly, or even better, should have software in place that automates the process. Copies of data should be stored in a secure place; for the highest level of security, this could mean an off-site location. Data stored on network drives managed by the institutional IT department are typically backed up on a daily basis.
Electronic Surveillance
Computers were first introduced into most hospitals in the form of large main-frame computers used for financial systems; computers were out of reach of most healthcare epidemiologists. When PCs became available, they were quickly adopted for use in IC. Schifman reported in 1985 on the use of custom software at the Tucson Veterans Administration Medical Center (VAMC); microbiology data were entered by hand to generate positive culture rates by ward by site (sputum blood, urine), and the results for each month were compared to the previous 12-month average [11]. The landmark Study of the Efficacy of Infection Control (SENIC) [12] demonstrated the value of surveillance in IC. In addition, 1985 was the year that the first precursor program to Epi Info was released. This convergence of the demonstrated need to do surveillance for HAIs and the increased availability of relatively inexpensive computers and software to assist the process led to the steady integration of computers into hospital IC programs.
For most healthcare facilities, PCs are used to perform retrospective analysis on data that have been entered. Some amount of data may be pulled from a variety of electronic sources, but typically, all of this information constitutes “denominator data” for analysis. “Numerator data” (i.e., which patients actually have HAIs) are typically determined by manual surveillance performed by ICPs. Computers may aid their data entry, but much of the work of detection and recording is still manual. Although this remains largely true today, efforts continue to shift more of this work onto computers. Such shifts require that a substantial amount of data be readily available electronically.
At the most basic level, computers can provide surveillance assistance that is relatively simple but can provide considerable time savings. One such basic assistance is simple aggregation of microbiology data [13]. Many hospitals provide such reports to their IC departments; these reports are generated by the hospital's laboratory systems and either printed or sent electronically to IC.
Some hospitals have been able to provide even more computer assistance. The Later Day Saints (LDS) hospital in Salt Lake City, Utah, developed the Health Evaluation through Logical Processing (HELP) system, a ground-breaking computer support system for clinicians. By applying a variety of rules to available information (e.g., microbiology data, patient date of admission), the HELP system could detect patients likely to have an HAI [14,15,16]. At Barnes Hospital (now BJC Hospital) in St. Louis, Missouri, the GermWatcher system was developed in the early 1990s [17,18,19]. Positive microbiology cultures are evaluated by an expert system that ranks them as likely contaminants or not and prioritizes cultures for ICPs (e.g., a sputum culture growing an acid fast bacteria is higher priority than a urine culture with E. coli). This assistance allows the ICPs to focus on patients most likely to have significant HAIs and not have to spend time scrutinizing all positive cultures. A data warehouse developed in Chicago that included data from Cook County Hospital and two smaller hospitals was recently described [20]. One goal in developing this system was to facilitate similar computer-assisted surveillance. Other facilities have developed similar systems [21,22].
In addition to the systems that have been developed at large academic medical centers as described, similar capabilities now are available to hospitals that do not have the expertise to build their own systems. Some of the commercial IC software described earlier, such as MedMined, SETNET, QC Pathfinder, and Infection Control Assistant, have components that offer similar surveillance assistance [23,24,25,26].
Beyond simple assisted surveillance, efforts have been made to fully automate surveillance for nosocomial bloodstream infections (BSIs). A retrospective evaluation of blood cultures from six hospitals in the Boston area suggested that a simple system of rules applied to microbiology data alone had a sensitivity of 64% and a specificity of 98% [27]. A similar study using data from neonatal intensive care units in six New York City hospitals reported a sensitivity of 79% and a specificity of 96% [28]. In both instances, the negative predictive value was much better than the positive predictive value. This suggests the possibility that although neither system would be capable of fully automated BSI surveillance, both could eliminate patients not likely to have true BSIs, thus aiding ICPs in selecting patients whose charts should be reviewed. A study in two Chicago hospitals evaluated a variety of rule-based approaches to BSI, compared with ICP review, compared with “gold standard” review by physician investigators [29]. The best computer algorithm had a sensitivity of 81% and a specificity of 72%. However, the temporal trends of the computer algorithm over time tended to track well with the reference standard. This suggests that although the automated method may not give a “true” rate, it may be possible to follow trends over time without any active surveillance by an ICP. If rates become unexpectedly high, a specific investigation could be launched.
As with BSIs, efforts have been made to automate a significant portion of the effort required for SSI surveillance. Studies in Boston evaluated use of hospital discharge diagnoses combined with inpatient antibiotic administration data to detect SSIs after cesarean section [30]. A combination of this information had a positive predictive value of 94% with a sensitivity of 59%. This method was later
P.127
validated for coronary artery bypass graft (CABG) surgery at hospitals in Boston and Israel [31] and at a number of U.S. hospitals for CABG, breast, and cesarean section surgeries [32]. Overall sensitivities in the latter study ranged from 93% to 97%, but positive predictive values were only 20% to 42%. Predictive value was particularly sensitive to antibiotic prescribing practices at the various hospitals. As with some of the BSI approaches mentioned, such methods may still be very valuable by allowing ICPs to narrow their focus to the patients most likely to really have an SSI, thus making surveillance more efficient. The Boston investigators also have reported using similar surveillance methods to detect SSIs in patients after hospital discharge by evaluating health maintenance organization (HMO) data [33]. Given the difficulty of postdischarge surveillance, such methods hold some promise but may be limited by the availability of records. Another approach to assist (but not fully automate) SSI surveillance was instituted in a Paris hospital where computer surveillance for positive cultures from surgical patients was used as a screen [34].
In addition to HAI surveillance, hospital computers can assist with other IC tasks. Many hospitals have systems in place in which patients who are known to be carriers of antimicrobial-resistant organisms, such as methicillin-resistant Staphylococcus aureus (MRSA), can be flagged in a hospital computer system. This flag then triggers automatic placement into appropriate isolation precautions if the patient is readmitted. Typically, turning such flags on and off is itself a manual task. However, some automated systems (such as GermWatcher and others) can highlight patients who have positive cultures with such organisms, making it less likely that they will be missed and hence not flagged. Implementation of such a system in Geneva, Switzerland, resulted in significantly more patients being isolated at the time of admission [35]. An even more automated system for ensuring that patients are placed in appropriate isolation was described at Columbia-Presbyterian Medical Center. Automated computer protocols evaluated chest radiograph reports using a natural-language processor. The automated system also evaluated patients for evidence of immunocompromised status by evaluating laboratory data for evidence of human immunodeficiency virus (HIV) infection and pharmacy data for use of medications used only to treat HIV. This automated system was able to identify patients who should have been on isolation but were not [36].
Not all attempts to use computers to assist or automate HAI surveillance have been successful. In particular, the use of electronic hospital claims data alone has been shown to correlate poorly with true HAIs [37,38,39].
What Not to Automate
Computers have many advantages over older paper systems in areas of data analysis, data sharing, and data backups. However, computers crash and may not always be available, entering data can be time-consuming, and sometimes data sharing can mean multiple copies of the same information that have been changed by different people and are now incompatible. At many hospitals, old paper-based systems continue to work just fine. An example is keeping track of patients with antimicrobial resistant organisms using an index card system. Such a system may be accessed more readily than launching a database program, searching for the patient's entry, and then closing the program. For hospitals that have such older systems, the decision to put the information into electronic format may be a complicated one, especially if a large amount of legacy information will have to entered into the computer. A split system, whereby only new data, or only limited older data, are entered into the computer may be an option if old information only rarely needs to be accessed. Likewise, changes to newer versions of software, or changes to “more automated” versions of processes that are already somewhat computer assisted may lead to changes in work flow that actually make the overall process less efficient. Any decision to “computerize” or upgrade an existing process should be based on a careful analysis of expected benefits and a realistic assessment of the negative aspects of the change. Changes should not be made simply for the sake of change or to have the “latest technology.” Unfortunately, in some instances, decisions by software vendors to no longer support older products will force a hospital epidemiology program to upgrade even if there are no real benefits to the process.
Conclusions
Computers are ubiquitous in modern life, including in healthcare epidemiology. As computers become more powerful and as more information becomes available electronically, healthcare epidemiologists can anticipate an even greater use of automated tools. Still, computers are only tools. At their best, they can perform the tedious portions of surveillance and calculations for us, but it will still take a person to interpret the data and use it to improve patient care.
Acknowledgment
Thanks to Ashleigh Goris for her assistance with this chapter.
References
P.128