that is important because the committee did not want to factor in the health outcomes occurring in the same people repeatedly.
The 24 major cohort studies of Gulf War veterans and their derivative studies have contributed greatly to our understanding of veterans’ health, but they are beset by limitations that are commonly encountered in epidemiologic studies, including lack of representativeness, selection bias, lack of control for potential confounding factors, self-reports of health outcomes, outcome misclassification, and self-reports of exposure. The committee members read each study carefully and noted the findings and limitations of each study.
The foremost limitation is lack of representativeness, which limits one’s ability to generalize results to the entire population of interest; for example, about half the cohorts focus on groups of veterans that are selected for study according to where they served in the military (a military-unit-based study). Military-unit studies are not representative of all Gulf War veterans with respect to their duties and location during deployment, their military status during the war (active duty, reserves, or National Guard), their military status after the war (active duty, reserves, or discharged), their branch of service (Army, Navy, Air Force, or Marines), or ease of ascertainment (IOM 1999b). The most representative studies are population-based: the cohorts are selected on the basis of where their members reside. In population-based studies of Gulf War veterans, the cohort might be the entire deployed population, as in studies of Canadian and Australian veterans, or a random selection from the population of interest, as in several studies of US and British veterans. The committee, in evaluating major cohort studies, gave greater weight to Gulf War studies that were population-based.
A study’s representativeness, even if it is population-based, can be compromised by low participation rates. Low participation rates can introduce selection bias, for example, when Gulf War veterans who are symptomatic choose to participate more frequently than those who are not symptomatic. Nondeployed veterans, who might be healthier, might be less inclined to participate. In some studies, researchers not only try to measure selection bias by comparing participants with nonparticipants from both deployed and nondeployed populations, but also make adjustments to overcome it, for example, by oversampling nondeployed populations as in the study by Eisen and colleagues (2005).
Selection bias might also occur through the so-called healthy-warrior effect. That form of bias has the potential to occur in most of the major cohorts that compare deployed veterans with nondeployed personnel. The healthy-warrior effect is a form of selection bias insofar as chronically ill or less fit members of the armed forces might be less likely to have been deployed than more fit members. That is, there might have been nonrandom assignment of those selected and not selected for deployment. Some of the best studies attempt to measure the potential for selection bias and adjust for it in the analysis.
A recurrent limitation is that most cohort studies rely on self-reporting of symptoms on questionnaires. Symptom self-reporting potentially introduces reporting bias, which occurs when the group being studied (such as deployed veterans) reports more frequently what it remembers than a comparison group (such as nondeployed veterans). Reporting bias, in this example, would lead to an overestimation of the prevalence of symptoms or diagnoses in the deployed population.