subject to recall or reporting bias (see Chapter 3) even when objective measures of health status are collected. The potential for bias is increased in studies in which both exposure and health-outcome information is based on self-reports. The use of self-reported exposure information was unavoidable in most Gulf War literature, but a number of attempts have been made to compare self-reported exposures with other estimates of exposure. Although such alternative estimates might appear to be more objective, for most comparisons between self-reported exposure and other measures there is no “gold standard” of exposure. Accordingly, the studies simply report on comparisons of different estimates of exposure rather than provide an objective assessment of the validity of self-reports. Exceptions might include exposure to vaccinations, for which records are available, and exposure to depleted uranium (DU), which can be verified with biologic monitoring. The following sections describe comparisons between self-reported and other measures of exposure.
A number of studies have, at least indirectly, examined the validity of self-reported exposures. With respect to the oil-well fires, Lange et al. (2002) reported moderate correlations (r = 0.4 and 0.5) between self-reports of low and high exposures to oil-fire smoke as assessed with a dispersion model linked to troop-unit location information. At each level of self-reported exposure (based on the number of days exposed), the modeled exposures were highly variable. Cowan et al. (2002) reported a low interclass correlation coefficient (kappa) of 0.13 for self-reported exposure to oil-fire smoke vs cumulative modeled exposure (according to the model used by Lange et al.) to oil-fire smoke or days with high modeled exposure. Wolfe et al. (2002) reported that responses to a yes-no question regarding oil-fire exposure in the Fort Devens cohort did not correlate well with modeled particle exposures. Higher correlations were found when information regarding the self-reported frequency, duration, and intensity of exposure was considered.
The strongest analysis of reporting bias with regard to vaccine exposure was conducted by Mahan et al. (2004) in their study of anthrax vaccination. Veterans were asked whether they received anthrax vaccination or were uncertain about receiving it. In a cohort of 11,441 Gulf War veterans who completed a health and exposure survey, 352 respondents also were on a Department of Defense (DOD) list of 7,691 people who were vaccinated at least once. The list was compiled from several sources and is the largest compilation of Gulf War veterans identified as receiving anthrax vaccination. In the full cohort, 4,601 (40%) reported receiving the vaccine, 2,979 (26%) reported not receiving it, and 3,861 (34%) were uncertain. Of the subset of 352 who were on the DOD vaccination list, 260 (74%) reported receiving the vaccine, 34 (10%) reported not having received it, and 58 (16%) reported that they were uncertain. This comparison indicates a 26% false-negative rate, but the lack of a documented “nonvaccinated” group makes it impossible to determine the false-positive rate. The study also provides some evidence of reporting bias.
Although immunization history was self-reported in most studies, Unwin et al. (1999) asked survey respondents to refer to their own vaccination records, if available, in a study of UK veterans deployed to the gulf compared with those deployed to Bosnia or other Gulf-War-era veterans. Some 32% of the Gulf War veterans in the survey reported that they had vaccination