Horner and colleagues (2003) conducted a nationwide, epidemiologic case-ascertainment study to determine if Gulf War veterans have elevated rates of ALS relying on both active and passive methods of case ascertainment. Active methods included screening of inpatient, outpatient, and pharmacy medical databases of VA or DoD. Passive methods included establishment of a toll-free telephone number, solicitations through relevant Internet sites, and mass mailings of study brochures to practicing neurologists in the VA and to members of the American Academy of Neurology. The ALS diagnosis in study participants was verified by medical record review.
Among nearly 2.5 million eligible military personnel active during the Gulf War period, nearly 700,000 had been deployed to the Gulf War between August 1990 and July 1991. Mostly based on active ascertainment methods, this study identified 107 eligible cases of ALS (40 in the Gulf War deployed and 67 in the nondeployed troops), for an overall occurrence of 0.43 per 100,000 persons per year between August 1990 to December 1999. The ALS risk was estimated to be about twofold for all deployed compared to the nondeployed military personnel (RR 1.92, 95% CI 1.29-2.84) with an attributable risk (that is, the risk difference or excess risk) due to deployment estimated as 18% (95% CI 4.9-29.4). The foremost limitation of the study was potential underascertainment of cases, particularly among nondeployed veterans, because nondeployed veterans had less incentive to participate. Because of the rarity of ALS, underascertainment of a few cases, particularly if it is greater among the nondeployed, can substantially exaggerate the risk among the deployed by comparison. The finding that the incidence of ALS in deployed veterans was actually lower than that of an age-matched sample from Washington state (McGuire et al., 1996) contributed to this concern; however, such a difference between military personnel and the general population might also reflect a healthy warrior bias. Another but less important study limitation raised in a letter by another ALS researcher was failure to consider smoking as a possible confounding factor in the study design under the assumption that smoking is a risk factor for the development of ALS (Armon et al., 2004; Nelson et al., 2000). In response, the authors of the original study pointed out that there is little reason to believe that smoking rates are different among deployed and nondeployed veterans reducing the likelihood for confounding bias.
More importantly, however, the same authors undertook a secondary analysis to address concerns about differential case ascertainment among deployed versus nondeployed veterans. In this secondary analysis (Coffman et al., 2005) they assessed case ascertainment bias by estimating the occurrence of ALS employing three capture–recapture analysis methods: log-linear models, sample coverage, and ecologic models. The investigators concluded that there might have been some modest underascertainment of cases in nondeployed military personnel but little underascertainment in the deployed. After correcting the rates for underascertainment, the investigators still found a higher age-adjusted risk of ALS among the deployed than among the nondeployed (RR 1.77, lower bound 1.21, with log-linear model). These analyses confirmed the original findings of an increase in ALS among deployed veterans assuming that the modeling assumptions they had to make for this exercise are correct. See Table 4-6 for a summary of the primary ALS studies discussed above.