The effect of disease misclassification on the ability to detect a gene-environment interaction: implications of the specificity of case definitions for research on Gulf War illness

Background: Since 1997, research on Gulf War illness (GWI) has predominantly used 3 case definitions—the original Research definition, the CDC definition, and modifications of the Kansas definition—but they have not been compared against an objective standard. Methods: All 3 case definitions were measured in the U.S. Military Health Survey by a computer-assisted telephone interview in a random sample (n = 6,497) of the 1991 deployed U.S. military force. The interview asked whether participants had heard nerve agent alarms during the conflict. A random subsample (n = 1,698) provided DNA for genotyping the PON1 Q192R polymorphism. Results: The CDC and the Modified Kansas definition without exclusions were satisfied by 41.7% and 39.0% of the deployed force, respectively, and were highly overlapping. The Research definition, a subset of the others, was satisfied by 13.6%. The majority of veterans meeting CDC and Modified Kansas endorsed fewer and milder symptoms; whereas, those meeting Research endorsed more symptoms of greater severity. The group meeting Research was more highly enriched with the PON1 192R risk allele than those meeting CDC and Modified Kansas, and Research had twice the power to detect the previously described gene-environment interaction between hearing alarms and RR homozygosity (adjusted relative excess risk due to interaction [aRERI] = 7.69; 95% CI 2.71–19.13) than CDC (aRERI = 2.92; 95% CI 0.96–6.38) or Modified Kansas without exclusions (aRERI = 3.84; 95% CI 1.30–8.52) or with exclusions (aRERI = 3.42; 95% CI 1.20–7.56). The lower power of CDC and Modified Kansas relative to Research was due to greater false-positive disease misclassification from lower diagnostic specificity. Conclusions: The original Research case definition had greater statistical power to detect a genetic predisposition to GWI. Its greater specificity favors its use in hypothesis-driven research; whereas, the greater sensitivity of the others favor their use in clinical screening for application of future diagnostic biomarkers and clinical care.