Robert W. Haley, M.D.
Epidemiology Division, Department of Internal Medicine,
University of Texas Southwestern Medical Center, Dallas, Texas
Two Federally sponsored studies of birth defects compared
rates of birth defects in all 695,000 Gulf War veterans with rates in
a 50% random sample of the non-deployed era veterans. Tests of statistical
significance were calculated assuming that the groups represented random
samples from an infinite population, or a "super-population"
Since the conditions to which Gulf War veterans were exposed in the war
were unique and a primary purpose of the studies was to determine whether
Gulf War veterans are to be compensated for illnesses and birth defects,
the populations must be treated as finite populations, at least for settling
the question of compensation. Consequently, estimates of variance for
risk measures must be adjusted for the probability of selection. Since
all members of the deployed group were studied, the risks of birth defects
measured in this group are known exactly, with no sampling error, and
since a 50% random sample of the non-deployed were studied, the risks
in this group are known with only 30% of the usual sampling error. Cowan
et al. found the relative risk of congenital hip dysplasia associated
with deployment to be 1.31 in the offspring of the men and 1.38 in those
of the women. The authors, assuming a super-population model, reported
both relative risks as not statistically significant, with 95% confidence
intervals (CI) of 0.94-1.83 for the men and 0.67-1.81 (sic) for the women.
After adjusting for the finite population correction, both relative risks
were statistically significant (95% CI 1.10-1.56 for the men and 1.01-1.88
for the women). With the effect being homogeneous across gender (p=.9
by Wolfe's test for homogeneity), the combined relative risk was right
at statistical significance even without adjusting for the finite population
correction (Mantel-Haenszel [MH] RR 1.32, 95% CI 0.98-1.79; MH X2=3.24,
df=1, p=.07) and highly statistically significant after adjusting for
the finite population correction (MHfpc RR 1.32, 95% CI 1.15-1.51; MHfpc
X2=16.11, df=1, p=.00006). Given that the authors tested for the effects
of war exposure on the six most common birth defects in the men and the
seven most common in the women (32), for a result to be statistically
significant despite the multiple hypothesis testing one might require
its p value to be below the most conservative Bonferroni-adjusted value
of p=.0038 (.05/13), and it remains statistically significant by this
criterion. Araneta et al. reported a 3-fold higher risk of Goldenhar syndrome
to be non-significant (RR 3.03; 95% CI 0.63-20.57); however, when adjusted
for the finite population correction it is statistically significant (RRfpc
3.03, 95% CI 1.14-8.08). Given that ill Gulf War veterans disproportionately
left the military soon after the war and their offspring were selectively
lost to followup by the surveillance confined to military hospitals, the
excess risks of birth defects were underestimated. I conclude that offspring
of the deployed veterans had higher risks of some birth defects than those
of the non-deployed.
Meta-Analysis; Birth Defects; Finite Population Correction
No support but findings relevant to Federally supported
Gulf War veterans illnesses research
|