Application of the case-crossover design to reduce unmeasured confounding in studies of condom effectiveness.

This analysis examined how unmeasured confounding affects estimates of the effectiveness of condoms in preventing sexually transmitted infections. Data were analyzed from a prospective cohort study of 1,122 female sexually transmitted disease clinic patients in Alabama (1992-1995), wherein participants were evaluated for sexually transmitted infections at six 1-month intervals. Associations between condom use and incident gonorrhea and chlamydia infection were compared between case-crossover and cohort analyses. In a case-crossover analysis of 228 follow-up visits ending in gonorrhea/chlamydia ("case intervals") and 743 self-matched follow-up visits not ending in gonorrhea/chlamydia ("noncase intervals") (183 women), consistent condom use without breakage or slippage was associated with significantly reduced risk of infection relative to nonuse (adjusted risk odds ratio = 0.49, 95% confidence interval: 0.26, 0.92). Conversely, a cohort analysis of 245 case intervals and 3,896 noncase intervals (919 women) revealed no significant reduction in infection risk from consistent use of condoms (adjusted risk odds ratio = 0.79, 95% confidence interval: 0.53, 1.17). Dose-response relations between the number of unprotected sex acts and infection were stronger in the case-crossover analysis (p for trend = 0.009) than in the cohort analysis (p for trend = 0.18). These findings suggest that epidemiologic studies confounded by unmeasured differences between condom users and nonusers underestimate condom effectiveness against these infections. The case-crossover method provides an additional technique for reducing unmeasured confounding in studies of condom effectiveness.