| In this study, we correct two existing estimates of the effect of highly active antiretroviral therapy (HAART) on time to AIDS or death for two known types of bias. Using data from the AIDS Clinical Trial Group (ACTG) 320 study, we correct for noncompliance in a randomized clinical trial. Using data from the Multicenter AIDS Cohort Study (MACS), we correct for exposure misclassification by calendar period in an age-period analysis of an observational study. In 1996-7, the ACTG 320 study randomized 1,156 HIV-infected patients to HAART or combination therapy with equal probability. 96 patients incurred AIDS or died, 51 dropped out, and 290 dropped out or stopped their assigned therapy for reasons other than toxicity during a 52 week follow-up. Comparing HAART to combination therapy, the intent-to-treat hazard ratio was 0.51 (95% confidence limits [CL]: 0.33, 0.77). To correct for noncompliance and to ameliorate non-proportional hazards in the standard analysis, we apply Robins and Finkelstein's inverse probability-of-censoring weights. Comparing HAART to combination therapy, the inverse probability-of-censoring weighted hazard ratio was 0.45 (95% CL: 0.27, 0.74).;We report on 614 HIV+ homosexual men followed from 1984 to 2007 participating in the MACS. 268 of 614 men incurred AIDS, 49 died, and 90 were lost to follow-up during 5,321 person-years. Comparing the non-HAART calendar period (<1996) to the HAART calendar period (≥1996), the naive rate ratio was 3.62 (95% CL: 2.67, 4.92). To correct the bias due to misclassification of HAART use by calendar period, we adapt an instrumental variable estimator. Comparing the non-HAART calendar period to the HAART calendar period, the instrumental variable rate ratio was 5.02 (95% CL: 3.45, 7.30).;Finally, we measured the validity of self-reported antiretroviral therapy use in 68 HIV+ men from the Baltimore/Washington site of the MACS. Using prescribed therapies as a gold standard, the estimated sensitivity, specificity, positive predictive value, negative predictive value, and kappa of self-report were 0.94 (95% CL: 0.84, 0.99), 0.80 (95% CL: 0.52, 0.96), 0.94 (95% CL: 0.84, 0.99), 0.80 (95% CL: 0.52, 0.96), and 0.74 (95% CL: 0.55, 0.91), respectively. Therefore, the validity of self-reported HAART use is relatively high. |
