Far from MCAR: Obtaining Population-Level Estimates of HIV Viral Suppression

• Activity Number: 545 - Machine Learning and Nonparametric Methods in Causal Inference
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Laura B. Balzer* and SEARCH Collaboration
• Companies: UMass Amherst and Makerere University - UC San Francisco
• Keywords: causal inference; HIV; machine learning; missing data; Super Learner; TMLE

Abstract

Population-level estimates of disease prevalence and control are needed to assess prevention and treatment strategies. However, available data are often subject to differential missingness. We discuss 3 sets of assumptions to identify population-level HIV viral suppression: the proportion of all HIV+ persons who are suppressing viral replication. Using data on nearly 100,000 participants in the SEARCH Study (NCT01864603) in rural Kenya and Uganda, we compare estimates from an unadjusted approach assuming data are missing-completely-at-random (MCAR); stratification on age group, sex, and community; and, targeted maximum likelihood estimation (TMLE) with Super Learner to adjust for baseline and time-updated factors. Despite high levels testing, estimates varied by approach. Unadjusted estimates were most optimistic: 50% at baseline, 80% at Year 1, 85% at Year 2, and 85% at Year 3. Stratification on baseline predictors yielded slightly lower estimates, and full adjustment reduced estimates meaningfully: 42% at baseline, 71% at Year 1, 76% at Year 2, and 79% at Year 3. Estimates relying on the MCAR assumption or baseline stratification should be interpreted with caution.