HIV prevalence over time is a critical metric for understanding the effectiveness of programs aiming to prevent HIV. Prevalence is often measured using surveillance of clinic patients, which can lead to selection bias: clinics located in areas of high HIV prevalence are often the first to be monitored by the surveillance systems, distorting the estimated HIV prevalence based on clinic data. To help understand the impact of selection bias on the estimation of HIV prevalence trends, researchers compared the efficacy of two approaches for handling selection bias.
The researchers used simulations to compare and contrast various approaches and models for handling selection bias in clinic-based HIV surveillance, focusing on two methods: complete-case analysis and multiple imputation. They then applied these methods to surveillance data collected by the National AIDS Control Organization of India between 2002 and 2008.
Data simulations suggested that selection bias can lead to biased estimates of HIV prevalence trends and inaccurate evaluation of the impact of intervention programs. The two methods studied differed in their ability to handle selection bias, with complete-case analysis providing relatively unbiased and stable estimates, in contrast with multiple imputation.
Because selection bias is common in clinic-based HIV surveillance, the authors conclude that appropriate adjustment methods need to be used when analyzing HIV prevalence data. However, the choice of adjustment method matters, with complete-case analysis performing better than the other adjustment methods studied. When analyzing data from clinic-based surveillance systems, failure to take selection bias into account can lead to biased estimates of the magnitude of decline in HIV prevalence and the impact of an intervention program.
Ng M, Gakidou E, Murray CJL, Lim SS. A comparison of missing data procedures for addressing selection bias in HIV sentinel surveillance data. Population Health Metrics. 2013; 11:12.