APH 1999, 57, 89-105:

Adjusting for confouding when estimating a time trend in HIV prevalence based on pooled serum samples.

S. Vansteelandt, E. Goetghebeur, and T. Verstraeten

Keywords: epidemiologic research design, HIV seroprevalence, sentinel surveillance, seroepidemiological methods, standardization, statistical models   

Over the last decade, many epidemiological studies have demonstrated the successful use of pooled sera for screening purposes or HIV risk estimation (1-1l). The method was originally designed as a cost-reductive tool, but also appears to lower the error rates associated with diagnosis and in low prevalence areas it produces, at most, a slight loss in the estimation accuracy. In this paper, we use test results on pooled sera to estimate a time evolution in HIV prevalence where we need to account for the presence of important confounders. An adjusted time trend estimate is proposed, assuming confounding variables are measured for each subject, but there is only one diagnostic test result per pool.

Experimental design is important if one is to achieve a precise and cost-efficient estimate of an evolution of risks over time. Specifically, the distribution of known prognostic factors over the pools is influential and can be influenced. Choosing pools to be covariate-homogeneous increases the amount of information at virtually no additional cost. The cost-precision balance is further optimized by calculating pool sizes in function of the distribution of covariates over the pools.

Our study was motivated by the planning of a growing database to monitor a time trend in HIV prevalence (14). The methods are used to adjust for age as a confounder in data obtained from Kenyan pregnant women. Analysis of the Kenyan data shows that age homogeneous pools of optimal size reduce cost to 44% of the original price, whilst precision remains close to the one obtained from non-pooled samples.