Objectives: To evaluate a method for adjusting estimates of HIV prevalence based on data from a program for the prevention of mother-to-child transmission (PMTCT) of HIV infection for the potential bias attributable to refusal of PMTCT-related testing.
Methods: Age-specific logistic regression models were used to estimate the HIV risk coefficients for 10 predictor variables among women who accepted the PMTCT-related testing (n = 1874) at an antenatal clinic in northern Uganda. These risk coefficients were used to predict the prevalence among women who were not tested (n = 1719) and to adjust the PMTCT-based prevalence for nonparticipation bias. Crude and adjusted PMTCT-based prevalence estimates were compared with the prevalence among women who were anonymously tested as part of routine sentinel surveillance (n = 2225).
Results: The PMTCT-based prevalence represented an underestimate compared with that based on anonymous surveillance in 2004 (9.0% vs. 10.5%); in 2005, it constituted an overestimate (11.8% vs. 10.9%). Adjusting the PMTCT-based prevalence reduced the difference attributable to nonparticipation bias by approximately 70% in both years, so that the adjusted prevalence (10.1% in 2004 and 11.2% in 2005) was similar to the surveillance-based prevalence.
Conclusions: The adjustment method was effective in reducing the nonparticipation bias. Further studies are needed to assess the utility of PMTCT program data for HIV surveillance.