|
Abstract:
|
To monitor and combat the spread of HIV and other infectious diseases, public health workers need accurate incidence estimates. Cohort studies are expensive and time-consuming, and may not provide reliable incidence estimates because of various biases. Instead, sequential testing of blood sera for biomarkers using multi-assay classification algorithms (MAAs) can be used to detect recent infections in cross-sectional studies. However, biomarker-based studies are not free from sampling complications; in particular, nonrandom missingness of biomarker data requires careful handling. Even if individual biomarkers are missing at random, the overall outcome of an MAA may be missing not at random. We present a method for handling such a missingness pattern, which affects the estimates of the incidence-estimation parameters known as mean window period, shadow, and false-recent rate. We also compare several alternative definitions for the false-recent rate, examine how these differences lead to different estimators for this parameter, and determine how these estimators need to be adjusted for the missingness pattern. We apply these methods to a multi-country study of HIV clade C infections.
|
