To confront the spread of HIV and other infectious diseases, public health workers need accurate incidence estimates. Cohort studies are expensive, time-consuming, and vulnerable to several sources of bias. Instead, multi-assay algorithms (MAAs) for detecting recent infections from blood serum biomarkers can be used to estimate incidence from cross-sectional studies. However, before being applied to a target population, MAAs must be calibrated for that population, using a data set of biomarker measurements with known durations of infection. Ideally, the calibration data set is sampled from a population similar to the target population, such as the same demographic area at an earlier time point; however, as an epidemic and its corresponding public health response evolve over time, data might not be available from a sufficiently similar population to enable direct calibration. We present a method of adjusting the calibration analysis for differences in covariate distributions between the calibration data set and target population. We assess the accuracy and robustness of our adjustment method using a simulation framework, and we apply it to a recent study of HIV subtype B infections.