A statistical calibration framework for subnational health surveillance from routine data and survey audits

In many low- and middle-income countries, where civil registration and vital statistics remain incomplete, decision makers rely on estimates derived from large international household surveys. Those surveys are rarely powered for frequent subnational estimates, and the contraction in international aid has made their future uncertain. Small area estimation has supplemented them with increasingly available routine health records and satellite imagery to improve their precision, but the survey estimate remains the response variable. We propose a statistical calibration framework that inverts the roles of the two data sources, whereby the routine records provide a biased baseline estimator continuously over time at the subnational level of interest, while a probability survey is used intermittently to audit and adjust for the selection bias of the direct register estimator. We develop the corresponding modelling and estimation methods, including diagnostic and analysis tools for the transport assumption that the out-of-survey subpopulations and time points require. The proposed framework is applied in four African countries to estimate undiagnosed HIV positivity by district and the selection bias by age and sex, and we explain how it applies to other health surveillance needs with similar data settings.