Linking household surveys and health facility assessments to estimate intervention coverage for the Lives Saved Tool (LiST)

Abstract Background Calls have been made for improved measurement of coverage for maternal, newborn and child health interventions. Recently, methods linking household and health facility surveys have been used to improve estimation of intervention coverage. However, linking methods rely the availability of household and health facility surveys which are temporally matched. Because nationally representative health facility assessments are not yet routinely conducted in many low and middle income countries, estimates of intervention coverage based on linking methods can be produced for only a subset of countries. Estimates of intervention coverage are a critical input for modelling the health impact of intervention scale-up in the Lives Saved Tool (LiST). The purpose of this study was to develop a data-driven approach to estimate coverage for a subset of antenatal care interventions modeled in LiST. Methods Using a five-step process, estimates of population level coverage for syphilis detection and treatment, case management of diabetes, malaria infection, hypertensive disorders, and pre-eclampsia, were computed by linking household and health facility surveys. Based on data characterizing antenatal care and estimates of coverage derived from the linking approach, predictive models for intervention coverage were developed. Updated estimates of coverage based on the predictive models were compared, first with current default proxies, then with estimates based on the linking approach. Model fit and accuracy were assessed using three measures: the coefficient of determination, Pearson’s correlation coefficient, and the root mean square error (RMSE). Results The ability to predict intervention coverage was fairly accurate across all interventions considered. Predictive models accounted for 20–63% of the variance in intervention coverages, and correlation coefficients ranged from 0.5 to 0.83. The predictive model used to estimate coverage of management of pre-eclampsia performed relatively better (RMSE = 0.11) than the model estimating coverage of diabetes case management (RMSE = 0.19). Conclusions The new approach to estimate coverage represents an improvement over current default proxies in LiST. As the availability of reliable coverage data improves, impact estimates generated by LiST will improve. This study underscores the need for continued efforts to improve coverage measurement, while bringing to the fore the importance of health facility assessments as complementary data sources