A case-control sampling strategy for zero-inflated models with an application to female sex worker mapping in sub-Saharan Africa

Music City Center 

Eastern and Southern Africa bear a disproportionately large share of the global HIV burden. Although this region comprises only about 6.2% of the world's population, it accounts for 45% of new HIV infections worldwide. Historically, HIV programs in these regions have focused almost exclusively on the general population, with limited attention to the distinct needs of key populations. With a recent shift towards more targeted interventions, our work leverages the unique PLACE datasets to produce granular size estimates of female sex workers, facilitating more effective and efficient HIV program planning. Because of social stigma and discrimination towards FSW, it is difficult to measure or estimate the size and location of FSW at any spatial resolution, especially a fine-scale resolution. In this study, we develop a generalized linear mixed-effect model to estimate the female sex worker population at the grid-cell level and propose a case-control sampling strategy to speed up the computation of the model fitting. This sampling approach broadly applies to zero-inflated models with large sample sizes. We demonstrate the approach's efficiency and accuracy through simulation studies and analyses of PLACE data, and we establish the theoretical properties of the optimal sampling procedure. With our proposed model and identified demographic variables, we obtain a reasonable estimation of female sex workers' distribution across four Eastern and Southern African countries.

Cause-Specific Mortality