Flexible Statistical Models with Applications in Reproductive Health

Abstract

Statistical modeling consists of approximating or representing the data generating process through mathematical models while incorporating a set of distributional assumptions for the sampled data. Our research was focused on a class of two-part models that takes into account necessary complexities such as two or more data generating processes and heterogeneity. The models were applied to data on facility-based maternal mortality and HIV/AIDS in Mozambique. Often, the analysis of data on maternal mortality involves the use of counts of maternal deaths as outcome variable, which usually presents an excessive number of zeros. Classical model approaches to these type of data fails to accommodate both the fact that the data presents more zero counts than expected and the overdispersion (variance greater than the mean) induced by it or by group or individual heterogeneity, in order to have unbiased estimates and valid inferences. We extended the zero-inflated models to deal with excessive zero counts while taking additional hierarchical data structures into account. Zero-inflated models assume that for each observation there are two possible data generating processes with different probabilities: one generates a zero (not-at-risk sub-population) and the other the counts (at-risk sub-population). One key finding of our analysis with facility-based maternal mortality rate was its dependence on the location of the health facility in relation to the district capital. This points mainly to differences in healthcare service utilization between rural and urban (or semi-urban) areas specially in the South and North of Mozambique, an area of research that is currently very active in the country. There is a need to further investigate the association between delay to receive pregnancy-related treatment and facility-based maternal mortality. Our research showed that in well-resourced health facilities the expected survival time of patients is shorten by an increase in waiting time to received treatment.