Dissertation Title: "Modeling Health Outcomes in Resource-limited Settings"
This dissertation presents methods for improving estimates of important health outcomes in resource-limited settings. Paper 1 develops a new approach to estimating health-state valuations from ordinal responses, which may be easier to collect than cardinal information. The model combines paired comparison and time trade-off data in a single likelihood. This approach offers a solution to the scaling problem common to existing ordinal models for health-state valuation and highlights the importance of considering the variability in respondents’ valuations when using ordinal methods. Simulations indicate that ordinal designs that employ this multi-method approach may have comparable efficiency to cardinal ones and could therefore facilitate the measurement of health-state valuations in populations with low numeracy.
Paper 2 proposes modifications to the dynamic model used by UNAIDS to generate annual updates on the global HIV/AIDS epidemic. The current model fails to reproduce recent trends in HIV/AIDS prevalence in several countries. To address this, I explicitly model changes in average infection risk over time, using penalized B-splines in a Bayesian framework, and include an informative prior distribution for infection risk beyond the last year of data, which enhances the plausibility of short-term extrapolations. This more flexible model produces better fits than the current model to recent HIV/AIDS prevalence trends in several Sub-Saharan African countries and can be incorporated easily into the UNAIDS modeling framework.
Paper 3 employs Heckman-type selection models to correct for selection on unobserved factors when imputing missing HIV status for those who did not participate in HIV testing for 18 Demographic and Health Surveys in Sub-Saharan Africa. Estimates of national HIV/AIDS prevalence that were corrected for selection on unobserved factors were substantially higher for some countries but did not show a consistent pattern across surveys in comparison to estimates from conventional imputation. The results indicate that important uncertainty remains around HIV/AIDS prevalence estimates in many Sub-Saharan African countries and more emphasis should be placed on increasing HIV testing participation in surveys that aim to establish national HIV/AIDS prevalence rates.