Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier

https://orcid.org/

0000-0002-0261-3615

AccessType

Open Access Dissertation

Document Type

dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Public Health

Year Degree Awarded

2022

Month Degree Awarded

February

First Advisor

Leontine Alkema

Abstract

Estimation of health indicators globally is complicated because of data sparsity and data quality issues, especially in low and middle income countries without well-functioning registration systems. This dissertation introduces Bayesian methods for the estimation of stillbirth rates and adult mortality in data-limited settings.

Motivated by statistical challenges in the estimation of stillbirth rates globally, we develop a Bayesian hierarchical temporal sparse regression model (BHTSRM). Bayesian hierarchical temporal regression models combine a hierarchical regression model with a temporal smoothing process. This type of model has been used for estimating health indicators for multiple populations in data-sparse settings to track high-quality data while producing covariate-driven estimates for populations with limited or no data. To extend its usage to settings where the number of candidate covariates is large relative to data availability, we propose the use of BHTSRMs that impose sparsity by using horseshoe priors on regression coefficients. We also develop a method to adjust observations with alternative stillbirth definitions and account for varying levels of uncertainty associated with different data sources in fitting the BHTSRM to stillbirth data. The proposed model has been used by the United Nations to estimate stillbirth rates globally.

To facilitate prediction based on BHTSRMs, we propose an associated variable selection method: horseshoe shrinkage parameter reference distribution variable selection (HSS-VS). We check the performance of the new method through simulation exercises and use it for variable selection in the estimation of stillbirth rates.

In low and middle income countries without well-functioning registration systems, sibling survival history (SSH) data can be used to estimate adult mortality but it may be subject to substantial reporting errors. We propose a new Bayesian survival model to estimate age-cohort specific survival probabilities from SSH data while accounting for bias and uncertainty introduced by SSH reporting errors. In the model, the cumulative hazard function is captured with a two-dimensional spline function. We apply it to estimate adult survival in Senegal.

DOI

https://doi.org/10.7275/26867436.0

Share

COinS