Toward Nonparametric Propensity Score Estimation With Guaranteed Covariate Balance

<p><em> Establishing clear causality between exposures and outcomes is often complicated </em><em>by confounders in observational studies, which leads to imbalance in covariate distributions </em><em>between treatments and biased treatment effect inference. The propensity score </em><em>(PS) has been widely used to adjust this covariate imbalance in observational data. However, </em><em>the propensity score analysis methods rely on a correctly specified parametric PS </em><em>model. When the model is misspecified, the covariate imbalance may occur, which leads </em><em>to biased estimation of the treatment effect. Therefore, it is necessary to study how to </em><em>improve the model misspecification in propensity score analysis. </em></p> <p><em></em><em> My Ph.D. dissertation consists of three aims. In Aim 1, we examined whether the </em><em>optimization of global balance — the mean balance of covariates or their transformations </em><em>in the overall study population, can circumvent the need for correct propensity score </em><em>model specification, and whether the use of a propensity score model further improves </em><em>the estimation performance compared to methods without modeling propensity score. </em><em>In Aim 2, we developed a propensity score analysis framework, the propensity score </em><em>with local balance (PSLB), which incorporates nonparametric propensity score models </em><em>and improves the balancing property of the estimated propensity score compared to </em><em>existing methods that only optimize the global balance. In Aim 3, we developed a </em><em>subgroup analysis method that is robust to propensity score model misspecification. </em><em>Specifically, we proposed a new algorithm, the guaranteed subgroup balancing propensity </em><em>score (G-SBPS), to ensure exact subgroup balance — the mean balance of covariates in </em><em>each subgroup. In addition, we implemented kernel methods in G-SBPS to improve </em><em>the propensity score model fitting. For each of the aim, we provided theoretical and </em><em>simulation-based justification for the research question or proposed methodologies, and </em><em>applied the proposed methods to the right heart catheterization (RHC) data to estimate </em><em>the length of hospital stay or the diabetes self-management training (DSMT) data to </em><em>evaluate the hospitalization rate within three years.</em></p>

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Thesis (MS)

Year dc:date.available

2023

Authors dc:creator

• Li, Yan
• <p>https://orcid.org/0000-0001-9952-9737</p>

Contributors dc:contributor

• Liang Li
• Yu Shen
• Ying Yuan

• Observational study
• Propensity score
• Nonparametric modeling
• Inverse probability weighting
• Covariate balancing
• Subgroup analysis
• Biomedical Informatics

Repository record dc:identifier

https://digitalcommons.library.tmc.edu/utgsbs_dissertations/1291

OAI identifier oai:identifier

oai:digitalcommons.library.tmc.edu:utgsbs_dissertations-2348