Show simple item record

dc.contributor.advisorSchwartz, Joel D.
dc.contributor.authorWang, Yan
dc.date.accessioned2018-12-20T13:45:16Z
dash.embargo.terms2020-05-01
dc.date.created2018-05
dc.date.issued2018-04-09
dc.date.submitted2018
dc.identifier.urihttp://nrs.harvard.edu/urn-3:HUL.InstRepos:37945607*
dc.description.abstractExposure to fine particulate matter (PM2.5) has been associated with many adverse health outcomes, including increased risk of mortality and hospital admissions. There have been criticisms challenging whether the reported associations are valid estimates of the underlying causal effect of PM2.5, with confounding bias as one of them. Confounding bias may occur when the researchers fail to identify confounders among measured covariates, misspecify the functional form for the confounders in the outcome regression model, or fail to measure some of the confounders. This dissertation develops and applies novel biostatistical methods to relax the assumptions and address the confounding issue in order for the estimated effect of PM2.5 to be causal. The first study tackles the confounder selection problem, when the number of candidate covariates is large and the sample size is extraordinarily large. The study develops a novel divide-and-conquer adaptive least absolute shrinkage and selection operator (LASSO) to select variables and fit sparse Cox proportional hazards models for big datasets for short-term PM2.5 on readmission or death hazard among heart failure patients. The second study relaxes the assumption of correct specifying an outcome regression and develops doubly robust additive hazards models (DRAHM) for the effect of long-term PM2.5 on mortality, where either a correctly specified propensity score model or outcome regression model can lead to a causal estimate. The third study relaxes the assumption of no unmeasured confounding and gives the causal relative risk estimate of PM2.5 on mortality using a difference-in-differences approach. The fourth study presents a standard analysis using Cox proporitional hazards model for long-term exposure to PM2.5 and mortality with extensive adjustment for confounding. The studies in the dissertation present novel biostatistical methods that relax the assumptions required for the estimate of PM2.5 on health outcomes to be causal, while relying on other assumptions including consistency, positivity, no exposure measurement error, and no interference. The applications suggest that exposure to PM2.5 causally increases the risk of mortality in a long term and the risk of readmission or death due to heart failure in a short term.
dc.format.mimetypeapplication/pdf
dc.language.isoen
dash.licenseLAA
dc.subjectHealth Sciences, Epidemiology
dc.subjectBiology, Biostatistics
dc.subjectEnvironmental Sciences
dc.titleToward a Causal Health Effect of Fine Particulate Air Pollution
dc.typeThesis or Dissertation
dash.depositing.authorWang, Yan
dash.embargo.until2020-05-01
dc.date.available2018-12-20T13:45:16Z
thesis.degree.date2018
thesis.degree.grantorHarvard T.H. Chan School of Public Health
thesis.degree.levelDoctoral
thesis.degree.nameDoctor of Science (SD)
dc.contributor.committeeMemberDominici, Francesca
dc.contributor.committeeMemberZigler, Corwin M.
dc.type.materialtext
thesis.degree.departmentEnvironmental Health
dash.identifier.vireohttp://etds.lib.harvard.edu/hsph/admin/view/226
dc.description.keywordsFine particulate air pollution; Mortality; Causal modeling; Divide-and-conquer; Cox proportional hazards model; Additive hazards model; Difference-in-differences;
dash.author.emailyaw719@mail.harvard.edu


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record