- Title Information
- Title
- Topics in Heterogeneous Treatment Effect Estimation
- Type of Resource (primo)
- dissertations
- Name:
Personal
- Name Part
- Yang, Jiabei
- Role
- Role Term:
Text
- creator
- Name:
Personal
- Name Part
- Schmid, Christopher
- Role
- Role Term:
Text
- Advisor
- Name:
Personal
- Name Part
- Steingrimsson, Jon
- Role
- Role Term:
Text
- Advisor
- Name:
Personal
- Name Part
- Dahabreh, Issa
- Role
- Role Term:
Text
- Reader
- Name:
Corporate
- Name Part
- Brown University. Department of Biostatistics
- Role
- Role Term:
Text
- sponsor
- Origin Information
- Copyright Date
- 2022
- Physical Description
- Extent
- xvi, 143 p.
- digitalOrigin
- born digital
- Note:
thesis
- Thesis (Ph. D.)--Brown University, 2022
- Genre (aat)
- theses
- Abstract
- Heterogeneous treatment effect estimation is commonly performed by exploring treatment covariate interactions in regression models or by performing pre-specified subgroup analyses using data from randomized controlled trials. Such analyses come with several challenges: 1) when participants belong to multiple subgroups defined by different covariates, it is hard to estimate the treatment effects for individual participants; 2) investigators need to pre-specify the covariates that define subgroups; 3) the sample size of clinical trials is usually underpowered to detect heterogeneous treatment effects. In this dissertation, we propose methods to address these challenges. N-of-1 trials are single participant crossover trials suitable for estimating individual-specific treatment effects. The first aim derives sample size calculations for a series of n-of-1 trials. We illustrate the calculations in this dissertation and implement the procedure in a Shiny app. Observational studies and electronic health record database often contain rich longitudinal data that can be used for more detailed estimation of heterogeneous treatment effects. However, in observational studies treatment assignment is not randomized; in longitudinal data 1) treatment assignment may vary over time, 2) repeated measurements on participants are not aligned, and 3) there are missing data due to missed clinic visits and dropout. The second aim develops the Causal Interaction Tree (CIT) algorithms which extend the classification and regression tree algorithm and incorporate the inverse probability weighting, g-formula and doubly robust estimators for subgroup identification using observational data. The third aim develops the Longitudinal Identification of Subgroup Algorithm (LISA) which incorporates the targeted likelihood estimators (TMLE) for subgroup identification using longitudinal observational data. We evaluated the performance of the CIT algorithms in simulations and applied the CIT and the LISA algorithms to analyze data from an observational study that evaluated the effectiveness of right heart catheterization on critically ill patients and an electronic health record database for HIV participants to evaluate the effect of dolutegravir-containing antiretroviral therapies on weight, respectively.
- Subject
- Topic
- Machine Learning
- Subject
- Topic
- causal inference
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/00864429")
- Topic
- Clinical trials
- Subject
- Topic
- heterogeneity of treatment effects
- Subject
- Topic
- N-of-1 Trials
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01091986")
- Topic
- Recursive partitioning
- Language
- Language Term (ISO639-2B)
- English
- Record Information
- Record Content Source (marcorg)
- RPB
- Record Creation Date
(encoding="iso8601")
- 20220706