Causal inference methods for cluster-randomized trials under complex selection mechanisms Dane Philip Isenberg

Format:

Book

Thesis/Dissertation

Author/Creator:

Isenberg, Dane Philip, author.

Contributor:

University of Pennsylvania. Epidemiology and Biostatistics., degree granting institution.

Language:

English

Subjects (All):

Biostatistics.

Statistics.

Epidemiology.

0308.

0766.

0463.

Local Subjects:

Biostatistics.

Statistics.

Epidemiology.

0308.

0766.

0463.

Genre:

Academic theses

Physical Description:

1 online resource (187 pages)

Contained In:

Dissertations Abstracts International 87-12B

Place of Publication:

Ann Arbor : ProQuest Dissertations and Theses, 2026

Language Note:

English

Summary:

Individually randomized trials are viewed as the benchmark for assessing individual-level treatment effects. However, to mitigate contamination or logistical concerns, studies may instead randomly assign entire groups of individuals, such as hospitals, to treatments. Hence, these studies, known as cluster-randomized trials (CRTs), require certain assumptions to analyze individual-level treatment effects. In this three-part dissertation, we develop causal inference frameworks for CRTs with selection mechanisms that further complicate these analyses. In the first two parts, we focus on CRTs that assess a non-mortality outcome, but some study participants may not survive to follow-up. Therefore, death precludes defining a complete measurement of their non-mortality outcome. To address this issue of "truncation by death", we target the survivor average causal effect (SACE), a well-defined subgroup treatment effect represented via principal stratification. In part one, for parallel-arm CRTs, we establish two sets of causal assumptions for point identification of the SACE, yielding two weighting estimators. In part two, we target the SACE for the cluster-randomized crossover (CRXO) design, an extension of the parallel-arm design, by employing a Bayesian modeling approach. The CRXO design, in which clusters are assigned to sequences of alternating treatments, is typically more statistically efficient than the parallel-arm design, making it an appealing option when the number of clusters is small. In part three, we propose a method for estimating the average treatment effect in parallel-arm CRTs with misclassified outcomes and non-random internal validation subsets. Here, validation selection is a post-treatment variable (akin to survival), and we show how to use these validation data to correct for outcome misclassification while avoiding selection bias. We evaluate the finite-sample performance of our three methods with extensive simulations and subsequently apply each method to the CRT that motivated its development

Notes:

Source: Dissertations Abstracts International, Volume: 87-12, Section: B.

Advisors: Mitra, Nandita Committee members: Harhay, Michael O.; Hubbard, Rebecca A.; Li, Fan; Schaubel, Douglas E.

Ph.D. University of Pennsylvania 2026

Vendor supplied data

Local Notes:

School code: 0175

ISBN:

9798247973140

Access Restriction:

Restricted for use by site license