Back to search

Boston University

Methods in subgroup analysis: estimation of risk and implications for randomized controlled trial design

Abstract

dc:description.abstract

Estimation of exposure-specific risks (ESRs) using estimates of the overall risk and relative risk of disease given exposure has been performed in previous studies, but the performance of such an estimator has not been assessed nor has a variance for such an estimate been proposed. In this project I evaluated the performance of a simple product-based ESR and its variance derived using the delta method. I used the variance to estimate the 95% confidence interval. I found that this point estimate was biased and that the accompanying 95% confidence interval did not attain 95% coverage. I also proposed a revised product-based estimator and found that this estimator was unbiased. I used the delta method to derive a variance for this estimator and estimated the 95% confidence interval. The coverage of this interval attained 95% coverage in most situations. According to the CONSORT statement, subgroup analyses in randomized controlled trials (RCTs) should be pre-planned and accompanied with a formal test of interaction. I considered the interaction between treatment and a dichotomous prognostic factor with a continuous outcome. I examined the impact of misspecifying the distribution of the prognostic factor on power and sample size for interaction effects. I found that power for the interaction test was decreased when the misspecification of the distribution of the prognostic factor was away from a balanced design. I also proposed three methods for improving the power under misspecifications. Quota sampling maintained the power at 80%, but trial completion may be delayed under misspecifications. Modified quota sampling improved the power, but results were related to the proportion of trials switching to the quota sampling procedure. Sample size re-estimation improved the power, but did not always attain 80% power. All three methods maintained appropriate type I error. Lastly, I examined the impact of unplanned cross-over on power and sample size for interaction effects in RCTs. Unplanned cross-over is common in surgical trials and can diminish the magnitude of the interaction effect. Due to this, the sample size re-estimation procedure performed better than quota sampling and modified quota sampling in the presence of unplanned cross-over.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Biostatistics
Grantor dc:publisher
Boston University
Year dc:date.issued
2012

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Reichmann, William Michael

Subjects

dc:subject × 2

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Dc Identifier Barcode
11719032088918
Dc Identifier Bibid
99199970870001161

Chain of custody

source
Harvested from
Boston University
Base URL
open.bu.edu/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Reichmann, William Michael. Methods in subgroup analysis: estimation of risk and implications for randomized controlled trial design. doctoral thesis, Boston University, 2012. https://hdl.handle.net/2144/32047