Comparability of heterogeneity of treatment effect findings from two randomized clinical trials datasets: application of machine-learning based algorithms
Principal Investigator
Name
Motohiko Adomi
Degrees
M.D., M.S.
Institution
Harvard T.H. Chan School of Public Health
Position Title
PhD student
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1119
Initial CDAS Request Approval
Sep 5, 2023
Title
Comparability of heterogeneity of treatment effect findings from two randomized clinical trials datasets: application of machine-learning based algorithms
Summary
This study aimed to examine whether the heterogeneity of treatment effect (HTE) finding from one RCT can be validated in another RCT, using machine-learning (ML) based algorithms. We will use the participant dataset from NLST and divide it to participants from the American College of Radiology Imaging Network (ACRIN) study and participants from the Lung Screening Study group (LSS) study.
First, we use one dataset to assess HTE using several ML-based algorithms such as causal forest, Bayesian additive regression trees, and XG Boost. Next, as a validation step, we create subgroups in the second dataset using HTE findings from the first dataset and check whether the estimated conditional average treatment effect across subgroups is replicated in the second dataset. We also check whether the replicability of HTE findings can be validated across several ML-based algorithms.
First, we use one dataset to assess HTE using several ML-based algorithms such as causal forest, Bayesian additive regression trees, and XG Boost. Next, as a validation step, we create subgroups in the second dataset using HTE findings from the first dataset and check whether the estimated conditional average treatment effect across subgroups is replicated in the second dataset. We also check whether the replicability of HTE findings can be validated across several ML-based algorithms.
Aims
To assess HTE in NLST dataset using ML-based algorithms
To compare the HTE findings from LSS dataset and ACRIN dataset
To compare the replicability of HTE findings across several ML-based algorithms
Collaborators
Kosuke Inoue, M.D., Ph.D., Department of Social Epidemiology, Graduate School of Medicine, Kyoto University, Kyoto, Japan