Analyzing PSA-based Prostate Cancer Screening using Causal Inference Methods
Principal Investigator
Name
Robert Tibshirani
Degrees
PhD
Institution
Leland Stanford Junior University
Position Title
Professor of Biomedical Data Science and of Statistics
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-1625
Initial CDAS Request Approval
Jul 22, 2024
Title
Analyzing PSA-based Prostate Cancer Screening using Causal Inference Methods
Summary
Building on previous work by Shoag J et al. (2015) and others, this project aims to advance the use of causal inference methods and their methodological development in medical contexts. Regression discontinuity designs (RDDs) are a quasi-experimental causal inference technique that enables estimation of causal effects from observational data. RDDs exploit the fact that patients just above a clinical threshold are comparable with those just below, addressing the issue of confounding in observational data. Despite the potential of quasi-experimental methods, they often lack methodological extensions that allow researchers to apply them directly.
As part of our analysis, we will explore the use of machine learning to enhance causal inference methods in medical settings. This is relevant for prostate-specific antigen (PSA)-based screening, in which we can analyze survival outcomes and have access to covariates. The PLCO prostate cancer screening trial is ideally suited for this by its diverse cohort, large sample size, and the capture of long-term, cause-specific outcomes.
[1] Shoag J, Halpern J, Eisner B, Lee R, Mittal S, Barbieri CE, Shoag D. Efficacy of Prostate-Specific Antigen Screening: Use of Regression Discontinuity in the PLCO Cancer Screening Trial. JAMA Oncol. 2015 Oct;1(7):984-6. doi: 10.1001/jamaoncol.2015.2993. Erratum in: JAMA Oncol. 2015 Oct;1(7):989. doi: 10.1001/jamaoncol.2015.3943. PMID: 26291583.
As part of our analysis, we will explore the use of machine learning to enhance causal inference methods in medical settings. This is relevant for prostate-specific antigen (PSA)-based screening, in which we can analyze survival outcomes and have access to covariates. The PLCO prostate cancer screening trial is ideally suited for this by its diverse cohort, large sample size, and the capture of long-term, cause-specific outcomes.
[1] Shoag J, Halpern J, Eisner B, Lee R, Mittal S, Barbieri CE, Shoag D. Efficacy of Prostate-Specific Antigen Screening: Use of Regression Discontinuity in the PLCO Cancer Screening Trial. JAMA Oncol. 2015 Oct;1(7):984-6. doi: 10.1001/jamaoncol.2015.2993. Erratum in: JAMA Oncol. 2015 Oct;1(7):989. doi: 10.1001/jamaoncol.2015.3943. PMID: 26291583.
Aims
Aim 1: Improve the applicability of causal inference methods with a focus on quasi-experimental methods for cancer screening using machine learning.
Aim 2: Assess the usefulness of new causal inference frameworks for treatment effect estimation in settings with multiple outcomes, downstream treatments and patient subgroups.
Collaborators
The requestor and PI will have access and analyze the data.