Statistical methods to detect and estimate etiologic heterogeneity in epidemiologic studies

Principal Investigator

Name
Anil Chaturvedi

Degrees
PhD

Institution
National Cancer Institute

Position Title
Senior Investigator

Email
chaturva@mail.nih.gov

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCO-2038

Initial CDAS Request Approval
Apr 13, 2026

Title
Statistical methods to detect and estimate etiologic heterogeneity in epidemiologic studies

Summary
Epidemiologic associations from standard regression models (e.g., logistic, Cox regression) represent overall average effects. Such average effects could underlie etiologic heterogeneity, with strong signal from an unknown subset of outcomes/cases. Lack of recognition of such heterogeneity could mute or mask associations.

We propose to use the PLCO lung cancer data to evaluate statistical methods to detect and estimate etiologic heterogeneity. Using the association of smoking across lung cancer histologies as the gold standard, we will evaluate the performance of competing risk Cox regression models and finite mixture models to detect and estimate heterogeneous etiologic effects. We will also evaluate the performance of these methods in simulation studies to estimate the impact of number of heterogeneous outcome groups, relative prevalence of groups, and outcome sample sizes.

Aims

1. Using the association of smoking across lung cancer histologies as the gold standard, we will evaluate the performance of competing risk Cox regression models and finite mixture models to detect and estimate heterogeneous etiologic effects.
PLCO lung cancer data will be used to estimate hazard ratios (HRs) for major histologic subtypes (squamous cell carcinomas, adenocarcinomas, large cell, and small cell)

2. We will also evaluate the performance of these methods in simulation studies to estimate the impact of number of heterogeneous outcome groups, relative prevalence of groups, and outcome sample sizes.

Our project aims to develop methods for sensitivity analyses in epidemiologic studies to investigate the presence of heterogeneous etiologic effects.

Collaborators

Anil Chaturvedi National Cancer Institute
Fangya Mao National Cancer Institute
Li Cheung National Cancer Institute