Modeling multi-level survival data in multi-center epidemiological cohort studies: Applications from the ELAPSE...

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Public Health

Background:

• The ELAPSE project pooled data from 14 cohorts to study associations between residential exposure to low levels of air pollution (PM2.5 and NO2) and health outcomes.
• Health outcomes investigated included natural-cause mortality and incidence of cerebrovascular disease, coronary heart disease, and lung cancer.

Purpose of the Study:

• To evaluate and compare different statistical methods for analyzing multi-level survival data in multi-center epidemiological studies.
• To assess the impact of accounting for different levels of clustering (cohort and small residential areas) on effect estimates.

Main Methods:

• Five approaches were applied within a multivariable Cox model to handle cohort-level clustering: no adjustment, indicator variables, strata, frailty terms, and random intercepts.
• Second-level clustering due to residential area characteristics was addressed using random intercepts or variance correction.
• Simulations were conducted to evaluate the stratified, frailty, and mixed Cox models under various heterogeneity scenarios.

Main Results:

• Effect estimates remained stable when adjusting for cohorts but varied significantly when no adjustment was made.
• Adjusting for small area grouping increased standard errors of effect estimates.
• Simulations showed identical results for stratified and frailty models, supporting their reliability.

Conclusions:

• It is essential to account for between-cohort heterogeneity in multi-center studies using pooled individual-level data.
• A stratified multivariable Cox model was selected for the ELAPSE project to manage cohort heterogeneity, given the study's data limitations.