# Comparison of Methods for Analyzing Environmental Mixtures Effects on Survival Outcomes

## TL;DR

This paper compares different statistical methods for analyzing how environmental mixtures affect survival outcomes in epidemiology.

## Contribution

The study evaluates and compares five methods for modeling environmental mixtures on survival data using simulations.

## Key findings

- Log-linear models had low coverage for effect estimation under high exposure correlations.
- Flexible models improved coverage but introduced bias and high variability.
- Flexible models outperformed constrained models in most scenarios but face real-world limitations.

## Abstract

Estimating the effect of environmental mixtures on survival outcomes is common in epidemiological studies, yet the applicability and performance of advanced mixture modeling methods in this context remains underexplored. In this review, we identify available methods for this context and evaluate their performance via simulations.

We compared five methods – Cox Proportional Hazards (with/without penalized splines), Cox Elastic Net, Bayesian Additive Regression Trees (BART), and Multivariate Adaptive Regression Splines (MARS). Simulations showed log-linear models achieved low coverage when estimating individual exposure and mixture effects, especially under high exposure correlations and proportional hazards violations. More flexible models exhibited higher variability but improved coverage in effect estimation.

While flexible models were better able to estimate mixture effect on survival outcomes compared to more constrained models for most simulation scenarios, they still introduced bias and often had high variability. Given real-world constraints like limited sample sizes and high censoring, there likely remains significant complexities for the application of flexible modeling for environmental mixtures for the survival analysis contexts. We recommend evaluating if findings are consistent across methods.

The online version contains supplementary material available at 10.1007/s40572-025-00500-y.