Partial Effects in Environmental Mixtures: Evidence and Guidance on Methods and Implications.

Area of Science:

• Environmental Epidemiology
• Statistical Methods in Public Health
• Toxicology

Background:

• Assessing the health impacts of mixed environmental exposures is crucial for public health.
• Distinguishing between positive and negative partial effects of individual exposures within a mixture is methodologically complex.
• Accurate estimation of partial effects is vital for designing effective public health interventions.

Purpose of the Study:

• To evaluate the performance of existing and novel statistical methods in estimating positive and negative partial effects of environmental mixtures.
• To investigate the bias-variance trade-offs associated with different approaches under various exposure scenarios.
• To provide guidance on selecting appropriate methods for partial effects estimation in mixture research.

Main Methods:

• Comparison of quantile g-computation (QGC) and weighted quantile sums regression (WQSr) with and without sample-splitting.
• Evaluation of novel methods including a priori approaches (QGCAP, WQSAP), model-averaging (QGC-MA), and elastic net regularization (QGC-Enet).
• Simulation studies assessing performance under varying exposure correlations, sample sizes, negative effect distribution, and effect imbalance.

Main Results:

• Estimation of partial effects (both positive and negative) is increasingly biased and has higher root mean squared error with higher exposure correlation, smaller sample sizes, wider spread of negative effects, and greater imbalance between positive and negative effects.
• These findings were illustrated using real-world examples involving oxidative stress biomarkers and telomere length.
• No single method demonstrated optimal reliability across all tested exposure scenarios, emphasizing the context-dependent nature of method performance.

Conclusions:

• The accuracy of estimating partial effects in environmental mixtures is sensitive to the characteristics of the exposure mixture and the study design.
• Prior knowledge-based methods (a priori) can be highly effective when applicable, but generalizability is limited.
• Practitioners should carefully consider simulation-informed guidance when selecting methods for estimating partial effects to ensure reliable results.