Environmental Mixture-Health Associations: Current Analytical Practice and Reporting Recommendations

Analytical decisions for environmental mixture exposure remain under-standardized. The aim was to outline current analytical practice across studies and propose a reporting checklist.

An empirical methodological review was conducted for original environmental mixture-exposure studies using Bayesian kernel machine regression (BKMR), weighted quantile sum (WQS), or quantile-based g-computation (QGC), published in four high-impact journals between 2021 and 2025 for exposure types, sample size, and statistical methods. A three-stage analytical framework was outlined. Summaries were analyzed using R software, and descriptive statistics were used to characterize methodological usage and preferences.

Of the 97 studies, the median sample size was 729 [interquartile range (IQR): 396-1,992], and the median number of exposures analyzed was 9 (IQR: 6-15). Most studies focused on internal exposures 86/97 (88.7%); 58/86 (67.4%) reported an explicit detection-rate threshold. Feature reduction was performed in 25/97 studies (25.8%). BKMR, WQS, and QGC were used in 78/97 (80.4%), 35/97 (36.1%), and 30/97 (30.9%) studies, respectively. Overall, 41/97 studies (42.3%) used at least two methods. Post-hoc analyses were common, with 87/97 (89.7%) studies assessing variable importance, 78/97 (80.4%) examining interactions, and 87/97 (89.7%) assessing nonlinearity. Common reporting gaps included incomplete model-parameter specification, insufficient documentation of preprocessing decisions, and limited justification for mixture-method selection.

Among published studies, analytical choices were heterogeneous, and key reporting details were incomplete. Justification of method choice, transparent documentation of preprocessing decisions, and standardized reporting are required to improve reproducibility and interpretability. The checklist supports planning and reporting environmental mixture analyses, and provides transparency in analytical decisions, model specifications, and post-hoc interpretations.