Global phthalate occurrence and risk assessment in drinking water: A comprehensive Bayesian and probabilistic...

Phthalate plasticizers are widely detected in finished drinking water and are frequently discussed in the context of endocrine disruption. Yet global syntheses rarely integrate (i) between-study heterogeneity, (ii) co-occurrence structure, and (iii) probabilistic screening-level risk characterization under contemporary regulatory practice.

To synthesize global occurrence patterns of phthalates in finished drinking water, quantify co-occurrence relationships relevant to monitoring design, and estimate probabilistic non-cancer risk for multiple demographic groups with explicit uncertainty attribution.

We compiled study-level concentration summaries from 87 peer-reviewed finished drinking-water studies covering 38 phthalate analytes. Global concentration patterns were quantified using robust descriptive statistics and a Bayesian hierarchical model to accommodate substantial between-study heterogeneity and data sparsity. Co-occurrence structure was evaluated using rank-based correlation and a Gaussian Bayesian Network (GBN) fitted on complete cases (n = 20) to infer conditional dependency relationships and to explore monitoring prioritization and missing-data imputation (not causal inference). Probabilistic risk was assessed via Monte Carlo simulation (10,000 iterations per scenario) across four demographic groups, using established U.S. EPA IRIS reference doses to compute compound-specific Hazard Quotients (HQ) and mixture Hazard Index (HI), followed by Sobol sensitivity analysis to identify dominant uncertainty drivers.

DEHP, DBP, and DEP were the most frequently detected analytes, appearing in 81.6%, 80.5%, and 70.1% of studies, respectively, with median concentrations of 0.819, 0.514, and 0.317 μg/L. The GBN identified a stable dependency chain DEHP → DEP → BBP → (DBP, DMP), with BBP as the highest-connectivity node, supporting its use as a high-leverage analyte for tiered monitoring when analytical panels are constrained. Across 28 compound-demographic scenarios, no HQ exceeded 1 at the 95th percentile; DEHP produced the highest upper-tail HQ, remaining below 0.15 even in the highest-exposure demographic scenario. Mixture analysis yielded mean HI values of 0.009-0.012 across groups, with 95th-percentile HI ranging from 0.05 to 0.07 and 0% exceedance probability (HI > 1). DEHP contributed 67.9% of cumulative mixture risk, and Sobol analysis indicated that concentration variability accounted for 77.9% of overall risk uncertainty, far exceeding demographic or assumed RfD-uncertainty contributions.

Under an ingestion-only, RfD-based framework, current global finished drinking-water phthalate concentrations appear unlikely to exceed conventional non-cancer risk thresholds for either individual compounds or mixtures. The dominant role of concentration variability as an uncertainty driver underscores the need for improved analytical harmonization and reporting comparability across monitoring programs. Interpretation should remain bounded by the study design: inputs are study-level central tendencies, non-detects were treated as missing, and the GBN results describe conditional dependencies (n = 20 complete cases) rather than causal mechanisms - motivating their use for monitoring optimization rather than etiologic claims.