Estimating mouthing exposure to chemicals in children's products

Children frequently mouth products (toys, teethers, pacifiers) that contain various additives (e.g., plasticizers, flame retardants, antimicrobials). Because many additives are not covalently bound to polymers, they can migrate into saliva during mouthing, potentially exposing children—a sensitive population with distinct behaviors and higher exposure per body weight—to harmful chemicals. Mouthing is often poorly quantified or neglected across exposure and risk assessment frameworks. Existing exposure estimations rely on experimentally measured migration rates that are chemical- and material-specific and available for only a limited set of combinations, rendering many assessments infeasible. This study aims to develop a high-throughput mouthing exposure model to estimate migration into saliva and subsequent exposure for diverse chemical-material combinations, identify key properties influencing migration, and derive risk metrics under average and upper bound mouthing scenarios for infants and toddlers.

Prior work has primarily focused on a limited set of chemicals (notably phthalates and brominated flame retardants) and few materials (e.g., PVC, silicone), despite evidence that material type substantially influences migration. Empirical correlations have linked migration to initial chemical concentration and inversely to molecular weight, octanol-water partition coefficient, and water solubility, but these single-parameter relationships do not explain the wide variability seen across chemicals and materials. Food-packaging research highlights internal diffusion coefficients in materials and material–food partition coefficients as key drivers of migration; analogous parameters are hypothesized to drive product-to-saliva migration. Existing exposure assessments require measured migration rates, mouthing area and time, and body weight, but standard test methods and consistent units are lacking. There is a need for predictive, generalizable models to estimate migration across many chemical–material combinations and support high-throughput screening.

• Data collection and harmonization: Conducted a systematic review of peer-reviewed studies reporting chemical migration into saliva from children's products, including initial chemical concentration in the product and migrated amount. Standardized migration rates to µg/10 cm²/min (10 cm² as typical mouthing area). When not directly reported, migration rates were derived from experimental settings (contact time, sample size) and results (saliva concentration, fraction migrated). Built a harmonized dataset with migration rate, initial concentration, material type, and test conditions. Materials included wood, PVC, polypropylene (PP), ethylene-vinyl acetate (EVA), and silicone. - Analysis of influencing properties: Explored correlations between migration and chemical/material properties: initial concentration, molecular weight, log Kow, and combined material–chemical parameters: diffusion coefficient in material (Dp) and material–saliva partition coefficient (Kms). Dp was estimated via QPPR as a function of material and MW. Kms was estimated via QPPR as a function of material, log Kow, and assumed ethanol equivalency (EtOH-eq) representing saliva’s affinity (tested 20% for static in vitro conditions and 50% for in vivo or agitated conditions). - Mechanistic migration model: Adapted a high-throughput food-packaging migration model to saliva contact. The model uses Dp and Kms and accounts for short-term diffusion-dominated release and longer-term saturation considering partitioning, switching at a deviation time estimated as a function of a dimensionless parameter (alpha). The model predicts the fraction migrated to saliva over time, which is converted to a migration rate by dividing by mouthing area and duration. No parameter fitting was applied (fully predictive). - Regression-based model: Built multiple linear regression with forward selection to predict log10 migration rate from candidate predictors (initial concentration, Dp, Kms, log Kow, MW). The final model included log10 Dp, log10 initial concentration, and log10 Kow. - Exposure scenarios: Defined daily mouthing durations based on age (3 to <6 months; 2 to <3 years) and product type (pacifier for long-term mouthing; doll for occasional mouthing), considering average and 99th percentile durations. Allocated materials to product types (PVC, PP, wood as dolls; silicone and EVA as pacifiers). Predicted daily mouthing exposure (µg/kgBw/d) using mouthing duration, contact area, predicted migration rate, and body weight. - Risk characterization: Calculated hazard quotients (HQ = exposure/RfD) using oral reference doses from authoritative sources when available and QSAR-predicted values otherwise. Conducted cross-validations to assess predictive performance, including leave-chemical-group-out and leave-experiment-group-out strategies for the regression model.

• Dataset: Compiled 437 experimental migration rates covering 66 chemical–material combinations (60 chemicals; 5 materials). Rates spanned ~7 orders of magnitude (1.7×10−6 to 32.7 µg/10 cm²/min). Plasticizers in PVC showed the highest migration; DINP in PVC ranged 0.5–11.1 µg/10 cm²/min across 62 data points. - Influencing properties: Migration rates positively correlated with initial concentration (R² = 0.74 on log scale) but with residual variability exceeding three orders of magnitude. Normalized migration correlated strongly with Dp (R² = 0.70) and inversely with Kms (R² = 0.43); MW or log Kow alone were insufficient to explain variability. - Mechanistic model performance: Predicted migration rates matched experiments well (R² = 0.85; standard error S = 0.79 on log scale) without parameter fitting, outperforming simple concentration-only correlations. - Regression model: Final form log10 Rmgr = 3.23 + 0.73 log10 Dp + 0.92 log10 C0 − 0.06 log10 Kow. In-sample performance R² = 0.89; S = 0.68. However, cross-validation errors increased (S ≈ 0.96–1.04) and were worse than the mechanistic model for new chemical groups or experimental sets. - Exposure estimates: Predicted daily mouthing exposure spanned 4×10−8 to 252 µg/kgBw/d across scenarios. Highest exposures were for DBP in PVC dolls (≈21.7–253 µg/kgBw/d) and propylparaben in EVA pacifiers (≈5.8–224 µg/kgBw/d). - Risk: For 3 to <6 months average mouthing, three combinations exceeded HQ ≥ 1: 2,2,4-trimethyl-1,3-pentanediol diisobutyrate (PVC dolls; HQ ≈ 1.25), diisononyl cyclohexane-1,2-dicarboxylate (PVC dolls; HQ ≈ 1.06), and propylparaben (EVA pacifiers; HQ ≈ 1.39). No HQ > 1 for average scenario in 2 to <3 years. For upper bound mouthing, 70 combinations (younger group; HQ up to 15.2) and 58 (older group; HQ up to 10.1) exceeded HQ ≥ 1. The regression model produced similar patterns but typically 3-fold lower exposures.

The study addressed the lack of broadly applicable tools for estimating chemical migration into saliva by adapting a mechanistic model from food packaging to children’s mouthing exposure. Using chemical–material diffusion and partitioning parameters, the model predicts migration without requiring chemical- and material-specific migration tests, achieving strong agreement with experimental data across diverse chemicals and materials. This enables high-throughput screening of children’s products for potential mouthing exposure and risk and supports integration into broader exposure frameworks such as the Product Intake Fraction approach. The regression alternative performs comparably within the training domain but shows reduced generalizability when predicting new chemical groups, reinforcing the mechanistic model’s suitability for unseen combinations. Exposure and risk analyses show that average mouthing behavior typically poses limited risk, but upper bound behaviors can generate concern for several plasticizers and parabens, emphasizing the importance of considering high-end users in safety assessments. Comparisons with literature indicate that mouthing exposures are often comparable in magnitude to other pathways (e.g., dust ingestion, hand-to-mouth), highlighting mouthing as a relevant contributor to total exposure for young children.

A fully predictive mechanistic mouthing exposure model, adapted from food-packaging migration, reliably estimates chemical migration into saliva and children’s mouthing exposure across multiple materials and chemicals. A complementary regression model offers similar performance within its training scope. Results underscore that while average mouthing often yields low risk, upper bound mouthing behaviors substantially elevate risk for certain plasticizers and parabens, necessitating consideration of high-end users in product safety and regulatory decisions. The model provides a practical tool for high-throughput screening, life cycle impact assessment, and chemical alternatives assessment to inform safer product design. Future work should expand empirical data on diffusion and partition coefficients across more materials, standardize saliva migration testing protocols, and improve the availability and quality of toxicity reference values to reduce uncertainty.

• Scope limited to organic chemicals; inorganic substances (e.g., metals) are not covered by the current mechanistic migration framework. - Dependence on predicted material–saliva partition coefficients (Kms), which are sensitive to log Kow estimation and assumed ethanol equivalency; uncertainties in these inputs propagate to migration estimates. - Does not account for ingestion of detached particles or object–hand–mouth contacts, which can contribute to total mouthing exposure. - Potential need for model adaptations for textiles and coated surfaces where transfer mechanisms and partitioning may differ. - Experimental heterogeneity (e.g., saliva simulant composition, pH, temperature, agitation) complicates cross-study comparisons; many parameters could not be standardized beyond unit harmonization and EtOH-equivalency assumptions. - Toxicity assessment limited by availability of experimental RfDs; many chemicals relied on QSAR-predicted values, introducing additional uncertainty. - Regression model shows reduced predictive power for chemical classes not represented in training and potential sensitivity to underlying QPPRs for Dp and Kms.