We revised a previously published probabilistic one-compartment TK model by Lynch et al. (2023)^[43]. Model revisions predict the distribution of population serum PFAS concentrations attributable to drinking water exposure over the life course for six PFAS. The population PFAS TK model is designed as a tool to support clinical, policy, and other decision-making by providing a quantitative linkage between exposure to PFAS in drinking water and the distribution of resultant internal serum PFAS concentrations for populations, with explicit consideration of sensitive subgroups, including infants. Exposure pathways include ingestion of contaminated drinking water, PFAS transfer from maternal serum at birth and to breast milk, and non-drinking-water sources [Figure 1]. Chemical-specific TK parameters and age-specific exposure factors are represented with probabilistic parameter distributions, which are combined using Monte Carlo simulation to generate population serum PFAS distributions over time. All model parameters and inputs are listed in Supplementary Table 1 and described in the methods below.

We demonstrate the model capabilities by applying reverse dosimetry to estimate drinking water concentrations predicted to elevate serum PFAS concentrations in infants to the NASEM clinical action level. This approach provides a quantitative framework to evaluate the relationship between internal and external PFAS exposure and inform public health decision-making.

The model by Lynch et al. (2023) is intended for use by individuals and predicts age groups separately, with one model for infants/children less than 6 years old and one for children older than 6 and adults^[43]. Model revisions predicting the distribution of population serum PFAS concentrations, fall into four main categories including: (1) modifications to exposure factor distributions and exposure scenario flexibility to allow population-level predictions; (2) changes to the modeling framework for infants and young children and to enable continuous prediction from birth; (3) model processing changes to increase computational efficiency, and (4) the addition of parameters and modeling capability for PFDA and PFHpA. Model revisions are listed and compared with the Lynch et al. (2023) model in Supplementary Table 2, and parameters for PFDA and PFHpA are described in Table 1^[43].

An overview of the exposure sources, TK modeling, and model outputs is shown in Figure 1. Serum PFAS is calculated over time using the integrated rate law for the one-compartment system with zero-order intake and first-order elimination [Supplementary Figure 1]^[44]. Serum PFAS for a specified population is modeled across the entire life course, beginning with calculation of serum PFAS at birth (C₀) [Equation (1)]^[43]. For each date, the current serum PFAS and associated parameter values are used to calculate serum PFAS for subsequent dates, accounting for exposure from drinking water or breast milk, exposure from non-drinking water or “background” sources, and PFAS elimination [Equation (2)].

(1) $$ C_0=C_M * PTF $$

Where:C_M is the maternal serum PFAS concentration at birth (µg/L),and PTF is the placental transfer factor (unitless ratio).

(2) $$ C_{i+1}=B_{i+1}+(C_i-B_i)e^{-k_{el}(t_{i+1}-t_i)}+\frac{LI_i*AF_i*LC_i}{k_{el}*V_d*10^6}(1-e^{-k_{el}(t_{i+1}-t_i)}) $$

Where:B_i+1 is the serum PFAS “background” on the subsequent (i+1)^th date (µg/L),C_i is the total serum PFAS on the i^th date (µg/L),B_i is the serum PFAS “background” on the i^th date (µg/L),k_el is the elimination rate constant (1/days), calculated from half-life k_el = ln(2)/(T_1/2*365.25),t_i is the number of days after birth on the i^th date,t_i+1 is the number of days after birth on the subsequent (i+1)^th date,LI_i is the mean daily body weight-adjusted liquid intake starting on the i^th date [mL/(kg-day)],AF_i is the adjustment factor for liquid intake starting on the i^th date (unitless ratio),LC_i is the concentration of PFAS in the liquid consumed (ng/L),V_d is the weight-adjusted volume of distribution (L/kg), andthe factor of 10⁶ converts the third term’s units to µg/L.

Modeled populations consume PFAS in either drinking water (adult) or breast milk (infant). Breast milk PFAS concentrations are determined by maternal serum PFAS concentrations at birth and decrease over time [Equation (3)].

(3) $$ BMC_i=C_M*LTF*e^{-k_{milk}*t_i}*1000 $$

Where:BMC_i is the breast milk PFAS concentration on the i^th date (ng/L),C_M is the maternal serum PFAS concentration at birth (µg/L),LTF is the lactational transfer factor (unitless ratio),k_milk is the rate constant for breast milk decline over time (1/days),t_i is the number of days after birth on the i^th date, andthe factor of 1000 converts units to ng/L.

User-defined inputs are organized in Microsoft Excel and calculations are performed in R (Version 4.5.1 for this analysis^[58]). Monte Carlo simulations combine exposure and TK parameter distributions to generate a population serum PFAS distribution. The user-specified exposure scenario includes selecting the modeling approach for maternal serum (C_M), population sex(es), age(s), duration of breastfeeding, exposure period(s) specified as dates, and PFAS drinking water concentration(s) during the exposure period(s). The population serum distribution is calculated monthly before age 1 and annually thereafter, as well as when water exposure changes.

Each iteration of the simulation draws a single value from chemical-specific parameter distributions that are constant across the life course, including T_1/2, V_d, PTF, LTF, and k_milk. The age-dependent parameter distributions, AF and B, are drawn independently for each age range [Supplementary Table 1].

We demonstrate the utility of the population PFAS TK model by: (1) showing predicted population serum PFAS concentrations over the life course at a constant drinking water exposure; (2) evaluating model performance using studies that report paired drinking water and serum PFAS concentrations; and (3) deriving drinking water concentrations associated with serum PFAS levels that reach the NASEM clinical action level of 20 ng/mL in sensitive populations. These three model applications are described below. We ran all modeling scenarios with 35,000 Monte Carlo iterations. This number of iterations is higher than the typical 500 to 1,000 iterations in other simulation studies and was selected to increase the reproducibility of model results across the distribution of predicted results^[59]. Serum concentrations are presented in ng/mL or μg/L (1 ng/mL = 1 μg/L) and water concentrations are presented as ng/L.

To demonstrate the model’s ability to capture population serum trajectories over time, we modeled a male population from birth to age 70 with constant drinking water exposure of 20 ng/L for each PFAS [Supplementary Table 6]. We used the model feature to enter a known maternal serum PFAS distribution. The maternal distribution is the predicted serum PFAS distribution of a modeled female population at age 40 with lifetime exposure to 20 ng/L PFAS in drinking water. Breastfeeding duration was set to 6 months for the male population. The only difference between modeling male and female populations is the contribution from “background”, which is higher for males than females for most PFAS [Supplementary Table 5]; no other model parameters differ by sex.

We performed a local sensitivity analysis to assess how comparable shifts in select input parameters affect predicted 50th and 90th percentile serum concentrations using the output for six-month-old exclusively breast-fed male infants. For the sensitivity analysis, we used the 35,000 values of each parameter drawn for the baseline model run. A comparable shift in the mean for each variable was determined first by expressing all variables on a normal scale with lognormally distributed variables log-transformed to approximate normality; then shifting the mean value up or down by one standard deviation, yielding a new mean for the distribution. Lognormal parameters were back-transformed to their original scale. Identical positions drawn from the original variable distribution were selected from the shifted distribution, and the model was re-run with each parameter shifted up or down by one standard deviation on the normalized scale. The relative change in 50th and 90th percentile values per standard deviation change in the input parameter is compared. This approach allows for consistent comparison of parameter influence across normal and lognormal distributions and avoids Monte Carlo sampling error while maintaining the probabilistic structure of the model.

To assess performance, the model was used to predict serum PFAS concentrations for published data from studies of paired drinking water and serum PFAS concentrations and studies reporting late pregnancy maternal serum PFAS levels and infant serum PFAS levels. Only studies reporting sufficient information about exposure duration, population demographics, drinking water PFAS concentration(s), and serum PFAS concentrations in the study population could be used to reconstruct a scenario for input into the model. Five suitable studies reporting results from nine populations of adults and older children were used in this evaluation^[35,60-63]. The modeling scenarios for each cohort, including drinking water PFAS concentrations and exposure durations, are described in the Supplementary Table 7.

The second type of study used for model evaluation reported late pregnancy maternal serum PFAS levels and infant serum PFAS levels. Two suitable studies were identified^[64,65]. The distribution of maternal late pregnancy PFAS levels was directly entered into the model, and measured infant PFAS concentrations at 6 months were compared to model-predicted results. Details on the infant populations along with the modeled scenario for each study are described in the Supplementary Table 8.

All evaluation scenarios were run with 35,000 Monte Carlo iterations to produce the model-predicted population distribution of serum PFAS levels. Summary statistics for serum PFAS levels reported for the study population were compared with the corresponding summary statistic or percentile from the modeled distribution of serum PFAS levels. For studies that reported a minimum or maximum measured concentration, these values were compared to the 5th and 95th percentiles of the modeled distribution, respectively. All studies reported at least one central tendency (median or GM) and upper percentile value (maximum, 90th, or 95th percentile).

Model performance was considered good when predicted serum concentrations fell within 2-fold of the observed values, a criterion that has been used to assess the performance of other PFAS TK models^[44,66]. Root mean square error (RMSE) was calculated for the modeled vs. measured summary statistics for each study and was used to evaluate the relative performance of the model for different PFAS and at different percentiles, among other comparisons. A lower RMSE indicates a better fit between predicted and measured serum PFAS concentrations.

The population PFAS TK model was implemented to estimate population lifetime drinking water concentrations corresponding to the NASEM clinical action level of 20 ng/mL, where additional health screening is recommended^[42] for each of the six PFAS (PFOA, PFOS, PFHxS, PFNA, PFDA, PFHpA) regulated in MA. The drinking water concentration corresponding to the NASEM clinical action level is termed the Clinical Guidance-based Water Concentration (CGWC). The CGWC provides an indication of lifetime drinking water exposure levels where enhanced clinical action could be recommended. Consistent with public health approaches, we made choices in the modeling scenario (described below) to calculate health-protective CGWCs, including for sensitive populations, based on an upper-end estimate of the relationship between drinking water and serum levels; the modeling scenario was not designed to estimate the median CGWC for the US population.

We calculated CGWC for a sensitive population of breastfed infants and also calculated adult values for comparison. Infants are a sensitive population because (1) we expect infants to have higher internal serum concentrations than adults at the same drinking water exposure due to high body weight-adjusted liquid intake relative to adults [Supplementary Table 4]; and (2) infants have increased susceptibility to the adverse effects of PFAS at this life stage^[45,67-70]. The higher likelihood of both exposure and adverse health effects makes the CGWC particularly relevant for infants. Calculating CGWC for adult populations as well enables a comparison of differences between general population estimates and estimates for sensitive populations.

The infant population we selected for modeling is breastfed male infants. Due to efficient transfer from breast milk for some PFAS, breastfed infants have higher exposure than formula-fed infants; thus, using modeled breastfed infants is also protective of formula-fed infants. While breastfed infants are predicted to have higher serum PFAS levels than bottle-fed infants, breastfeeding remains important for the development of the child. We modeled male infants because the mean and GSD “background” for male children is higher than corresponding female values for most modeled PFAS [Supplementary Table 5]. As a result, drinking water concentrations corresponding to clinical action levels for breastfed male infants produce an estimate that is protective of female and formula-fed infants.

Using these two defined populations - breastfed male infants and male adults - we determined the relationship between drinking water and serum PFAS levels by modeling the population distribution of serum PFAS levels for each PFAS over a series of constant drinking water concentrations between 4 and 90 ng/L [Supplementary Table 6]. While the breastfed infant population does not consume drinking water, drinking water is used to model maternal serum concentrations which then set the initial infant serum PFAS concentration and the breastmilk PFAS concentration. For this modeling scenario, maternal serum PFAS concentrations are from a population of 40-year-old women who were breastfed for three months as infants; the infant population used to calculate CGWC had a breastfeeding duration of six months to estimate an upper-end exposure scenario [Supplementary Table 6]. This population of infants was modeled to age 40 with constant exposure at each drinking water concentration [Supplementary Table 6]. For each population, this step produced a distribution of serum PFAS concentrations for each lifetime drinking water exposure level.

We then fit a linear regression model to the 50th and 90th percentiles of the output population serum PFAS distributions with drinking water as the predictor variable. This produced two linear regression models for each population: one describing the relationship between drinking water and 50th percentile serum PFAS concentrations and the second describing the relationship between drinking water and 90th percentile serum PFAS concentrations. We used these regression models to calculate the drinking water serum concentration predicted to result in a serum level of 20 ng/mL at the 50th percentile of the distribution and at the 90th percentile of the distribution.

We defined CGWCs as the PFAS drinking water concentration predicted to result in the 90th percentile of a modeled population’s serum reaching the NASEM 20 ng/mL clinical action level. We selected the 90th percentile to align with standard USEPA practice for calculating enforceable drinking water standards using 90th percentile drinking water ingestion rates^[40,71].

CGWCs were calculated separately for PFOA, PFOS, PFHxS, PFNA, PFDA, and PFHpA. The NASEM clinical guidelines apply to the sum of seven PFAS regularly monitored in NHANES which does not include PFHpA (though PFHpA has been monitored previously in NHANES). We chose to extend the NASEM guidelines to PFHpA because the NASEM report indicates that the additive approach could be expanded to apply to additional PFAS^[42] and PFHpA health outcomes overlap with outcomes observed for some of the NASEM PFAS, which is the justification for its inclusion in state-specific drinking water regulations^[13].

The NASEM clinical guidelines apply to the sum of seven PFAS measured in serum. Cumulative risk could be evaluated using the chemical-specific CGWC by applying mixture assessment methods based on chemical additivity such as the hazard index (HI) approach^[72,73]. A HI can be calculated using Equation (4), by dividing the measured water concentration of each PFAS (DWC_PFAS) by the CGWC for each PFAS (CGWC_PFAS)^[72]. A HI equal to or greater than one (1) indicates that the sum of the predicted serum PFAS concentrations exceeds the NASEM clinical action level of 20 ng/mL.

(4) $$ HI=\frac{DWC_{PFHxS}}{CGWC_{PFHxS}}+\frac{DWC_{PFOA}}{CGWC_{PFOA}}+\frac{DWC_{PFDA}}{CGWC_{PFDA}}+\frac{DWC_{PFNA}}{CGWC_{PFNA}}+\frac{DWC_{PFOS}}{CGWC_{PFOS}}+\frac{DWC_{PFHpA}}{CGWC_{PFHpA}} $$

Where:HI is the hazard index,DWC_PFAS is the measured concentration for each PFAS, andCGWC_PFAS is the PFAS-specific Clinical Guidance-based Water Concentration.

Model-predicted population serum PFAS concentrations change over time based on chemical-specific parameters and age-specific exposure factors [Figure 2]. For all six PFAS, model predictions show serum concentrations rise sharply by a factor of 2-5 between birth and 6 months due to high PFAS levels in breast milk and high liquid intake relative to body weight [Supplementary Table 9 and Supplementary Figure 2]. With lifetime drinking water exposure at 20 ng/L, breast milk concentrations are 2-6 times higher than the drinking water concentration for PFOA, PFOS, PFHxS and PFDA, and similar to the drinking water concentration for PFNA and PFHpA at the median of the distribution [Supplementary Table 9]. As a result, the median modeled infant serum PFAS concentrations after 6 months of exclusive breastfeeding are 1.2-3 times higher than the median modeled maternal serum PFAS concentrations [Supplementary Table 9]. Sensitivity analysis results indicate that infant serum concentrations are most influenced by maternal serum levels, breast milk intake rate, volume of distribution, and lactational transfer. Similar patterns are observed for all six PFAS at both the median and 90th percentile of the distribution [Supplementary Figure 3].

Peak serum concentrations for all six PFAS occur within the first 3 years of life [Figure 2]. After this peak, serum PFAS concentrations decline as the declining body weight-adjusted drinking water intake reduces exposure [Supplementary Figure 2]. Key trends in modeled lifetime serum PFAS concentrations, including peak concentrations in early life, have been observed in both epidemiological and TK modeling studies. Longitudinal epidemiological studies measuring early life serum concentrations show trajectories similar to the modeled lifetime trajectory, with PFOA, PFOS, PFNA, and PFDA generally increasing during the first 6-11 months followed by decreases at later timepoints^[64,74]. For PFHxS, findings differ across studies: one reported increasing concentrations between birth and 6 months, followed by decreases^[64], while another observed declining concentrations after birth^[74].

Additional evidence indicates that early life serum concentrations of PFOA, PFOS, PFHxS, and PFNA are significantly higher than those measured in mid- and late childhood^[75]. The early life increase observed for most PFAS corresponds to the nursing period^[64,74], indicating that breastfeeding is an important exposure pathway in early life. Consistent with this interpretation, multiple studies have reported positive associations between breastfeeding duration and PFAS concentrations in infants and children, especially for PFOS and PFOA^[64,74-78]. TK models similarly predict peak serum concentrations of PFOA, PFOS, and PFHxS within the first 1-2 years of life for both breastfed and formula-fed infants, with the latter peak resulting from high body weight-adjusted drinking water intake for young children^[18,49]. In addition to these temporal trends, some studies report changes in population variability similar to the lifetime trajectory modeling. One longitudinal epidemiological study measuring serum PFOA, PFOS, PFHxS, and PFNA concentrations between ages 1-10.5 observed decreasing variability in the distribution of measured concentrations over time along with decreasing median concentrations^[75]. Together, these findings support the ability of the population PFAS TK model to capture both temporal trends in serum PFAS concentrations over time and changes in population-level exposure factors during sensitive life stages.

The population PFAS TK model performed well in predicting the distribution of population serum concentrations reported in empirical studies with paired drinking water and serum data for the six modeled PFAS, across exposure levels spanning 3-4 orders of magnitude [Figure 3]. Modeled central tendency and upper percentile serum PFAS concentrations mostly fell within two-fold error of the observed concentrations reported in each study. When results were combined for all PFAS and modeled study populations, RMSE values were similar for predictions of both central tendency and upper percentile values from the observed serum PFAS distributions [Figure 3A and B]. The comparable predictive performance at the center and upper percentiles of the distribution indicates the model captures population variability in TK and exposure factor parameters.

Less human TK data are available for PFNA, PFDA, and PFHpA, and data gaps for these chemicals were filled with a combination of data from animal studies, read-across from other PFAS, and imputation in the case of breastmilk elimination rate constants for PFNA and PFDA, as described in the methods. Overall, model error was not consistently higher for PFAS with more limited human TK data compared with data-rich PFAS. For example, imputed breast milk elimination rate constants for PFNA and PFDA, combined with less available data on breast milk transfer factors, did not result in consistently higher model prediction error for these PFAS. PFHpA, which has the least human TK information among the six modeled PFAS, showed the highest model error at central tendency values but comparable error to the data-rich PFAS at the upper percentile values [Supplementary Table 10]. These results suggest that the methods used to address PFAS-specific TK data gaps, especially for modeling infant exposure, can capture population-level TK for comparatively data-poor chemicals. However, serum PFAS distributions for compounds with less certain TK parameters are associated with greater modeling uncertainty because both the central estimates and variability of the input parameters are less well-characterized. In contrast, for PFAS with TK parameter distributions informed by human data, the resulting serum distribution is more representative of human variability and therefore associated with less uncertainty. Additional sources of uncertainty intrinsic to the model include the application of a single T_1/2 distribution for modeling all populations, including the single-sex populations from Arnsberg, DE, and uncertainty in exposure factors used for modeling.

Model error was lower for infant populations than for adults and older children across PFAS and at both the central tendency and upper percentiles of the distribution [Figure 3A and B, Supplementary Table 10]. For infant modeling scenarios, maternal serum PFAS concentrations are the exposure input that dictates the starting infant serum PFAS concentrations and the exposure via breast milk for exclusively breastfed infants. The availability of more complete exposure information and shorter modeled time periods for infant populations likely contributed to reduced model error.

Model accuracy depends on both an accurate reconstruction of the exposure scenario used for evaluation and the accuracy of the TK parameters and exposure factors used in the model. In addition to intrinsic model uncertainties discussed above, incomplete or inaccurately captured exposure information from sources including drinking water concentrations, exposure duration, and background exposure can contribute to discrepancies between predicted and observed concentrations.

For example, the model underpredicts empirical data from two study populations - the Ronneby, SE low exposure cohort and the Pittsboro, US cohort - at both the central tendencies and upper percentiles from the observed serum PFAS distributions. Uncertainty in reconstructing the exposure scenarios likely contributed to these discrepancies. For the Pittsboro, US cohort, lifetime drinking water exposure was modeled using constant PFAS concentrations reported in the study. However, PFAS concentrations in the Haw River, which serves as the drinking water source for the Pittsboro, US cohort, have varied substantially over time^[62,79].

Similarly, for the adult population modeled from Ronneby, SE, the potential underestimation of exposure combined with the long modeling timeframe may have contributed to model underprediction. In Ronneby, one drinking water source was highly contaminated with PFAS (up to 8,000 ng/L for PFOS) while another source had exposure levels several orders of magnitude lower^[35]. Although the evaluation cohort resided in areas served by the lower-exposure source, their proximity to areas supplied by the highly contaminated water source raises the possibility of unaccounted exposure through consumption of contaminated water or other exposure pathways not captured in the modeling scenario. Beyond the scenario-specific uncertainties addressed above, the use of 2017-2018 and 2013-2014 NHANES data for adults and children, respectively, for all modeling scenarios could contribute additional uncertainty, especially when modeling historical exposures. Importantly, the flexible model inputs that accommodate temporal and concentration variability over time in drinking water and background serum PFAS levels, among other inputs, allow for the reconstruction of complex historical exposures.

The population PFAS TK model was applied to estimate CGWC for the sensitive population of breastfed infants. Adult values were also calculated for comparison.

Figure 4 shows the relationship between increasing drinking water PFAS concentrations and the 90th percentile serum PFAS concentrations for six-month-old, exclusively breastfed infants. For this population, drinking water concentrations determine the distribution of maternal serum PFAS concentrations, which in turn define both the starting point for infant serum PFAS levels and PFAS concentrations in breast milk. For all six PFAS, increasing drinking water concentrations produced a linear increase in modeled serum concentrations.

The slope of the regression line at the 50th and 90th percentiles of the population serum distribution for infants and adults derived using this model compared with literature values are shown in Table 2. These slopes are the change in serum PFAS concentration (μg/L) per one ng/L increase in drinking water PFAS concentration. Comparing model-derived slopes for the six modeled PFAS, differences in slope across chemicals highlight the importance of elimination half-life for the relationship between drinking water exposure and serum levels in breastfed infants: PFHxS with the longest half-life has a large regression coefficient while PFHpA with the shortest half-life has the smallest.

The relationship between drinking water and serum PFAS concentrations has been investigated using a variety of methods in empirical studies. Regression analyses using either measured data^[33,80,81] or modeled data^[82] to relate drinking water and serum PFAS concentrations can control for potentially confounding variables such as age and sex. Another approach is to calculate the serum-to-water ratio from empirical data^[83-86], or similar relationships predicted with TK modeling^[39,87] enabling examination of the distribution of serum to water ratios across a study population. Both regression analysis and serum-to-water ratios yielded similar interpretations, showing an estimated increase in serum PFAS concentrations for each unit increase in drinking water PFAS concentration, providing some comparability to the slope derived from the regression analysis from the TK modeling in this work.

The median and maximum of available literature regression coefficients for PFOA, PFOS, PFHxS, and PFHpA were within 2-fold of the regression coefficients at the median and 90th percentile modeled for adult male populations in this study [Table 2]. For the regression coefficients in infant populations, regression coefficients at the median and 90th percentile were similarly within 2-fold of literature-reported values for PFOA and PFOS, and were larger than literature-reported values for PFHxS and PFHpA. The single available literature regression coefficients for PFNA and PFDA may be biased by methods to impute data points below limits of detection^[80]; regression coefficients at the median of the adult male population modeled for this study were higher for both chemicals. Despite the methodological variability in literature-based values, modeled regression coefficients from this analysis were overall similar to those reported in the literature for multiple PFAS. While literature approaches provide useful empirical summaries of the relationship between drinking water and serum concentrations, they do not explicitly represent the combined effects of life-stage-specific exposure pathways and interindividual variability in TK parameters. By contrast, the population PFAS TK model framework used here integrates these factors to estimate how drinking water exposure translates into population distributions of serum PFAS concentrations over time.

The linear regression equation from the relationship between drinking water and serum PFAS concentrations at the 90th percentile of the distribution was used to calculate the CGWC for breastfed infants and adults. The CGWC is defined as the drinking water concentration for each PFAS predicted to result in 90th percentile serum concentrations reaching the NASEM clinical action level of 20 ng/mL.

Table 3 shows the PFAS CGWCs calculated for each PFAS for breastfed infants and adults. CGWCs are lower for breastfed infants than adults for all PFAS. For both populations, PFHxS has the lowest CGWC and PFHpA has the highest CGWC, underscoring the importance of chemical half-life in determining the relationship between drinking water exposure and serum PFAS levels.

For comparison, we entered the PFOA CGWC into a recently revised, transgenerational PFOA TK model that takes deterministic TK parameters and exposure factors to calculate a reasonable maximum exposure scenario with a combination of central tendency and upper percentile values^[49]. The projected serum concentrations at the CGWC under parallel exposure scenarios for infants and adults were 22.5 and 19.5 ng/mL, respectively, showing good agreement with our model results. The slightly higher infant serum concentration likely stems from the use of a higher LTF and lower volume of distribution used by Greene et al. (2024)^[49].

The CGWC for each PFAS in Table 3 could be used in Equation (4) to evaluate a mixture of PFAS concentrations in drinking water compared to the NASEM clinical action level of 20 ng/mL. CGWCs are not health-protective drinking water values but rather concentrations associated with elevated health risk sufficient to prompt clinical follow-up. The CGWC could be used as a screening tool for clinicians and regulatory agencies to screen for exposure scenarios where enhanced clinical action, health monitoring, and/or exposure interventions are warranted.