The U.S. PFAS exposure burden calculator for 2017–2018: Application to the HOME Study, with comparison of epidemiological findings from NHANES

Abstract

Background:

The 2017–2018 U.S. PFAS exposure burden calculator was designed to provide a summary exposure score for per- and polyfluoroalkyl substances (PFAS) mixtures using targeted PFAS analyte data. Its aim was to place PFAS burden score estimates onto a common scale based on nationally representative U.S. reference ranges from 2017–2018, enabling comparisons of overall PFAS burden scores across studies even if they did not measure the same set of PFAS analytes.

Objective:

To use the U.S. PFAS exposure burden calculator for comparing the same mixture of PFAS compounds in similarly aged adolescents and their associations with cardiometabolic outcomes in the HOME Study and NHANES between 2015–2018.

Methods:

We applied the PFAS burden calculator to 8 PFAS analytes measured in the serum of adolescents from the HOME Study (Cincinnati, Ohio; age range 11–14 years; years: 2016 – 2019; n = 207) and NHANES (US; age range 12–14 years; years 2015–2018; n = 245). We used the non-parametric Mann-Whitney U test and chi-squared test to compare the two study samples. In both studies, we examined associations of PFAS burden scores with the same cardiometabolic outcomes, adjusted for the same core set of covariates. We conducted sensitivity analyses to verify robustness of exposure-outcome associations, by accounting for measurement error of PFAS burden scores.

Results:

PFAS burden scores were significantly different (p=0.004) between the HOME Study (median: 0.00, interquartile range −0.37, 0.34) and the NHANES samples (median: 0.04, IQR −0.11, 0.54), while no significant difference was found for PFAS summed concentrations (p=0.661). In the HOME Study, an interquartile (IQR) increase in PFAS burden score was associated with higher total cholesterol [7.0 mg/dL, 95% CI: 0.6, 13.4]; HDL [2.8 mg/dL, 95% CI: 0.4, 5.2]; LDL [5.9 mg/dL, 95% CI: 0.5, 11.3], insulin [0.1 log(mIU/L), 95% CI: 0.01, 0.2], and HOMA-IR [0.1, 95% CI: 0.01, 0.2]. In NHANES, an IQR increase in PFAS burden score was associated with higher diastolic blood pressure [2.4 mmHg, 95% CI: 0.4, 4.4] but not with other outcomes. Sensitivity analyses in the HOME Study and NHANES were consistent with the main findings.

Conclusions:

Performance of the U.S. PFAS exposure burden calculator was similar in a local versus national sample of adolescents, and may be a useful tool for the assessment of PFAS mixtures across studies.

Introduction:

Nearly every person in the U.S. has detectable levels of per- and polyfluoroalkyl substances (PFAS) in their blood.(Liu et al. 2022; Calafat et al. 2007) PFAS are synthetic and environmentally persistent chemicals(Fenton et al. 2021) used in oil and water-resistant products, firefighting foams, non-stick coatings, and industrial processes.(Glüge et al. 2020) Manufactured since the 1940s, PFAS make their way into the human body through contaminated drinking water, food, and consumer products.(Buck et al. 2011; Cousins et al. 2020; Sunderland et al. 2019) PFAS are linked to negative health outcomes, including higher cholesterol, decreased immune functioning, and some cancers.(Sunderland et al. 2019) There have been calls to regulate PFAS as a chemical class(Kwiatkowski et al. 2020; Trager 2021) instead of one at a time because many PFAS impact similar human health endpoints. Thus, assessing total burden to PFAS may be a more informative approach for health risk assessments.

No gold standard exists for estimating total body burden of PFAS. Most human biomonitoring and epidemiological studies estimate human PFAS exposure by measuring targeted PFAS analytes in serum. The National Academies of Science, Engineering and Medicine (NASEM) recently released clinical recommendations (National Academies of Sciences 2022) for health monitoring based on a sum of 7 PFAS analyte concentrations. While this is an important first step that recognizes the importance of having a summary measure for PFAS, using a simple summation as a proxy of exposure burden has limitations. Using summed PFAS concentrations as a measure of cumulative PFAS exposure may not be sensitive to health effects as compared with using IRT-derived burden scores; for empirical findings that show this, please see Liu et al. 2022.(Liu et al. 2022) Further, various studies measure different sets of PFAS analytes; using a simple sum would require researchers and clinicians to analyze a common set of PFAS analytes. Thus, a standardized measure of PFAS exposure could more easily be compared across studies and facilitate harmonization across studies, consortia, and meta-analyses, thus enabling observations of trends in exposure, assemble larger studies, and making inferences across a more heterogenous population.

We previously developed a novel exposure burden score method for chemical mixtures using item response theory (IRT) and applied it to PFAS analytes.(Liu et al. 2022) As a proof-of-concept, we developed a U.S. PFAS exposure burden calculator for 2017–2018. Building on this work, here we used our U.S. PFAS exposure burden calculator to put PFAS burden metrics on the same scale across studies. We assessed cross-sectional associations of adolescent PFAS burden with cardiometabolic outcomes in the Health Outcomes and Measures of the Environment (HOME) Study(Braun et al. 2017) and made comparisons with similarly aged participants with similar calendar years of serum collection, in the U.S. National Health and Nutrition Examination Survey (NHANES). We compared exposure-outcome associations from those using the PFAS burden calculator as the summary measure to those found when using a simple summation of PFAS concentrations as the summary measure.

Methods

Datasets

HOME Study Sample

We used data from the Health Outcomes and Measures of the Environment (HOME) Study (Braun et al. 2017; Braun et al. 2020), a prospective pregnancy and birth cohort, to demonstrate the utility of the PFAS exposure burden calculator. In the HOME Study, pregnant women from the greater Cincinnati, Ohio region from 2003 to 2006 were enrolled in the longitudinal follow-up of their children at multiple time points. Among the children from a total of 468 pregnant women, 207 had follow-up PFAS data collected at an average age of 12 years (range from 11–14 years), collected in 2016–2019. (Braun et al. 2017; Braun et al. 2020)

Pediatric phlebotomists collected child blood samples via venipuncture. After separating serum from whole blood, the biospecimens were stored at −80 °C until PFAS quantification. The CDC laboratory staff quantified serum concentrations of N-methylperfluoroctane sulfonamidoacetic acid (Me-PFOSA-AcOH), perfluoroundecanoic acid (PFUA), perfluorodecanoic acid (PFDeA), perfluorohexane sulfonic acid (PFHxS), perfluorononanoic acid (PFNA), n-perfluorooctanoic acid (n-PFOA), branch perfluorooctanoic acid isomers (Sb-PFOA), n-perfluorooctane sulfonic acid (n-PFOS), and perfluoromethylheptane sulfonic acid isomers (Sm-PFOS) using online solid phase extraction coupled to high performance liquid chromatography-isotope dilution tandem mass spectrometry.(Kato et al. 2011) PFAS concentrations were measured in a 100 μL serum aliquot diluted with formic acid and fortified with isotopically-labeled PFAS internal standards.(Kuklenyik, Needham, and Calafat 2005; Kato et al. 2011) Each analytic batch included reagent blanks and low- and high-concentration quality control materials that were evaluated using standard statistical probability rules. The limits of detection (LOD) for PFAS were ~0.1 to 0.2 ng/mL. Concentrations below the LOD were replaced with the LOD/√2. (Hornung and Reed 1990)

Cardiometabolic outcomes included serum total cholesterol (TC, mg/dL), high-density lipoprotein (HDL, mg/dL), non-high-density lipoprotein (non-HDL, mg/dL), low-density lipoprotein (LDL, mg/dL), triglycerides (mg/dL), systolic blood pressure (SBP, mmHg), diastolic blood pressure (DBP, mmHg), serum glucose (mg/dL), insulin (mIU/L), HbA1c, Homeostatic Model Assessment of Insulin Resistance (HOMA-IR) and insulin secretion index (HOMA-B). Blood-based cardiometabolic biomarkers were measured with valid and reliable assays(Li et al. 2021) on fasting blood samples (fasting time>8 hours). Three sitting blood pressure measurements were taken one minute apart using a Dinamap Pro100 automated monitor.(Braun et al. 2020; Perloff et al. 1993; Gillman and Cook 1995)

Non-HDL was calculated by subtracting HDL from TC, HOMA-IR was calculated by using the equation of $\text{HOMA}-\text{IR}=\text{fasting insulin}(\text{uU}/\text{mL})\times \frac{\text{fasting glucose}(\text{mmol}/\text{L})}{22.5}$ (Wallace, Levy, and Matthews 2004; Matthews et al. 1985), HOMA-B was calculated by using the equation of $\text{HOMA}-\text{B}=20\times \frac{\text{fasting insulin}(\text{uU}/\text{mL})}{\text{fasting glucose}(\text{mmol}/\text{L})/-3.5}$ (Liu et al. 2016).

We used previous studies to identify potential confounders that may be associated with both adolescent PFAS exposure and adolescent cardiometabolic risk, but which were not causal intermediates or colliders.(Kingsley et al. 2018) Core covariates included child sex, age (in months), race/ethnicity groups (non-Hispanic White, non-Hispanic Black, Hispanic/Other), household income, and body-mass-index (BMI, kg/m²). An expanded set of covariates, used in sensitivity analyses because these variables were not available in the NHANES sample, included breastfeeding status (weeks of breastfeeding), adolescent pubertal stage, and adolescent Healthy Eating Index score. Adolescents self-evaluated their pubertal staging (stages I-V), using standardized instructions based on pubic hair growth, in a private room with a full-length mirror.(Yayah Jones et al. 2021) The Healthy Eating Index was calculated from 3 24-hour food recalls (2 weekdays and 1 weekend day) using the Nutrition Data Systems for Research software and foods database (University of Minnesota, MN). They provide a measure of diet quality, based on concordance with federal dietary guidance.(“University of Minnesota Nutrition Coordinating Center (NCC) Food and Nutrient Database” ; Guenther et al. 2013)

NHANES dataset

We compared findings from the HOME Study with similarly aged adolescents in the National Health and Nutrition Examination Survey (NHANES). NHANES is a publicly available cross-sectional survey of the US population conducted by the National Center for Health Statistics (NCHS) (Zipf et al. 2013). To compare PFAS data collected in a similar time period to that of the HOME Study sample, thus avoiding differences in PFAS exposure profiles due to calendar time, we combined NHANES samples from 2015–2016 and 2017–2018. PFAS analyte concentrations of Me-PFOSA-AcOH, PFUA, PFDeA, PFHxS, PFNA, PFOA, and PFOS were quantified by the CDC laboratories. In addition, we only included NHANES participants aged 12–14 years (n=299). After excluding participants missing PFAS analyte concentrations (n=54), 245 eligible participants in the NHANES data were included in our analysis.

The same cardiometabolic outcomes from the HOME Study sample were also quantified in the NHANES sample. Further, for triglycerides, glucose, insulin, HOMA-IR and HOMA-B, we only included participants whose fasting time was ≥ 8 hours in the NHANES sample. We did not use sample weights for NHANES.

PFAS exposure burden scores in the HOME Study sample and the NHANES sample

We previously applied IRT methods to the 2017–2018 NHANES data and calculated an IRT model to quantify latent PFAS burden score. (Liu et al. 2022) We developed a PFAS exposure burden calculator via an R shiny app (https://pfasburden.shinyapps.io/app_pfas_burden/) to allow researchers to calculate PFAS exposure burden scores on a common scale (a standard normal scale), so that PFAS burden estimates can be compared across studies, even if researchers did not measure exactly the same set of PFAS analytes. (Liu et al. 2022) The PFAS burden calculator uses survey-weighted (using PFAS subsample weights) decile cutoffs for categorizing each PFAS analyte. The PFAS burden score is on a standard normal scale (mean 0, standard deviation 1). A score of 0 means that a person has average PFAS burden for the population, while a score of 1 means a person has PFAS burden that is one standard deviation above average, and a score of −1 means a person has PFAS burden one standard deviation below average. In general, a negative PFAS exposure burden means the participant’s PFAS exposure is below average, based on population-level reference ranges from 2017–2018.

Using this calculator, we calculated PFAS burden scores for the HOME Study sample and for the NHANES sample from 7 PFAS analytes: Me-PFOSA-AcOH, PFUA, PFDeA, PFHxS, PFNA, PFOA and PFOS. We calculated PFOS by summing linear PFOS (n-PFOS) and monomethyl branched isomers of PFOS (Sm-PFOS), and calculated PFOA by summing linear PFOA (n-PFOA) and branched isomers of PFOA (Sb-PFOA). Detection limits and detection rates of the PFAS analytes for the two samples are presented in Supplementary Table 1.

For comparison to the PFAS burden scores, we also calculated summed PFAS concentrations (the sum of all seven PFAS analytes) in the HOME Study sample and NHANES sample. The PFAS assays for both the NHANES and HOME Study samples were conducted at the Centers for Disease Control and Prevention laboratories; thus, the same set of PFAS analytes were available for both studies and we used data from all available PFAS analytes.

Comparing the two study samples

We used Mann–Whitney U tests to compare the exposure to PFAS mixtures, cardiometabolic outcomes, and the matched covariates between the two study samples.

Association of PFAS burden scores with cardiometabolic outcomes

We assessed the association of PFAS exposure burden scores or summed PFAS concentrations with serum TC, HDL, non-HDL, LDL, triglycerides, SBP, DBP, serum glucose, insulin, HbA1c, HOMA-IR and HOMA-B. In regression analyses, triglycerides, insulin, HOMA-IR and HOMA-B were natural log-transformed to account for skewness. Regression analyses were adjusted for the core set of covariates, so that the same set of covariates were used in the HOME Study sample and in the NHANES sample. We adjusted for adjusted for age, sex, race/ethnicity groups, midpoint of household income category, and BMI in the HOME Study sample, and adjusted for age, sex race/ethnicity, ratio of family income to poverty, and BMI in the NHANES sample. We compared the adjusted associations corresponding to a 1 interquartile range (IQR) increase in the PFAS burden scores or the summed PFAS concentrations. Summed PFAS concentrations were natural log-transformed in regression analyses to account for skewness.

Sensitivity Analyses

1. Sensitivity analysis: Accounting for measurement error of PFAS burden scores to verify the robustness of PFAS burden score-cardiometabolic outcome associations

PFAS exposure burden scores are estimated using expected a posteriori (EAP) estimation, which have associated standard errors. (Carroll 2006) Of note, the measurement error of PFAS burden scores varies across levels of the burden score, and depends on the test information curve. Our previous work on the 2017–2018 U.S. PFAS exposure burden calculator shows that errors are minimized for −1 to 2 standard deviations of exposure burden.(Liu et al. 2022) To verify robustness of PFAS exposure burden-cardiometabolic outcome associations, we accounted for measurement errors in estimating PFAS burden scores using a resampling approach. We sampled 1,000 values from the normal distribution with mean equal to the estimated PFAS burden score and standard deviance equal to the standard error of the estimated PFAS burden for each individual. We treated the 1,000 values for each participant as the new estimated PFAS burden scores and assessed the association of the scores with the cardiometabolic outcomes that were found to be significant in our main findings. We then assessed whether the significant findings were consistent in 1,000 resamples.

2. Sensitivity analysis: Accounting for measurement error of summed PFAS concentrations to verify the robustness of summed PFAS-cardiometabolic outcome associations

Summed PFAS concentrations can also be treated as a latent variable measured with error. This idea is the fundamental premise of classical test theory (CTT), a precursor to item response theory.(Lord, Novick, and Birnbaum 1968) In CTT, the same standard error (SE) is used for each sum score, and the standard error is a function of the reliability of the entire set of items (i.e., all PFAS analyte concentrations). Reliability may be estimated in several ways, but a common choice is Cronbach’s alpha,(Guttman 1945; Cronbach 1951) which reflects the internal consistency of the set of items. For a set of $k$ PFAS analytes with observed concentration $X_i$ for biomarker $i$, and observed concentration $X_j$ for biomarker $j$,

$$alpha=\frac{k}{k-1}(1-\frac{{\sum }_{i}var\left(X_i\right)}{{\sum }_{i}var\left(X_i\right)+{\sum }_{i\ne j}cov\left(X_i,X_j\right)})$$

and the standard error of sum scores $\left(\sum X_i\right)$ equals

$$SE\left(\sum X_i\right)=\sqrt{var\left(\sum X_I\right)(1-alpha)}.$$

We calculated a constant error for the summed PFAS concentrations using Cronbach’s alpha. We used the same resampling framework as the sensitivity analysis corresponding to PFAS burden scores, to verify exposure-outcome associations while accounting for measurement error of summed PFAS concentrations.

3. Sensitivity analysis: Association of PFAS burden scores with cardiometabolic outcomes adjusting for a larger set of covariates in the HOME Study sample

We also assessed the associations between PFAS burden score and cardiometabolic outcomes after adjusting for additional covariates (core covariates plus breastfeeding status, pubertal stage and Healthy Eating Index) in the HOME Study sample.

4. Sensitivity analysis: Calibrating a separate IRT model in the HOME Study sample, and associations with cardiometabolic outcomes using a larger set of covariates

The PFAS burden calculator was developed using 2017–2018 U.S. nationally representative data. (Liu et al. 2022) To verify robustness of study findings, we calibrated a new IRT model for the HOME Study sample using the same survey-weighted cutoffs from the 2017–2018 NHANES PFAS data. We then estimated HOME Study-specific PFAS burden scores, calculated from the HOME Study IRT model, and assessed associations with cardiometabolic outcomes, using the larger set of covariates (core covariates plus breastfeeding status, pubertal stage and Healthy Eating Index).

Results

Comparison of study participant profiles in the HOME Study sample vs. NHANES sample

We observed no significant difference in child sex for the HOME Study sample and the NHANES sample (p=0.868) (Table 1). In contrast, children in the HOME Study sample were younger than those in NHANES (p<0.001). Median age in the HOME Study sample was 12.3 years (IQR [11.9, 12.8]), and in the NHANES sample was 13 years (IQR [12, 14]). Children in the HOME Study sample had lower BMI (p<0.001), lower HbA1c (p<0.001), lower serum glucose (p<0.001), lower triglycerides (p=0.036), lower systolic blood pressure (p<0.001), and higher total cholesterol (p=0.048) compared with children in the NHANES sample.

Table 1:

Summary statistics of the HOME Study and NHANES samples.

We presented the median (interquartile range) continuous variables and presented the percentage for categorical variables. Chi-squared test was used to compare categorical variables and Mann–Whitney U test was used to compare continuous variables between the two study samples.

Table 1A: Socio-demographic characteristics of the HOME Study and NHANES samples.
	HOME Study sample (n=207)	Number of missing	NHANES sample (n=245)	Number of missing	P
Sex (%)		0		0	0.868
Male	52.7		51.4
Female	47.3		48.6
Race/ethnicity
Race/ethnicity (3 groups in HOME Study sample) (%)		0
Non-Hispanic White	59.4
Non-Hispanic Black	34.8
Hispanic/Other	5.8
Race/ethnicity origin (5 groups in NHANES sample) (%)	NA			0
Mexican American			20.4
Other Hispanic			12.2
Non-Hispanic White			29.0
Non-Hispanic Black			21.2
Other Race - Including Multi-Racial			17.1
Age (in years) (median [interquartile range])	12.3 [11.9, 12.8]	0	13.0 [12.0, 14.0]	0	<0.001
SES (median [interquartile range])		0		21	NA
Midpoint of household income category at all time points	75000 [35000, 150000]		NA
Family income to poverty ratio	NA		1.7 [1.1, 2.9]
Table 1B: Summary statistics of outcomes and other covariates in the HOME Study and NHANES samples
	HOME Study sample (n=207)	Number of missing	NHANES sample (n=245)	Number of missing	P
HbA1c	5.2 [5.0, 5.3]	8	5.3 [5.1, 5.5]	144	<0.001
Glucose (mg/dL)	91.1 [88.0, 95.5]	4	98.0 [94.0, 101.0]	148	<0.001
Insulin (mIU/L)	12.3 [8.2, 18.8]	3	11.4 [7.8, 17.7]	148	0.401
HOMA-IR	2.8 [1.9, 4.2]	4	2.8 [1.9, 4.4]	150	0.995
HOMA-B	0.2 [0.1, 0.2]	4	0.1 [0.1, 0.2]	150	0.001
Total cholesterol (mg/dL)	154.5 [142.2, 174.6]	4	148.5 [134.0, 172.3]	1	0.048
HDL (mg/dL)	52.6 [44.6, 59.8]	4	51.0 [44.0, 60.3]	1	0.935
non HDL	102.0 [86.6, 121.5]	4	98.0 [79.0, 121.3]	1	0.130
LDL	83.3 [72.9, 99.4]	4	85.0 [66.0, 100.0]	144	0.559
Triglycerides (mg/dL)	76.3 [59.7, 98.4]	4	68.0 [46.0, 95.0]	150	0.036
Systolic BP (mmHg)	101.0 [96.0, 108.0]	0	106.0 [102.0, 112.7]	10	<0.001
Diastolic BP (mmHg)	57.0 [53.0, 60.0]	0	58.0 [51.3, 65.3]	10	0.133
BMI	19.9 [17.6, 23.1]	0	22.2 [18.9, 26.5]	2	<0.001
HOME specific variables
FBMI z-score	0.2 [−0.4, 0.9]	6
LBMI z-score	−0.2 [−1.1, 0.6]	6
Healthy eating index	43.9 [37.3, 52.8]	2
Breastfeeding status (weeks of breastfeeding)	16.0 [1.5, 39.5]	12
Pubic Hair Tanner Stage (%) *
1	9.8
2	25.4
3	28.3
4	21.5
5	15.1

^*At age 12 years, we assessed pubertal development of participants using the Tanner staging system with a range from stage 1 indicating no signs of initiating puberty to stage 5 indicating fully mature status. Female participants self-assessed Tanner staging of breast and pubic hair development and males were assessed pubic hair development by completing a self-reported questionnaire using line drawings and descriptions understandable to children. (Yayah Jones et al., 2021)

Participants in the HOME Study sample had lower serum Me-PFOSA-AcOH (p<0.001), PFDeA (p<0.001), PFNA (p<0.001), PFUA (p<0.001), and higher PFOA (p<0.001) concentrations than participants in the NHANES sample. Serum PFDeA and PFOS concentrations did not significantly differ between the HOME Study and NHANES samples. Median summed PFAS concentrations was 5.14 (IQR [3.88,6.96]) in the HOME Study sample and 5.18 (IQR [3.88,7.24]) in the NHANES sample. Median PFAS burden was 0.00 (IQR [−0.37, 0.34]) in the HOME Study sample and 0.04 (IQR [−0.11, 0.54]) in the NHANES sample. Summed PFAS concentrations did not significantly differ between the HOME Study sample and NHANES sample, while the HOME Study sample had significantly lower PFAS exposure burden (p=0.004). (Supplementary Table 2) The distributions of PFAS exposure burden and summed PFAS concentrations in the two study samples are presented in Figure 1. Pearson correlation coefficients between PFAS burden scores and summed PFAS concentrations were 0.75 (p<0.001) in the HOME Study sample and 0.79 (p<0.001) for the NHANES sample (Figure 2). Pearson correlation coefficients between PFAS burden scores and log transformed summed PFAS concentrations were 0.96 (p<0.001) in the HOME Study sample and 0.95 (p<0.001) for the NHANES sample.

Figure 1:

Distributions of PFAS exposure burden and summed PFAS concentrations in the HOME Study and NHANES samples

Figure 2:

Scatterplot of PFAS exposure burden vs. summed PFAS concentrations in the HOME Study and NHANES samples. Pearson correlation coefficients are presented.

Association of PFAS burden scores and cardiometabolic outcomes in the HOME Study sample vs. the NHANES sample:

In the HOME Study, PFAS burden was associated with five cardiometabolic outcomes (total cholesterol, HDL, LDL, insulin, and HOMA-IR); Figure 3, Supplementary Table 3. An IQR increase in the PFAS burden score was associated with 7.0 (95% CI: [0.6, 13.4]) higher total cholesterol (mg/dL), 2.8 (95% CI: [0.4, 5.2]) higher HDL (mg/dL), 5.9 (95%CI: [0.5, 11.3]) higher LDL (mg/dL), 0.1 (95% CI: [0.01, 0.2]) higher insulin (log(mIU/L)), and 0.1 (95% CI: [0.01, 0.2]) higher HOMA-IR. In NHANES, PFAS burden was associated with higher DBP. An IQR increase in PFAS burden score was associated with 2.4 mmHg (95% CI: [0.4, 4.4]) higher DBP.

Figure 3:

Adjusted associations corresponding to 1 IQR increase in PFAS exposure burden with cardio-metabolic outcomes in the HOME Study (adjusted for age, sex, race/ethnicity groups, midpoint of household income category, and BMI) and NHANES (adjusted for age, sex race/ethnicity, ratio of family income to poverty, and BMI). For consistency of adjusted associations across the HOME Study and NHANES, adjusted associations were calculated using the IQR of PFAS burden in NHANES (IQR=0.65).

Association of summed PFAS concentrations and cardiometabolic outcomes in the HOME Study sample vs. the NHANES sample:

In the HOME Study, PFAS summed concentrations were associated with two cardiometabolic outcomes (insulin and HOMA-IR); Figure 4; Supplementary Table 4. An IQR increase in log-transformed summed PFAS was associated with higher insulin [0.1 log(mIU/L), 95% CI:0.01, 0.1] and higher HOMA-IR [0.1, 95% CI: 0.01, 0.2]. In NHANES, PFAS summed concentrations were associated with higher DBP. An IQR increase in log-transformed summed PFAS was associated with 3.2 mmHg (95% CI: [0.9, 5.5]) higher DBP.

Figure 4:

Adjusted associations between PFAS summed concentrations (natural log transformed) and cardio-metabolic outcomes in the HOME (adjusted for age, sex, race/ethnicity groups, midpoint of household income category, and BMI) and NHANES (adjusted for age, sex race/ethnicity, ratio of family income to poverty, and BMI) study samples. We used the IQR of log PFAS sum concentration [IQR=0.62 log(ng/mL)] of 12–14 years old children in 2015–2016 and 2017–2018 NHANES cycles to calculate 1 IQR effect sizes.

Sensitivity analysis:

1. Sensitivity analysis: Accounting for measurement error of PFAS burden scores to verify the robustness of PFAS burden score-cardiometabolic outcome associations

Sensitivity analysis verified the robustness of the adjusted associations between PFAS burden score associations and cardiometabolic outcomes (Supplementary Figure 1). In the HOME Study, PFAS burden (calculator) was consistently associated with total cholesterol (p<0.05 in 862 of the 1000 resamples, p<0.1 in 999 of the 1000 resamples), HDL (p<0.05 in 941of the 1000 resamples, p<0.1 in all of the 1000 resamples), LDL (p<0.05 in 812 of the 1000 resamples, p<0.1 in 978 of the 1000 resamples), insulin (p<0.05 in 993 of the 1000 resamples, p<0.1 in all of the 1000 resamples), and HOMA-IR (p<0.05 in 957 of the 1000 resamples, p<0.1 in all of the 1000 resamples).

In NHANES, PFAS burden was associated with higher DBP (p<0.05 in all 937 of the 1,000 resamples, p<0.1 in all 1000 resamples).

2. Sensitivity analysis: Accounting for measurement error of summed PFAS concentrations to verifying the robustness of summed PFAS-cardiometabolic outcome associations

In the HOME Study sample, summed PFAS concentrations were associated (p<0.05) with higher insulin in 83.7% of resamples, and with higher HOMA-IR in 64.8% of resamples (Supplementary Figure 2). In NHANES, summed PFAS concentrations were associated with DBP in 100% of the resamples.

3. Sensitivity analysis: Accounting for a larger set of covariates in the HOME Study sample to study PFAS burden-cardiometabolic outcome associations

After adjusting for a larger set of covariates in the HOME Study, PFAS burden scores were no longer significantly associated with higher total cholesterol or LDL (Figure 5 and Supplementary Table 5). An IQR increase in PFAS burden continued to be associated with higher HDL 3.0 mg/dL (95% CI: [0.5, 5.5]), insulin 0.1 log mIU/L (95% CI: [0.01, 0.1]), HOMA-IR 0.1 (95% CI: [0.003, 0.2]). PFAS burden was borderline significantly associated with higher HOMA-B (0.1, 95% CI: [−0.01, 0.2], p=0.062). Supplementary Figure 3 shows the results of associations accounting for measurement error.

Figure 5:

Sensitivity analysis that shows adjusted associations corresponding to 1 IQR increase in PFAS exposure burden with cardio-metabolic outcomes in the HOME after adjusting for a larger set of covariates, for PFAS burden scores calculated from the U.S. 2017–2018 PFAS burden calculator, and HOME-specific burden scores calculated using an IRT model calibrated in the HOME Study sample. Associations were adjusted for age, sex, race/ethnicity, SES, BMI, breastfeeding status, pubertal status, and Healthy Eating Index). For consistency of adjusted associations across the HOME Study and NHANES, adjusted associations were calculated using the IQR of PFAS burden in NHANES (IQR=0.65).

4. Sensitivity analysis: Calibrating a separate IRT model in the HOME Study sample, and associations with cardiometabolic outcomes using a larger set of covariates

We calibrated a separate IRT model using HOME Study data; Supplementary Table 6 presents the item parameters; Supplementary Figure 4 provides the item information curves. HOME Study-specific burden scores were highly correlated with PFAS burden scores calculated from the U.S. 2017–2018 PFAS exposure burden calculator (Pearson correlation = 0.97; p<0.001); Supplementary Figure 5. HOME Study-specific PFAS burden scores were still significantly associated with higher HDL, insulin, HOMA-IR, and HOMA-B (Figure 5; Supplementary Table 5). An IQR increase in HOME Study-specific PFAS burden scores was associated with 1.5 (95% CI: [0.1, 2.8]) higher in HDL (mg/dL), 0.1 (95% CI: [0.01, 0.1]) higher insulin (log(mIU/L)), 0.1 (95% CI: [0.002, 0.1]) higher HOMA-IR, and 0.1 (95% CI: [0.01, 0.1]) higher HOMA-B. Supplementary Figure 6 shows the results of associations accounting for measurement error.

Discussion:

We employed the U.S. PFAS exposure burden calculator for 2017–2018 in the HOME Study to examine cross-sectional associations of adolescent PFAS burden and cardiometabolic outcomes, and compared our findings with similarly aged participants in NHANES. Distributions of PFAS burden were similar across both samples. We found that in the HOME Study, higher PFAS burden was significantly associated with higher total cholesterol, HDL, LDL, insulin, HOMA-IR and borderline significantly associated with triglycerides and HOMA-B. In NHANES, higher PFAS burden was associated with higher DBP. We compared our findings of using PFAS burden to using natural log transformed summed PFAS concentration. Log summed PFAS concentration was not significantly associated with HDL, and was borderline associated with LDL. We adjusted for additional covariates (breastfeeding, pubertal stage, and Healthy Eating Index score) when assessing the associations between PFAS burden and outcomes. Associations between PFAS burden and HDL, insulin, HOMA-IR and HOMA-B were consistent in the HOME sample after adjusting for additional covariates unique to the HOME Study, while PFAS burden was no longer associated with LDL.

Our findings differed somewhat between the NHANES and HOME Study samples. There are several potential reasons for this. One potential reason is the different population make-up of the studies, with the HOME Study based in the greater Cincinnati, Ohio region while the NHANES study is a sample across the United States and the geographic sample differs in different NHANES cycles. The HOME Study children reside in a region with historical PFOA contamination, which could have influenced their PFAS exposure profiles and vulnerability. Second, the HOME Study population differs from the NHANES population on population characteristics including the outcome distributions, which may influence vulnerability. In prior analyses of the HOME Study, we found significant associations between exposure to PFAS during pregnancy and adolescent cardiometabolic outcomes. Higher maternal serum PFOA during pregnancy was significantly associated with higher HOMA-IR, visceral adiposity area, and leptin to adiponectin ratio in children at 12 years old.(Li et al. 2021) In addition, prenatal serum PFOA concentration was associated with altered BMI trajectories between birth and 12 years of age, as well as increased risk of being overweight or obese. (Braun et al. 2022; Liu et al. 2020) Previous results from the HOME Study did not suggest that individual postnatal PFAS concentrations were associated with cardiometabolic risk markers at age 12 years.(Liu et al. 2020) The lack of associations may have been because we did not consider measures of aggregate exposure like the burden calculation employed in the present study.

Previous literature has identified associations of individual PFAS and cardiometabolic outcomes (Hoyer et al. 2015; Frisbee et al. 2010; Maisonet et al. 2015; Lin et al. 2009; Geiger, Xiao, and Shankar 2014; Geiger et al. 2014), rather than studying the exposure burden to the PFAS mixture. Several studies also found significant cross-sectional associations between individual PFAS and cardiometabolic outcomes in adolescents among the US population using NHANES survey data.(Lin et al. 2009; Geiger, Xiao, and Shankar 2014; Geiger et al. 2014) Geiger et al found that higher serum PFOA concentrations were associated with increased total cholesterol, increased LDL cholesterol, and decreased HDL cholesterol among participants aged ≤18 years in 1999–2008 NHANES.(Geiger et al. 2014) Using data from 474 adolescents in 1999–2000 NHANES, Lin et al assessed the associations between exposure to 4 PFAS (PFOA, PFOS, PFNA and PFHxS) and metabolic syndrome. They reported a significant positive association between PFNA and insulin, prevalence of blood glucose above 5.55 mmol/l or a self-report of taking antihyperglycemic medications, negative association between log PFNA and prevalence of HDL below 1.04 mmol/l. They reported a negative association between log PFOA and prevalence of blood glucose above 5.55 mmol/l or a self-report of taking antihyperglycemic medications.(Lin et al. 2009) Due to regrettable substitution, and because they may exert similar health effects, it may be helpful to study PFAS as a mixture. An exposure burden score to the mixture can also be helpful towards report-back in studies to help summarize an individual’s PFAS burden and how it compares to others to the community. By placing the PFAS burden summary metric on the same scale, which was calibrated in a nationally representative set of the U.S. population, we can then compare burden across individuals in the U.S. In this study, both samples used the CDC labs to quantify PFAS analytes, and thus the same set of biomarkers could be studied. However, there may be other settings in which different studies or laboratories may quantify different sets of PFAS analytes, and we have previously shown that the burden calculator could be used even if studies do not assay the same set of PFAS.(Liu et al. 2022) As our PFAS burden calculator was developed for 2017–2018, further research is needed to extend the PFAS burden calculator over calendar time to address regrettable substitution and time trends in phase out of some PFAS and replacements by others.

Our previously developed PFAS burden calculator can calculate burden scores in two ways: 1) using the 7 PFAS analytes (Me-PFOSA-AcOH, PFUA, PFDeA, PFHxS, PFNA, PFOA and PFOS) in the NASEM clinical recommendation and 2) using the 5 PFAS (Me-PFOSA-AcOH, PFUA, PFDeA, PFHxS, PFNA) and linear and branched isomers of PFOA (n-PFOA, Sb-PFOA was not included in the IRT model because the detection rate of Sb-PFOA was too low) and PFOS (n-PFOS and Sm-PFOS).(Liu et al. 2022) In our previous study, we showed that the PFAS burden calculator we developed can calculate PFAS burden scores even if not all the 7 PFAS analytes were available in the data, and the associations between PFAS burden and health outcomes would be consistent whether or not all 7 PFAS analytes were measured. Therefore, our PFAS burden calculator may be a useful tool to compare PFAS burden across studies. Further studies are needed to evaluate the adaptability of the PFAS burden calculator when PFAS analytes are measured in different laboratories.

The analysis had some strengths and limitations. Our analysis, which only includes targeted PFAS analytes, will not capture the contribution of unmeasured PFAS to total body burden. Additionally, we did not validate the calculated burden index against biological markers of total PFAS body burden, like extractable organic fluorine. A key strength is that burden scores from an IRT model independently calibrated in the HOME Study sample was highly correlated with burden scores from the U.S. PFAS burden calculator for 2017–2018. In addition, we were able to adjust for a comparable set of covariates in both datasets. Another limitation is the cross-sectional nature of these data. However, we did not set out to make causal inferences from these data, instead trying to show the utility of the PFAS body burden index. In the NHANES sample, there were a large number of participants with missing self-reported fasting time. We restricted the regression analyses on relevant outcomes to those with fasting time greater or equal to eight hours, and as a result had a much smaller sample size for some outcomes compared with the HOME Study. For the glucose, insulin, HOMA-IR, HOMA-B and triglycerides regression models, sample size was much smaller in NHANES, so perhaps we did not have adequate sample size to detect significant associations. Although our PFAS burden score calculator used NHANES sample weights to calculate survey weighted decile cutoffs of the PFAS analytes when calibrating the IRT models, we did not use sample weights to assess the association between PFAS exposure and outcomes in the NHANES sample, as we wanted to make comparisons to the HOME Study which had a similar sample size. Thus, our association findings in the NHANES sample cannot be generalized to the US population. There may be unmeasured or residual confounding in the reported calculations and associations. We analyzed the two samples separately in our study and did not conduct meta-analysis given our goal was to compare the results for two independent samples.

Because the goal of this paper is to demonstrate the utility of the PFAS exposure burden calculator in two studies, we did not compare findings with that from other supervised mixtures methods, such as quantile g-compuation, Bayesian kernel machine regression, and weighted quantile sum regression. Because these methods are all supervised methods, this means that the exposure mixture is modeled simultaneously with the outcome, and hence the identified “mixture” may change with different outcomes. Meanwhile, the PFAS exposure burden score is unsupervised, calculated independently of health outcomes, so that the mixture exposure metric stays constant across outcomes. Therefore, we feel that they cannot be directly compared because supervised and unsupervised approaches are not equivalent and although they are two approaches to understand the effects of a mixture, they do it in different ways.

Regulatory agencies and clinicians may increasingly monitor aggregate levels of PFAS; for example, the NASEM panel has recommended that clinicians use the sum of seven PFAS analytes for human biomonitoring. Further research is needed to identify the optimal ways to quantify and compare cumulative PFAS burden in human biomonitoring studies. Here, we found that performance of the U.S. PFAS exposure burden calculator was similar in a local versus national sample of adolescents, and may be a useful tool for the assessment of cumulative exposure to PFAS mixtures across studies.

Highlights:

• A PFAS exposure burden calculator estimate a person’s overall exposure burden to PFASs

• In the NHANES sample, adolescent PFAS exposure burden was associated with higher cholesterol and insulin resistance

• In the HOME Study sample, adolescent PFAS exposure burden was associated with higher diastolic blood pressure

• A standardized PFAS exposure burden metric can be useful for cross-study comparisons