Participants and study design
The study consist of 16,224 individuals aged 20-39 years recruited in the health surveillance program offered to the community of Veneto Region who was exposed for several decades to PFAS via drinking water distributed by contaminated public waterworks. The health surveillance program has been described in more detail elsewhere . In brief, the target population was initially constituted by people born between 1951 and 2002 and residing in the municipalities where were identified as in the area served by PFAS-contaminated waterworks. Surveillance involved the active invitation of the eligible population and the free offer of health examinations including: I) a questionnaire on personal health history and lifestyle habits, socio-demographic characteristics, self-reported height and weight; II) measurement of blood pressure; and III) non-fasting blood and urine samples.
Pregnant women (n=327), and individuals with missing information on the selected covariates (n=111) were excluded, leaving in the analyses 15,786 subjects. No missing data on exposure and outcome variables were present (See Supplementary Figure1, Additional File 1).
Serum concentrations of twelve PFAS were measured by HPLC MS/MS (Shimadzu UFLC XR 20 Prominence coupled to Sciex API 4000): perfluorooctanesulfonate (PFOS), perfluorooctanoic acid (PFOA), perfluorohexanesulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluoroheptanoic acid (PFHpA), perfluorobutanesulfonic acid (PFBS), perfluorohexanoic acid (PFHxA), perfluorobutanoic acid (PFBA), perfluoropentanoic acid (PFPeA), perfluorodecanoic acid (PFDeA), perfluoroundecanoic acid (PFUnA), and perfluorododecanoic acid (PFDoA). Details of the analytical method have been described elsewhere .
Method performances allow analytes to be detected as low as 0.1 ng/mL (LOD) and to be quantified above 0.5 ng/mL (LOQ). Only four PFAS quantifiable in at least 40% of samples were considered for the analyses: PFOA (detected in 99.86% of people), PFOS (detected in 99.70% of people), PFHxS (detected in 96.72% of people), PFNA (detected in 49.86% of people). Samples below the LOQ were assigned a value equal to LOQ/√2. The most extreme outliers of PFAS serum levels were removed as followings: (PFOA>700 mg/L (n=6), PFOS >50 mg/L (n=9), PFHxS>100 mg/L (n=3), PFNA >10 mg/L (n=1)).
Blood pressure and hypertension
Blood pressure (BP) was measured by trained nurses with participants first sitting at rest for at least five minutes, according to the European Society of Hypertension recommendations . A validated semi-automatic sphygmomanometer with an appropriate cuff size for the arm circumference was used. When the first measure was ≥140 mmHg for systolic blood pressure (SBP) or ≥90 mmHg for diastolic blood pressure (DBP), a second measurement was taken at least two minutes apart. In general, 1714 subjects went through the second measurement. When the second measurement was within the cut offs, the second measurement was used (n=1078). Otherwise, the mean of the two measurements was considered when both measurements were above the cut-offs (n=636).
Medical history data were collected directly from participants by trained nurses via structured software-based questionnaire using in-person interviews at the study enrolment. The questionnaire included items on personal health history ("Which diseases do you suffer from?") and medications ("Do you take any medication on a regular basis?" "If yes, which medications do you take?").
Hypertension was defined considering any self-reported diagnosis of hypertension, reported use of antihypertensive medications, or raised SBP (≥140 mmHg) or DBP (≥90 mmHg).
We obtained information on age, gender, country of birth, education level, smoking habits, body mass index (BMI), physical activity, history of certain diseases, medication, alcohol consumption, and food intakes including salt habit. Standard data checks and cleaning procedures (e.g. range and consistency checks) were used to minimize errors and missing values and to maximize data quality. Data on food consumption were transformed from number of serving per day/week/month to number of serving per week for all the food categories to create harmonized diet pattern classiﬁcation. After checking the accuracy of data on BMI regarding height and weight, BMI was recalculated and classified as underweight (<18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), obese (≥30 kg/m2). Alcohol consumption was categorized in 0, 1-2, 3-6, 7+ alcohol units per week. Smoking status was subdivided into current smokers, previous smokers and non-smokers. Degree of physical activity (Light, Moderate, or Heavy) was defined based on an algorithm that combined information reported by the subject on intensity, duration, and frequency of all types of physical activity practiced during the week . Countries of birth were classified in two categories based on geographical areas including: Italy plus other Highly Developed Countries, and High Migratory Pressure Countries. The time-lag between the beginning of the study (1st January 2017) and the date of enrollment was calculated for each subject and included as possible covariate (number of months). Information on the center in charge of the BP measurements was considered as possible confounder in statistical analyses.
Covariates to be included as potentially confounders of the BP/PFAS association were selected from the available variables, based on related literature, through the construction of a directed acyclic graph (DAG) representing the identification of a minimally sufficient set of variables to control confounding. The minimally sufficient adjustment set was identified using DAGitty v1.0 (www.dagitty.net) implemented in R (R Development Core Team 2010, R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL: http://www.R-project.org/).
The serum concentrations of PFAS by gender were expressed as arithmetic mean, standard deviation (SD) and percentiles. Since data on PFAS were markedly skewed to the right, concentrations were natural log (ln) transformed in order to improve normality of the data distribution. Spearman's correlation (ρ) was used to describe pair-wise relations between the PFAS.
Our main outcomes are continuous SBP and DBP. For these analyses, participants with self-reported diagnosis of hypertension or under treatment with antihypertensive medications (n=406) were excluded, leaving 15,380 subjects in analysis (See Supplementary Figure1, Additional File 1).
We used generalized additive models (GAMs) to analyze the relation between each (In) PFAS and BP outcomes, adjusted for the potential set of confounders. In order to explore the shape of possible associations between PFAS and BP levels the models used thin plate spline smooth terms  for the exposures and continuous covariates, and plotting the predicted values. Degree of smoothing was selected by generalized cross validation as implemented in the R package mgcv . Since the spline analysis showed associations compatible with a linear relationship on the In PFAS, linear regression coefficient (β) and 95% confidence intervals (CI) were reported. Serum PFAS levels were also categorized into quartiles, in order to limit the influence of extreme values, with the exception of PFNA for which the large proportion below the LOQ did not allow the quartiles subdivision.
For the analyses on PFAS associations with hypertension prevalence, a binomial link function was used in the models and Odds Ratios (ORs) were calculated, together with their 95% confidence intervals (95%CI).
All analyses were fully adjusted for the established set of covariates: age, BMI, time-lag between the enrolment and the beginning of the study (all continuous variables modelled using thin plate spline) and categorical covariates including gender, physical activity, smoking habits, food consumption (tertiles or quartiles of fruit/vegetables, milk/yogurt, cheese, meat, sweet/snacks/sweet beverage, eggs, fish, bread/pasta/cereals per week), salt habit, country of birth, alcohol consumption, education level and center in charge of the BP measurement (Lonigo, Legnago, San Bonifacio, and Noventa Vicentina).
All the above analyses have been also stratified according to gender and an interaction term between gender and ln-PFAS was also added to the main models.
Since PFAS are predominantly excreted by the kidney through glomerular filtration and impaired kidney function is associated with raised BP , to assess for possible confounding a sensitivity analysis was conducted adjusting all models for estimated glomerular filtration rate (eGFR with cut-off <90 mL/min) calculated according to the CKD-EPI equation .
Finally, we analysed SBP and DBP associations with PFAS excluding subjects with raised BP (SBP≥140 mmHg or DBP ≥90 mmHg).
The procedures of the health surveillance program changed over time: until 31 December 2017, blood sampling, and interview and BP measurement were carried out in the same session for each participant, whilst thereafter they were performed in two different sessions roughly one month apart in order to be able to provide blood test results on the day the participant came for the interview and the BP measurement. To explore whether this organizational change may have affected the PFAS/BP associations, analyses were also restricted to the subgroup of 10,656 individuals recruited after 31 December 2017.
The level of statistical significance was set at 0.05. The statistical software STATA/SE version 13.0 (Stata Corp LP, College Station, TX, USA) and R (R Development Core Team 2010, R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL: http://www.R-project.org/) was used for statistical analyses.