Ethical statement

The study was approved by the Ethics Committee of Uppsala University, and all the participants gave their informed consent prior to the study.

Cohorts

Two population-based Swedish cohorts were utilized for the present study. The Epihealth study is a large cohort study including men and women in the age range 45–75 years from the Swedish general population, as described in detail previously [13]. Briefly, participants were randomly selected from the population registries of the Swedish cities Malmö and Uppsala between 2011 and 2016 with the response rate of approximately 20%. The present study includes a subset of 2373 individuals for whom data exists on both PFAS measurements (metabolomics) and prevalent diabetes between the years 2011 and 2016.

The Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study is a longitudinal investigation over 10 years. A total of 1016 subjects (50% women), aged 70 years and living in Uppsala, were investigated at baseline during 2001–2004. The participants were invited to follow-up examinations at age 75, (n = 822), and at 80 years (n = 603). During the first 5 years of the study, 52 individuals passed away and 142 withdrew. During the next 5 years, 106 individuals passed away and 113 subjects withdrew. All measurements were carried out with essentially the same protocol at age 70, 75 and 80 years. More detailed information on the study population can be found in Lind et al. 2005 [14].

Physical examinations

Across both cohorts, body mass index (BMI) was calculated as weight in kilograms divided by the square of body height in meters (kg/m2). In EpiHealth, weight was measured on a scale that uses bioelectrical impedance analysis to also calculate total fat mass. Total fat mass was then divided by total body weight (Tanita body composition analyzer BC-418MA, Tokyo, Japan).

Questionnaires

In EpiHealth, a web-based questionnaire about medical and family history and symptoms as well as lifestyle factors, including diet (e.g., fish intake), was filled in by all participants. Fish intake was converted to grams per day using the consumption frequencies of different kinds of fish (cod, tuna, salmon, etc.) using standardized portion sizes. Participants also reported medication usage, leisure time, and physical activity in five levels from low (level 1) to strenuous physical activity (level 5). They also reported age, sex, alcohol intake (drinks per week), education length (up to 9 years, 10–12 years, or >12 years), and current smoking habit (smoked years in life).

In PIVUS, the participants were asked to answer a questionnaire regarding their socioeconomic status, medical history, physical activity, smoking habits and regular medication. Unfortunately, no data on dietary intake was collected in the PIVUS study.

Healthy lifestyle index (HLI) in EpiHealth

We have previously constructed a healthy lifestyle index (HLI) to use as a confounder in our analyses. The HLI includes physical activity (levels 4 and 5 considered healthy), healthy diet (upper 25% of individuals in adherence to a previously described healthy dietary pattern), sleeping habits (7–9 h of sleep considered healthy), alcohol intake (>1 drink/day considered unhealthy), stress (never or rarely stressed considered healthy), smoking (smoked <2 years in life defined as non-smoker).

PFAS analyses

In Epihealth, 100 mL of blood was taken, and 90 mL was prepared into plasma, serum and whole blood (for later DNA extraction) and stored in −80 °C in a biobank facility for later PFAS analysis. Blood samples from 2373 individuals were randomly chosen for PFAS analysis and PFAS levels could be detected in >95% of the study population. PFAS (PFHxS, PFOA and PFOS) were analyzed in plasma by non-targeted metabolomics (Metabolon Inc, Morrisville, NC; UAS). After the analysis, the PFAS values were normalized and expressed in relative concentrations. In a third cohort (the POEM study, Uppsala, Sweden, 502 males and females, all aged 50 years) PFAS were measured both with the same relative technique as used in EpiHealth and with the same quantitative technique using standards as used in the PIVUS study. When linear regression analysis were performed, the fit between the relative and quantitative technique was very good with correlations coefficients ranging from 0.83–0.96 for PFOS, PFOA and PFHxS. From those linear regression models, formulas could be derived to estimate quantitative values for the values being initially measured on the relative scale. The same formulas were then used in the EpiHealth study to give a rough estimate of the median levels of PFOS, PFOA and PFHxS in that cohort.

In the PIVUS study, blood serum and plasma were collected in the morning (8–10 am) after an overnight fast and stored in freezers (−70 °C) until later analysis. The current study evaluated six PFAS for which >75% of the study population showed measurable levels above the lower limit of detection (LOD); PFHxS, PFOA, linear isomer of PFOS, PFNA, PFDA, and PFUnDA. PFAS levels were analyzed by UPLC-MS/MS as previously described [15]. The method detection limits (MDLs) for all three investigations ranged from 0.01–0.18 ng mL−1 depending on the analyte. PFAS values below LOD were replaced by LOD/√2.

Outcome assessment

In EpiHealth, prevalent diabetes was defined as either taking diabetes medication or having fasting glucose levels of ≥7 mmol/L. In PIVUS, we could not use incident diabetes as outcome due to the limited number of cases. Plasma glucose levels were analyzed by routine laboratory methods at Uppsala University Hospital, Uppsala, Sweden after an overnight fast.

Statistical analyses

STATA 16 was employed for all computations (Stata Corp, College Station, TX, USA).

Analyses of EpiHealth data

The three PFAS were inverse-ranked normalized to ensure a normal distribution and to obtain all the PFAS on the same SD-scale. Thus, the ORs reported for diabetes are for a 1 SD change of each measured PFAS. For the EpiHealth data, logistic regression analysis was used to evaluate the relationships between the relative concentrations of PFHxS, PFOA and PFOS, and prevalent diabetes. The fully adjusted model was adjusted for age, sex, participation date, total fat mass, smoking, education, physical exercise, alcohol use, creatinine as a proxy for kidney function and fish intake. An additional adjustment was also made for HLI and for plasma levels of the marine omega-3 fatty acids eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA).

A sex*PFAS interaction term was inserted in the main models in order to investigate potential sex differences. If the interaction term was significant, the analyses were also performed in men and women separately. Further, age-interactions were investigated by inserting an age*PFAS interaction term in the main models. Finally, potential non-monotonic relationships were examined by inserting a squared term of the PFAS in the models. We divided the nominal p-value by 3 to account for 3 different measured PFAS (0.05/3 = 0.016666), which resulted in a significance level of p < 0.016666 for the associations between PFAS and prevalent diabetes in the primary analysis. This Bonferronni-adjustment for multiple testing was done to limit the number of false positive findings frequently reported in the literature.

Analyses of PIVUS data

Due to the right-skewed distribution, plasma levels of all PFAS and plasma glucose levels were ln-transformed in order to obtain a more normal distribution of the data. First we wanted to see if there was a significant change in glucose levels over the 10-year follow-up period. Thus, the change in plasma glucose levels over 10 years (three measurements) was assessed by mixed-effects linear regression models with fasting glucose as the dependent variable, time as the independent variable, and sex as confounder (age was the same in all subjects). The same analysis has been performed also for the changes in PFAS levels over 10 years and the results have been previously described [16].

Thereafter, we examined associations between the changes over 10 years in plasma levels of both total PFAS and individual PFAS: PFHxS, PFOA, linear isomer of PFOS, PFNA, PFDA, and PFUnDA (three measurements) and the changes over 10 years in fasting glucose levels (three measurements). The theory and assumptions behind this model as well as the detailed formula can be found in [17]. The general formula is; Yij = Zibeta0 − Xi1betaC + (Xij − Xi1)betaL + eij, where Y is fasting glucose levels, X is the specific PFAS, i is the individual, j the time, betaC is the coefficient for the first observation and betaL, is the coefficient for the change over time. Confounders and the intercept are given as Zibeta0. The general formula used in STATA for the calculations of the present study was; mixed fasting glucose PFASchange PFAS1 sex | |id:, where 1 denotes the first observation at age 70.

The fully adjusted model was adjusted for sex, BMI, insulin and diabetes medication, smoking, physical exercise and glomerular filtration rate, (age was the same in all individuals). Additional adjustment was made for relative plasma levels of the marine omega-3 fatty acids EPA and DHA. The PIVUS study was regarded as a supportive analysis and p < 0.05 was thus regarded as statistically significant.