Hypothesis

Counties with higher PFAS concentrations in public drinking water (EPA UCMR3/UCMR5 monitoring data) have elevated gestational diabetes mellitus (GDM) rates (CDC WONDER natality records), with Department of Defense AFFF-contaminated sites serving as instrumental variables for causal identification.

What We Know

The epidemiological signal is now robust. The largest meta-analysis to date — India-Aldana, Yu, Valvi et al. (2025) in Lancet eClinicalMedicine (DOI: 10.1016/j.eclinm.2025.103747), covering 129 studies — found PFOS associated with GDM at OR 1.13 per exposure doubling (8 prospective studies, I²=0.0%), PFOA at OR 1.23 (6 nested case-control studies), and PFNA at OR 1.21 (5 studies). Wang et al. (2025, BMC Pregnancy and Childbirth, DOI: 10.1186/s12884-025-07551-x) found PFOA high-exposure OR 1.51 (95% CI: 1.25-1.83) and a dose-response of 0.3% increased risk per 1 ng/mL PFOA. The PETALS cohort (Peterson et al. 2023, DOI: 10.1186/s12884-023-05953-3) showed mid-pregnancy PFOS OR 1.41 (1.17-1.71) per IQR.

The biological mechanism is established: PFAS disrupt PPARγ signaling, competitively bind liver fatty acid binding protein (L-FABP) causing free fatty acid accumulation, impair insulin receptor substrate phosphorylation, and reduce pancreatic β-cell glucose sensing. HOMA-IR worsens with PFOS exposure (β=0.06, 95% CI: 0.01-0.12) per the Lancet eClinicalMedicine meta-analysis.

The Gap

Every study above measured PFAS in blood. Nobody has linked EPA's actual drinking water monitoring data to GDM outcomes at population scale. This matters because:

Zhu & Bartell (2020, Environmental Epidemiology, DOI: 10.1097/EE9.0000000000000107) built the exact UCMR3-to-CDC-WONDER county-level pipeline — for birthweight, not GDM. They linked 551 counties (87 with PFAS detections) and found significant birthweight reductions. Li et al. (2025, J Expo Sci Environ Epidemiol, DOI: 10.1038/s41370-024-00742-2) used the same UCMR-to-registry linkage for cancer incidence. The GDM application is an untouched extension of a validated methodology.

Meanwhile, two Swedish drinking-water studies found null or inverse PFAS-GDM associations — Ebel et al. (2023, Environ Res, DOI: 10.1016/j.envres.2023.117316) in Ronneby (OR 1.03, CI: 0.67-1.58) and Savé-Söderbergh et al. (2025, Environ Int, DOI: 10.1016/j.envint.2025.109415) reporting an inverse OR 0.72 (0.61-0.84) nationally. This US-Sweden tension is unresolved. A US ecological study using the same exposure pathway (drinking water) would directly address whether the Swedish null results reflect true biology or exposure assessment differences.

Testable Predictions

1. Cross-sectional dose-response: Negative binomial regression of county-level GDM birth counts on mean UCMR3 PFOA concentration (≥1,500 counties, controlling for maternal age, race, BMI prevalence, poverty, insurance, urbanicity via ACS) yields IRR ≥1.10 per log-unit PFAS increase (p<0.05). Power: >90% at this effect size given ~3.9 million annual US births and 7.8% background GDM prevalence.

2. UCMR3→UCMR5 difference-in-differences: Counties newly detecting PFAS in UCMR5 (2023-2025) that were below detection in UCMR3 (2013-2015) show a larger GDM rate increase over the same period than counties with no detection in either wave, after adjusting for secular GDM trends (US GDM rose from 6.0% in 2016 to 7.9% in 2024). This exploits UCMR5's expansion from 6 to 29 PFAS analytes and from ~4,900 to ~10,300 water systems as a natural experiment.

3. Instrumental variable confirmation: 2SLS using proximity to DOD PFAS-contaminated installations (630 confirmed sites) as an instrument for county PFAS concentration recovers a causal GDM effect estimate ≥1.15, following the identification strategy validated by Jacqz, Somuncu & Voorheis (2024, Census Bureau WP CES-WP-24-72) who used DOD fire training area AFFF adoption timing for birthweight.

4. Chemical specificity: Among UCMR5's 29 PFAS analytes, compounds with demonstrated insulin-disrupting activity (PFOA, PFOS, PFNA) show stronger GDM associations than structural analogs without established metabolic effects (e.g., short-chain PFBS, PFHxA), providing a mechanistic fingerprint beyond generic PFAS exposure.

Falsification

If negative binomial regression across all linkable counties (n>500) shows no significant association (IRR 95% CI crossing 1.0) between UCMR3 PFOA/PFOS concentrations and county-level GDM rates after demographic adjustment, AND the DiD estimator for UCMR3→UCMR5 newly-detected counties is null, the hypothesis is falsified. Birth certificate GDM sensitivity (~65% vs. medical records) introduces measurement error that biases toward the null — a positive finding despite this attenuation is conservative.

Data Sources (All Public, All Free)

• EPA UCMR3/UCMR5: County-level PFAS concentrations in public water systems (epa.gov/dwucmr)
• CDC WONDER Natality: County-level births cross-tabulated by gestational diabetes, 2016-2024 expanded dataset
• Census ACS: County demographics for confounder adjustment (age, race, poverty, insurance)
• CDC PLACES: County-level obesity/BMI prevalence estimates
• DOD PFAS Sites: 630+ confirmed AFFF-contaminated installations for instrumental variable construction
• UCMR5 ZIP Code Files: UCMR5_ZIPCodes.txt for geographic linkage to counties

Why This Matters

GDM affects ~300,000 US pregnancies annually and is rising fast. If drinking water PFAS explains even a fraction of this increase, it converts an individual-level biomonitoring problem into an infrastructure policy lever — water utilities can act on UCMR data directly. The UCMR3→UCMR5 temporal design provides quasi-experimental evidence stronger than cross-sectional correlation alone, and the 2SLS framework addresses the endogeneity concern that healthier counties invest more in water treatment.