Associations of hurricane exposure and forecasting with impaired birth outcomes

The study addresses how exposure to hurricanes and the accuracy and timing of storm forecasts affect birth outcomes. While early warnings aim to reduce harm, forecasts can induce psychological stress and behavioral changes via media dissemination and public interpretation (e.g., the widely shared but often misinterpreted Cone of Uncertainty). Prior work links disasters and stress during pregnancy to adverse neonatal outcomes, and suggests mechanisms including in utero stress, environmental toxins, and disrupted healthcare access. However, no large-scale analysis has linked forecast accuracy to health outcomes in hurricane-threatened populations. The authors examine Hurricane Irene as a natural experiment to quantify impacts on birth weight, gestation length, and incidences of low and preterm births among more than 700,000 births in North Carolina, hypothesizing that in utero exposure reduces birth weights and gestation periods and increases adverse birth outcomes. They also explore mechanisms, expecting groundwater contamination and intense rainfall and wind to contribute to effects.

The paper situates its research within literature documenting that disasters and stressful events can impair birth outcomes via maternal stress, environmental exposures, and disrupted healthcare. Media-driven disaster coverage is linked to PTSD symptoms and mental health effects before and after hurricanes, suggesting that forecast dissemination itself may be a public health risk when exposures do not materialize (type I forecasting errors). Prior studies have shown links between in utero exposures and reduced birth weight and gestation, and to abnormal neonatal conditions, with downstream consequences for disease prevalence, mental health, education, and wages. Despite this, empirical isolation of mechanisms and the role of institutional forecasting in shaping birth outcomes has been lacking. The study builds on hurricanes-and-birth outcomes research and broader fetal origins literature by focusing on forecast uncertainty and healthcare disruption as potential drivers.

Design: Observational study leveraging Hurricane Irene (landfall August 27, 2011) as a natural experiment. Data: North Carolina Department of Health and Human Services vital statistics, all live and still births from August 26, 2006 to June 14, 2012 (710,186 birth records) with georeferenced residential addresses. Outcomes: birth weight (g), gestation length (weeks), and binary indicators for low birth weight (<2500 g), very low birth weight (<1500 g), preterm (<37 weeks), and extreme preterm (<34 weeks). Prenatal care indicators: month prenatal care began and total number of prenatal visits. Medical risk factors: prepregnancy hypertension, gestational hypertension, eclampsia, and prior poor pregnancy outcomes. Exposure measurement: Maximum 1-day rainfall at each residence between August 14 and September 4, 2011 from PRISM was used as a proxy for exposure intensity; wind categories were also characterized. Forecast exposure: Number of 6-hour NOAA NHC Cone of Uncertainty advisories that included the residence during the forecast period (advisories #7 to #30A spanning Aug 22–27, 2011). Addresses were geocoded via ArcGIS API; distances to storm track and Cone overlays were computed using NOAA NHC GIS data. Private well water tests (>17,000 samples statewide) for arsenic, cadmium, chromium, lead, manganese, and nitrate from the NC State Laboratory of Public Health were merged for the same treatment/control windows to probe contamination mechanisms. Sample construction: Treatment defined by expected birth date e = c + 280 days (c = conception); women with expected dates within 280 days after Aug 25, 2011 (Irene disaster declaration) were in treatment; control comprised expected dates in the 5 years (1825 days) before Aug 25, 2011 within the same zip codes. This avoids mechanical bias from actual birth date-based windows given disaster effects on gestation. Empirical strategy: Regression with zip code and month fixed effects and a linear year trend; standard errors clustered at the county level. Main specification: Yiymz = β0 + β1 Eiymz + β2 ln(Riymz) + β3 Eiymz × ln(Riymz) + μm + Year + γz + εiymz, where Eiymz indicates treatment and ln(Riymz) is log 24-h max rainfall intensity. Nonlinear impacts across rainfall intensity were examined; predicted impacts and CIs were plotted against rainfall and overlaid with wind-speed exposure distributions. Effects of additional 6-h advisory windows were estimated within strata of rainfall intensity (Rain > 2 in; 1–2 in; ≤1 in) using linear models with zip and month fixed effects. Software: Stata/MP 16.1; county-clustered SEs. Ethical oversight: ECU IRB 17-000354; informed consent waived; STROBE guidelines followed.

Average in utero exposure to Hurricane Irene was associated with: • Birth weight reduction of 12.7 g (95% CI: 5.4–20.0 g), a 0.17–0.61% decrease on the sample mean (3263.7 g). Largest reduction ~14.4 g for hurricane-force winds with >10 in 1-day rainfall; smallest ~10.1 g for <1 in rainfall with mild winds. • Gestation shortened by 0.10 weeks (95% CI: 0.07–0.14), a 0.18–0.36% decrease on the sample mean (38.5 weeks). Effects were similar across exposure intensity (largest ~0.11 weeks for high wind/rain; smallest ~0.09 weeks for mild wind/low rain). • Increased incidence (percentage point changes, with percent increases relative to means): - Low birth weight: +0.56 pp (95% CI: 0.22–0.90), a 2.52–10.34% increase (mean 0.087). - Very low birth weight: +0.38 pp (95% CI: 0.23–0.52), a 15.33–34.67% increase (mean 0.015). - Preterm (<37 wks): +0.96 pp (95% CI: 0.53–1.38), a 5.20–13.53% increase (mean 0.102). - Extremely preterm (<34 wks): +0.56 pp (95% CI: 0.35–0.78), a 12.07–26.90% increase (mean 0.029). • Prenatal care disruptions: First prenatal visit delayed by 0.24 months (95% CI: 0.18–0.30; ~1 week), a 6.92–11.54% delay on mean start month (2.60). Total prenatal visits decreased by 0.63 (95% CI: 0.37–0.89), a 3.03–7.29% reduction on mean 12.21 visits. Disruptions varied little across exposure intensity. • Forecast exposure (Cone advisories): - Heavy rainfall group (Rain > 2 in; mean 15.9 advisory periods ≈95 h): An additional 6-h advisory had no meaningful effect on birth weight or gestation, but significantly decreased preterm risk by 0.00126 (SE 0.000496), ≈1.2% reduction relative to mean 0.103; other outcomes not significant. - Light rainfall group (Rain ≤ 1 in; mean 6.5 periods ≈39 h): An additional 6-h advisory decreased birth weight by 4.111 g (SE 1.449; p<0.01), and was associated with higher risks: low birth weight +0.00206 (SE 0.00105; p<0.1), very low birth weight +0.000846 (SE 0.000484; p<0.1), preterm +0.00114 (SE 0.000666; p<0.1), and a positive but not conventionally significant increase in extreme preterm +0.00139 (SE 0.000703). • Mechanisms and null checks: No meaningful relationship between exposure intensity and private well contamination rates at EPA MCLs or state detection limits; no increases in gestational hypertension or eclampsia; no evidence of differential selection via prediagnosed risk factors. Overall, adverse birth effects were consistent across intensity bands, suggesting drivers beyond direct physical exposures, with institutional/anticipatory disruptions implicated.

The findings indicate that in utero exposure to Hurricane Irene degraded birth outcomes across North Carolina and that these effects did not scale with physical storm intensity, pointing to mechanisms beyond direct physical impacts such as wind, rainfall, or groundwater contamination. Robust evidence shows prenatal care was disrupted—delays in initiation and fewer visits—across exposure intensities, implicating anticipation of severe weather and institutional responses (e.g., appointment cancellations) as key contributors. Forecast accuracy mattered: avoiding type II errors (failing to warn when severe exposure occurs) provided benefits to highly exposed women via reduced preterm risk with additional warning. However, type I errors (warning of severe exposure that does not materialize) were more prevalent and associated with small but measurable deteriorations in birth outcomes and prenatal care among lightly exposed populations. These results align with the broader disaster literature linking prenatal stressors to reduced birth weight and increased prematurity and provide novel evidence that uncertain hurricane forecasts can disrupt healthcare delivery. This raises important implications for public warning systems: while intended to mitigate harm, forecasts can inadvertently exacerbate health risks via behavioral and institutional pathways.

This study demonstrates that prenatal exposure to Hurricane Irene is associated with reduced birth weight, shorter gestation, and higher risks of low and preterm births, and that forecast-driven anticipation can disrupt prenatal care and affect outcomes, especially where severe physical impacts do not occur. Contributions include: (1) quantifying adverse birth effects across a statewide population using a rigorous quasi-experimental framework; (2) separating impacts of physical exposure from those arising through forecast anticipation and healthcare disruption; and (3) documenting how additional warning time can benefit highly exposed women while potentially harming lightly exposed populations via type I forecasting errors. Future research should examine heterogeneity in effects (e.g., by prior hurricane experience, neighborhood characteristics), trimester-specific vulnerability, linkage of residential histories to past storms, and the behavioral/psychological impacts of delayed but higher-accuracy forecasts. Broader generalizability to other storms with different predictability and hazard mixes should be tested, and cost-benefit analyses of forecast timing and accuracy should be developed to inform public warning policies.

The analysis focuses on a single event (Hurricane Irene) in North Carolina, which may limit generalizability to storms with different tracks, intensities, or predictability. Mechanisms cannot be fully isolated: while groundwater contamination and direct physical exposures were probed with null findings, unmeasured environmental or psychosocial factors may contribute. The study cannot quantify psychological impacts of alternative forecast timing or ambiguity. Although selection into treatment is argued to be minimal using fixed effects and controls, residual confounding cannot be entirely excluded in an observational design. Statistical significance for some advisory-related effects (e.g., extreme preterm in lightly exposed areas) is marginal or not conventional, and clinical significance of small average birth weight changes remains uncertain.