Mortality caused by tropical cyclones in the United States

Although natural disasters attract widespread attention, their full societal impacts—particularly on health—are not well understood because health outcomes are influenced by many confounding factors such as behavior, healthcare systems, and pollution. Traditional approaches focus on enumerating direct deaths where the disaster is the immediate cause (e.g., drowning), potentially misrepresenting total mortality because disasters can trigger complex cascades that cause additional future deaths. To date, the full excess mortality effect has not been characterized at population scale for any class of disaster, nor is the broader health burden of living in disaster-prone environments well understood. This paper develops a long-run estimate of the overall effect of individual tropical cyclones (TCs) and the TC climate on all-cause mortality across populations in the contiguous United States (CONUS). TCs are frequent and cause numerous socio-economic and environmental disruptions that may affect health through complex, delayed pathways, often unrecognized by affected individuals. The research question is whether TCs cause delayed excess mortality and how large and persistent this effect is across populations, places, and time.

Prior work suggests that disasters can have long-run economic and health effects, but comprehensive population-scale estimates of delayed mortality are lacking. Studies have documented economic damages, relocation, disruptions, ecological change, reduced access to services, increased pollution, and political responses following TCs, all of which could affect health. Previous analyses often focus on sub-populations, specific events (e.g., Katrina, Maria), short time windows, or direct deaths, providing suggestive evidence of post-TC health impacts without capturing the full indirect mortality burden. Broader climate-health literature documents temperature-related mortality and adaptation to heat over the twentieth century, raising questions about whether similar adaptation occurs for TCs. Evidence on unequal environmental vulnerability is mixed, though descriptive work often indicates greater harm to minority and socially vulnerable populations. Overall, the literature hypothesizes substantial indirect and delayed mortality impacts from TCs but has not quantified them comprehensively across the full population and over long horizons.

Design: The study uses a natural-experiment framework leveraging the quasi-random timing, location, and intensity of TC exposures across US states over time. The key idea is to compare mortality within the same state before versus after TC exposure, controlling for confounders, thereby approximating treated vs. control conditions. Exposure reconstruction: The Limited Information Cyclone Reconstruction and Integration for Climate and Economics (LICRICE) model is used to reconstruct monthly maximum ground-level wind speed at 0.125° × 0.125° resolution for each TC from 1930–2015. State-level monthly TC incidence is computed as land-area-weighted average wind speed. This metric correlates with storm damages and rainfall and serves as a consistent, though imperfect, proxy for storm physical intensity across the entire sample. Outcome data: Monthly state all-cause mortality rates (1950–2015) are compiled from CDC Mortality Statistics Annual Volumes, Multiple Cause of Death files, and the Underlying Cause of Death database. Analyses also stratify by age (infants <1, 1–44, 45–64/65, 65+), race (Black, white), and cause-of-death categories when available. Econometric approach: Mortality is modeled as the superposition of impulse responses to TC incidence over up to 240 months (20 years) post-landfall, deconvolving overlapping effects from multiple storms. TCs are represented as scaled impulses (Dirac deltas) by monthly state-level wind speed. The model non-parametrically estimates the impulse-response function for mortality following a TC. Controls and identification: The specification includes (1) state fixed effects; (2) state-specific seasonal fixed effects; (3) nonlinear state-specific trends; (4) national month-of-sample effects; (5) state-by-month-specific linear trends; and (6) controls for the nonlinear effect of temperature on mortality. These controls address time-invariant differences, seasonality, demographic/economic trends, national shocks (e.g., influenza), and month-specific state trends. Identification relies on quasi-random TC exposure timing conditional on controls. Validation and robustness: Model fit is high (R^2 ≈ 0.93) for predicting state-month mortality. Four randomization-based placebo experiments (total randomization; within-state; within-month; across-state block shuffles) show no spurious associations, indicating unbiased estimates. Robustness checks include alternative weighting schemes (population-based), count models, region-by-month shocks, and temporal stability tests; results hold. The study also examines heterogeneity by age, race, cause, and climatological risk (average TC incidence), and tests for within-state spatial adaptation and temporal changes in responses. Aggregation and burden estimation: State-level impulse responses are aggregated over months and across storms to estimate cumulative excess mortality per storm, monthly national mortality flows, total mortality burden (1950–2015), annual averages, and shares of total deaths. Decomposition analyses attribute long-run trends to climatology, population spatial shifts, and demographic growth/aging via simulations with counterfactual population distributions.

- TC impacts on mortality persist for approximately 15 years (significant increases over 172 months post-landfall). - Immediate effect: In landfall month, mortality rises by 0.033 (±0.012) deaths per 100,000 per m s−1 of state-level wind speed (t=2.78, P<0.05). - Peak monthly excess rate ~0.042 per 100,000 occurs ~68.6 months post-landfall (quadratic fit). - Cumulative effect: 5.37 (±1.8) excess deaths per 100,000 per m s−1 after 172 months (t=2.94, P=0.0038). With average state-level event wind of 6.9 m s−1, this implies ~37.05 (±12.4) excess deaths per 100,000. - Average per-storm mortality: Roughly 7,170–11,430 excess deaths per TC (far exceeding NOAA’s average of 24 direct deaths per storm; 22 excluding Katrina). - Total burden (1950–2015): All TCs combined produced an estimated 4,600–7,300 excess deaths per month; 501 TCs (1930–2015) generated ~3.6–5.7 million excess deaths. Annual burden is ~55,280–88,080 deaths, equal to ~1.9–3.1 per 100,000 annually or 3.2–5.1% of all deaths. - Age heterogeneity: Highest cumulative excess mortality risk for infants (<1 year): 49.8 per 100,000 per m s−1 (±11.3; t=4.41, P<0.0001). Ages 65+: 22.8 (±10.0; t=2.28, P<0.05). Ages 1–44: 2.49 (±0.53; t=4.70, P<0.0001). Ages 45–64/65: 3.50 (±2.4; t=1.46, P>0.05). By counts, 65+ account for ~46% of excess deaths; 1–44 ~32%; infants ~14%; 45–64 ~8%. - Race heterogeneity: For the same TC exposure, Black populations experience 13.53 (±5.51) per 100,000 per m s−1 (t=2.46, P<0.05) vs white 4.19 (±1.56) (t=2.68, P<0.05). However, 66% of cumulative excess deaths are among white individuals and 34% among Black individuals due to population shares. - Causes of death: Majority of TC-related excess deaths are recorded under ‘other’ causes (58.9%, 2.27 ± 0.59 per 100,000 per m s−1). Cardiovascular diseases: 36.0% (1.30 ± 0.86). Neoplasms: 11.6% (0.46 ± 0.48). Infectious diseases, respiratory diseases, and motor vehicle accidents show no linkage. - Climatological risk/adaptation: States with least frequent TC incidence have higher vulnerability—10.4 per 100,000 per m s−1 (±4.11; t=2.53, P<0.05) vs 3.49 (±0.74; t=4.71, P<0.0001) in more frequently affected states (~2.8× larger effect). Upper quartiles show similar vulnerability, indicating limits to adaptation. No evidence of within-state spatial adaptation altering mortality responses. - Spatial burden: Highest proportion of deaths attributable to TCs in southeastern states: Florida (~13%), North Carolina (~11%), South Carolina (~9%), Louisiana (~8%). - Demographic burden shares: TCs explain ~25% of infant mortality and ~15% of mortality for ages 1–44; for 65+, TC-related deaths are a smaller fraction (~3.5%). Black populations bear a relatively larger burden proportionally (15.6% of all Black mortality vs 3.1% of all white mortality). - Temporal dynamics: After 2001, TC-related mortality trend accelerates (+43.3 vs +9.2 deaths per month previously), driven by more frequent storms (17 vs 14 yr−1) despite slightly lower maximum state-level intensities (23.6→21.4 m s−1). Average incidence to CONUS populations increased (0.125→0.143 m s−1). - Trend decomposition: Long-run increase in TC-related mortality is attributed ~12% to climatological factors, ~7.5% to population shifting toward coasts, and ~80.5% to population growth and aging.

The study demonstrates that tropical cyclones substantially elevate all-cause mortality for up to 15 years after landfall, confirming the hypothesis that disaster impacts extend far beyond immediate direct deaths. By deconvolving overlapping storm effects and controlling for confounding geographic, temporal, and climatic factors, the analysis provides causal evidence that TCs induce large, delayed excess mortality across the US population. These findings reframe the public health significance of the TC climate, highlighting particularly high vulnerability among infants and Black populations, and showing that infrequently exposed states suffer greater mortality impacts from otherwise similar storms. The results explain part of the persistent mortality differences between TC-exposed and non-exposed states, especially for ages 0–44, and indicate that the TC climate is an important, previously unmeasured determinant of mortality risk distribution across the United States. The observed post-2001 acceleration is primarily due to increased storm frequency and demographic change rather than increased storm intensity, suggesting that health system planning and policy must account for cumulative and long-lived disaster effects.

This paper provides the first population-scale, long-horizon estimate of excess mortality attributable to tropical cyclones in the contiguous United States. It reveals that indirect, delayed deaths from TCs far exceed official direct-death counts, constituting roughly 3.2–5.1% of all deaths and totaling millions over the study period. Vulnerability is concentrated among infants, younger adults (1–44), and Black populations, and is higher in states with infrequent TC exposure. The mortality burden helps explain geographic disparities in mortality and has increased in recent decades largely due to more frequent storms and demographic changes. Future research should identify the mechanisms driving delayed mortality—economic disruption, social network changes, public finance adjustments, environmental exposures, and stress pathways—to enable targeted interventions. Policymakers should recognize TCs as a major public health driver and develop strategies that address long-term health impacts, not just immediate emergency responses.

- Exposure measurement uses reconstructed maximum wind speed as a proxy; storm surge, rainfall, and flooding are not modeled explicitly. Their effects are only partially captured via correlation with wind. - Mortality data are at the state-month level with limited demographic detail; finer spatial and temporal resolution over the full horizon is unavailable. - Interactions of closely spaced multiple TCs or interactions with other non-TC events are not modeled explicitly, though their average effects are embedded in estimates. - Non-fatal health outcomes are not captured, so overall public health burdens may be understated. - Migration across state borders after TCs is not tracked for individuals; while population adjustments are accounted for within origin states and overall TC-induced migration is small on average, migration effects cannot be fully ruled out and were notable in specific cases (e.g., Katrina).