From 1999 to 2024, US death certificates listed cold as a cause in 45,301 deaths and heat in 23,951. That reverses the National Weather Service's ranking, which puts heat first among weather hazards and cold seventh; the two count different things. The gap has narrowed: in 2023 heat reached 95% of cold. As winter starts, cold still leads, and its age-adjusted death rate more than doubled from 1999 to 2022, according to a study in JAMA.
In the weather service's count, heat killed 5,366 people from 1997 to 2025, 2.5 times as many as tornadoes; cold, with 849, is seventh. Over 1999–2024 it recorded about a fifth as many heat deaths as death certificates list, and far fewer still for cold.
What these numbers are
The weather service's count. Local forecast offices enter each weather event in Storm Data with the deaths they attribute to it, each death to one event type, and the National Weather Service's annual summaries add them up. Under the NWS's instructions to its offices (the July 2021 version; earlier years were entered under earlier versions), a heat death goes in when a medical examiner or coroner finds heat was the primary or a major contributing factor, and generally only if the weather met that office's own heat-advisory or warning threshold. Cold works much the same way, and a hypothermia death during a snowstorm may be booked under "winter storm" instead; even adding winter storms and ice to cold gives 1,573 deaths in 1997–2025, far below heat.
Death certificates. We count any mention of X30 (excessive natural heat) or T67 (effects of heat and light), and X31 (excessive natural cold), T68 (hypothermia) or T69 (other effects of reduced temperature). T67/T68 are contributing-cause codes, not underlying-cause codes. We do not exclude certificates carrying W92/W93, excessive heat/cold of man-made origin, so these are not exclusively weather deaths. The heat and cold groups are not mutually exclusive and should not be added together. These are the cold codes Liu and colleagues used; this extract reproduces their 40,079 cold deaths for 1999–2022 exactly. ICD-10 has been used for deaths since 1999, so the codes are the same across the series; how often certifiers write heat or cold on a certificate may not be.
Why the two are not directly comparable. These ratios compare differently defined records, not ascertainment rates. NWS distinguishes direct from indirect fatalities; that distinction is not equivalent to underlying versus contributing causes on death certificates. The annual summaries do not document that distinction sufficiently to assume comprehensive inclusion of indirect deaths. NWS also covers territories, whereas these WONDER queries cover residents of the 50 states and DC.
The latest weather-service year, 2025, is preliminary, and death-certificate data for 2025 is provisional. Neither enters the matched 1999–2024 comparison; the NWS ranking includes preliminary 2025. An NCHS study of 2006–2010 reported slightly lower totals than this extract for both heat and cold; DATA.md gives the comparison.
The weather service's ranking
Heat leads the weather service's count by a wide margin. Part of that lead is one state: Arizona accounts for 1,990 of the 5,366 heat deaths, almost all of them since 2019. Without Arizona, heat is 3,376, still first.
On death certificates, cold comes first every year
Cold appeared on more death certificates than heat in each of the 26 years from 1999 to 2024. Both have risen, heat faster: heat deaths in 2020–2024 were 2.2 times those in 2015–2019, while cold deaths rose 38%. So cold, 2.4 times heat in 2015–2019, was 1.5 times heat in 2020–2024.
These are counts of what certifiers wrote, not of deaths caused. This data cannot separate more deaths from heat or cold from more complete recording of them.
Why the weather service's count is smaller
The weather service's count links deaths to forecast weather events by design: a death enters it only when a local office ties it to one. So what gets counted depends on place. Apart from 51 in 2005, Arizona recorded at most 9 heat deaths a year in the weather service's count before 2019; in 2020 it was 337 of the 350 nationwide. Even now the count is not Arizona's full toll: in 2023 Maricopa County, which includes Phoenix, counted 645 heat deaths on its own, more than the weather service recorded for the whole country that year. No source found says why Arizona's deaths began reaching the weather service's count in 2019.
Arizona dominates the recent NWS heat count. These data cannot distinguish changes in reporting from changes in mortality, so the series alone cannot establish a national mortality trend. For cold the weather service's count is smaller still, relative to death certificates, and this data cannot say why.
Neither count is complete
Both counts depend on someone recording heat or cold as a cause. Studies that compare deaths on hot or cold days with deaths on ordinary days find far more deaths linked to temperature than either count records. For 1997–2006, Weinberger and colleagues estimated about 5,608 deaths a year attributable to heat in 297 US counties holding about 62% of the population. In 135 US cities over 1985–2006, Gasparrini and colleagues attributed 5.51% of deaths to cold and 0.35% to heat. Cold and heat mean temperatures below and above each location's minimum-mortality temperature; most attributed mortality involved moderate cold, not extreme cold spells. Those are statistical estimates of a different burden, not counted deaths, and this data does not test them.
What the counts here do show is that the familiar ranking, heat first and cold far behind, is a property of the weather service's record. On death certificates, the record the weather service points to as the official source, cold came first in every year from 1999 to 2024, and in 2023 heat came close.
How this was made
The weather-service figures come from the DataPressr dataset us-natural-hazard-statistics, extracted from the NWS's annual PDF summaries, and from the NWS's per-state heat summaries; both were rechecked against the NWS website on 8 November 2026 and are unchanged. The death-certificate figures come from CDC WONDER, queried through its API on 8 November 2026; WONDER suppresses counts of 1–9, and this post uses only national yearly counts, all far above that. fetch.mjs writes the snapshot (weather service, Arizona, death certificates); calculate.mjs checks the files' hashes and computes every number in this post into statistics.json; heat-deaths-make-charts.mjs builds the charts, and re-running it reproduces them exactly. Sources, licences and choices: PROVENANCE.md and DATA.md. From the download folder, run:
npm ci
node heat-deaths-src/calculate.mjs
node heat-deaths-make-charts.mjs