Phoenix and Las Vegas Alone Took Nearly 12% of U.S. Heat-Coded Deaths

National heat-coded mortality is no longer a flat background risk. CDC WONDER tallies that treat ICD-10 codes for exposure to excessive natural heat and related heat effects as an underlying or contributing cause show 2,325 U.S. deaths in 2023, up from 1,008 in 2018 — a 131% rise in six years. The geography of those certificates is not evenly spread. A small set of Sun Belt metros, led by Phoenix and Las Vegas, absorb a disproportionate share of the national count.

This piece asks a narrow question: which large Sun Belt metros account for the largest share of recent CDC heat-coded deaths, and how does that metro concentration sit against the national climb? The dashboard answers with metro share bars, a national trend panel, a rate-versus-summer-temperature scatter, and an attribution stack that holds five desert and Texas cores against the rest of the country.

How heat shows up on a death certificate

CDC heat-coded deaths are not the same object as “everyone who died on a hot day.” The operational definition used in the JAMA 2024 research letter and related WONDER extracts flags certificates that list X30 (exposure to excessive natural heat), T67 (effects of heat and light), or P81.0 (environmental hyperthermia of newborn) as underlying or contributing causes. Man-made heat (W92) is typically excluded when the goal is natural-heat exposure.

That coding choice matters for metro comparisons. Desert counties with aggressive medical-examiner protocols and long outdoor-exposure case histories print more heat codes. Humid Gulf and Southeast metros can see large all-cause mortality bumps during heat waves without a matching surge in X30/T67 mentions. Shares calculated from coded deaths therefore measure where heat is written on the certificate, not the full excess-mortality burden of heat.

Local medical-examiner series can be even wider. Maricopa County’s heat-associated death tallies have reached hundreds in a single recent summer under a case definition that includes heat as a contributing environmental factor beyond the narrow ICD set. Those local counts are useful for city operations; they are not interchangeable with WONDER when computing national shares. This dashboard stays on the CDC-coded side of that line.

The national climb after 2016

Joinpoint work published in JAMA shows a structural break: age-adjusted heat-related mortality rates were roughly flat-to-down through the mid-2010s, then rose sharply. Anchored annual counts for the recent window are:

2018, 2021, 2022, and 2023 are disclosed CDC / JAMA / USAFacts anchors. 2019 and 2020 are desk interpolations consistent with the published trend and should be treated as directional. Even with that caveat, the path is unambiguous: the United States roughly doubled heat-coded deaths between 2018 and 2023, and the 2023 print is the highest in the modern WONDER series.

The six-year sum used for metro shares in this desk is 8,727 national heat-coded deaths. That denominator is the reference for every share percentage below.

Phoenix and Las Vegas dominate the metro share table

When county WONDER compilations for 2018–2023 are read as metro proxies, two Southwest cores lead by a wide margin:

• Phoenix (Maricopa County): 648 deaths7.43% of the national period total.
• Las Vegas (Clark County): 391 deaths4.48% of the national period total.

Together, those two counties alone account for about 11.9% of all U.S. heat-coded deaths in the window. No other large metro in the panel clears four percent. Tucson (Pima) adds 113 deaths (1.3%). A five-metro Sun Belt core — Phoenix, Las Vegas, Tucson, Dallas–Fort Worth (Dallas + Tarrant), and Houston (Harris) — sums to 1,390 deaths, or 15.9% of the national total.

That concentration is the headline. The United States added more than a thousand heat-coded deaths between 2018 and 2023; a meaningful slice of the cumulative ledger sits in a handful of desert and inland Southwest counties.

Rate versus share: Lake Havasu is not Phoenix

Share and rate answer different questions. Mohave County (Lake Havasu City–Kingman) logged 239 heat-coded deaths over the same window — fewer than Las Vegas, but against a population near 217,000. Its crude window rate is about 18.4 deaths per 100,000 residents, far above Phoenix’s 2.4 or Houston’s 0.31.

High-rate micropolitans matter for local public health. They do not dominate the national share ledger, because the denominator is population-scaled death volume. Phoenix’s share lead comes from combining a high desert climate with a metro population above 4.4 million. Lake Havasu’s rate lead comes from extreme heat against a thin population base. The scatter panel plots NOAA summer daily highs against window rates with bubble size proportional to deaths so both stories stay visible.

Yuma County prints a similar pattern: 66 deaths and a 5.2 per-100,000 window rate — deadly for residents, modest in the national share ranking.

Texas and Florida metros: large populations, thin coding

Dallas–Fort Worth’s principal-county pair (Dallas + Tarrant) sums to 149 coded deaths — more than Tucson in absolute terms, but a window rate near 0.5 per 100,000 and a national share under 2%. Harris County (Houston) shows 89 deaths and a 0.31 rate despite a population larger than Maricopa’s. Derived estimates for Miami and Atlanta sit even lower on the coded ledger.

Those humid metros are not “safe” in an epidemiologic sense. Excess all-cause analyses and recent media modeling routinely argue that official heat codes understate Gulf and Southeast burdens, especially where death investigators are less likely to list heat on the certificate. For share-of-coded-deaths, however, the Southwest still concentrates the ink. Treating Houston or Miami as “low risk” because their WONDER cells are small would reverse the meaning of the metric.

NOAA Climate Normals put summer daily highs in Phoenix and Las Vegas above 100°F, while Houston and Miami sit in the low 90s with much higher humidity. Dry-bulb extremes and heat-index extremes do not map one-to-one onto certificate coding. The dashboard’s climate axis is a context layer, not a causal claim that every degree of mean summer high produces a fixed share of national deaths.

Reading the Sun Belt stack against the national trend

The stack panel attributes each national year using the fixed period share of the five-metro Sun Belt core (15.9%). That is an attribution device, not a claim that those metros’ year-by-year WONDER extracts were pulled separately. It shows what a constant Southwest core share implies as the national total climbs from roughly 1,000 to 2,300 deaths: the absolute attributed core rises even when the percentage is held flat.

If coding intensity or homelessness-linked outdoor exposure in Maricopa and Clark rose faster than the national average — a pattern suggested by local examiner reports — the true recent-year share of those metros would be higher than the period-average attribution. The desk does not force that acceleration into the chart without year-by-metro WONDER extracts; the caveat is the point.

Caveats that change how you should use the numbers

Several limits are structural, not cosmetic:

1. Definition mismatch. WONDER ICD coding ≠ local “heat-associated” medical-examiner tallies ≠ excess all-cause heat mortality.
2. Geography proxies. Several rows use principal counties as CBSA stand-ins. Peripheral counties can add deaths; San Bernardino is not folded into the Riverside row.
3. Derived cells. Miami, Atlanta, and San Antonio figures are illustrative scalings where top-county lists did not disclose a clean CBSA totalmarked as such in the data module.
4. Suppression and small cells. CDC suppresses or flags unreliable rates for sparse counties; humid metros may look artificially quiet.
5. COVID-era confounding. Excess-mortality methods struggle in 2020–2022; cause-specific heat codes avoid some of that noise but still inherit changes in investigation practice.
6. Climate normals ≠ event heat. A 30-year summer mean does not capture a single 31-day streak above 110°F. Event timing drives spikes that annual metro sums blur.

Use the share ranking to answer “where do coded deaths concentrate?” Use the rate ranking to answer “where is risk densest per resident?” Use the national trend to answer “is the country getting worse?” Do not collapse those three questions into one bar.

What the metro map implies for adaptation desks

If nearly 12% of national heat-coded deaths already sit in two Southwest counties, cooling centers, street outreach, shade, and housing policy in Phoenix and Las Vegas are national-scale mortality interventions, not merely local amenities. If humid metros under-code heat, their emergency-management case for resilience funding should not wait for X30 counts to catch up with desert metros.

The interactive filters let you toggle large metros only, Sun Belt only, or desert stations with summer highs ≥100°F, and sort by deaths, share, or rate. Start on the share panel for the concentration story; switch to the scatter when someone asks why Lake Havasu outranks Houston on risk density; use the national trend when the question is whether 2023 was an outlier or the new slope.

Bottom line: CDC heat-coded deaths surged nationally from 1,008 (2018) to 2,325 (2023). Inside that climb, Phoenix and Las Vegas alone absorbed about 11.9% of the 2018–2023 national total, and a five-metro Sun Belt core held roughly one in six coded deaths — a metro concentration that the national trend line alone cannot show.