Associations Between Extreme Temperatures and Cardiovascular Cause-Specific Mortality: Results From 27 Countries

The researchers tested whether exposure to extremely hot or cold ambient temperatures (compared to the "minimum mortality temperature" — the temperature at which death rates are lowest in each location) was associated with increased risk of dying from specific cardiovascular causes.

**Intervention (exposure):** Days with extreme temperatures, defined as:

Extreme heat: temperatures at or above the 99th percentile of local temperature distribution

Extreme cold: temperatures at or below the 1st percentile of local temperature distribution

They also examined a range of less extreme thresholds (97.5th, 95th, 90th percentiles for heat; 2.5th, 5th, 10th percentiles for cold)

**Comparator:** Days at the minimum mortality temperature (MMT) — the temperature associated with the lowest death rate in each city, which varied by location (e.g., ~20°C in temperate cities, ~30°C in tropical cities).

**Outcome measures:**

Death from any cardiovascular cause (ICD-10 codes I00-I99)

Death from ischemic heart disease (heart attacks; I20-I25)

Death from stroke (I60-I69)

Death from heart failure (I50)

Death from arrhythmia (I47-I49)

The study analyzed death records from 567 cities across 27 countries on 5 continents, spanning overlapping periods from 1979 to 2019. The total sample included:

**32,154,935** deaths from any cardiovascular cause

**11,745,880** deaths from ischemic heart disease

**9,351,312** deaths from stroke

**3,673,723** deaths from heart failure

**670,859** deaths from arrhythmia

Countries included: Australia, Brazil, Canada, Chile, China, Colombia, Costa Rica, Ecuador, Estonia, Finland, France, Germany, Greece, Guatemala, Iran, Ireland, Italy, Japan, Moldova, Norway, Paraguay, Portugal, South Africa, South Korea, Spain, Taiwan, Thailand, UK, USA, Vietnam. This covers tropical, temperate, continental, and arid climate zones.

The population is the general population of these cities — not a selected experimental group. Age, sex, and socioeconomic status varied across locations but were not individually analyzed (the study design controls for these at the individual level).

**Temperature data:** Daily ambient temperatures (in °C) were obtained from:

Meteorological weather stations in national and regional networks

Climate reanalysis models (which combine weather station data with satellite and atmospheric model data to estimate temperatures at locations without stations)

Where multiple stations existed in a city, the daily average across all stations was used

**Mortality data:** Death certificates from national and regional death registries, coded using the International Classification of Diseases (ICD-9 and ICD-10). The researchers extracted the "underlying cause of death" — the disease that initiated the chain of events leading to death.

**Other environmental data (used as confounders):**

Relative humidity (24-hour average, %)

Air pollutants: PM10, PM2.5, ozone (maximum 8-hour average), nitrogen dioxide (24-hour average)

**Country-level data:** Gross domestic product (GDP) per capita from the World Bank.

**Climate zones:** Each city was classified using the Köppen-Geiger climate classification system (e.g., tropical, temperate, continental, arid).

**Study design:** This is a two-stage meta-analysis of observational data using a case-crossover design at the city level.

**Stage 1 — City-level analysis:**

For each of the 567 cities, the researchers fit a conditional quasi-Poisson regression model — a statistical model designed for count data (daily deaths) that accounts for overdispersion (more variability than expected in a simple Poisson model). This model included a three-way interaction term between year, month, and day of the week, which serves as a flexible alternative to a traditional case-crossover design.

**Why the case-crossover design matters:** In a case-crossover study, each person serves as their own control. The logic is: for each person who died on a given day, you compare the temperature on that day (the "case" day) to the temperature on other days when that person did not die (the "control" days). Because the same person is compared to themselves, this automatically controls for all time-invariant confounders — age, sex, genetics, smoking history, diet, socioeconomic status, etc. This is a major strength because these factors are notoriously difficult to measure and adjust for in traditional observational studies.

**Temperature modeling:** The researchers used distributed lag nonlinear models (DLNMs). This is a sophisticated approach that allows the temperature-mortality relationship to be nonlinear (not just a straight line) and to have delayed effects. They modeled:

Exposure-response: quadratic B-splines with knots at the 10th, 75th, and 90th percentiles of temperature for each city

Lag-response: natural spline with 3 knots equally spaced on the log scale

Lag period: up to 14 days (to capture delayed effects and check for "mortality displacement" — where deaths are only advanced by a few days)

**Stage 2 — Pooling across cities:**

The researchers used a hierarchical extended mixed-effects meta-analysis framework. This means:

City-specific estimates were pooled, accounting for the fact that cities are nested within countries and climate zones

Meta-predictors included: mean summer temperature, mean winter temperature, and country-level GDP per capita (to account for adaptation and socioeconomic factors)

Random effects allowed cities from the same country and climate zone to "borrow information" from each other

**Minimum Mortality Temperature (MMT):** For each city and each cause of death, the researchers empirically identified the temperature associated with the lowest mortality risk, without imposing constraints on where that temperature falls. The MMT varied by location — reflecting local adaptation (e.g., people in hot climates have a higher MMT than those in cold climates).

**What this design can prove:**

That extreme temperatures are associated with increased cardiovascular mortality

That the association is likely causal, because the case-crossover design controls for individual-level confounders, and because temperature exposure precedes death (temporal sequence)

That the effect varies by cause of death (heart failure being most vulnerable)

**What this design cannot prove:**

It cannot prove individual-level causation (this is ecological data — we don't know the temperature each individual was actually exposed to)

It cannot distinguish between death caused by temperature vs. death merely accelerated by a few days (mortality displacement)

It cannot identify mechanisms (why temperature causes death — e.g., dehydration, blood clotting, strain on the heart)

It cannot account for behavioral confounders that vary day-to-day (e.g., people might stay indoors on extremely hot days, reducing exposure)

It cannot tell us about non-fatal cardiovascular events (heart attacks that people survive)

**Major methodological weaknesses:**

Temperature data is city-wide average, not personal exposure (people may be indoors with air conditioning or heating)

Death certificate coding can be inaccurate (especially for heart failure vs. other causes)

The study cannot separate the effect of temperature from the effect of air pollution, which often correlates with temperature

The 14-day lag window may not capture longer-term effects of sustained heat waves or cold spells

The study only examines mortality, not morbidity (non-fatal events)

**Primary outcome — Any cardiovascular death:**

Extreme heat (99th percentile vs. MMT): Relative risk (RR) = 1.09 (95% empirical CI [eCI], 1.07–1.12) — meaning 9% higher risk of death on extreme heat days

Extreme cold (1st percentile vs. MMT): RR = 1.22 (95% eCI, 1.20–1.25) — meaning 22% higher risk of death on extreme cold days

Excess deaths per 1,000 cardiovascular deaths: Heat (above 97.5th percentile) = 2.2 excess deaths (95% eCI, 2.1–2.3); Cold (below 2.5th percentile) = 9.1 excess deaths (95% eCI, 8.9–9.2)

**Secondary outcomes — Cause-specific:**

**Ischemic heart disease (heart attacks):**

Extreme heat: RR = 1.08 (95% eCI, 1.05–1.11)

Extreme cold: RR = 1.18 (95% eCI, 1.15–1.22)

Excess deaths per 1,000 IHD deaths: Heat = 1.8 (95% eCI, 1.6–2.0); Cold = 8.2 (95% eCI, 7.8–8.5)

**Stroke:**

Extreme heat: RR = 1.10 (95% eCI, 1.06–1.14)

Extreme cold: RR = 1.24 (95% eCI, 1.19–1.29)

Excess deaths per 1,000 stroke deaths: Heat = 2.2 (95% eCI, 1.9–2.5); Cold = 10.1 (95% eCI, 9.5–10.6)

**Heart failure:**

Extreme heat: RR = 1.12 (95% eCI, 1.07–1.18)

Extreme cold: RR = 1.37 (95% eCI, 1.29–1.45)

Excess deaths per 1,000 heart failure deaths: Heat = 2.6 (95% eCI, 2.4–2.8); Cold = 12.8 (95% eCI, 12.2–13.1)

**Arrhythmia:**

Extreme heat: RR = 1.05 (95% eCI, 0.96–1.15) — NOT statistically significant (confidence interval crosses 1.0)

Extreme cold: RR = 1.15 (95% eCI, 1.03–1.28)

Excess deaths per 1,000 arrhythmia deaths: Heat = 0.7 (95% eCI, -0.4 to 1.9); Cold = 6.2 (95% eCI, 4.3–8.1)

**Dose-response pattern:** The risk increased progressively as temperatures became more extreme. For example, at the 90th percentile (moderate heat), the RR for any cardiovascular death was 1.03, while at the 99th percentile (extreme heat), it was 1.09.

**Heterogeneity:** There was substantial variation across cities and countries. Cities in colder climates showed greater vulnerability to heat, while cities in warmer climates showed greater vulnerability to cold (suggesting local adaptation matters).

To translate these numbers into plain English:

**Cold is roughly 4 times more deadly than heat** for cardiovascular deaths. For every 1,000 cardiovascular deaths, extreme cold causes ~9 excess deaths while extreme heat causes ~2.

**Heart failure is the most temperature-sensitive condition.** On extreme cold days, a person with heart failure has a 37% higher risk of dying compared to a day at the optimal temperature. This is roughly equivalent to the increased risk from smoking 5–10 cigarettes per day (depending on the study).

**The absolute risk is small but population-wide.** If you have heart failure and live in a city with 10 extreme cold days per year, your annual risk of dying from heart failure increases by about 0.13% (12.8 excess deaths per 1,000 heart failure deaths, spread across multiple cold days). However, because cardiovascular disease is the #1 cause of death worldwide, even small relative risks translate into thousands of excess deaths.

**For comparison:** The 22% increased risk from extreme cold (any cardiovascular death) is roughly similar to the increased cardiovascular risk from being sedentary vs. moderately active, or from having a BMI of 30 vs. 25.

**What the authors acknowledge:**

Temperature data is city-wide average, not personal exposure — people may use air conditioning, heating, or spend time indoors

Death certificate coding may misclassify specific cardiovascular causes

The study cannot fully separate temperature effects from air pollution effects (though they adjusted for pollutants where data was available)

The 14-day lag window may not capture longer-term effects

Data was not available for all countries (e.g., Africa, South Asia, Middle East are underrepresented)

The study only examines mortality, not non-fatal cardiovascular events

**What a critical reader would note:**

**Ecological fallacy:** The study uses city-level temperature averages, but individual exposure varies enormously. A person who stays in air-conditioned environments on hot days may have zero exposure, while an outdoor worker has full exposure.

**Confounding by behavior:** On extremely hot or cold days, people change their behavior (stay indoors, reduce physical activity, change diet). These behavioral changes could independently affect cardiovascular risk.

**Confounding by air pollution:** Hot days often coincide with high ozone levels; cold days often coincide with high particulate matter from heating. Despite adjustment, residual confounding is possible.

**Mortality displacement:** Some of the excess deaths on extreme temperature days may be people who would have died within a few days anyway (the "harvesting" effect). The 14-day lag window partially addresses this, but longer-term displacement is possible.

**No individual-level data:** The study cannot tell us about individual risk factors (age, medications, comorbidities) that might modify the temperature effect.

**Publication bias:** Countries with null findings may be less likely to contribute data to the MCC network.

**Generalizability:** The 27 countries are mostly high-income or upper-middle-income. Results may not apply to low-income countries with different housing, healthcare, and occupational exposures.

For someone running their own n=1 experiment to understand how temperature affects your cardiovascular health:

### What to test

**Primary test:** Compare your resting heart rate, blood pressure, and subjective symptoms (chest discomfort, palpitations, shortness of breath) on days when outdoor temperature is below the 1st percentile for your location (extreme cold) vs. days near the minimum mortality temperature (typically 18–25°C depending on your climate).

**Secondary test:** Compare the same metrics on extreme heat days (above 99th percentile) vs. comfortable temperature days.

**Dose-response test:** Track these metrics across a range of temperatures (e.g., every 5°C increment) to see if there's a gradual relationship.

### Minimum meaningful duration

**At least 30 days** of daily measurements to capture a range of temperatures (ideally spanning both a cold spell and a warm spell).

**For seasonal comparison:** 3–6 months to capture both winter and summer extremes.

**For causal inference:** You need at least 3–5 extreme temperature days (below 1st or above 99th percentile) to have enough data points for a meaningful comparison.

### What to measure (specific metrics)

**Resting heart rate** (measured at the same time each morning, before getting out of bed, using a heart rate monitor or pulse oximeter)

**Blood pressure** (morning and evening, using a validated home blood pressure monitor, after 5 minutes of seated rest)

**Heart rate variability (HRV)** if you have a wearable device (e.g., Garmin, Apple Watch, Whoop) — specifically the standard deviation of NN intervals (SDNN) or root mean square of successive differences (RMSSD)

**Subjective symptoms** on a 0–10 scale: chest discomfort, palpitations, shortness of breath, fatigue, dizziness

**Daily outdoor temperature** at the time of measurement (use a local weather station or personal thermometer)

**Indoor temperature** (where you actually spend your time — this may be very