Measuring heat-related mortality

Measuring heat-related mortality is not a technical exercise in epidemiology, it is the basis for action at every scale. Robust figures generate political will, direct resources to where the burden falls hardest, and set the baseline against which interventions can be judged. Because heat deaths are routinely miscoded, for example recorded as cases of cardiac arrest or renal failure rather than attributed to heat, the burden visible to decision-makers is a fraction of the true toll. Improving measurement is therefore itself an intervention, making every subsequent action possible.

Why counting heat deaths is genuinely hard

Heat mortality is hard to measure because there is rarely a single moment of obvious causation; a death can be the product of heat, pre-existing illness, housing conditions and consecutive sleepless nights acting together, so deaths attributable to heat show up only as a statistical excess, visible in aggregate and usually only in retrospect. This measurement is further complicated by factors such as air pollution or seasonal influenza, which often peak alongside extreme temperatures, making it difficult to isolate the independent effect of heat. Furthermore, mortality displacement (or 'harvesting') poses a consistent analytical challenge: while heatwaves trigger sharp spikes in death rates, these may represent earlier deaths for individuals already at high risk, rather than entirely 'excess' deaths that would not have occurred in the absence of a heatwave. Because of these variables, acute deaths attributable to heatwaves are more easily measurable than chronic harm accumulated over months or years, such as the kidney disease documented among outdoor agricultural workers in Central America.[1] Deaths attributable to heatwaves are also hard to identify in real time; most mortality evidence is reconstructed weeks or years after the fact, which is adequate for building an evidence base but inadequate for triggering an emergency response. Historic reconstruction is a fundamentally different exercise from future projection, a projection of 60 additional deaths per 100,000 in the Sahel by 2050 is a scenario conditional on assumptions, not a forecast. And there is no agreed definition of dangerous heat: a recent scoping review of 237 studies found 116 different definitions of extreme heat in use, and thresholds calibrated on European or North American populations routinely misrepresent risk when applied elsewhere.[2]

A brief overview of methods

No single method captures the full burden. Excess mortality (as frequently seen in the headlines) compares observed deaths against an expected baseline, sidestepping the miscoding problem, but requires continuous death registration data that remain sparse across much of the Global South. Time-series and case-crossover studies, typically using distributed lag non-linear models (DLNM), underpin most of the global epidemiological literature, but share the same data dependency. Burden of disease modelling, used by the Global Burden of Disease (GBD) and the Lancet Countdown, generates the figures most often cited in policy, but relies on temperature-mortality relationships calibrated in data-rich countries and extrapolated globally. In settings without civil registration, verbal autopsy (structured interviews reconstructing cause of death) is the best available tool, but current instruments are not designed to capture heat as a contributing factor. Analogue climate transfer, applying risk relationships from climatologically similar, better-studied locations, offers a pragmatic stopgap but assumes a similarity in baseline health and adaptive capacity that often does not hold.

Community-based reporting offers a practical bridge where death registration is weak. In India, the Mahila Housing Trust has reported that networks of community health workers and local women’s groups can identify and document heat-related illness and mortality, even when formal cause-of-death systems fail to capture them. Furthermore, in field settings and for verbal autopsy, asking about time outdoors, sleep disruption, housing conditions and hydration makes heat-related attribution more feasible than relying on standard cause-of-death categories alone.

Two recent developments show this evidence being put to use. The UK's Climate Change Committee, in A Well-Adapted UK (2026), demonstrates what becomes possible once robust mortality data exists: a quantified cost-of-inaction case for cooling investment, built on decades of reliable excess-mortality evidence. HERA's City Heat Solutions Calculator (2026) attempts something closer to what the Global South needs, a cost-benefit tool spanning over 11,000 cities worldwide that estimates heat mortality and economic impact using simplified exposure-response methods and proxy data where direct evidence is absent. Both illustrate the same lesson: better mortality evidence does not just describe the problem, it makes the investment case for solving it.

Across Sub-Saharan Africa and South Asia, incomplete death registration, inadequate diagnostic systems where heat stroke is misread as malaria or sepsis, sparse weather station networks, and chronic underinvestment in environmental health research capacity produce an inverse relationship between burden and measurement. This is not a neutral data gap; it shapes which populations appear in the evidence informing global health policy.

The experience of the COVID-19 pandemic offers an instructive parallel: official death counts proved a fraction of the true total, and it was excess-mortality methods, applied retrospectively, that reconstructed the real toll (often three times higher than reported). The same logic applies to heat. The absence of counted deaths is not evidence that deaths are not occurring. The physiology, the modelling and the limited direct evidence available, including He et al.'s 2025 finding of rising nighttime heat mortality across Sub-Saharan Africa, all indicate that heat is killing people at a scale invisible to current data systems.

Vulnerability framework for at-risk populations

Where mortality cannot be directly measured, the exposure-sensitivity-adaptive capacity framework introduced in How heat kills & who is at risk offers another path: predicting risk rather than waiting to count deaths. Heat vulnerability assessments operationalize this framework through geospatial mapping; selected indicators for each dimension are combined into a composite score visualized across neighbourhoods or city districts, identifying which populations face the highest risk. The specific indicators depend on context and purpose: a mortality-focused assessment might draw on the prevalence of diabetes or cardiovascular disease, access to water, or housing type, while an assessment built to anticipate productivity loss or guide emergency response might weigh different factors entirely. Where mortality reduction is the goal, the assessment is only as reliable as its indicators: it should be grounded in factors with a well-evidenced relationship to heat mortality where this evidence exists, rather than risk factors imported wholesale from other contexts.

This is where the principal opportunity for the Global South lies. Developing a robust picture of exposure, sensitivity and adaptive capacity requires granular data, including local climate conditions, housing quality, population health, sociodemographics, urban greenery and access to services, and these high resolution data are frequently unavailable in the low-income countries and cities where heat risk is greatest. There is no standardized methodology for building these assessments, and given how widely the purpose and local context vary, a one-size-fits-all approach is unlikely to work in any case. The more useful path is pragmatic: building assessments from whatever reliable indicators are available, focused on actionability rather than completeness. In addition, participatory heat risk assessments  that integrate community-led data collection with scientific mapping to identify extreme heat vulnerabilities offer a way to plug data gaps, generate awareness and galvanize action.

A well-constructed assessment, built this way, becomes in effect a mortality risk model: it identifies the populations and places where exposure, sensitivity and constrained adaptive capacity converge to make deaths attributable to heat most likely, turning a broad statistical concern into a map that can direct cooling investment, target outreach and tailor a response, even in advance of, or in the absence of, complete mortality data. Building the systems that can both count deaths directly and predict risk where counting remains impossible is the subject of this report's concluding section.