The Way Scientists Measure Heat Changes the Death Toll It Predicts

When a heatwave strikes Europe, the question of how many people it kills has no single answer. It depends, surprisingly, on how researchers choose to define heat itself. A new study spanning five European cities from the Mediterranean to the Arctic Circle shows that the choice of heat metric can materially change estimates of heat-related mortality, a finding with direct consequences for how policymakers plan adaptation strategies and how the public understands the health toll of a warming continent. The research, published in the journal Environmental Health, examined mortality and climate data from Heraklion in Greece, Bologna in Italy, Geneva in Switzerland, Augsburg in Germany and Oulu in Finland for the years 2020 through 2023, covering a deliberately diverse range of European climate conditions.

The study was conducted by an international team led by Alexandrina Stoyanova of the University of Barcelona, working with colleagues from institutions including the European Centre for Medium-Range Weather Forecasts, the University of Bologna, the CMCC Foundation, the University of Oulu, the University of Geneva, University Hospital Augsburg and the University of Crete, under the TRIGGER consortium project funded by the European Union’s Horizon Europe programme. Their central question was deceptively simple: does it matter whether heat is measured as raw air temperature, as a composite thermal stress index, or as a binary heatwave flag? The answer, their results demonstrate, is yes, and the differences are large enough to influence how cities prioritize heat action plans.

To understand why the definition matters, it helps to consider what different heat metrics actually capture. Air temperature, the most familiar measure, records the thermal state of the atmosphere but ignores humidity, wind and radiation, all of which shape how the human body experiences and dissipates heat. The Universal Thermal Climate Index, or UTCI, goes further by combining air temperature, humidity, wind speed and radiation into a single measure of thermal stress on the body. The wet-bulb globe temperature, or WBGT, similarly integrates humidity and radiant heat and has long been used in occupational health to gauge heat stress in workers. Each metric captures a different dimension of heat: intensity, duration, the degree to which a population has acclimatized to its local climate, and the combined physiological burden of multiple environmental factors acting at once.

The researchers applied a statistical framework known as distributed lag non-linear models, or DLNMs, to harmonized daily mortality and climate data across all five sites. DLNMs are a standard tool in environmental epidemiology because they can model exposure-response relationships that are non-linear, meaning the risk of death does not rise in a straight line with temperature, and lagged, meaning the effect of a hot day on mortality can unfold over several subsequent days. By fitting continuous exposure-response functions across the full observed range of air temperature, UTCI and WBGT, the team could estimate how mortality risk changes at every level of heat, not just during officially designated heatwave events. They also analyzed binary heatwave indicators separately under the same framework, including the excess heat factor, the warm spell duration index and percentile-based definitions, allowing a direct comparison between continuous and event-based approaches.

The results revealed a striking north-south gradient in heat-mortality associations. Cumulative relative risks, which express how much more likely death is on hot days compared with cooler reference days, indicated stronger heat-mortality associations in Heraklion and Bologna, moderate associations in Geneva, and non-significant associations in Oulu, the Finnish site at the northern edge of the study. The Augsburg data, which came from hospital-based records rather than population-based mortality registries, also yielded non-significant associations, a methodological detail that itself illustrates how data provenance shapes what studies can detect. These spatial patterns align with expectations: southern European populations face more intense and frequent heat, while northern populations, though potentially vulnerable during unusual heat episodes, experience fewer extreme events within the study period.

But the relative risks told only part of the story. When the team converted their statistical associations into attributable mortality burdens, the actual number of deaths that could be attributed to heat exposure, the picture became more nuanced. The burdens showed similar spatial patterns to the relative risks, with the heaviest tolls in the southern sites, but they also varied according to how heat exposure was defined and captured by different metrics. A city’s estimated heat death toll could shift depending on whether researchers used air temperature, UTCI, WBGT or a binary heatwave flag, even though all were applied to the same mortality data over the same years. This metric-dependence is the study’s most consequential finding, because it means that any single heat metric provides an incomplete and potentially biased view of heat’s health impact.

One consistent pattern emerged across all metrics and sites: continuous models generally produced higher burden estimates than binary heatwave definitions. The explanation lies in what each approach can see. Binary heatwave indicators, by design, flag only officially defined heatwave events, such as periods when temperatures exceed a local percentile threshold or when the excess heat factor crosses a critical value. But mortality risk is not confined to these flagged episodes. Heat kills across the full range of exposure, including moderately hot days that never meet heatwave criteria, and the cumulative effect of many such days can rival or exceed the toll of a single dramatic event. Continuous exposure-response functions capture this ambient risk across the entire observed range, while event-based definitions miss the deaths that occur outside flagged heatwaves. The difference between the two approaches is therefore not a statistical artifact but a reflection of fundamentally different conceptions of what constitutes dangerous heat.

The study also identified which populations bear the greatest risk. Older adults and women consistently emerged as the most vulnerable groups across the five sites, a finding consistent with a large body of epidemiological evidence. Older adults face reduced cardiovascular reserve, diminished thirst perception and greater prevalence of chronic conditions that impair thermoregulation, while physiological and social factors, including differences in body composition and living circumstances, contribute to the heightened vulnerability of women in many settings. These demographic patterns held across different heat metrics, suggesting that while the magnitude of estimated mortality may shift with the choice of definition, the identity of the most at-risk populations is robust.

The implications for policy are significant. Heat action plans, early warning systems and adaptation strategies across Europe are typically built on specific heat definitions, whether national meteorological heatwave thresholds or percentile-based triggers. If the underlying metric underestimates the health burden, as binary definitions tend to do relative to continuous models, then warning systems calibrated to those metrics may fail to activate during dangerous conditions that fall outside official heatwave criteria. Conversely, metrics that overestimate burden in a particular setting could misdirect resources. The authors argue that using multiple complementary heat metrics, rather than relying on any single definition, provides a more complete characterization of heat-related health risks and supports fairer comparisons across settings with different climates.

The authors also recommend that studies report both relative risks and attributable mortality burdens, since each answers a different question. Relative risks express the strength of the heat-mortality association and are useful for comparing how populations respond to heat, while attributable burdens express the actual number of deaths and are more directly relevant to policymakers weighing adaptation investments. Reporting both, the team argues, can help identify locally appropriate heat adaptation strategies, from urban greening and cooling centers in Mediterranean cities to targeted outreach for isolated older adults in northern populations. As Europe continues to warm faster than the global average, and as heat events grow more frequent and intense, the study’s message is clear: the way heat is defined is not a technical footnote but a fundamental determinant of what societies understand about heat’s human cost, and getting that definition right is a prerequisite for protecting the most vulnerable.

Subject of Research: How heat exposure is defined matters: evidence from multi-metric mortality estimates across Europe

Article Title: How heat exposure is defined matters: evidence from multi-metric mortality estimates across Europe

Article References: How heat exposure is defined matters: evidence from multi-metric mortality estimates across Europe. (n.d.). https://doi.org/10.1186/s12940-026-01349-w

Image Credits: AI Generated

DOI: 10.1186/s12940-026-01349-w

Keywords: heat, exposure, defined, matters, evidence, multi-metric, mortality, estimates, across, Europe, scientific research