How Heat Vulnerability Indices Are Developed, Validated, and Mapped: A Systematic Review

Heat Vulnerability Maps Are Everywhere. A Systematic Review Finds Many Still Struggle to Predict Who Will Die

As heatwaves become more frequent, longer-lasting and more intense, cities around the world are turning to a deceptively simple tool: the heat vulnerability index, or HVI. By combining demographic, economic, health and environmental data, an HVI produces a map showing which neighborhoods may be least able to withstand extreme heat. The promise is powerful. A city could use such a map to position cooling centers, target welfare checks, improve emergency warnings, plant trees or direct medical resources before temperatures become dangerous. But a systematic review of the scientific literature has now exposed a central weakness in the approach. Although areas with higher HVI scores generally experience greater heat-related health risks, the relationship is often weak. In some studies, the maps explained very little of the geographic variation in deaths or hospital use. The finding does not make HVIs useless, but it shows that a colorful risk map should not automatically be mistaken for a precise prediction of who will be harmed.

The review, led by Yanlin Niu of the Chinese Center for Disease Control and Prevention and colleagues, examined research on the development and validation of heat vulnerability indices. The investigators searched PubMed, Web of Science, ScienceDirect, China National Knowledge Infrastructure and Wanfang Data for peer-reviewed articles published in English or Chinese between January 2010 and October 2020. Their initial search returned 941 records. After duplicates, irrelevant studies and papers that did not test their index against observed health data were removed, only 13 studies met all the criteria. This distinction is crucial: dozens of studies had created or applied an HVI, but far fewer had checked whether the index actually corresponded to real-world outcomes such as mortality, hospital admissions, emergency visits or ambulance callouts. Most of the included studies came from the United States, with additional work from the United Kingdom, Canada, China and South Korea, revealing how heavily the evidence base depends on a small number of countries with detailed health and census data.

An HVI is usually built around a broad definition of vulnerability. In climate science, vulnerability is not simply the amount of heat a person experiences. It combines exposure, sensitivity and adaptive capacity. Exposure describes the heat hazard itself: temperature intensity, the number of hot days, duration, humidity or other measures of thermal stress. Sensitivity refers to how strongly a person or community may respond to that hazard, including age, chronic illness or disability. Adaptive capacity describes the ability to avoid or reduce harm, through air conditioning, access to health care, income, social support, transportation, information and effective public institutions. The reviewed studies grouped their indicators into five major categories: hazard exposure, demographic characteristics, socioeconomic conditions, the built environment and underlying health. In practice, however, researchers often selected variables according to local data availability, previous publications or informed judgment, rather than a universally accepted scientific framework.

Across the studies, the number of indicators included in an index ranged from four to 19. Demographic, socioeconomic and built-environment variables appeared in almost every analysis, while health-related indicators were included in seven of the 13 studies. Heat exposure itself was represented in only five. The factors most frequently used were social cohesion, race or ethnicity, landscape, age and economic status. Social cohesion commonly appeared through measures such as living alone, which can become especially important when an older person is isolated during a heat emergency. Age captures physiological vulnerability: infants and older adults can have less effective thermoregulation, while older people are also more likely to live with cardiovascular, respiratory or metabolic disease. Income and poverty reflect the ability to pay for cooling, relocate temporarily or obtain transportation and medical care. Race and ethnicity may act not as biological explanations, but as markers of structural inequality, unequal access to services, discrimination and historical patterns of housing and environmental disadvantage.

The physical shape of a neighborhood can amplify or reduce heat exposure. Dense areas dominated by asphalt, concrete and dark roofs absorb solar energy and release it slowly, creating urban heat islands that remain hot after sunset. Vegetation can cool surroundings through shade and evapotranspiration, the process by which plants move water into the atmosphere and consume energy as that water evaporates. Several reviewed studies therefore used land cover or vegetation as proxies for environmental protection. Air conditioning was another important adaptation indicator, because it can lower indoor temperatures and reduce heat stress. Yet the researchers emphasized that environmental measurements are not interchangeable. Land-surface temperature, often estimated from satellites, describes the temperature of the ground or roofs, whereas air temperature measures the atmosphere surrounding people. These quantities can diverge because of wind, shading, clouds, surface materials, solar radiation and the angle from which a sensor observes the landscape. Human health is influenced by both radiative heat from hot surfaces and convective heat from the air, so relying on only one can distort the exposure component of an index.

The dominant statistical technique used to construct the indices was principal component analysis or factor analysis, applied in 11 of the 13 studies. These methods transform many correlated variables into a smaller number of mathematical components. For example, poverty, unemployment and low car ownership may load onto a common component interpreted as socioeconomic disadvantage. The resulting component scores can then be combined into an index and mapped across neighborhoods. This approach is attractive because it reduces dimensionality and can limit arbitrary decisions about how much weight each variable should receive. But it also creates interpretive problems. The mathematical components may be difficult to explain in practical terms, and the final index can change substantially when researchers alter the input variables, geographic scale or study population. A neighborhood may receive a high score not because it has one dominant risk, but because several indicators happen to align statistically.

The geographic unit used to calculate an HVI also matters more than it may appear. Eight of the 13 studies relied on census-based areas, including census tracts, block groups and their British equivalents. Other analyses used counties, postal codes or administrative districts. One Shanghai study used a 500-meter grid, offering a much finer spatial resolution. These choices can produce different conclusions because conditions within a single neighborhood are rarely uniform. A census tract may contain both tree-lined streets and heat-exposed apartment blocks, or residents with very different access to cooling and health care. This is a classic problem in spatial analysis known as the modifiable areal unit problem: patterns can change when the same data are grouped into different boundaries or scales. A map constructed at county level may be useful for broad resource planning but miss small pockets of extreme risk. A high-resolution map may locate vulnerable blocks more precisely, but it can also create a false impression of certainty if the underlying health data are sparse.

Validation was the most revealing part of the review. Ten of the 13 studies used mortality data, while others examined morbidity, ambulance callouts, hospital admissions or emergency-department visits. The health datasets covered periods ranging from one to 17 years, with an average duration of eight years, and the number of geographic units ranged from 159 to 4,765 where sample sizes were reported. Researchers commonly used Poisson, logistic or negative-binomial regression to test whether higher HVI scores were associated with more adverse outcomes. Most studies found the expected direction of association: as vulnerability scores rose, health risks tended to rise as well. But the strength of the relationship varied and was usually not strong. In one Dallas analysis, a coefficient of determination, or R², of 0.03 indicated that the index explained only about 3 percent of the variation in total deaths. Other work using supervised and unsupervised principal component methods also produced very low R² values when tested against deaths occurring on extreme-heat days.

That weakness may result from several scientific and practical problems rather than from one failure of the HVI concept. Heat-related mortality is influenced by weather duration, nighttime temperatures, humidity, air pollution, public warnings, access to transportation, social contacts, medication use, housing quality and individual behavior. These factors can vary rapidly and may not be visible in annual census statistics. Health outcomes may also occur at a different location from a person’s home, while deaths can be recorded without enough detail to identify the precise circumstances of exposure. Short validation periods may capture unusual events, and data from one city may not transfer to another with different housing, climate, health systems or patterns of social inequality. The index itself may also be circular if variables are selected because they are already known to correlate with past deaths. A map can therefore appear scientifically sophisticated while still failing to predict how a particular heatwave will affect a particular community.

The review points toward a more demanding future for heat-risk mapping. New indices should incorporate better measurements of actual heat exposure, including both air temperature and land-surface temperature, as well as humidity and the duration and timing of heat events. Underlying health conditions and access to medical services deserve greater attention, because the same temperature can produce very different consequences in populations with different levels of chronic disease or care access. Governance is another overlooked factor. The capacity of local authorities to issue warnings, open cooling centers, conduct outreach, maintain electricity and water supplies and respond rapidly may determine whether vulnerability becomes illness or death. Public awareness also matters: people need to recognize danger, understand protective guidance and have the means to act on it. The researchers argue that validation should use longer periods, more locations and higher-quality health data, while protecting privacy. For now, an HVI is best understood as a decision-support tool—a way to identify where preventive action may be needed—not as a crystal ball. Its maps can save lives when combined with local knowledge and flexible emergency planning, but their predictions must be tested continuously against what happens on the ground.