Urban heat stress: novel survey suggests health and fitness as future avenue for research and adaptation strategies

The administrative area of Berlin, the capital of Germany, is inhabited by 3.45 million people over an area of 892 km². It is the largest city in Germany and represents one of the major metropolitan areas in Europe, with a clear urban heat island effect (Dugord et al 2014). Berlin is located in the temperate climate zone (52°28'N, 13°18'E), affected by both maritime and continental climatic characteristics. The annual average temperature is 9.5 °C and the annual rain fall is 590 mm. The proportions of green spaces and open water, as well as building type and building density, vary largely within the city boundaries. Inner-city areas are compact and densely populated. In Berlin, these areas are dominated by late-19th century housing block developments. Residents have varying levels of social status, similar to most European cities (Honold et al 2012). In 2006 and 2010, Berlin was struck by severe heat waves, which were accompanied by strong increases in death counts (Schuster et al 2014).

Survey design

We developed a cross-sectional, quantitative heat stress survey to investigate individual-level heat impairment and potential risk factors. For questions regarding the survey please contact the main author.

We conceptualized our analysis along the general risk concept common in the climate change community, where heat stress risk is defined as a function of hazard and vulnerability (Romero-Lankao et al 2012, Scherer et al 2013). We developed a model for the concept based on variables that interact with one another along a functional chain of urban heat stress (figure 1). While hazard is determined by the heat impact (the local climate) an urban area is exposed to, vulnerability is determined by the exposure, the sensitivity, and the adaptive capacity of the urban population. Exposure describes how individuals or population samples are exposed to the heat island in terms of its intra-urban differentiation as induced by variations in land use (Dugord et al 2014) or individual living space exposure. Sensitivity refers to individuals' physiological constitution and describes how sensitive people are when exposed to the heat impact. Individual adaptive capacity describes the degree to which people are capable of implementing adaptive measures to counterbalance exposure or sensitivity driven heat impacts based on awareness and resources.

Zoom In Zoom Out Reset image size

To assess urban heat stress, the survey included a series of questions on participants' appraisal and self-assessed consequences of summer heat, referring to the heat waves in 2013. We investigated self-assessed heat impairment as a dependent variable by querying the participants for their perceived level of heat impairment during hot summer days, since self-assessed health is regarded as the most informative health measure in population studies (Jylhä 2009, Maheswaran et al 2015). It was assessed on a five-point Likert scale from 1 = not impaired at all to 5 = very strongly impaired, following standard survey procedures like the SF-36 health survey to assess general physical health (Cohen et al 2015). We further assessed personal emotions towards summer heat and the quantity of related health symptoms as control variables.

We aimed at arranging a questionnaire that allows for comprehensive exploration of potential risk factors. The extensive set of risk factors covered all three vulnerability dimensions (tables S9–S12). Risk factors were not designed for further causal inference but rather for locating empirical individual-level dynamics on heat stress risk factors for future studies. The variables were selected based on the heat and health literature at different spatial scales and thematic detail. We designed additional questions about potential risk factors that we assumed to be influential as well, as arose from continuous exchange with experts from different scientific fields (see the acknowledgements).

Sampling design, implementation, response rate

Sample characteristics and data preparation

For our exploratory data analysis, we studied the dependent variable self-assessed heat impairment using a five-point Likert scale. We checked for its general agreement with other potential dependent variables such as personal emotions towards summer heat, the individual awareness of heat risks, and the total quantity of reported symptoms by calculating the correlation. We used a three-step approach for the exploration of risk factors.

1). Variable reduction

2). Bivariate correlation assessment

3). Regression modeling

The dependent variable self-assessed heat impairment correlated both in magnitude and significance with the control variables emotions towards summer heat (rho 0.73, p-value < 2.2 × 10⁻¹⁶) and the quantity of health symptoms (rho 0.46, p-value < 2.2 × 10⁻¹⁶) as a quantitative variable.

Physical sensitivity variables returned the strongest correlations with heat impairment (table S1, available at stacks.iop.org/ERL/12/044021/mmedia). We observed particular strong associations for health and fitness variables. Among others these included self-assessed health and self-assessed fitness, body mass index (figure 2), pre-existing diseases (online supplementary figure S1), and amount of cycling or active travel. The association with age, however, was weaker. There was no association with gender. Ordered linear regression (OLR), variable cross correlation, and odds ratio calculation indicated that self-assessed fitness was the most predictive of all variables in the health domain (online supplementary table S4, S5).

Zoom In Zoom Out Reset image size

Environmental exposure variables only returned significant associations when linked to the personal living space, not to the neighborhood environment (online supplementary table S2). We found the strongest association for the geographic orientation of bedroom windows (figure 3) and further ones for apartment story and building types. The results of the OLR further supported these observations (table S4, S5).

Zoom In Zoom Out Reset image size

Adaptive capacity variables returned relatively strong associations for risk awareness and adaptation measures. Interestingly, these were positive (online supplementary table S3). We found significant negative correlations and relevant odds ratios (table S5) for some resource-related variables. Classic socio-economic variables were insignificant.

We particularly investigated variable cross-correlations for age and health/fitness.

All variables related to health and fitness were strongly correlated with each other, as described in the supplementary material, 1.3.

Age had relatively strong negative associations with health/fitness variables (online supplementary table S6). For male participants, we observed an interesting bi-directional trend of impairment and age (figure 4). While younger persons reported medium levels of heat impairment, there was a tendency for older ones to report levels on the two ends of the scale, either impaired or non-impaired (figure 4(a) and (b). For males aged 65 and older (figure 4(c)), the highest level of heat impairment was not associated with the oldest respondents. Age also returned significant correlations with most relevant exposure and adaptive capacity variables. However, bedroom orientation (exposure) and household size (adaptive capacity) were not correlated with age (online supplementary table S7).

Zoom In Zoom Out Reset image size

Our study uses self-assessed heat stress to measure the vulnerability of the population on the individual level. It represents a self-assessed approximation score. The commonly used dependent variable is heat-related excess mortality, a statistical estimation, which is used for comparison with aggregated population counts. We observed that self-assessed heat impairment was in agreement with the emotional appraisal of summer heat and the quantity of heat stress symptoms reported by the participants (see section 2.3). Also, we observed similar patterns for risk factors that have already been investigated in previous mortality studies (mainly age and diseases) (Romero-Lankao et al 2012). This agrees with studies that regarded health assessments as top predictors of future mortality, more reliable than physiologic measurements (Mossey and Shapiro 1982, Miilunpalo et al 1997). Ultimately, it indicates general validity of using questionnaires on self-assessed impairment to investigate heat stress risk.

The study of vulnerability dimensions revealed that variables related to physical sensitivity returned by far stronger and more numerous associations with heat impairment than the ones related to environmental exposure or adaptive capacity. Our analysis on environmental exposure indicated significant and apparently non-confounded associations for bedroom orientation as individual living space exposure, but no relevant associations for neighborhood exposure.

Physical sensitivity: health and fitness most important

Environmental exposure: relevant for personal living space

Adaptive capacity: low relevance and difficult to assess

Age is widely considered to have a major (direct) effect on heat vulnerability (Oudin Åström et al 2011, Romero-Lankao et al 2012), even though some mortality studies did not find clear associations with age (O'Neill et al 2003, Xu et al 2013).

Our dataset also showed that age had an association of high statistical significance with heat impairment. Yet, the association was not as strong as for health/fitness variables and we observed diverging impairment patterns with increasing age (figure 4). We assume that males in their early seventies that reported strong impairment could be the ones that die earlier as a result of weak physical constitution, confirming the assumption of premature death of the weakest in terms of health capital (Robine et al 2012).

There is evidence from laboratory-based physiological studies that the elderly may have decreased thermoregulation capacities due to decreased blood circulation, and reduced thirst and sweating capacity (Kenney and Munce 2003, Kenny et al 2010). However, this applies only to the elderly at the end of their lives. More generally, when people age, the reduced heat tolerance is believed to be mainly caused by a decrease in aerobic fitness that often, but not necessarily, accompanies aging (Hajat and Kosatky 2010, Kenny et al 2010). 'In fact, studies that have attempted to separate the effects of chronological age from concurrent factors, such as fitness level, body composition, and the effects of chronic disease, have shown that thermal tolerance appears to be minimally compromised by age.' (Kenney and Munce 2003).

Our study provides first empirical findings based on an individual-level survey suggesting that health and fitness could be just as linked to heat stress than age is. Supported by existing knowledge from laboratory studies we believe that physical sensitivity is primarily determined by health and fitness, not by age. While age may still serve as a sound statistical indicator, decision makers should not focus on age in the development of adaptation strategies (see the following section).

Adaptation strategies lack sufficient empirical evidence on heat stress risk as a fundament and respective policy development seems to be underdeveloped (Kinney et al 2008, Lesnikowski et al 2011, Yardley et al 2011, Wolf et al 2014). Previous recommendations on heat stress adaptation mainly focused on (a) heat exposure reduction via changes in the urban structure (e.g. urban green) or (b) managing health risks through improved heat warning systems and emergency response planning (Wilhelmi and Hayden 2010, Huang et al 2013). Measures that target (a) may reduce people's actual heat exposure, without affecting the important sensitivity dimension. Measures that target (b) deal with symptoms only, but do not tackle any root cause.

Health and fitness has obvious adaptation potential, unlike other recognized risk factors like age. Adaptation could aim at improving population health behavior as a preventive and sustainable adaptation strategy. Adaptation through health prevention, however, is not included in official recommendations at all (EEA 2012, IPCC 2014). This might be a consequence of the perception of urbanization, global warming, and aging as root causes (Fernandez and Creutzig 2015) with limited adaptation potential.

Active travel (walking, cycling) could be a preventive measure (to be started in absence of a heat wave). This has been mentioned by only very few authors (Harlan and Ruddell 2011, Huang et al 2013), but never studied before. Active travel represents the most effective way to improve cardiovascular health by integrating physical activity into daily life routines (Bassuk and Manson 2005, Mackett and Brown 2011, Nazelle et al 2011, WHO 2002, Creutzig et al 2012). And it is already identified as a means for climate change mitigation (EEA 2009, IPCC 2014). Through reduced CO₂ emissions from urban transport, societies could reap double rewards by addressing two root causes of urban heat stress: population health and global warming.

A preventive solution in the exposure domain could be the consideration of solar heating effects induced by bedroom window orientations in building design. Heat sensitive people could be advised to avoid unfavorable bedroom orientations. For emergency response, bedroom heat exposure could be reduced for particularly vulnerable people, e.g. those in hospitals (table 1).

The survey design enabled the exploration of a massive set of potential risk factors using descriptive statistics. Variable types and interdependence limited the use of statistical rigorous models and we cannot exclude selection bias from the non-probabilistic survey conducted. However, there were actually striking similarities between the characteristics of our sample and the Berlin population, even though it was not our aim to find representative statistics from our sample. The similarity was not only for the general characteristics of the sample (age, income, gender) but also for the heat-stress relevant characteristics identified in our study (distribution of age, obesity, chronic diseases, household sizes, mobility behavior, etc) (see supplementary material, part 3). Therefore, we assume no relevant selection bias in our survey, for the purpose of this study.

Further, more extensive survey designs could build upon this explorative study to validate the hypotheses it generated. Larger surveys, probabilistic sampling designs, and subsequent more rigorous statistical approaches would allow for a more precise quantification of effect magnitudes and potential confounding effects and to establish functional relationships, e.g. for specific types of diseases or air pollution. The cross-sectional nature of our survey has limitations regarding causal interpretations that could be reduced by longitudinal survey designs. To analyze specific risk factor associations in greater detail, the survey design could be thematically focused, eventually using fully standardized designs.

There are limitations concerning the comparability with mortality studies. However, as discussed in section 4.1, we assume that self-assessed heat impairment and heat-related mortality studies yield similar risk factor patterns, since subjective health indicators are regarded as reliable predictors of future mortality (Mossey and Shapiro 1982, Miilunpalo et al 1997). The question of geographic transfer is also critical. Further research is needed to validate our results and understand prospects for generalizing our main findings by transferring the approach to other cities worldwide.