Estimating policy-relevant health effects of ambient heat exposures using spatially contiguous reanalysis data

While residents of New York State (NYS) have largely experienced moderate temperatures during the summers, the frequency and intensity of hot days and heat events have been increasing and are projected to continue to rise over the next several decades [1]. Many studies of health risks associated with extreme ambient heat exposures have been conducted in urban areas and rely on assigning exposure based on distance to a single monitor [2, 3]. Air monitors are sparsely located and tend to be concentrated in urban areas. In situ data therefore have limited utility in studying heat effects over large areas that may include suburban or rural areas which lack air monitoring stations [4]. The North American Land Data Assimilation System (NLDAS) which combines remote sensing, in situ and model data to provide fine scale temperature metrics at the near surface level [5, 6], makes it possible to assign more precise data statewide [7]. The NLDAS has been used to calculate relative temperature metrics [8]. The application of spatially resolved finer scale reanalysis data to estimate public health effects associated with extreme heat is a recent development for NYS climate research. This spatially contiguous temperature dataset allows for the quantification of areal heterogeneity in health effects related to heat.

Furthermore, the fine scale temperature data allow us to determine heat-health thresholds for excessive risk which can better inform initiatives to protect health during heat events, such as the National Weather Service (NWS) heat alert system. Regional NWS forecast offices issue excessive heat alerts (advisories, watches, and warnings) based on the maximum heat index forecasts over 24–72 h [9]. Heat index is a metric that combines temperature and humidity to estimate conditions most likely to affect human health. An advisory is issued when the heat index is forecast to exceed 37.8 °C (100 °F) for 2 h or more for upstate NYS regions [10]. Appropriate heat warnings may prevent morbidity due to heat exposures. However, current temperature thresholds for heat advisories and warnings in upstate NYS were established over 20 years ago and were not based on heat-health associations. Previous studies have also proposed the consideration of health events and diagnoses for identifying threshold temperatures [11, 12]. The NLDAS reanalysis dataset provides the opportunity to conduct heat-health analysis for all regions of NYS and reassess the criteria for heat advisories, so they are more relevant to temperatures experienced in NYS during the summer.

The aims of this study were to characterize morbidity risk related to summertime heat exposure using residential address-based health records for emergency department (ED) visits / hospitalizations (May through September, 2008–2012) in NYS. We incorporated reanalysis data (12-km NLDAS products) to define exposure-response functions for regions where observed data from air-monitors are unavailable. In addition, we evaluated current heat advisory criteria in the region and established heat-health thresholds for excessive risks of morbidity in order to inform health policy.

We examined the association between ambient summer temperatures and ED visits/hospital admissions for multiple health outcomes using fine-scaled environmental data over a broad geographic region. There has been a recent interest in using spatially resolved data to estimate health effects of heat in wider geographical regions [4, 27]. The use of a spatially contiguous exposure fields allows estimation of heat effects in small cities, rural and suburban areas that lack dense air monitoring networks. In addition, the case-crossover design minimized bias due to inter-individual variation and consequent uncontrolled confounding in epidemiological studies of heat exposures [22].

The goals of this study were to 1) assess the effect of heat on health using fine-scale reanalysis data and 2) use our findings to better inform policies that protect the health of NYS residents during periods of extreme heat. Our results show significant risks of heat stress, dehydration, and AKF from lags 0 through 6. We found that effects of heat can start at moderately high values of Heat Index much below the current thresholds for NWS advisories in the area. In its most recent directive for the Eastern United States region, the NWS has recognized that emergency department visits due to heat illness may begin well before NWS thresholds are met, thus forecasters have been directed to focus on issuing long-lead watches along with early, enhanced safety messaging via Public Information Statements and social media [10].

We observed a positive relationship between summer high temperatures and acute occurrences of heat stress and dehydration with lagged effect. The same-day exposure effects that we observed for heat stress and dehydration are expected due to the direct relationship between heat and these health outcomes; and are consistent with literature [28]. We found that significant risk of heat stress persisted up to 6 days after exposure. We found an elevated risk of acute renal illness from lag 0, with the highest risk at lag 1 for renal illness. We observed only a modest increase in risk of cardiovascular illness at lags 4–6. These findings are consistent with other studies of health effects in temperate climate [29, 30].

We also explored the potential effect modification by individual characteristics such as age, sex, race, ethnicity, rural/urban areas, and exposure month for all health outcomes. People aged 25–44 years were more susceptible to dehydration and heat stress compared to those between ages 5 and 24 years. The increased risk in the 25–44 group may be elevated because of subjects in this category have a higher likelihood of being exposed to heat through occupational and/or recreational exposures. Higher risk among older adults between 65 and 84 years old may be due to existing comorbid conditions that may make them more susceptible to heat stress [31].

The risk of dehydration in the transition months (May and September) is lower in magnitude than the hotter months of June, July and August. This may be due to the infrequency of events during this period. However, since the risk is still statistically significant, it suggests that when heat events occur in transition months, people are less prepared and likely to experience adverse health outcomes. There have been inconsistent findings for heat-health risk by gender [32–35]. In this study, we did not find significant differences between the risk of heat stress among females and males. But females were less likely to have dehydration-related ED visit/admissions compared to males. This is consistent with results from the Center for Disease Control and Preventions’ Environmental Public Health Tracking Program survey, where males were found to be more affected by adverse heat-related conditions than females [36]. These differences may be due to sex differences in sweat threshold, sweat rate, morphology (more muscle and less fat in men) and different levels of hormones regulating fluid balance [32]. They could also reflect the tendency for men to engage in more outdoor work than women, especially in the construction and agricultural industries [33–35].

Furthermore, we explored the spatial distribution of effects of temperature on heat-health outcomes in NYS. As expected, the metropolitan NYC area had the highest risk; likely due to its geographic location and the Urban Heat Island effect that results from building materials that retain heat and increased anthropogenic heat production [37, 38]. Rural areas in the region are perceived to be at lower risk due to a temperate climate and uncommon heat events. However, risk ratios for other urban centers and rural areas in NYS were comparable. We surmise that with increasing temperatures, a lack of climate mitigation measures such as access to home air-conditioners and public cooling centers, may lead to an increase in susceptibility to heat morbidity in these regions [37, 39].

Although we would expect that those with chronic health conditions may be at a higher risk, we observed an overall lower risk of hospitalization/ED visit for heat stress and dehydration in this group. It has been suggested that those with chronic long-standing conditions are more likely to avoid outdoor exposures during unpleasant or extreme weather, thus resulting in a null or negative correlation with temperature [24, 40]. A sub-analysis of comorbid conditions among those who visited the hospital for heat stress and dehydration during the study period supported this hypothesis (Additional file 1: Figure S2).

Heat warning systems utilize myriad approaches to characterize threshold temperatures for adverse effects of heat and form an important part of regional heat mitigation and climate adaptation policies [11, 41, 42]. Recognizing that the Excessive Heat Warning / Heat Advisory criteria should be based on regional climate variability and the effect of excessive heat on the local population, the NWS encourages regional offices to work with health departments and develop criteria based on scientific evidence derived from local data [43]. The NYSDOH has therefore worked closely with the local weather service offices to develop a recommendation to lower the heat advisory thresholds in our region based on results of this study. Other efforts in the northeastern US including NYC where criteria have been changed based on the predicted number of deaths attributable to forecast heat events were based on monitoring data from mostly urban centers [43, 44]. Prior to this study, the heat advisory threshold for the upstate NY region and surrounding areas was 37.8 °C (100 °F) or more for two consecutive hours. However, our research findings have shown that this criteria for issuing heat advisories may not be sufficiently protective for public health. The ERT for heat stress morbidity using the maximum heat index is 28.8 °C (83 °F) and the MRTs for other health outcomes in the study were between 24 and 26 °C (75.2–78.8 °F). We recommended that a conservative heat advisory threshold of 35 °C (95 °F) be considered for the general public, as that would capture a high proportion of heat events likely to result in significant morbidity, while avoiding warning fatigue if frequent advisories were issued at lower temperatures. Based on research findings and recommendations, four NWS offices (Albany, NY; Binghamton, NY; Buffalo, NY; and Burlington, VT) changed their heat advisory criteria for New York, effective on or about June 1st, 2018 to 35 °C (95 °F) or more for two consecutive hours. Lower advisory criteria for high risk populations such as outdoor workers, student athletes and school children are either already in place or being considered in the region [45–47]. The NWS heat-advisory threshold for NYC had been lowered previously [43]. We also recommend that heat awareness messaging be initiated early in the summer as our research suggests that significant morbidity risks can occur at temperatures lower than the NWS heat advisory thresholds.

Current public health interventions to increase awareness of heat exposure symptoms and provide advice for staying cool focus on public outreach to urban populations. Local health departments in rural areas are at the forefront of combating climate effects but are challenged with competing priorities and lack of tools and resources [48]. Using spatially resolved climate and health data from this project, NYSDOH has developed tools and infographics for each county in NYS to aid climate health surveillance and mitigation efforts [49]. Findings from this project have also been included in extreme heat messaging issued by the NYSDOH and disseminated through public websites and social media.

As with any epidemiologic analysis of administrative datasets, there are some limitations of our study that should be considered when interpreting results. We used hospital administrative data to identify individuals with the health outcomes assessed. Residential addresses were used to assign exposure temperatures in lieu of information on personal activities including exposure to indoor temperatures; and may not reflect the point of exposure, thereby leading to exposure misclassification if the exposure occurred elsewhere. However, the use of case-crossover design and same-day of the week selection for control days help in minimizing the bias introduced. We also do not know how behavioral adaptation relates to exposure such as access to air conditioning and cooling centers. Factors of the built environment that increase temperature such as housing type and socio-economic factors were not available. Occupational exposures were also not accounted for in this analysis. In urban areas, where land cover is predominantly non-vegetated, differences between the Land Surface Temperature (LST) and air temperature can be large. LST is the radiative skin temperature of the surface and is usually higher than air temperature. Our use of NLDAS air temperature data obviates this complication. NLDAS temperature data give estimates of the ambient temperature, derived from air monitor, remote sensing, and weather model data. The NLDAS air temperature compares relatively well with measured monitor data, where available [50]. Systematic differences exist in the association between temperature and health outcome based on time and type of temperature observations; and may consequently translate into differences in estimated temperature-health effects [51]. Although the spatial resolution of the NLDAS 1/8th-degree grid temperature data provides a more accurate measure of exposure than sparse monitoring networks, the urban scale features such as urban heat island effects may not be well-captured. Other studies comparing health effects using spatially resolved models with those estimated using monitoring data have demonstrated a better model fit probably due to better estimation of temperature variability in small areas [4]. Our study only estimated heat risks on morbidity. Previous studies evaluating heat risks of mortality have found lower risk ratios and higher threshold temperatures for mortality [11, 27].

In this project, we demonstrate the application of spatially resolved climate data in informing climate policy for a large region comprising urban and rural areas. Future efforts will be needed to assess health impacts on mortality. Our efforts have led to policy changes around definition and public health information on regional heat advisories. We continue to monitor how a change in heat advisory threshold will impact public health effects of heat in the region.