Impact of Heat on Respiratory Hospitalizations among Older Adults in 120 Large U.S. Urban Areas

Abstract

Rationale

Extreme heat exposure is a well-known cause of mortality among older adults. However, the impacts of exposure on respiratory morbidity across U.S. cities and population subgroups are not well understood.

Objectives

A nationwide study was conducted to determine the impact of high heat on respiratory disease hospitalizations among older adults (≥65 yr of age) living in the 120 largest U.S. cities between 2000 and 2017.

Methods

Daily rates of inpatient respiratory hospitalizations were examined with respect to variations in ZIP code–level daily mean temperature or heat index. For each city, we estimated cumulative associations (lag days 0–6) between warm-season heat (June to September) and cause-specific respiratory hospitalizations using time-stratified conditional quasi-Poisson regression with distributed lag nonlinear models. We estimated nationwide associations using multivariate meta-regression and updated city-specific associations via best linear unbiased prediction. With stratified models, we explored effect modification by age, sex, and race (Black or White). Results are reported as percentage change in hospitalizations at high temperatures (95th percentile) compared with median temperatures for each outcome, demographic group, and metropolitan area.

Results

We identified 3,275,033 respiratory hospitalizations among Medicare beneficiaries across 120 large U.S. cites between 2000 and 2017. Nationwide, 7-day cumulative associations at high temperatures resulted in a 1.2% (95% confidence interval, 0.4–2.0%) increase in hospitalizations for primary diagnoses of all-cause respiratory disease, driven primarily by increases in respiratory tract infections (1.8% [95% confidence interval, 0.6–3.0%]) and chronic respiratory diseases and/or respiratory failure (1.2% [95% confidence interval, 0.0–2.4%]). Stronger associations were observed when exposure was defined using the heat index instead of mean temperature. Across the 120 cities, we observed considerable geographic variation in the relative risk of heat-related respiratory hospitalizations, and we observed disproportionate burdens of heat-related respiratory hospitalizations among the oldest beneficiaries (≥85 yr of age) and among Black beneficiaries living in South Atlantic cities. During the 18-year study period, there were an estimated 11,710 excess respiratory hospitalizations due to heat exposure.

Conclusions

Results suggest that high temperature and humidity contribute to exacerbation of respiratory tract infections and chronic lung diseases among older adults. Geographic variation in heat-related hospitalization rates suggests that contextual factors largely account for disproportionate burdens, and area-level influences should be further investigated in multicity studies.

Exposure to extreme heat and humidity has a direct impact on human health and well-being (1–3) and is projected to result in a 370% increase in heat-related mortality among those 65 years and older by midcentury (2). Robust associations between heat and excess mortality among older adults are consistently reported in the literature. However, associations between heat and healthcare utilization, such as emergency department visits and hospitalizations, are often weaker, with inconsistent findings across study locations (4–9). Previous studies of heat and respiratory morbidity have focused primarily on all-cause respiratory events or major categories of respiratory disease (chronic obstructive pulmonary disease [COPD], asthma, and respiratory tract infections [RTIs]), with little attention given to less common airway diseases such as interstitial lung disease, pneumonitis, and pleural effusion. The impact of extreme heat on cause-specific morbidity is an important knowledge gap that must be addressed to optimize patient outcomes and increase the resilience of healthcare systems for future extreme heat events.

Older adults are among the most physiologically sensitive to prolonged heat exposure because of the effects of aging on thermoregulation and evaporative heat loss. In addition, older adults may have impaired mobility, have preexisting diseases, or take medications that affect body fluid balance or interact with hemodynamic regulation (3, 10). When prolonged heat exposure results in hyperthermia, older adults may experience increases in ventilation rate, tidal volume, and respiratory rate (14), which may damage the lung parenchyma and exacerbate underlying respiratory illnesses. Epidemiologic and mechanistic studies provide evidence that some heat-related pulmonary injury may also be mediated by inhaling hot air (11, 15–18). Proposed pathways include the activation of bronchopulmonary vagal afferent C fibers, increased cholinergic responses, and the stimulation of specific heat shock proteins leading to both epithelial barrier dysfunction and airway inflammation (11, 16, 17).

In addition to advanced age, numerous studies have demonstrated that disproportionate exposure and vulnerability to hazardous heat is influenced by complex interactions among municipal intervention efforts, socioeconomic inequality (19), structural racism (20, 21), access to health care, adaptive capacity, risk perception, and physiological acclimatization and susceptibility (6, 19–23). Urban populations have also been shown to be at increased risk of heat-related health outcomes because of the urban heat-island effect (24), in which the heat-retaining and heat-generating properties of the urban form (low-albedo surfaces, building configurations that trap heat, diverse sources of heat generation, decreased vegetation, and less evapotranspiration) result in higher air temperatures in urban environments compared with rural areas (25).

In this nationwide study of respiratory hospitalizations among Medicare beneficiaries (65 yr and older) living in large urban centers (LUCs), we assessed the role of high ambient heat on cause-specific respiratory morbidity, quantified the heat-related respiratory burden for specific disease outcomes, and assessed whether the burden of all-cause respiratory hospitalizations varied by demographic group and/or geographic setting. This work’s novel contributions include defining the relationship between heat and previously unexplored respiratory diagnoses, as well as evaluating the variation in heat-attributable burden and risk across 120 LUCs.

Some of the results of these studies have been previously reported in preprint form (https://doi.org/10.1101/2024.05.22.24307126).

Methods

Outcome Ascertainment

We obtained inpatient hospitalization data for all Medicare beneficiaries 65–114 years of age from 2000 to 2017 from the Centers for Medicaid and Medicare Services. Ethical approval was obtained from the U.S. Environmental Protection Agency Institutional Review Board (IRB00006222). We excluded “long-stay” or “skilled nursing facility” hospitalizations as well as hospitalizations that occurred within four days of discharge from a prior hospitalization. Medicare records included date of admission; International Classification of Diseases, Ninth and Tenth Revisions, diagnosis codes for billing claims; age; race; sex; an indicator of short-stay, long-stay, or skilled nursing facility; and ZIP code of patient residence. Five cause-specific respiratory groupings were considered (Table 1): all causes, asthma, COPD, RTI, and all other respiratory diseases that did not fit into the previous categories. We referred to this last group as chronic respiratory disease or respiratory failure (CRD/RF), as nearly 86.8% of these hospitalizations were for diagnoses of RF or progressive inflammatory and fibrotic pulmonary diseases (see Table E1 in the data supplement). We did not separate the CRD/RF category into more specific diseases, because of a small number of events for some cause-specific outcomes (e.g., interstitial lung disease, sinonasal disease) and because of nonoverlapping differential diagnostic definitions when mapping International Classification of Diseases, Ninth Revision, to International Classification of Diseases, Tenth Revision, codes. Health endpoints were also distinguished on the basis of whether respiratory events were reported in the principal diagnostic code position (principal diagnoses) or the first three diagnostic code positions (first three diagnoses).

Table 1.

International Classification of Diseases, Ninth and Tenth Revisions, codes for respiratory disease outcomes

Outcome	ICD-9 Codes	ICD-10 Codes	Total Number of Primary Events, Warm Season
All-cause respiratory diseases	460–519	J00–J99	3,275,033
Asthma	493	J45	161,392
Chronic obstructive pulmonary disease	491, 492, 496	J41–J44	763,628
Respiratory tract infections	460–466, 480–488	J00–J18, J20, J21, J22	1,276,350
Pneumonia without flu	480–486	J12–J18	1,203,135
Chronic respiratory diseases (inflammatory and fibrotic) or respiratory failure*	470–478, 500–519, 490, 494, 495	J30–J40, J46–J49, J60–J70, J80–J86, J90–J99	1,073,663

Definition of abbreviations: ICD-9 = International Classification of Diseases, Ninth Revision; ICD-10 = International Classification of Diseases, Tenth Revision.

^*Among the events in this category, 86.8% are from respiratory failure, aspiration pneumonitis, pleurisy, pneumothorax, and postinflammatory pulmonary fibrosis or other interstitial pulmonary diseases (see Table E1).

Following previous work (24), we included hospitalizations of beneficiaries living in metropolitan statistical areas within the contiguous United States with a total 2010 population of 500,000 or more. From these metropolitan statistical areas, we identified 120 separate study locations (LUCs) (Figure 1; see Study Locations in the data supplement).

Figure 1.

Spatial representation of LUCs included in this study. LUCs are shown by U.S. division, and the color scale of the LUC indicates its International Energy Conservation Code climate zone. LUC = large urban center.

Ambient Meteorological Data

Weather station observations of daily ambient temperature and relative humidity were obtained from the Global Surface Summary of the Day dataset maintained by the National Oceanic and Atmospheric Administration’s National Centers for Environmental Information. Meteorological parameters from 3,208 weather stations were interpolated to U.S. census tract population centroids (the mean location of residence for a census tract population) for each day using thin-plate spline regression (24). Meteorological variables included daily mean, minimum, and maximum temperature (°C), daily mean relative humidity, and daily mean and maximum heat index (°C) calculated using the weathermetrics (https://github.com/geanders/weathermetrics/) package in R (https://www.r-project.org/). The analysis was restricted to June through September, the four warmest months in all LUCs during the study period (26). For more details, see Exposure Ascertainment in the data supplement.

Statistical Analyses

We applied a two-stage modeling approach to estimate associations between daily ZIP code–level ambient heat and respiratory hospitalizations within each LUC and nationwide. In the first stage, associations were estimated for each LUC using time-stratified conditional quasi-Poisson regression with distributed lag nonlinear models (27, 28), conditioning on ZIP code of residence, year, month, and day of the week of hospitalization (29) (see Detailed Statistical Analyses in the data supplement). A maximum lag window of 7 days (lag days 0–6) was chosen a priori to account for the effects of the exposure distributed over time (30–35). The primary exposure metric was daily mean temperature while controlling for same day relative humidity. Results for secondary exposure, daily heat index, are included in the data supplement. All models controlled for federal holidays and residual temporal trends and were fitted for each exposure metric and outcome. In addition, associations between heat and all-cause respiratory hospitalizations were estimated for the following demographic groups: sex (male and female), age group (65–74, 75–84, and 85+ yr), and race (Black and White).

In the second stage, we pooled the risk estimates using multivariate meta-regression to obtain nationwide effect estimates (36, 37). The meta-regression included LUC average ambient temperature and temperature range (difference between minimum and maximum temperatures) as fixed effects to explain variation across LUCs, as done previously (24, 38, 39). We used the best linear unbiased prediction to update the estimates of the LUC-specific associations (24, 36). Sensitivity analyses assessed whether results were robust to model specification, and those results are reported in the data supplement.

For all models, the cumulative effect is the sum of the relative risk (RR) estimates across lag days 0–6 associated with a specific temperature value relative to the LUC-specific warm-season median temperature (reference exposure). We report lag-response associations, cumulative percentage change in hospitalizations (%Δ = [RR − 1] × 100) and 95% confidence intervals (CIs) comparing high warm-season heat (95th ambient heat percentile) with median warm-season heat (40–43).

Attributable Burden Analyses

We used the following attributable burden measures to quantify the respiratory health impacts of heat exposure at the national, LUC, and subpopulation levels: 1) attributable number, defined as the number of respiratory hospitalizations attributable to temperatures or heat index values above the 50th percentile, and 2) annual attributable rate (AR), defined as the average annual attributable number per 100,000 Medicare beneficiaries within a subgroup. For the attributable measures, interval uncertainty was obtained empirically through Monte Carlo simulations (44) and reported as 95% empirical CIs. Additional details on the attributable burden calculation are reported in the data supplement.

Results

Descriptive Results

In an open cohort of 24.4 million Medicare beneficiaries 65–114 years of age living in the 120 largest U.S. LUCs from 2000 to 2017, we identified 3,275,033 hospitalizations with primary respiratory diagnoses and 8,374,625 hospitalizations with respiratory diagnoses in the first three diagnostic code positions (Table 2). Within all-cause respiratory diseases, RTI and CRD/RF were the top two categories, with pneumonia without flu accounting for 94% of all RTIs (Table 1). White and Black individuals were the first and second largest racial and/or ethnic demographics in the study cohort (Table 2). Daily hospitalizations among Asians, Hispanics, and North American Natives were not sufficiently numerous for separate analyses by these races or ethnicities. Cause-specific respiratory hospitalizations had similar proportions by age, sex, and race, except for asthma, which had proportionally more women and Black beneficiaries compared with other outcomes (Table 2).

Table 2.

Summary of respiratory hospitalization data, overall and by individual modifying factors, for 120 large urban centers, June to September, 2000–2017

	Total Hospitalizations [n (%)]
	All-Cause Respiratory	Asthma	COPD	CRD/RF	RTI
Overall
Principal diagnosis	3,275,033	161,392	763,628	1,073,663	1,276,350
First three diagnoses	8,374,625	527,677	3,076,960	3,819,130	2,625,698
Subpopulation*
Sex
Male	1,438,224 (44%)	42,077 (26%)	321,316 (42%)	499,486 (47%)	575,345 (45%)
Female	1,836,805 (56%)	119,315 (74%)	442,312 (58%)	574,176 (53%)	701,002 (55%)
Age group
65–74 yr	963,430 (29%)	61,690 (38%)	274,411 (36%)	307,729 (29%)	319,600 (25%)
75–84 yr	1,338,840 (41%)	66,679 (41%)	339,342 (44%)	424,600 (39%)	508,219 (40%)
≥85 yr	972,763 (30%)	33,023 (20%)	149,875 (20%)	341,334 (32%)	448,531 (35%)
Race/ethnicity
Black	351,148 (11%)	32,824 (20%)	81,785 (11%)	117,614 (11%)	118,925 (9%)
White	2,709,189 (83%)	110,227 (68%)	641,362 (84%)	883,374 (82%)	1,074,226 (84%)

Definition of abbreviations: COPD = chronic obstructive pulmonary disease; CRD = chronic respiratory disease; RF = respiratory failure; RTI = respiratory tract infection.

^*Total numbers of hospitalizations within the sex, age, and race/ethnicity groups are based on principal causes of the health outcomes. Race/ethnicity percentages do not sum to 100% because all races and ethnicities are not represented in this table.

Across the study period and all LUCs, warm-season daily mean temperatures ranged from 0.1°C to 42.6°C (see Table E2). LUCs in the Pacific and Mountain U.S. divisions experienced the widest warm-season temperature ranges, while temperatures were higher, on average, in LUCs of the South Atlantic and West South Central divisions (see Table E2). Across all locations, daily mean warm-season temperatures were highly correlated with maximum and minimum temperature, and mean and maximum heat index (see Table E3). Spearman correlations between warm-season mean temperature and warm-season relative humidity within each LUC were more variable and ranged from −0.71 to 0.10 (mean [standard deviation] ρ = −0.28 [0.25]; see Table E3). LUC average warm-season temperature and temperature range helped explain the heterogeneity across locations, as meta-regression I² statistics were <16% for all outcomes and all population subgroups (see Table E4). LUC-specific daily mean temperature summary statistics are reported in Table E5.

Nationwide Risk and Burden

Nationwide, 7-day cumulative associations at high warm-season temperatures resulted in a 1.22% (95% CI, 0.42–2.03%) increase in hospitalizations for primary diagnoses of all-cause respiratory diseases, driven primarily by increases in RTI (1.84% [95% CI, 0.70–3.01%]) and CRD/RF hospitalizations 1.17% (95% CI, −0.09–2.44%) (Table 2 and Figure 2). For these outcomes, associations monotonically increased with increasing temperatures above the 50th percentile of the warm-season temperature distribution (Figure 3, left; see Figure E1). Nationwide, 7-day cumulative associations for all-cause respiratory disease, CRD/RF, and RTI were slightly elevated when using mean heat index instead of mean temperature as the exposure metric (see Table E6 and Figure E2). We did not observe associations between high warm-season heat and asthma or COPD hospitalizations (Figures 2, E1, and E2). For all outcomes, principal diagnoses had a stronger association with high warm-season mean temperature compared with the first three diagnoses (Figures 2, 3, and E1). Given these findings, only principal diagnosis outcomes were considered in subgroup and LUC-specific analyses.

Figure 2.

Nationwide 7-day cumulative percentage change in hospitalizations and 95% confidence intervals between high warm-season temperature and cause-specific respiratory hospitalizations pooled across 120 large urban centers, June to September, 2000–2017. Percentage increase in hospitalization compares hospitalizations on days of high warm-season temperature (95th percentile) versus median temperature (reference exposure). Percentage changes in red represent associations between ambient temperature and respiratory hospitalizations reported in the first diagnostic code position (principal cause of hospitalization), while percentage changes reported in gray represent associations between ambient temperature and respiratory hospitalizations reported in the first three diagnostic code positions.

Figure 3.

Relative risks and 95% CIs between temperature and all-cause respiratory hospitalizations pooled across 120 large urban centers, June to September, 2000–2017. (A) Overall 7-day cumulative, nationwide exposure–response relationship between increases in daily average temperature and all-cause respiratory hospitalizations. (B) Nationwide lag–response association for each lag day comparing a day of high warm-season temperature (95th percentile) versus median temperature (reference exposure). Associations for principal respiratory causes of hospitalization are reported in red. Associations for respiratory hospitalizations reported in the first three diagnosis codes are reported in gray. The dashed black line indicates the temperature percentile value used as the centering point for temperature contrasts (50th percentile). The dotted red line indicates the 95th percentile (high warm-season temperature). Color bands around solid lines represent 95% CIs. CI = confidence interval; Mean Temp. = daily mean temperature.

For all-cause respiratory hospitalizations, we observed a 2.1% (95% CI, 1.5–2.8%) increase in hospitalizations on lag day 0, followed by a decline in risk on lag day 1, possibly due to morbidity displacement, and weak or slightly negative risk across later lag periods (Figure 3, right). This pattern was also observed across RTI, COPD, and CRD/RF outcomes (see Figure E1). Across all study locations and years, we estimated 11,710 (95% empirical CI, 8,290–14,670) excess all-cause respiratory hospitalizations due to warm-season temperatures above the 50th percentile (Table 3). Results from sensitivity analyses are reported in the data supplement and in Table E7 and Figures E12 and E13).

Table 3.

Seven-day cumulative percentage change in relative risks, attributable number, and annual attributable rates between respiratory hospitalizations and high warm-season temperature for 120 large urban centers, June to September, 2000–2017

Outcome	Percentage Change (95% CI)*		Attributable Number (95% eCI)†		Annual Attributable Rate (95% eCI)‡
All-cause respiratory
Overall	1.22	(0.42, 2.03)	11,710	(8,290, 14,670)	2.66	(1.89, 3.34)
Age group
65–74 yr	1.40	(0.15, 2.68)	3,460	(2,540, 4,270)	1.55	(1.14, 1.91)
75–84 yr	0.77	(−0.41, 1.97)	3,280	(1,490, 4,930)	2.21	(1.00, 3.32)
≥85 yr	1.46	(0.00, 2.94)	4,320	(2,480, 5,810)	6.36	(3.65, 8.56)
Sex
Male	1.01	(0.01, 2.03)	4,730	(3,520, 5,790)	2.54	(1.89, 3.11)
Female	1.20	(0.17, 2.25)	6,560	(3,980, 8,790)	2.59	(1.57, 3.47)
Race/ethnicity
Black	0.94	(−0.98, 2.90)	730	(90, 1,290)	1.78	(0.21, 3.15)
White	1.06	(0.16, 1.96)	8,750	(5,640, 11,400)	2.44	(1.52, 3.17)
Cause-specific outcomes
Asthma	0.57	(−2.45, 3.69)	200	(−220, 560)	0.05	(−0.05, 0.13)
COPD	0.09	(−1.31, 1.51)	380	(−530, 1,070)	0.09	(−0.12, 0.24)
CRD/RF	1.17	(−0.09, 2.44)	3,590	(2,130, 4,760)	0.82	(0.48, 1.08)
RTI	1.84	(0.70, 3.01)	6,580	(4,950, 8,030)	1.50	(1.13, 1.83)

Definition of abbreviations: CI = confidence interval; COPD = chronic obstructive pulmonary disease; CRD = chronic respiratory disease; eCI = empirical confidence interval; RF = respiratory failure; RTI = respiratory tract infection.

^*Percentage change in relative risk of respiratory hospitalization is in relation to a change in temperature from the 50th percentile to the 95th percentile.

^† Attributable number represents the estimated number of excess hospitalizations for temperatures above the 50th percentile, rounded to the nearest 10.

^‡ Annual attributable rate is the annual attributable number per 100,000 beneficiaries.

Subgroup-Specific Risk and Burden

We did not observe effect measure modification in RR by age group, sex, or race/ethnicity at the national level for either mean temperature or heat index (Tables 3 and E6; see Figures E2–E4). Exposure–response curves and lag associations between mean temperature and all-cause respiratory disease by demographic group are reported in Figure E4. For most subpopulations, RR was largest on lag day 0, but risk among Black beneficiaries was elevated across lag days 0 and 1, with the highest risk observed on lag day 1 (see Figure E4). For mean temperature and mean heat index, the burden of heat-related respiratory hospitalizations, reflected by the annual AR, was largest among beneficiaries 85 years and older compared with other age groups (Tables 3 and E6). Differences in attributable burden by sex and race/ethnicity were not consistently observed across exposure metrics, but heat index results suggest higher burden rates among women and White beneficiaries (Tables 3 and E6). Because there were very few observed hospitalizations among Black beneficiaries, we were unable to estimate associations for three LUCs (Provo–Orem, Utah; Boise City, Idaho; and Portland–South Portland, Maine).

LUC-Specific Risk and Burden

We observed considerable geographic variation in the magnitude and direction of high temperature–related RR and AR across the study areas. LUC-specific RR estimates were very similar, though often more elevated, when heat index was specified as the exposure metric compared with mean temperature. Medicare populations living in some cities (e.g., Miami, Florida) seemed particularly responsive to high–heat index days (see Figure E5).

For high temperature–related all-cause respiratory diseases, percentage change in hospitalizations ranged from a 3.3% decrease to a 4.7% increase, and ARs ranged from −7.6 to 11.9 annual excess hospitalizations per 100,000 beneficiaries (Figure 4). Across the nine U.S. census divisions, the highest attributable burden rates were observed in LUCs located in the Pacific, East North Central, New England, and Mid-Atlantic (Figures 4 and E6; see Table E8). Although the 58 LUCs in these divisions accounted for only 56% (1,826,777) of the total number of respiratory hospitalizations, 98.5% (11,539) of all high temperature–related excess respiratory hospitalizations between 2000 and 2017 were reported in these LUCs. The lowest RRs and ARs were observed in LUCs in the South Atlantic region, especially Florida (Figures 4 and E6; see Table E8). RTI and CRD/RF accounted for most of the heat-related respiratory hospitalization burden across all LUCs (see Figure E7).

Figure 4.

Percentage change and annual attributable rate of high temperature–related all-cause respiratory hospitalizations for 120 LUCs by U.S. division, June to September, 2000–2017. LUC-specific percentage changes in respiratory hospitalizations are shown by LUC-specific annual AN of excess hospitalizations per 100,000 beneficiaries (annual attributable rate) for each U.S. division. For each location, the color of the circle is determined by LUC-specific annual attributable rate, and the size of the circle is determined by the number of Medicare beneficiaries within each LUC. Annotated study area names represent LUCs (n = 6) in the top 5% of annual attributable rates, where Medicare beneficiaries experience the greatest burden of heat-related respiratory morbidity. AN = attributable number; LUC = large urban center.

At the metropolitan level, we observed a disproportionate burden of heat-related all-cause respiratory hospitalization across age groups, sex, and race. In contrast to the overall population, the highest burden among Black beneficiaries was observed in LUCs of the South Atlantic and East South Central divisions (Figures 5 and E8). In the Mid-Atlantic and Pacific divisions, the burden was more evenly distributed among White and Black beneficiaries. In the Mountain, West North Central, and New England divisions, which have many fewer Black beneficiaries, the White population accounted for most of the burden (see Figure E8). Across age groups, the population 85 years and older accounted for most of the burden in almost every LUC (see Figure E9). On average, burden rates were similar between men and women across most U.S. divisions. However, we consistently observed higher ARs among men living in LUCs of the Mountain and West North Central divisions (see Figure E10).

Figure 5.

Percentage change and annual attributable rate of heat-related all-cause respiratory hospitalizations among White and Black beneficiaries for 120 large urban centers (LUCs) by U.S. division, June to September, 2000–2017. LUC-specific percentage changes in respiratory hospitalizations are shown by LUC-specific annual AN of excess hospitalizations per 100,000 beneficiaries (annual attributable rate) for Black and White beneficiaries for each U.S. division. Red circles represent risk estimates among Black beneficiaries, and gray circles represent risk estimates among White beneficiaries. The size of the circle is determined by the number of Medicare beneficiaries within each study location. AN = attributable number; LUC = large urban center.

Discussion

In this nationwide study, we characterized the relationship between short-term exposure to high warm-season heat and cause-specific respiratory morbidity among older populations living in the 120 largest U.S. urban centers. Notably, we observed that elevated ambient temperature led to excess hospitalizations for RF and chronic inflammatory and fibrotic diseases such as pneumonitis, pleural effusion, and interstitial pulmonary diseases. Across demographic groups, most of the burden was attributed to the oldest age group (≥85 years), and most heat-related excess hospitalizations were due to exacerbation of RTI and CRD/RF. Respiratory health impacts were more dramatic when using the heat index, which allowed estimation of the combined effect of temperature and humidity. Finally, demographic and LUC results suggest that heat-related health disparities were likely driven by complex social and contextual forces acting on individuals to intensify their exposure and respiratory health risk.

To our knowledge, this is the first population-based study in the United States to define a relationship between ambient high heat and hospitalizations for a combined grouping of outcomes such as RF, pneumonitis, pleural effusion, and interstitial lung disease. Because of the mixed composition of the CRD/RF outcome, it is unclear which specific disease processes are more sensitive to heat exposure and why. However, the most prevalent diseases represented in CRD/RF (see Table E1) share common features, including progressive inflammation and/or fibrosis and fluid accumulation in the lung. A nascent literature base focused on the biological effects of high ambient heat on respiratory injury points to a few key mechanisms involving proinflammatory responses and epithelial barrier permeability that may act independently or in concert to initiate and sustain injury (11, 16, 17, 45, 46). Elevated temperatures have also been shown to affect electrolyte balance and activate the cholinergic pathway and/or C fibers, leading to injury (47–49). Inflammatory pathways can participate in the response to high temperatures with 1) elevated concentrations of white blood cells, lymphocytes, and neutrophils and 2) activation of proinflammatory mediators (e.g., interleukins, tumor necrosis factors) (50). Increased indices of oxidative stress and activation of specific heat shock proteins have also been observed under heat-stress conditions and are being investigated as possible mechanisms of pulmonary injury (51, 52). Compounding the individual effects of increased heat on respiratory function, the addition of suboptimal amounts of humidity (less than 40% and more than 60%), which we would expect with the global rise in temperature (53–55), are also known to adversely affect respiratory health through promoting bacterial and/or fungal growth and transmission of viruses, irritation of the respiratory tract, and weakening the mucosal layer and epithelial junctions (56). In addition, other factors may be influencing the pathophysiology of respiratory disease after elevated temperatures, including the impact of better medical surveillance and more widely available and effective treatments for individuals with diagnoses of asthma and COPD (e.g., inhaled bronchodilators, corticosteroids) relative to individuals with RTIs and CRD/RF. Although the lungs are a target for heat stress, the mechanisms by which heat disrupts lung homeostasis remain largely unknown, and further elucidation is needed.

The results of this study underscore the importance of considering geographic and sociodemographic factors when assessing heat-related health risks. At the national level, we did not observe effect modification by sex, age, or race; however, LUC and regional analyses suggest a disproportionate burden of heat-related respiratory hospitalizations among Black beneficiaries compared with their White counterparts across LUCs in the South Atlantic and East South Central divisions. We also observed higher heat-related excess hospitalization rates among beneficiaries of advanced age (85 yr and older) compared with other age groups in most LUCs. Although age-related susceptibility to extreme heat is well established, this study is among the first to demonstrate the disproportionate burden of heat on the respiratory health of older adults across nearly every major U.S. metropolitan area. The differences by race/ethnicity and age group were not apparent when comparing only RRs. For outcomes with different baseline population risks, the use of an excess burden metric, such as annual AR, can help evaluate the total impact of heat on health and is important to consider in addition to RR estimates.

In the overall Medicare population, risks were generally lowest in the Southeast and Southwest, possibly because of higher prevalences of air conditioning (57, 58) and greater physiological adaptation to heat. Findings of lower heat-related mortality and morbidity risk in communities with hotter summers compared with milder summers are commonly reported in multicity, multinational studies (4, 15, 43). However, the southwestern and southeastern U.S. geographies are expected to experience more severe extreme heat over the coming decades, and human heat acclimatization capacity has a physiological ceiling that may not be able to overcome projected temperature increases without behavioral and technological adaptations (12, 13). This study also identified LUCs in the Pacific, the East North Central, New England, and Mid-Atlantic that are particularly vulnerable to the effects of extreme heat exposure, with high risk and high burden rates among the Medicare population (Figure 4). On average, LUCs in these U.S. divisions have historically lower prevalence of central air conditioning (57, 58), mild summers (see Tables E3 and E5), and a wide range of warm-season temperature (see Tables E3 and E5). In addition, a recent study suggested that high-temperature warning thresholds may be too high in colder U.S. locations and may not take into consideration the sensitivity of the local population (59).

Overall, the nationwide pooled risk estimates for all-cause respiratory disease were in the range of risk estimates reported in other heat-related hospitalization studies among Medicare beneficiaries. In this study, we observed a 1.22% increase in all-cause respiratory hospitalizations across lag days 0–6. Other Medicare-based studies, with similar cumulative lag considerations and outcome definitions, reported heat-related increases in all-cause respiratory hospitalizations ranging from 0.0% to 4.0% for a change in the previous week’s ambient high heat (6, 15). In contrast to previous studies, we did not detect a cumulative association between heat and COPD. However, we observed elevated COPD exacerbation risk on lag day 0 (1.70% [95% CI, 1.04% to 2.36%]), with a large negative association possibly due to morbidity displacement on lag day 1 (−1.47% [95% CI, −2.12% to −0.80%]) and no association across later lag periods. Differences in study locations, study period, heat exposure metrics, the functional form of temperature, and how we defined a health relevant change in temperature (95th percentile to median temperature) could explain differences in reported risk estimates.

The ability to evaluate heat-related respiratory hospitalizations was facilitated by an 18-year study period with rich patient-level data and fine-scale meteorological data. However, additional considerations should be acknowledged when interpreting results. First, this study considers only the older adult population; as such, we would not expect these findings to be generalizable to younger adult populations or pediatric populations. In addition, this study did not explicitly estimate urban heat-island effects and is not able to account for air conditioning prevalence or other important behavioral and housing characteristics, because of limited data availability. Populations that do not have access to air conditioning or cannot afford to adequately cool their homes likely experience higher degrees of heat exposure and could be at greater risk. As other studies have indicated, heat-related mortality and morbidity risk may be in competition in some locations. Thus, high rates of heat-related mortality could artificially create a protective effect when examining heat-related hospitalizations (6, 8, 60). Also, the use spatially interpolated weather station data aggregated to the ZIP code level is unlikely to capture important spatial gradients such as urban heat-island effects and could have resulted in exposure misclassification error. We expect this error to be nondifferential and to attenuate associations toward the null, as this analysis related daily changes in ambient temperature to health outcomes, and day-to-day variation in ambient temperature was well characterized. However, misclassification error could be greater for certain populations (e.g., Black beneficiaries, beneficiaries living in urban heat islands), which may have attenuated risk estimates within population subgroups. Finally, reported risks for the entire Medicare population are driven largely by the strength of the association among White beneficiaries living in large LUCs (see Figure E11) and may not be generalizable to other races and ethnicities or to rural populations.

Conclusions

This study considerably extends the current understanding of the relationship between heat and respiratory morbidity among older adults by characterizing novel respiratory endpoints that have not been previously examined in a large, multicity epidemiologic study. The findings raise new questions on the role of heat in relation to RF and airway diseases such as interstitial lung disease, pneumonitis, and pleural effusion. In addition, we observed geographically dependent disparities in the impact of heat on the respiratory health of population subgroups. These results suggest that the drivers of heat-related respiratory morbidity are likely due to the joint effects of physiological susceptibility and contextual forces acting on individuals to intensify their exposure and risk. Additional studies to further explore these disparities, as well as the effects of heat on understudied airway diseases, are needed to corroborate our findings. Mechanistic studies, especially controlled exposures, will be essential to determine whether the associations identified in this study are causative in nature.