Short-term associations between warm-season ambient temperature and emergency department visits for Alzheimer’s disease and related dementia in five US states

Abstract

Background:

Ambient temperatures are projected to increase in the future due to climate change. Alzheimer’s disease (AD) and Alzheimer’s disease-related dementia (ADRD) affect millions of individuals and represent substantial health burdens in the US. High temperature may be a risk factor for AD/ADRD outcomes with several recent studies reporting associations between temperature and AD mortality. However, the link between heat and AD morbidity is poorly understood.

Methods:

We examined short-term associations between warm-season daily ambient temperature and AD/ADRD emergency department (ED) visits for individuals aged 45 years or above during the warm season (May to October) for up to 14 years (2005 – 2018) in five US states: California, Missouri, North Carolina, New Jersey, and New York. Daily ZIP code-level maximum, average and minimum temperature exposures were derived from 1km gridded Daymet products. Associations are assessed using a time-stratified case-crossover design using conditional logistic regression.

Results:

We found consistent positive short-term effects of ambient temperature among 3.4 million AD/ADRD ED visits across five states. An increase of the 3-day cumulative temperature exposure of daily average temperature from the 50^th to the 95^th percentile was associated with a pooled odds ratio of 1.042 (95% CI: 1.034, 1.051) for AD/ADRD ED visits. We observed evidence of the association being stronger for patients 65-74 years of age and for ED visits that led to hospital admissions. Temperature associations were also stronger among AD/ADRD ED visits compared to ED visits for other reasons, particularly among patients aged 65-74 years.

Conclusion:

People with AD/ADRD may represent a vulnerable population affected by short-term exposure to high temperature. Our results support the development of targeted strategies to reduce heat-related AD/ADRD morbidity in the context of global warming.

1. INTRODUCTION

Climate change attributed to greenhouse gas emissions from human activities has contributed to multiple heat-related public health challenges [1–4]. As a consequence of climate change, ambient temperature is projected to increase in the future [5–7]. For the US, climate models have projected that the population exposed to high temperature will increases four- to six-fold by 2070 [8]. From previous epidemiologic studies, short-term exposure to high temperatures has been identified as a risk factor for various adverse health outcomes, including fluid and electrolyte imbalance, acute renal injury, circulatory disease, respiratory disease, heat stroke, septicemia, and gastrointestinal infections [9–13].

Dementia is a leading public health issue with a substantial financial and emotional burden, and nearly 50 million individuals are affected worldwide [14]. By 2050, this number is projected to triple to 152 million [15]. Dementia has many causes including Alzheimer’s disease (AD), cerebrovascular disease, Lewy body disease, frontotemporal lobar degeneration, Parkinson’s disease, hippocampal sclerosis, and of mixed pathologies [16]. Alzheimer’s disease (AD) contributes to the majority of dementia cases (60-80%) and is the sixth leading cause of death in the US; more than 6 million people in the US presently suffer from AD and this number is projected to reach approximately 12.7 million by 2050 in people aged 65 years and older [16, 17]. In addition to AD, frontotemporal, Lewy body, vascular, and mixed dementias are considered AD-related dementias (ADRD) which contribute to millions of dementia cases in the US [18, 19]. Various studies have showed that the prevalence of AD/ADRD differs by race in the US, with a higher prevalence observed for Blacks and Hispanics than for Whites [16, 18]. The substantial demands of medical care for people with AD/ADRD result in high economic burden for patients, caregivers, and society. In 2021, the medical expenditures directly associated with AD/ADRD was estimated to be $355 billion in the US among populations aged over 65 years [16]. With the growth of the elderly population, the direct cost is estimated to be $1 trillion by 2050 [20–22].

People with dementia suffer from reduced or lost cognitive function, which greatly influences their daily lives [22]. There is biological evidence indicating that thermoregulation is altered for people with dementia [23]. Several studies have reported that the core body temperature is increased in AD patients [24–26], which is hypothesized to be caused by the increased expression of cytokines in the brain and neuroinflammation [26]. Furthermore, due to their cognitive issues, some people with dementia have more difficulty interacting with their environmental conditions and communicating with their caregivers to reduce heat exposure [27]. These different hypothesized mechanisms related to patient physiology, behavior, and the social and built environments may potentially lead to heterogeneity in the effect of extreme temperature.

Recently, an increasing number of epidemiological studies have reported that exposure to heat waves or extreme temperature events increased mortality and morbidity associated with AD or dementia [28–32]. However, morbidity studies have predominantly focused on hospital admissions, have typically only examined AD, or have been restricted to the elderly population (≥65 years of age). Using hospital admissions as the measure of morbidity may only capture AD/ADRD cases resulting in severe health conditions, while missing a considerable portion of the disease burden and health care costs associated with emergency department (ED) visits that do not lead to hospitalization. Hence, there remains a need to more comprehensively examine associations between short-term exposure to high ambient temperature and dementia-related morbidity. This knowledge could play an important role in allocating healthcare resources and designing effective preventive strategies to reduce AD/ADRD morbidity due to heat.

In this work, we conducted time-stratified case-crossover analyses to estimate short-term health effects of high ambient temperature on both AD and ADRD ED visits in five US states during warm seasons: California, Missouri, North Carolina, New Jersey, and New York. We included both ED visits with AD/ADRD as a primary and secondary diagnosis, both ED outpatient visits and ED visits that resulted in inpatient admission, and included patients over a broad age range (≥45 years of age). To our best knowledge, this is the largest study to date on ambient temperature and AD/ADRD ED visits in the US.

2. METHODS

2.1. Emergency department visit and meteorological data

We obtained patient-level ED visit records from hospital associations or state health departments in five US states: California (2005 - 2016), Missouri (2005 - 2018), North Carolina (2011 - 2017), New Jersey (2005 - 2016), and New York (2005 - 2016). These five states were selected due to data availability and that they represent different climatic regions. The definition of an ED visit included patients seen in the ED and admitted to hospital, as well as patients discharged direct from the ED. Visit records included admission date, age in years, self-reported patient’s ZIP code of residence, and International Classification of Diseases (ICD) diagnosis codes. We applied the condition algorithm from the Centers for Medicare and Medicaid Services to identify AD/ADRD ED visits, using both primary and secondary diagnosis codes (Supplementary Table S1) [33]. Additional information about data sources are given in Acknowledgements.

To examine the association between high ambient temperature and AD/ADRD ED visits, the analysis was restricted to the warmest six months of the year, May to October. Daily maximum and minimum temperature in degrees Celsius, and water vapor pressure in pascals at 1 km × 1 km spatial resolution were acquired from Daymet [34]. Daymet model outputs 1 km gridded daily estimates of minimum temperature, maximum temperature, precipitation, shortwave radiation, vapor pressure, snow water equivalent, and day length using inputs from ground-based metrological stations over large regions including Canada, Mexico, the US, and Puerto Rico [34, 35]. Daily average temperature was calculated by taking the average of daily maximum and minimum temperature as a common practice. The daily 1km gridded Daymet data were then aggregated to the 2010 ZIP code tabulation areas (ZCTA) by spatial averaging. ED records with ZIP codes outside of the study states or that could not be assigned to a ZCTA were excluded (1.10%).

2.2. Statistical analyses

To estimate state-specific and overall short-term effects of various temperature metrics on AD/ADRD-related ED visits, we implemented a two-stage approach. In the first stage, state-specific associations were estimated using a conditional logistic regression model with a time-stratified case-crossover design, run separately for each state. In the case-crossover design, each AD/ADRD-related ED visit case was matched to multiple control days based on the patient’s residential ZCTA, visit year, month, and day of week. This case-only design automatically controls for individual-level time-invariant confounders. We assessed associations between AD/ADRD-related ED visits and daily minimum, average, and maximum temperatures at different lags (0-3) using distributed lag models. We examined these three temperature metrics because they are proxies of temperature exposures during different time period of a given day. For example, the daily minimum temperature could be a good proxy of nighttime exposure, and daily maximum temperature may reflect the exposure when people perform outdoors activities since it usually occurs in the afternoon. The daily average temperature is the average of the daily minimum and maximum temperatures. For our primary analysis, we reported cumulative 3-day effect obtained from the distributed lag model which permits effects of temperature at different lags to be different. This model was also selected because it was favored by the Akaike information criterion (AIC) [36]. In all models, the included temperature metric were modeled non-linearly using natural cubic splines with 4 degrees of freedom. All models included several time-varying confounders not accounted for in the case-crossover design. First, residual temporal trends were modeled using natural cubic splines on day of year (4 degrees of freedom) with interaction terms between spline coefficients and indicators of year. Temporal trends are needed to potentially control for within-month trends that are often present in ED visits, in contrast to mortality. Second, we controlled for dew-point (DP) temperature using natural cubic splines with 4 degrees of freedom. Because Daymet derived vapor pressure which is used for computing DP from night-time minimum temperature and precipitation variables, we did not control for DP when examining minimum temperature effects [37]. Finally, indicators representing days of federal holidays were included. To summarize the overall effects of temperature on AD/ADRD-related ED visits across states, in the second stage, state-specific spline coefficients of the exposure-response function were combined under a multivariate fixed-effect meta-analysis framework using weighted least squares.

Stratified analyses were conducted by patient age (45-64 years, 65-74 years, ≥75 years), restricting to AD/ADRD ascertained by primary diagnosis only, and by whether the ED visit resulted in hospitalization. These three age groups were chosen based on the cut points used to distinguish young-onset, early-onset, and late-onset dementia [38–40]. To explore whether the AD/ADRD patient population is more vulnerable to temperature compared to the general population utilizing the ED, we assessed and compared temperature associations among all ED visits that were not identified as AD/ADRD. Because of the large counts of non-AD/ADRD ED visits, a conditional Poisson model [41] was used following the same case-crossover design and confounder adjustment.

We implemented several sensitivity analyses to assess the robustness of our results, including exploring the association between ED visits and the 3-day moving average (same day and 2 preceding days) for all three temperature metrics, and examining cumulative 2-day effects of different temperature metrics using distributed lag models. Compared to the individual lag models containing 3-day moving average, the distributed lag models considering temperatures at lags 0-2 allow more flexible non-linearly associations between ED visits and temperature metrics at different lags. All data analyses were performed in R 4.0.2 [42]; the function clogit in R package survival were used to fit conditional logistic regression models, and the gnm function in R package gnm were used to fit conditional Poisson regression models.

3. RESULTS

3.1. Summary statistics

Table 1 presents the number of ED visits for AD/ADRD during the warm season (May - October) within each state by three age groups (45-64 years, 65-74 years, and ≥75 years). Additional stratifications are given in Supplementary Table S2. Overall, the study included 3,479,420 AD/ADRD ED visits. The majority of ED visits were among patients ≥75 years of age (83.24%), resulted in a hospital admission (63.04%), were coded as ADRD (74.63%), and were coded in the secondary diagnosis fields (94.50%). Descriptive statistics for daily minimum, average, and maximum temperature at lags 0, 1, and 2 are presented in Table 2. Differences in the means and variabilities of the temperature metrics across states reflect the different climatic regions that each state represents: West (California), upper Midwest (Missouri), Southeast (North Carolina) and North East (New Jersey and New York).

Table 1.

Total number of AD/ADRD ED visits stratified by age groups during May to October in five US states: California (2005 - 2016), Missouri (2005 - 2018), North Carolina (2011 - 2017), New Jersey (2005 - 2016), and New York (2005 - 2016).

State	Age Group	AD	ADRD	AD/ADRD
California	Ages 45-64	8,257	48,068	56,325
	Ages 65-74	34,340	121,841	156,181
	Ages 75+	298,334	862,546	1,160,880
	Overall	340,931	1,032,455	1,373,386
Missouri	Ages 45-64	2,657	13,652	16,309
	Ages 65-74	10,389	29,305	39,694
	Ages 75+	69,738	158,267	228,005
	Overall	82,784	201,224	284,008
North Carolina	Ages 45-64	3,478	25,001	28,479
	Ages 65-74	15,211	62,123	77,334
	Ages 75+	94,232	296,840	391,072
	Overall	112,921	383,964	496,885
New Jersey	Ages 45-64	3,844	15,014	18,858
	Ages 65-74	13,547	33,271	46,818
	Ages 75+	115,562	243,502	359,064
	Overall	132,953	291,787	424,740
New York	Ages 45-64	6,514	32,056	38,570
	Ages 65-74	23,203	81,067	104,270
	Ages 75+	183,366	574,195	757,561
	Overall	213,083	687,318	900,401
Total	Ages 45-64	24,750	133,791	158,541
	Ages 65-74	96,690	327,607	424,297
	Ages 75+	761,232	2,135,350	2,896,582
	Overall	882,672	2,596,748	3,479,420

Table 2.

ZCTA-level mean, standard deviation (SD), interquartile range (IQR), median, and 95^th percentile of daily minimum, average, and maximum temperature at lags 0, 1, and 2 in °C during the warm season (May-October).

State	Temperature metrics	Mean (SD)			IQR			Median			95th percentile
		Lag 0	Lag 1	Lag 2	Lag 0	Lag 1	Lag 2	Lag 0	Lag 1	Lag 2	Lag 0	Lag 1	Lag 2
California	Daily minimum	13.60 (4.49)	13.60 (4.49)	13.60 (4.50)	5.62	5.62	5.61	13.68	13.68	13.68	20.55	20.55	20.55
	Daily average	20.64 (4.97)	20.65 (4.98)	20.65 (4.98)	6.53	6.53	6.52	20.53	20.54	20.54	28.80	28.80	28.80
	Daily maximum	27.69 (6.24)	27.70 (6.24)	27.71 (6.24)	8.96	8.96	8.96	27.49	27.50	27.51	37.92	37.92	37.92
Missouri	Daily minimum	15.19 (5.83)	15.22 (5.79)	15.24 (5.75)	8.12	8.08	8.08	16.42	16.42	16.43	22.75	22.75	22.75
	Daily average	21.22 (5.60)	21.25 (5.56)	21.26 (5.53)	7.33	7.30	7.28	22.25	22.26	22.26	28.71	28.71	28.71
	Daily maximum	27.24 (5.93)	27.27 (5.87)	27.29 (5.87)	7.39	7.36	7.32	28.23	28.23	28.24	35.21	35.21	35.21
North Carolina	Daily minimum	16.33 (5.22)	16.37 (5.17)	16.41 (5.12)	7.20	7.15	7.11	17.37	17.38	17.39	22.91	22.91	22.91
	Daily average	22.07 (4.74)	22.10 (4.68)	22.13 (4.63)	6.59	6.55	6.52	22.98	22.98	22.99	28.29	28.29	28.29
	Daily maximum	27.80 (4.83)	27.83 (4.79)	27.86 (4.74)	6.45	6.41	6.39	28.62	28.62	28.63	34.42	34.42	34.42
New Jersey	Daily minimum	14.48 (5.44)	14.49 (5.41)	14.51 (5.39)	7.95	7.95	7.95	15.21	15.21	15.21	22.12	22.12	22.12
	Daily average	20.02 (5.24)	20.03 (5.21)	20.04 (5.19)	7.58	7.58	7.57	20.89	20.89	20.89	27.36	27.36	27.36
	Daily maximum	25.56 (5.58)	25.57 (5.56)	25.58 (5.54)	7.82	7.81	7.81	26.51	26.52	26.52	33.34	33.34	33.34
New York	Daily minimum	11.84 (5.79)	11.85 (5.77)	11.85 (5.76)	8.19	8.18	8.18	12.31	12.31	12.31	20.62	20.62	20.62
	Daily average	17.49 (5.54)	17.50 (5.52)	17.51 (5.50)	7.77	7.76	7.76	18.13	18.14	18.14	25.48	25.48	25.48
	Daily maximum	23.13 (5.93)	23.15 (5.91)	23.16 (5.89)	8.11	8.08	8.07	24.02	24.03	24.03	31.38	31.38	31.38
Overall	Daily minimum	13.69 (5.55)	13.71 (5.53)	13.72 (5.52)	7.59	7.58	7.57	14.15	14.15	14.16	21.88	21.88	21.88
	Daily average	19.84 (5.54)	19.86 (5.52)	19.87 (5.51)	7.46	7.44	7.43	20.36	20.36	20.37	27.91	27.91	27.91
	Daily maximum	25.99 (6.24)	26.00 (6.22)	26.02 (6.20)	8.18	8.16	8.15	26.55	26.56	26.56	35.34	35.34	35.34

3.2. Temperature effects

Figure 1 shows the estimated odds ratios for AD/ADRD ED visits associated with change in 3-day temperature exposure (lags 0-2) from their 50^th to the 95^th percentiles (numeric values are given in Supplementary Table S3). This exposure contrast was selected to represent the state-specific extreme heat effects on AD/ADRD ED visits. For all temperature metrics, temperature was consistently positively associated with AD/ADRD ED visits across the five US states during the warm season. The positive associations were also present using a different exposure contrast reflecting effects of typical heat exposure (an increase from the 25^th to the 75^th percentile, Supplementary Figure S1). For the pooled results, and for all state-specific results except New Jersey, associations with average temperatures at lags 0-2 were stronger than those using maximum and minimum temperature. For example, the estimated pooled odds ratio associated with daily average temperature increasing from 20.4°C to 28.0°C (95^th versus 50^th percentile contrast) for lags 0, 1, and 2 was 1.042 (95% CI: 1.034 - 1.051).

Figure 1.

Short-term associations between AD/ADRD ED visits and change in 3-day temperature exposure from the 50^th to the 95^th percentiles during the warm season in 5 states: California (CA), Missouri (MO), North Carolina (NC), New Jersey (NJ) and New York (NY). Three temperature metrics were examined: daily minimum (MIN), daily average (AVG), daily maximum (MAX).

Figure 2 (Panel A) shows the state-specific exposure-response functions of daily average temperature at lags 0-2. Specifically, Figure 2A) presents the state-specific estimated odds ratios associated with one unit change in daily average temperature at lags 0-2 from their 50^th percentiles. For the other two temperature metrics (daily maximum and minimum), the corresponding state-specific exposure-response functions are presented in supplementary materials (Supplementary Figure S2). The pooled exposure-response functions for the three temperature metrics are given in Figure 2 (Panel B). Overall, we found that daily temperature was associated with ED visits for AD/ADRD over the entire exposure distribution. As observed in Figure 1, daily average temperature showed the strongest pooled associations with ED visits compared to the other two temperature metrics.

Figure 2.

A) State-specific exposure-response functions and 95% pointwise confidence intervals (CI) for the associations between AD/ADRD ED visits and change in 3-day temperature exposure from their 50^th percentiles of state-specific distributions of the daily average temperature; B) Pooled exposure-response functions and 95% pointwise CIs for the associations of 3-day temperature exposure for daily minimum, average, and maximum temperature. Reference level of exposure-response function is set to the 50^th percentile of the distribution of temperature metrics across all states (MIN: 14.15 °C for lag 0, 14.15 °C for lag 1, and 14.16 °C for lag 2; AVG: 20.36 °C for lag 0, 20.36 °C for lag 1, and 20.37 °C for lag 2; MAX: 26.55 °C for lag 0, 26.56 °C for lag 1, 26.56 °C for lag 2).

3.3. Modifications of temperature effects

Results from stratified analysis by whether the ED visit led to a hospital admission and patient age are shown in Figure 3 and Supplementary Figure S3, respectively. Generally, we found evidence of stronger associations among AD/ADRD ED visits subsequently admitted to hospital for all three temperature metrics (Figure 3). In pooled analyses, we found stronger associations among younger patients (i.e., aged 45-64 and aged 65-74 years) compared to the oldest age group (i.e., ≥75). Notably, despite the smaller sample size compared to older age groups, in California and New York, we also identified positive associations between temperature and AD/ADRD ED visits among those ages 45 to 64 (Supplementary Figure S3). When restricting our analysis to associations between ED visits with AD/ADRD as primary diagnosis and temperature,we identified stronger positive associations compared with associations for AD/ADRD ED visits including both primary and secondary diagnosis (Supplementary Figure S4).

Figure 3.

Short-term associations between AD/ADRD ED visits stratified by decision on hospital admission and change in 3-day temperature exposure from the 50^th to the 95^th percentiles during the warm season in 5 states: California (CA), Missouri (MO), North Carolina (NC), New Jersey (NJ) and New York (NY). Three temperature metrics were examined: daily minimum (MIN), daily average (AVG), daily maximum (MAX).

3.4. Compare temperature effects on AD/ADRD vs. non-AD/ADRD ED visits

Finally, Figure 4 compares the pooled exposure-response functions for AD/ADRD and non-AD/ADRD ED visits among patient ages 65-74 years and ≥75 years. Across three exposure metrics, we found evidence of increased risks among AD/ADRD patient ages 65-74 years when the temperature metric increasing from their 50^th percentiles, and the increased risks are more pronounced at the highest temperatures. The increased risks at the highest temperatures were also observed among AD/ADRD patient ages ≥75 years and all patients (≥45 years) for all three exposure metrics (Figure 4 Panel B and Supplementary Figure S6). In pooled analysis, these differences between risks of AD/ADRD and non-AD/ADRD ED visits at highest temperatures are also reflected in the odds ratio for the 50^th to the 95^th percentile exposure contrast (Supplementary Figures S5 and S7).

Figure 4.

Pooled exposure-response functions and 95% pointwise CIs for the associations of 3-day temperature exposure for daily minimum, average, and maximum temperature. Reference level of exposure-response function is set to the 50^th percentile of the distribution of temperature metrics across all states among populations A) ages 65-74 years and B) aged ≥ 75 years (MIN: 14.15 °C for lag 0, 14.15 °C for lag 1, and 14.16 °C for lag 2; AVG: 20.36 °C for lag 0, 20.36 °C for lag 1, and 20.37 °C for lag 2; MAX: 26.55 °C for lag 0, 26.56 °C for lag 1, 26.56 °C for lag 2).

3.5. Sensitivity analysis

The positive associations between same-day temperature adjusted for the other two lags and AD/ADRD ED visits were consistently identified for daily minimum, average, and maximum temperature (Supplementary Figures S8 and S9). We observed attenuated lag-specific effects adjusted for other two lags for lags 1 and 2 in pooled analysis (Supplementary Figures S8 and S9). Numerical values of all estimated odds ratios, the corresponding 95% confidence intervals, and AIC from different models (Supplementary Table S4) are provided in Supplementary Materials.

4. DISCUSSION

Over 3.4 million ED visits for AD/ADRD from five US states across different climate regions were analyzed in our large population-based study. We found consistent positive short-term temperature effects on AD/ADRD ED visits for various temperature metrics (i.e., daily minimum, average, and maximum temperature) during the warm season. Our study provides new epidemiological evidence that short-term exposure to ambient temperature is linked to increased AD/ADRD ED visits for individuals aged 45 years or above, indicating an underestimated public health burden associated with global warming.

In this study, we observed regional differences in the risk of AD/ADRD ED visits across five US states reflecting different geographical and climatic regions. Compared to other states, the strongest positive associations between daily minimum temperature and AD/ADRD ED visits measured by odds ratio associated with a change in 3-day daily minimum temperature from their 50^th to 95^th percentiles was observed in California. It is known that the daily minimum temperature typically occurs at night time and is deemed as a proxy for nighttime exposure. We speculate that the strongest effects of the daily minimum temperature in California might because the percent of homes with air conditioner (AC) is probably the lowest among these five states [43]. People in California may tend to stay in warmer homes compared to people in other four states at night times, which may result in higher risk of being exposed to high temperature.

Three temperature metrics have been examined in this study, consistent positive effects of different temperature metrics on AD/ADRD ED visits were observed (Figures 1 and S1) with strongest effects observed for daily average temperature in pooled analysis and in most states. Models using daily average temperature capture both effects of daily minimum and maximum temperature. The observed strongest effect might because both minimum and maximum temperature have positive effects on AD/ADRD ED visits.

Although there has been no study conducted specifically on the association between short-term ambient temperature exposure and AD/ADRD morbidity as measured by ED visits, our results are consistent with previous studies on dementia-related hospital admissions and mortality [28, 30, 31, 44]. For example, a time-series study found an increase in daily maximum temperature at lag 1 over a certain threshold was associated with increased risk of dementia-related emergency hospital admissions in Madrid [28]. Fritze (2020) reported a short-term effects of heat waves on increasing mortality of people with dementia in Germany [31]. In this analysis, we have primarily focused warm-season months (May-October) and examined shortterm effects using exposures up to 2-day lag. However, previous studies with focuses on mental disorders reported cold effects when exploring longer lags during cold seasons [45], this finding warrants a future research of expanding our analysis to study cold effects.

While previous studies have showed that people with dementia represent a vulnerable subpopulation to high ambient temperature, the most studies have not directly compared AD/ADRD patient populations to the general population. The exposure-response functions estimated in this analysis (Figure 4 and Supplementary Figure S6) highlight that the difference of temperature effects on ED visits between AD/ADRD populations and general populations vary by age group. Specifically, the more pronounced difference is observed among individual aged 65-74 years. One possible explanation is that older populations commonly suffer from declined motor function [46, 47], which may impair their physical functioning, which prevent them from outdoors activities. In general, older populations may spend less time for outdoors activities during the day time and have more other comorbidities.

We further observed that the temperature effects for AD/ADRD-related ED visits varied by age group. Specifically, we found stronger positive associations in age groups (45-64 years and 65-74 years) compared to the 75+ age group. For example, the estimated pooled odds ratio associated with the 95^th versus 50^th percentile exposure contrast of 3-day cumulative exposure for daily minimum temperature was 1.065 (95% CI: 1.046,1.085) for ages 65-74 and 1.026 (95% CI: 1.019, 1.033) for ages ≥75. Older patients generally receive better care as the disease progresses may explain the larger temperature effects was observed among patients aged 65-74 years compared to patients aged 75+ [16]. For example, patients with more severe loss of cognitive functions reside in long-term care facilitates that have improved mitigation strategies against high temperature.

The more pronounced associations were observed for ED visits resulting in a hospital admission versus discharged ED visits (Figure 3). Compared to discharged ED visits, ED visits resulting in a hospital admission likely represent more severe AD/ADRD and/or AD/ADRD cases with more severe comorbidities. This stronger association indicates that patients having more severe outcomes (i.e., AD/ADRD patients that resulted in more sever health conditions) are more sensitive to high temperature. This result highlights the importance of identifying AD/ADRD patients that may be more vulnerable to severe heat-related outcomes. Specifically, further research is warranted to examine which specific patient characteristics and neighborhood factors may confer differential risks of extreme heat as done similarly for other outcomes [48–51]

One particular analysis is to assess effect modification by co-morbid conditions among AD/ADRD ED visits. A growing number of epidemiological and neuropathological studies have suggested the associations between cardiovascular risk factors (e.g., hypertension, diabetes, and obesity) and AD/ADRD [22, 52–54]. AD/ADRD patients are also known to have higher prevalence of other co-morbid conditions, including respiratory diseases and chronic kidney diseases [55–57]. We also found that the majority of ascertained AD/ADRD ED visits were identified with ICD-9/10 codes in secondary diagnosis field. For example, among our AD/ADRD ED visits, we found that more than 80% had a co-diagnosis of cardiovascular disease. Thus, the primary reason for many visits was likely for conditions other than AD/ADRD, and using only ED visits with a primary AD/ADRD diagnosis could be an underestimation of the AD/ADRD morbidity.

Another area of future research is to consider other types of extreme temperature exposures. For example, heat wave represents a sustained period of extreme heat (e.g., at least 2 days above the 98^th percentile). The frequency, intensity, and duration of extreme heat events are predicted to increase in the future [5, 6]. For the US, the number of days with extreme heat events and the duration of extreme heat events are both projected to be doubled by 2065 [7]. There are numerous epidemiological studies have shown that heatwaves were associated with higher risks of ED visits and hospital admission [58, 59]. Studies from extreme heat events have also identified the AD/ADRD patients as vulnerable population for mortality [60, 61].

Our analysis has mainly focused on assessing effects of ambient temperatures, however, other meteorological exposures (e.g., humidity, solar radiation, and precipitation) may also have impacts on the health outcomes. A prior study in Australia found that an increase in temperature and humidity is associated with an increase in the proportions of individuals with mental health conditions [62]. One possible future study is to examine joint effects of temperature and humidity on AD/ADRD ED visits.

As reported in the literature, AD accounts for 60% - 80% of all-cause dementia cases. However, in our study, the proportion of AD ED visits out of AD/ADRD-related ED visits were only around 20% - 30% across different states (Supplementary Table S2). This difference might be attributed to several reasons. First, we used diagnosis codes from bill records to identify AD and ADRD, which may suffer from differential under-reporting. Second, our case ascertainment may be impacted by disparities in patient’s health care. There is an increasing evidence base to suggest the existence of disparities in health care access for AD/ADRD (e.g., diagnosis, treatment) [22, 63], including differences between racial groups [64–66].

Our study has several limitations that are typical of similar population-based analyses of ambient temperature. First, we used ambient temperature as a surrogate to personal exposures. Specifically, using ambient temperature fails to capture individual heterogeneity in temperature measurements due to indoor temperatures and urban heat island. However, we utilized the Daymet temperature product that allowed us to account for spatial variation in exposure across patient ZIP codes, which has been shown to result in stronger health effect estimates compared to the use of coarser exposure assessment methods [9]. Further, we also assessed different temperature metrics (e.g., maximum and minimum), which may reflect different timing of exposures. Second, outcome misclassification may occur in identifying AD/ADRD ED visits based on diagnosis codes. Exposure measurement error may also occur due to the use of patient reported residential ZIP code. However, in our setting, exposure misclassifications are likely to be independent of ambient temperature and health outcome, which result in effect attenuation. Finally, the case-crossover design only allows for evaluating effects of ambient temperature on ED visits among people with AD/ADRD who had an encounter at the ED. Such subpopulations might not necessarily be representative of whole at-risk populations.

Specific strengths of our study include the consideration of both ED visits resulting in hospital admissions and discharged ED visits, the use of a spatially-resolved temperature product, and the inclusion of ADRD and all patients above the age of 45.

In conclusion, we found consistent associations between short-term exposure to heat and increased risk of AD/ADRD ED visits across five US states. People with AD/ADRD may represent a vulnerable population affected by short-term exposure to high temperature. Our results support the development of more targeted strategies to reduce heat-related AD/ADRD morbidity.

• Alzheimer’s disease (AD) and related dementia (ADRD) impact millions of Americans

• High ambient temperature may be a risk factor for AD/ADRD morbidity

• We analyzed over 3.4 millions AD/ADRD emergency department (ED) visits in 5 states

• We found positive associations between ambient temperature and AD/ADRD ED visits

• Associations were stronger for AD/ADRD ED visits resulting in hospital admission