Anticipating older populations’ health risk exacerbated by compound disasters based on mortality caused by heart diseases and strokes

The health of older populations in the Southeastern U.S. receives threats from recurrent tropical cyclones and extreme heat, which may exacerbate the mortality caused by heart diseases and strokes. Such threats can escalate when these extremes form compound disasters, which may be more frequent under climate change. However, a paucity of empirical evidence exists concerning the health threats of compound disasters, and anticipations regarding the health risks of older populations under future compound disaster scenarios are lacking. Focusing on Florida, which has 67 counties and the second-largest proportion of older populations among U.S. states, we calibrate Poisson regression models to explore older populations’ mortality caused by heart diseases and strokes under single and compound disasters. The models are utilized to estimate the mortality across future disaster scenarios, the changing climate, and the growing population. We identify that under multiple hurricanes or heat, current-month hurricanes or heat can affect mortality more heavily than previous-month hurricanes or heat. Under future scenarios, co-occurring hurricanes and extreme heat can exacerbate the mortality more severely than other disaster scenarios. The same types of compound disasters can coincide with an average of 20.5% higher mortality under RCP8.5-SSP5 than under RCP4.5-SSP2. We assess older populations’ future health risks, alerting health agencies to enhance preparedness for future “worst-case” scenarios of compound disasters and proactively adapt to climate change.

• RQ1 Can compound events of hurricanes and extreme heat influence older peoples' mortality (number of deaths) caused by heart diseases and strokes more severely than sole-event extremes (compound events refer to the multiple hits of similar weather extremes and co-occurrence of hurricanes and extreme heat events)?• RQ2 How would the older population's risk of mortality caused by heart diseases and strokes be exacerbated under future scenarios considering the possible occurrence of compound disasters of hurricanes and extreme heat?
To answer the research questions, we focus on the state of Florida, U.S., which has the second-highest proportion of older population among all the U.S. states (21.17% in 2021).The state also expects recurrent hurricanes and extreme heat events frequently 9,38,39 .We answer RQ1 with a quantitative longitudinal study, in which we analyze and compare the impacts of single and compound disasters of hurricanes and extreme heat on older population.The analysis focuses on monthly mortalities caused by heart diseases and strokes in each county of Florida from January 2000 to December 2019 (240 months in total).Along with the variables of climate and weather extremes, we also include (i) the climate variables, (ii) the sociodemographic variables, and (iii) the environment variables in the Poisson regression models to improve the comprehensiveness of the modeling.For RQ2, we develop a set of future scenarios that incorporate population growth, climate change, and the occurrence of single or compound disasters in 2050, 2070, and 2100.These prediction periods are adopted by the climate change projections of the U.S. Environment Protection Agency's (EPA) Locating and Selecting Scenarios Online 40 .With the models calibrated in the longitudinal study (RQ1), we further estimate the monthly mortality caused by heart diseases and strokes among older population in each Florida county under each scenario (RQ2).The predictions can help understand the impacts of climate change and compound disasters on populations' health at the local and regional levels.The risk assessment results shed light on the planning for the worst-case scenarios under climate change.Such efforts can increase the robustness of adaptation strategies that mitigate the health risks of older adults and other vulnerable populations.

Impacts of compound disasters on older population's mortality caused by heart diseases and strokes over the past two decades
Following the research procedure shown in Fig. 1, we investigate the impacts of different occurrences of hurricanes and extreme heat (i.e., single and compound disasters) on older populations' mortality (number of deaths) caused by heart diseases and strokes ( the "mortality") based on Poisson regression analysis (coefficients of regressions under different occurrences of extremes are shown in Fig. 2) with a total of 1326 events of hurricanes and extreme heat and 426 compound disaster events (Fig. 3).Based on the historical records hurricanes and extreme heat events from National Center for Environmental Information and Energy Information Administration 41,42 , hurricanes generally occurred between July and November in the studied period, extreme heat events occurred frequently between May and October.For compound disasters, multi-hurricane events tended to occur between July and October, multi-heat events tended to occur between August and September, and hurricanes and extreme heat tended to co-occur between August and September in the years of the studied period.During the studied period, the overall mortality among older populations in Florida was 917,168, among which the mortality of the months with compound disasters was 85,502 18 .
Our Poisson regression models investigate the influence of (i) the occurrence of single or compound disasters, (ii) climate variables, (iii) population variables, and (iv) environmental variables on the mortality (variables are listed in Table 1).Especially, we examine the variables' influence under six event types: (i) no disaster, (ii) single hurricanes, (iii) single extreme heat events, (iv) multiple-hits hurricanes, (v) multiple-hits extreme heat, and (vi) co-occurrence of hurricanes and extreme heat events.The last three types are compound disasters.We regard one variable as more influential if its coefficient is higher in a model calibrated for one type of event than values in other calibrated models.
In general, the disaster-related variables show significant ( P < 0.001 ) and positive associations with the mortality.By comparing the coefficients (with confident intervals, CI) of disaster-related variables in different calibrated models, we identify that hurricane-induced electricity disturbance and extreme temperatures can influence the mortality with different intensities among single-disaster and multi-disaster events.As shown in Fig. 2a, the mortality is significantly influenced by the hurricane-induced electricity disturbance (coefficient: 1.045; 95% CI 1.030, 1.060, P < 0.03 ) under single-hurricane events.If hurricane-induced electricity disturbances occur in both the current month and the previous month (i.e., forming multi-hurricane events), the mortality in the current month is also associated with the disturbance in the previous month.Specifically, under multi-hurricane events, both the electricity disturbance induced by the current-month hurricane can be associated with the mortality of the current month with a higher coefficient (coefficient: 1.083; 95% CI 1.044, 1.125, P < 0.001 ) than that induced by a single hurricane event (coefficient: 1.045; 95% CI 1.030, 1.060, P < 0.03 ).The association between the electricity disturbance induced by the previous-month hurricane and the mortality (coefficient: 1.052; 95% CI 1.014, 1.090, P = 0.006 ) is also stronger than that induced by a single hurricane event, but such association is weaker than the electricity disturbance induced by the current-month hurricane.
The mortality is also significantly associated with the occurrence of single-heat events (coefficient: 0.924; 95% CI 0.908, 0.942, P < 0.001 ).If extreme heat also occurs in the previous month and forms multi-heat events, the extreme heat events in the current month can significantly influence the mortality in the current month.Specifically, under multiple-heat events, the mortality is related to the extreme heat in the current month with a coefficient of 1.012 (95% CI 0.989, 1.036, P < 0.001 ), which is slightly higher than the coefficient of extreme heat in sole-heat events (coefficient: 0.924; 95% CI 0.908, 0.942, P < 0.001 ).Comparatively, the influence of extreme heat in the previous month on the mortality of the current month is much lower than the extreme heat in the single-heat disasters, as the coefficient is only 0.737 (95% CI 0.713, 0.762, P < 0.001 ) and the influence is not statistically significant.When extreme heat and hurricane-induced electricity disturbance co-occur in the same month, the electricity disturbance can still influence the mortality in the current month significantly (coefficient value: 1.107, 95% CI 1.078, 1.137, P < 0.001 ), which is slightly higher than their influence under sole-hurricane events (coefficient: 1.045; 95% CI 1.030, 1.060, P < 0.03 ).Differently, the influence of extreme heat on the mortality is much lower than the impact of extreme heat under single-heat events.Specifically, when co-occurring with hurricane-induced electricity disturbance, the extreme heat can influence the mortality in the current month with a coefficient of 0.635 (95% CI 0.571, 0.705, P < 0.001 ), which is much lower than the coefficient under single-heat events (coef- ficient: 0.924; 95% CI 0.908, 0.942, P < 0.001).
Meanwhile, the associations between the mortality and other variables also demonstrate some notable trends.The proportion of the older population (Fig. 2b), the proportion of smokers among older populations (Fig. 2e), and the availability of acute care beds (Fig. 2i) tend to have notable correlations with the mortality under weather extremes when multiple-hurricane events occur.The proportion of the minority population among older populations (Fig. 2c), the proportion of alcohol consumers among older populations (Fig. 2f), the proportion of obese people among older populations (Fig. 2g), the proportion of people without health insurance among older populations (Fig. 2h), and the concentration of PM10 (Fig. 2m) tend to have notable correlations with the mortality when multi-heat events occur.The proportion of females (Fig. 2e) and the average air temperature (Fig. 2k) tend to associate with the mortality with higher impacts during compound disaster events.

Anticipating older populations' health risks under the developed scenarios based on the mortality caused by heart diseases and strokes
Based on the relationships between the occurrence of single and compound disasters and the mortality, we anticipate older populations' mortality caused by heart diseases and strokes (RQ2) under future scenarios in 2050, 2070, and 2100 (Figs. 4, 5).The future scenarios are developed considering (i) the occurrence of single and compound disasters, (ii) the changing climate, and (iii) the trend of population growth (illustrated in Fig. 7).The x-axis represents the 67 counties of Florida, and the order of the counties is based on the size of their older populations in 2019.The counties on the right side of the figures have a larger population of older populations than the counties on the left.The absolute value of the estimated monthly mortality of older populations  -g), the medical resource availability variables (h,i), the climate-related variables (j,k), and environmental variables (l,m).In each plot, the x-axis shows the estimation and 95% CI for the exponential value of the coefficient.The scatters in each plot indicate the exponential value of the coefficient of the variables under different types of events, and the scope of the dashes indicates the range of 95% CI for the exponential value of the coefficient.The y-axis indicates different types of events, from sole events to co-occurring events (from bottom to top).Especially in (a), each type of compound disaster is related to two variables, e.g., "co-occurred hurricane (H1 in co-occurred events)" and "co-occurred heat (H3 in co-occurred events)".We use the same shape of scatters for variables under the same type of events.Notably, because disasterrelated variables have no value in the "No disaster" scenario, we did not include "No disaster" in the y-axis of (a).www.nature.com/scientificreports/due to heart diseases and strokes in each county is shown with the bars.We also draw lines based on the logbase-ten value of the anticipated mortality to facilitate a comparison of the mortality between large-population and small-population counties.To facilitate robust planning that considers both the mid-of-way and worst-case scenarios, we compare the mortality of different scenarios under RCP4.5-SSP2 and RCP8.5-SSP5.Specifically, "RCPs" stand for "representative concentration pathways", indicating the greenhouse gas concentration levels that are measured by the radiative forcing levels in 2100 27 ."RCP4.5" represents that the greenhouse gas concentration will lead to the radiative forcing of 4.5 Watts per meter squared in 2100, indicating the middle-of-way emission level 27 ."RCP8.5" that the greenhouse gas concentration will lead to the radiative forcing of 8.5 Watts per meter squared in 2100, indicating the high emission level 28 ."RCPs" are the climate change trajectories and the basis of existing climate models when projecting the future range of climate variables (e.g., temperature and precipitation).Meanwhile, "SSPs" stand for "shared socioeconomic pathways", i.e., trajectories of socioeconomic developments that are characterized by population growth and economic development under future climate change scenarios 43 .There are five SSPs, i.e., SSP1 to SSP5 43 .Among the five SSPs, SSP2 represents middle-of-theroad development (e.g., middle-level population growth), while SSP5 represents fossil-fueled development (e.g., high-level population growth) 43 .The scenarios of RCP4.5-SSP2 can show the potential mortality under different weather extreme events, with sustainable pathways for population growth and greenhouse gas emissions.On the contrary, the scenarios of RCP8.5-SSP5 show the potential mortality under different weather extreme events, with rapid population growth and the high greenhouse gas emissions associated with aggressive fossil-fuel use.Under the scenarios with the same combination of RCPs and SSPs (i.e., within the same figure of Figs. 4, 5), our anticipated outcomes show that the co-occurrence of hurricanes and extreme heat is expected to be concurrent with markedly higher-level mortality of older populations from heart diseases and strokes compared to the other weather extreme events in most counties, especially the counties with a large population of older populations.For example, in Miami-Dade County (459,200 older adults in 2019), the co-occurrence of hurricanes and extreme heat under RCP4.5-SSP2can be accompanied by older populations' mortality caused by heart diseases and strokes that will reach 1800 for one month in 2100, which is significantly higher than the mortality under scenarios with other weather extreme events.Despite the co-occurrence of hurricanes and extreme heat, multihurricane and multi-heat events are expected to cause the mortality at the same level as the co-occurred weather extreme events, especially in medium-population counties.For example, in Indian River County (51,300 older adults in 2019), the multiple-heat events and the co-occurrence of heat and hurricanes are expected to bring the mortality to similar levels under RCP4.5-SSP2 in 2050, 2070, and 2100.Also, in Seminole County (73,300 older adults in 2019), multiple heat events are expected to cause monthly mortality of 150 in 2100 under RCP8.5-SSP5, which is much higher than the expected monthly mortality under the impact of the co-occurrence of extreme heat and hurricanes in 2100 under RCP8.5-SSP5(45 mortality).We also compare the effects of the same weather extreme events under different combinations of RCPs and SSPs (i.e., comparing across Figs. 4, 5).Our comparisons indicate that most events of weather extremes tend to be simultaneous with more severe mortality caused by heart diseases and strokes among older populations under RCP8.5-SSP5compared to RCP4.5-SSP2 (with averagely 20.5% of the increase), especially under multihurricane events and co-occurred hurricanes and heat events.Taking Miami-Dade County as an example, the mortality is expected to reach 365 under multi-hurricane events, RCP4.5-SSP2 in 2100.Comparatively, multi-hurricane events can potentially increase the monthly mortality to 893 (144.65%increase) in 2100 under RCP8.5-SSP5.Also, under RCP4.5-SSP2 in 2100, the mortality is expected to be around 1800 in Miami-Dade County if hurricanes and extreme heat events co-occur, while the mortality may increase to around 2200 under RCP8.5-SSP5 in 2100 (22.2% increase).Overall, with the tremendous increase of population and temperature rising in RCP8.5-SSP5, compound disasters (i.e., multi-disaster events or co-occurring disasters) can potentially intensify older populations' mortality caused by heart diseases and strokes to a level that is much higher than the expected mortality under middle-of-way scenarios.Counties from left to right are ordered based on the population size.Specifically, counties with less than 100,000 people are classified as "small-population counties", counties with 100,000 to 500,000 people are "mediumpopulation counties", and counties with more than 500,000 people are "large-population counties".The y-axis on the left is scaled for the value of projected mortality (represented by scatters and lines), helping to illustrate the difference between small-population counties and large-population counties.The y-axis on the right indicates the value of projected mortality without being scaled (represented by bars).

Discussion
Our longitudinal study and future-scenario anticipations indicate that older populations' risks of heart diseases and strokes can potentially be intensified under compound disasters of hurricanes and extreme heat events.We find that under multi-hurricane events in two continuous months, the older populations' mortality (number of deaths) within the same county caused by heart diseases and strokes can significantly increase when hurricanes occur in the second month.The influence of both two hurricane events is also more intense than the impact of sole-hurricane events.When extreme heat occurs in two consecutive months, the extreme heat of the second month can influence the monthly mortality of the second month with a higher effect than that of sole-heat events.Additionally, in the same county, under future scenarios with the same combination of RCPs and SSPs, the mortality is expected to be higher under the co-occurrence of hurricanes and extreme heat than the occurrence of other types of weather and climate extremes.This condition is particularly pronounced in counties hosting substantial aging populations.Most weather extreme events tend to be accompanied by a larger number of deaths caused by heart diseases and strokes among older populations under RCP8.5-SSP5compared to RCP4.5-SSP2, especially with multi-heat events and the co-occurrence of hurricanes and heat.Our findings can Figure 5. Projected monthly mortality caused by heart diseases and strokes in each county of Florida under future scenarios (RCP8.5,SSP5).The x-axis indicates the counties of Florida, which are divided into three groups to facilitate the comparison of predicted mortality levels with counties with similar levels of populations.Counties from left to right are ordered based on the population size.Specifically, counties with less than 100,000 people are classified as "small-population counties", counties with 100,000 to 500,000 people are "mediumpopulation counties", and counties with more than 500,000 people are "large-population counties".The y-axis on the left is scaled for the value of projected mortality (represented by scatters and lines), helping to illustrate the difference between small-population counties and large-population counties.The y-axis on the right indicates the value of projected mortality without being scaled (represented by bars).www.nature.com/scientificreports/provide compelling evidence regarding the association between compound disasters and the older population's mortality and inform future strategies for risk mitigation.Our study goes beyond previous studies in revealing the health impacts of compound disasters from multiple aspects.First, existing studies have mainly focused on the health impacts of weather extremes on the general population, lacking sufficient attention to older adults, who are among the most vulnerable to the negative effects of a rapidly changing climate 4,7,44 .To fill this gap, we focus on older populations and estimate the potentially high-level impacts of future weather extremes on their health.Our findings highlight the necessity of investigating the health risks of different groups of older populations to promote a comprehensive understanding of the impacts of climate change and weather extremes.
Second, the results of existing studies on compound disasters mainly involve limited cases, e.g., Hurricanes Rita and Katrina 33 and the co-occurrence of the COVID-19 pandemic and Hurricane Ida 25 .The findings from these limited cases may not provide a reference for understanding the impacts of other types of weather extremes, such as multi-heat events.Comparatively, the temporal and spatial scales of our longitudinal study are much broader, covering hurricanes and extreme heat events in 67 Florida counties over 20 years.With sufficient numbers of both single-disaster events and compound disasters, our longitudinal study can contribute to promoting the understanding of the associations between older populations' health risks and different weather extremes under more uncertain and diverse circumstances.Communities with similar context, e.g., other counties in the Southeastern U.S. confronting extreme heat and tropical cyclones, may use our findings as a reference for understanding the impacts of climate change locally.Our longitudinal findings of compound disaster events can also provide some insights regarding how the influence of multiple hazards can be altered when they are compounded in a narrow temporal and spatial frame.
Third, the existing literature has rarely anticipated the health consequences of compound disasters under future scenarios 45 .Findings solely based on historical cases without considering the uncertain trends of climate change and population growth 12 , such as Cherry et al. 33 and Fuhrmann et al. 3 , may not reflect the impact of compound disasters under future scenarios.Our scenario-based anticipation of older populations' mortality caused by heart diseases and strokes fills this gap and provides a more intuitive estimation of the plausible risk levels of specific counties under single or compound disasters.The anticipated outcomes can provide a basis for developing regional and community strategies for both the "business-as-usual" situations and the "worstcase" compound disasters.The cross-scenario assessment framework of vulnerable populations' health risks can guide local communities and researchers to specify the future scenarios they may confront and facilitate them to disentangle the uncertain impacts of future climate change and weather extremes on local populations and the built environment.
Our study has some limitations that present opportunities for future research.First, our longitudinal findings may not apply to scenarios where temperature and precipitation are outside the range covered by our historical data and the employed climate projections.For example, the monthly maximum temperature in our historical data and employed climate projections were below 98 Fahrenheit degrees, but climate change may result in more extreme temperatures 44 .The relationship between the temperature of such extreme values and the mortality may deviate from our longitudinal findings.Future studies could employ datasets with broader value ranges of temperature and precipitation to capture more accurate relationships between these factors and the mortality.Second, we measured the health risks and sociodemographic characteristics of older populations at the population level, while the individual-level characteristics of older adults were not studied.Future studies can leverage individual-level information and characteristics, such as medical insurance records and social media data 46,47 , to reveal the compounding influence of social determinants of health on older adults' health risks under climate change.However, our projections of the mortality at the population level can still provide a macro assessment of the health impacts caused by large-scale hazard events.Third, we developed future scenarios based on current projections for climate change and population growth from EPA 40 and Hauer 48 , while the projected population may not be reached under the increase of projected temperature.Older populations' relocation behaviors motivated by the increasing temperature and their changing capacity to adapt to high-temperature weather may result in different population pathway 49,50 .Future studies may adopt more up-to-date projections of population growth and climate change for estimating the mortality of older population.Fourth, additional environmental factors, such as air pressure and sea surface temperature, may also influence older adults' health risks under future climate change 28 , but were not included in our assessments.Specifically, we considered the climate variables (i.e., temperature and precipitation) as suggested by existing climate projection models, such as the Geophysical Fluid Dynamic Laboratory (GFDL) and the Hadley Centre Global Environment Model (HadGEM2) 51,52 .Also, the regional-level air quality change we adopted in the future scenarios may not accurately reflect air quality at sub-regional scales, such as the county level 53,54 .Future studies could utilize local data collection and place-based projections to address the needs of subregional risk assessment.Additionally, this study focuses on compound disasters of hurricanes and/or extreme heat in Florida.Future studies can extend the investigation to other combinations of disaster types in other geographic regions, such as wildfire events and droughts in California.
Hurricanes and extreme heat events are expected to exacerbate older populations' mortality caused by heart diseases and strokes severely if they occur as compound disasters.Our study reveals the potentially exacerbated risks of heart diseases and strokes among the older populations under the future scenarios of climate change and compound disasters.The results highlight the interactions between different weather extremes that form compound disasters, which can amplify the influence of specific weather extremes (e.g., hurricanes) on the older populations' health risks significantly.Our findings provide a reference for developing robust risk mitigation strategies that are targeted to the health threats of different occurrences of weather extremes.Specifically, based on our scenario-based anticipations, local communities can develop risk mitigation strategies that are tailored for both single and compound disaster scenarios, avoiding the potential overlapping of disaster preparedness, response, and recovery.Our findings also suggest strengthening the collaboration among the facilities of public health, lifeline services, and disaster management, designing cross-departmental risk mitigation strategies for maintaining vulnerable populations' access to medical services during compound disasters.By comprehensively capturing the uncertain threats of compound disasters to the human population, adaptation strategies can mitigate the health risks and improve the well-being of older populations and other vulnerable populations effectively.

A longitudinal study of compound disasters' impacts on older populations' mortality caused by heart diseases and strokes
Florida has the second-highest percentage of senior citizens among U.S. states (21.17% in 2021) and expects recurrent hurricanes and extreme heat events frequently 9,38,39 .Older populations' mortality (number of deaths) caused by heart diseases and strokes in Florida is high, reaching 50,616 deaths in 2019 18 .The longitudinal study was conducted with the monthly data of 67 Florida counties from January 2000 to December 2019, as 2000 was the first year when the monthly data of the mortality caused by heart diseases and strokes in each county of Florida became available.The data since 2020 when the COVID-19 pandemic started was not considered, as the pandemic can disturb the trend of older populations' mortality in counties of Florida.The longitudinal data on older populations' monthly mortalities caused by heart diseases and strokes in Florida was obtained from the open-access datasets collected by the CDC WONDER.Our study did not contain any experiments with human participants or animals and followed the data usage policies of FDOH.An "event" represents the occurrence of weather extremes in one county within one month.Each event includes the records of the mortality, climate characteristics (temperature and precipitation), air quality, population characteristics, and the type and intensity of weather extreme events.As the events have a short time step, i.e., one month, the longitudinal study based on the events can capture the immediate impacts of extreme heat events and hurricanes on older populations' mortality caused by heart diseases and stroke 55, 56 .
The longitudinal study covered six types of weather extreme occurrences and compared their impacts on older populations' mortality caused by heart diseases and strokes, including (i) no disaster, (ii) single hurricanes, (iii) single extreme heat events, (iv) multiple-hits hurricanes, (v) multiple-hits extreme heat, and (vi) co-occurrence of hurricanes and extreme heat events.The last three types are compound disasters.The historical records of hurricanes and extreme heat events were collected from the U.S. National Center for Environmental Information (including the records of extreme heat) and the U.S. Energy Information Administration (including the records of hurricane-induced power outages) 41,42 ."Multiple hits" disasters are defined as several weather extremes of the same type affecting the same county in two continuous months.The "co-occurrence" refers to the events in which hurricanes and extreme heat affect the same county within the same month.The events were extracted from the monthly data of 67 Florida counties and categorized into six groups based on the six types of weather extreme occurrences.For each group of events, we quantified the impacts of each type of weather extreme event on older populations' mortality caused by heart diseases and strokes respectively.Especially, the severity of hurricanes and extreme heat events was represented by the occurrence of hurricane-induced electricity disturbance and monthly maximum temperature.The occurrence of hurricane-induced electricity supply disturbance was set with a binary value, i.e., 0 for "not occurring" and 1 for "occurring".To represent the severity of extreme heat, we used 0 to represent the situation in which no extreme heat events occur, and we represented the severity with the absolute value of the monthly maximum temperature if extreme heat events occur.
The longitudinal study is based on multivariate Poisson regression (Eq.1).β 0 refers to the constant, and β m,n (e.g., β 1,i ) refers to the coefficient of each independent variable in the Poisson regression.Poisson regression can explain the statistical relationships between multiple factors and the count of small-probability events occurring (e.g., disease-induced mortality) 57 .Poisson regression has been utilized to capture the association between the mortality of heat-related illnesses and specific determinants 45,58 .The dependent variable is the county-level monthly mortality c heart diseases and strokes among older populations ( HD ), i.e., the older adults who died because of heart diseases or strokes in a specific month within a specific county.The independent variables in our Poisson regression models are described in Table 1.Each of these variables shows a relatively independent nature from the others while exhibiting discernible correlations with the dependent variable, as depicted in Supplementary Figs.S1 and S2.Specifically, the impact of weather extreme events is represented by the disaster-related variables (i.e., H n ), covering the occurrence of hurricane-induced electricity supply disturbance and the monthly maximum temperature.Both variables can represent the intensity of hurricanes and extreme heat on influencing the risks of heart disease and stroke among older populations 59,60 .We did not include variables representing the duration of weather extreme events, which may not reflect the duration that older adults were exposed to the weather extremes in the studied period 27 .The explanatory variables also include the climate variables ( C i , including the average temperature and cumulative precipitation), air pollution ( E m , including the concentrations of PM2.5 and PM10), population factors ( P j ), and health resource availability ( M k ) as the explanatory variables.Specifically, the population factors include the proportion of older populations, the proportion of minority populations among older populations, the proportion of females among older populations, the proportion of smokers among older populations, the proportion of alcohol consumers among older populations, and the proportion of obesity population among older populations 61,62 .We selected these variables because according to the "Heart Disease and Stroke Statistics-2022 Update" from the American Heart Association, the risks of heart disease and stroke were significantly different among populations of different genders, minority groups, and obesity 63   www.nature.com/scientificreports/Also, the use of alcohol was the leading behavioral factor of mortality caused by heart disease and stroke 63 .We also considered the proportion of people with health insurance among older populations and the count of acute care beds among 1,000 people to represent the accessibility of health resources among older populations.For each type of weather extreme occurrence, we trained specific Poisson regression models based on the historical records of that weather extreme occurrence and the value of independent variables in related months (the historical trend of each independent variable is shown in Supplementary Fig. S3).

Developing disaster scenarios for anticipating future older population's mortality caused by heart diseases and stroke
The calibrated models of longitudinal studies are the basis for anticipating older populations' mortality caused by heart diseases and strokes at the county level under future scenarios over 2050, 2070, and 2100.We adopted the prediction periods of the U.S. Environment Protection Agency's (EPA) Locating and Selecting Scenarios Online 40 , i.e., 2050, 2070, and 2100.The future scenarios incorporate the dimensions of climate change, population growth, and the occurrence of weather extremes (Fig. 6).We considered different combinations of the three dimensions in the developed scenarios, aiming to capture the possible range of older populations' mortality caused by heart diseases and strokes in the uncertain future.
(1) log Similar to the longitudinal study, future scenarios cover six event types (the values of disaster-related variables are shown in Table 2).The severity of future climate and weather extremes is represented by the occurrence of hurricane-induced electricity supply disturbance and maximum temperature.The climate change in our scenarios is represented by the change in local average temperature, precipitation, and air pollution concentrations under RCPs 40,53,54 .Specifically, the temperature and precipitation change are estimated based on the projections from the Geophysical Fluid Dynamic Laboratory (GFDL) under CMIP5-based RCP4.5 and RCP8.5 (shown in Fig. 7a) 47 .The resolution of the projections is 1/16°, i.e., around 6.9-km resolution 40 .We calculated the average values of temperature change and precipitation change among the land cells within each county.For the air pollutant concentration, there is no widely accepted projection about the fine-scale air quality change under RCPs in Florida from 2050 to 2100.Based on the projected trends of air pollution concentrations under climate change 53,54 , we set the volume of air pollutant concentration under RCP4.5 as zero (i.e., air pollutants can be ignored) to represent the low-emission scenarios under the future emission controls.Under RCP8.5, we set the volume of air pollutant concentration as the same as in 2019, representing the business-as-usual scenarios.For the population growth, we adopted Hauer's projections 48 of the county-level population of older adults (Fig. 7b), minority, and female older populations.We can identify an increase trend of older populations under both the scenarios of SSP2 and SSP5 in 2050, 2070, and 2100.We adopted the basic assumptions of population growth under SSP2 and SSP5 in Hauer's projections 48 .Specifically, for high-income OECD countries (e.g., the U.S.), the population growth under SSP2 could follow the medium levels of fertility, mortality, and migration in historical records.In contrast, the population growth under SSP5 could follow the high levels of fertility and migration, while the mortality levels would be low levels based on historical records.Particularly, the increasing trend of older populations under SSP5 would be more tremendous than the increasing trend under SSP2.For example, the older population in Miami-Dade County is projected to increase by 1.996 million under SSP2 in 2100, while the projection is 2.781 million under SSP5 in 2100.The increasing trend of the older population highlights the necessity of investigating their health risks under future scenarios.Additionally, we regarded the proportion of smokers, alcohol consumers, population with health insurance, the obese population, and the count of acute care beds among 1000 population in the future scenarios as the same as the value in 2019.All the assumptions we adopted when applying regression outcomes to future scenarios are listed in Supplementary Table S1.
With the three dimensions in Fig. 6, we developed a set of future scenarios (shown in Fig. 8).Each scenario is assigned a unique code S i,j ."i" of S i,j represents the combination of SSPs and RCPs (e.g., S 1,j represents scenarios under RCP4.5-SSP2), and "j" of S i,j represents the type of weather extreme occurrences (e.g., S i,6 represents scenarios of co-occurred hurricanes and extreme heat).Overall, the scenarios are under the four combinations of SSPs and RCPs: SSP2-RCP4.5,SSP2-RCP8.5,SSP5-RCP4.5, and SSP5-RCP8.5.We assigned one of the six event types for each scenario, represented by "j" of S i,j .Using the calibrated Poisson regression models in the previous section, we estimated the monthly mortality caused by heart diseases and strokes among older populations in each Florida county under each of the developed scenarios.We compared the anticipated mortality levels across scenarios to investigate if higher-level mortality caused by heart diseases and strokes among older adults occurs in compound disasters than in single extreme events.

Figure 1 .
Figure 1.Schematic diagram of the research framework.

Figure 2 .
Figure 2. Poisson regression outcomes.The range of coefficient values for disaster-related variables (a), the population-related variables (b-g), the medical resource availability variables (h,i), the climate-related variables (j,k), and environmental variables (l,m).In each plot, the x-axis shows the estimation and 95% CI for the exponential value of the coefficient.The scatters in each plot indicate the exponential value of the coefficient of the variables under different types of events, and the scope of the dashes indicates the range of 95% CI for the exponential value of the coefficient.The y-axis indicates different types of events, from sole events to co-occurring events (from bottom to top).Especially in (a), each type of compound disaster is related to two variables, e.g., "co-occurred hurricane (H1 in co-occurred events)" and "co-occurred heat (H3 in co-occurred events)".We use the same shape of scatters for variables under the same type of events.Notably, because disasterrelated variables have no value in the "No disaster" scenario, we did not include "No disaster" in the y-axis of (a). https://doi.org/10.1038/s41598-023-43717-3

Figure 3 .
Figure 3. Historical trend (a) and monthly frequency (b) of hurricane and extreme heat events in Florida from January 2000 to December 2019.Plot (a) is a scatter plot, and each scatter represents the count of hurricane or extreme heat events in Florida in a specific month.Plot (b) is a histogram, and each bar represents the frequency of hurricane or extreme heat events that happened in a specific month.The compound disasters generally happened between August and October, which is highlighted with dashed lines in the histogram (b).Data source: EPA 43 and NHC 44 .

Figure 4 .
Figure 4. Projected monthly mortality caused by heart diseases and strokes in each county of Florida under future scenarios (RCP4.5,SSP2).The x-axis indicates the counties of Florida, which are divided into three groups to facilitate the comparison of predicted mortality levels with counties with similar levels of populations.Counties from left to right are ordered based on the population size.Specifically, counties with less than 100,000 people are classified as "small-population counties", counties with 100,000 to 500,000 people are "mediumpopulation counties", and counties with more than 500,000 people are "large-population counties".The y-axis on the left is scaled for the value of projected mortality (represented by scatters and lines), helping to illustrate the difference between small-population counties and large-population counties.The y-axis on the right indicates the value of projected mortality without being scaled (represented by bars). https://doi.org/10.1038/s41598-023-43717-3 . https://doi.org/10.1038/s41598-023-43717-3

Figure 6 .
Figure 6.Three types of compound disasters (a-c) and scenario development across climate change, population growth, and the occurrence of weather extremes.Plots (a,b,d) are illustrations of the three types of compound disasters: multi-hit of extreme heat, multi-hit hurricanes, and co-occurring hurricanes and extreme heat.Plot (d) shows the three dimensions of future scenarios.The grey squares in the bottom surface indicate the four types of climate change pathways in the United States, including "SSP2-RCP4.5", "SSP5-RCP4.5","SSP2-RCP8.5", and "SSP5-RCP8.5".Plot (d) also includes an example scenario, which is linked to the climate change pathways and occurrences of weather extremes with dash lines.

Figure 7 .
Figure 7. Projection data of (a) temperature and (b) older population growth.The darker colors represent the higher values of temperature change and population growth.This figure is produced using ArcGIS Pro Version 2.5, provided by ESRI (https:// www.esri.com/ en-us/ arcgis/ produ cts/ arcgis-pro/ trial).Data source: EPA 40 and Hauer 41 .

Figure 8 .
Figure 8. Developed scenarios: S i,j .i represents the combinations of RCPs and SSPs, j represents the types of weather extreme occurrences.

Table 1 .
Description of factors in the developed scenarios.

Table 2 .
Values of disaster-related factors.