**Increased Medical Visits and Mortality among Adults with Cardiovascular Diseases in Severely A**ff**ected Areas after Typhoon Morakot**

**Hsin-I Shih 1,2,3 , Tzu-Yuan Chao <sup>4</sup> , Yi-Ting Huang <sup>2</sup> , Yi-Fang Tu 2,5 , Tzu-Ching Sung <sup>6</sup> , Jung-Der Wang <sup>3</sup> and Chia-Ming Chang 2,7,\***


Received: 19 June 2020; Accepted: 2 September 2020; Published: 8 September 2020

**Abstract:** Natural disasters have negative health impacts on chronic diseases in affected populations. Severely affected areas are usually rural areas with limited basic infrastructure and a population have that has limited access to optimal healthcare after a disaster. Patients with cardiovascular diseases are required to maintain quality care, especially after disasters. A population-based case-control study enrolled adults from the National Health Insurance Registry who had ischemic heart disease and cerebrovascular disease histories and lived in the area affected by Typhoon Morakot in 2009. Monthly medical visits for acute cerebrovascular and ischemic heart diseases markedly increased at approximately 1–2 months after the typhoon. Survival analysis during the two years following the typhoon indicated a significant increase in mortality in adults with an acute ischemic heart disease history who lived in the severely affected area. Mortality hazard analysis showed that among affected adults with previous cerebrovascular diseases and acute ischemic heart diseases, patients with diabetes (adjusted hazard ratio [HR]: 1.3–1.7), Chronic Kidney Disease (CKD) (adjusted HR: 2.0–2.7), chronic obstructive pulmonary diseases (COPD) and asthma (adjusted HR: 1.7–2.1), liver cirrhosis (adjusted HR: 2.3–3.3) and neoplasms (adjusted HR: 1.1–2.1) had significantly increased mortality rates. Consequently, high-quality and accessible primary healthcare plans should be made available to maintain and support affected populations after disasters.

**Keywords:** disaster; typhoon; flood; elderly; cardiovascular diseases; cerebrovascular diseases

#### **1. Introduction**

The number of reported weather-related natural disasters has been increasing since the 1960s. These disasters result in over 60,000 deaths each year worldwide [1,2]. Increasingly variable rainfall

patterns have increased the frequency and intensity of flooding, which has had great impacts on human health. People may be forced to move, which heightens the risk of health effects on mental disorders, communicable diseases, and chronic comorbidities. Heavy floods cause substantial infrastructure damage and lead to massive economic and personal losses for communities [3,4]. These disasters require large-scale, multinational coordination to provide urgent humanitarian aid and continuous relief [5] to the affected areas.

Severely affected areas are usually rural and coastal areas with limited basic infrastructure and a small population. Rural and coastal areas are more likely than urban areas to have an inequitably high burden due to healthcare disparities. These disparities contribute to the inadequate provision of basic healthcare services that arise from few medical facilities, a minimal number of providers, few specialty practices, and a lack of accompanying technical innovations and health promotion programs [6–8]. In medically underserved rural settings, people experiencing barriers to primary healthcare tend to have a low income, low education level and high rate of unemployment [9]. Patients with chronic comorbidities need to maintain a delicate balance of care to decrease complications during and after disasters. A previous study found that 24.3% of visits to emergency treatment facilities after Hurricane Katrina were for chronic diseases or related conditions [10]. Following the March 2011 earthquake and tsunami in Japan, increased mortality was observed in some patients with chronic comorbidities [11]. Increased numbers of visits for acute ischemic heart disease and stroke following natural disasters have been identified after natural disasters. Survivors also have to cope with stressors such as searching for food and shelter, relocating, crowding, financial hardship, and navigating social services [12]. Stressors from hurricanes and other natural disasters can cause chronic and acute mental stress, which can trigger cardiovascular events. Affected communities, especially those who are severely affected and in a rural area, are faced with a myriad of disparities, each posing a barrier to timely response and complete recovery from a disaster.

The Asia-Pacific region has been recognized as the region with the highest risk of major disasters [13]. Moreover, Taiwan may be the location most vulnerable to natural hazards, with 73% of its land and population exposed to three or more hazards [14]. Typhoon Morakot affected the West Pacific Region from August 6 to 11 August 2009. On August 7, it caused copious amounts of rainfall in Taiwan, peaking at 2777 mm (109.3 in). This extreme amount of rain triggered massive mudflows and severe flooding throughout southern Taiwan, causing 702 deaths and roughly US\$6.76 billion in financial losses [15,16]. Almost all southern Taiwan experienced record-breaking heavy rain, and the associated flood damage required people to relocate to temporary shelters for approximately six months to one year. After Typhoon Morakot, multidisciplinary, large-scale reconstruction and relief actions were initiated. Residents who lived in the affected areas before the disaster were identified by the Ministry of Health and Welfare and received partial medical expense reimbursement for 3 months. Medical teams and clinics were established in the temporary shelters to provide primary basic healthcare in the severely affected areas for at least 6 months. A series of recovery and relief measures were implemented and initiated in the following years. To evaluate the parameters affecting health outcomes of cardiovascular and cerebrovascular diseases among different levels of affected areas (Appendix A Table A1), we conducted a longitudinal follow-up study to assess health parameters among adult residents in different areas affected by Typhoon Morakot, a major disaster.

#### **2. Materials and Methods**

#### *2.1. Data Source*

Data from the Taiwan National Health Insurance (NHI) Database from the National Health Research Institute (NHRI) was analyzed. All claims submitted by physicians to the National Health Insurance Registry must include a diagnostic code based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) developed by the World Health Organization which allows the NHI Registry to verify claims and generate statistics about causes of illness and death [17]. The NHRI recorded data included residents' demographic data, medications, treatments (including operations), and disease diagnoses at four different levels of healthcare facilities. After Typhoon Morakot, the NHI Bureau identified residents who had lived in the affected areas before the disaster and provided partial reimbursement of their medical expenses based on the Regulation "National Health Insurance Reimbursement for Medical Expenses after Typhoon Morakot". Adult residents living in the affected area were enrolled for further analysis.

#### *2.2. Study Design*

A population-based longitudinal case-control study was conducted. In total, historical data of 715,244 adults who had lived in the affected areas were obtained for this study. Adult patients with healthcare facility visit records were analyzed. The index date was 8 August 2009, the date that Typhoon Morakot struck Taiwan. The adults were separated into non-elderly and elderly groups (18–64 years and 65 years and older, respectively). Patients with chronic medical conditions requiring long-term follow-up, such as diabetes mellitus (DM), hypertension, chronic kidney disease (CKD), heart disease, hyperlipidemia, chronic obstructive pulmonary disease (COPD), liver cirrhosis and neoplasms, were identified from long-term medical records, and records of their most recent follow-up visit for their chronic medical conditions were reviewed to assess their comorbidities after the disaster. The observation period of the study was from January 2008 to December 2011.

Patients with underlying medical comorbidities, including hypertension (ICD-9-CM: 401–405), DM (ICD-9-CM: 250), asthma (ICD-9-CM: 493), chronic heart failure (ICD-9-CM: 428, 410–414), COPD (ICD-9-CM: 491–496, excluding 493, 495), liver cirrhosis (ICD-9-CM: 571.5), neoplasms (ICD-9-CM: 140–239) and CKD (ICD-9-CM: 585, 582, 583.9), that were recorded before the index date were selected and noted. All medical service utilization records after the index date were reviewed. The calculated Charlson Comorbidity Index (CCI) values indicated the severity of these patients' comorbidities. The patients' socioeconomic statuses were based on their income records, as reported to the NHI bureau.

The rates of acute ischemic heart diseases and cerebrovascular diseases in adults who lived in the affected areas were analyzed. The study analyzed patients with acute ischemic heart disease (ICD-9-CM: 410–414), intracranial hemorrhage (ICH) (ICD-9-CM: 430–432) and ischemic stroke (ICD-9-CM: 433–437) before and after Typhoon Morakot. The above diseases that occurred after the index date were evaluated to determine patients' health outcomes before and after Typhoon Morakot. The monthly visits of acute illnesses were measured from the patients' medical records. For each person, the occurrence of the same event within the same one-month period was counted only once. Hospitalization records after Typhoon Morakot were also analyzed to evaluate the health hazards, heart failure (ICD-9-CM: 428); common infections such as lower respiratory tract infection (ICD-9-CM: 480–488), urinary tract infection (ICD-9-CM: 590, 595, 597, 599), and skin and soft tissue infections (ICD-9-CM: 680–686); and trauma and injury (ICD-9-CM: 850–959).

The validated definition of mortality was adapted from the NHI database and was based on the insurance status among those affected adults enrolled in our study. Mortality cases were defined as enrolled adults withdrawn due to death or critical discharge with the diagnosis of one of the most common twenty causes of death (ICD-9-CM) of National Statistics in Taiwan, and without any medical records after withdrawal date, missing for more than six months, and were disqualified as an insurance applicant of the NHI program such as immigration and the expiration of the duration of stay of aliens [18]. All causes of death for the affected adults enrolled in this study between August 2009 and December 2011 were enrolled for further survival and hazard ratio analysis.

#### *2.3. Spatial-Temporal Analysis*

The classification of township and metropolitan areas in Taiwan was adapted from the 2005 Taiwan Social Change Survey, which tracks and provides insights into long-term societal trends and developments in each town or metropolitan area in six prospective areas through nationally representative survey data obtained through cluster sampling; data obtained included the proportion of the working population employed in the service sector, the proportion of the industrial population, the proportion of the population 15–64 years old, the proportion of the elderly population, the proportion of the population that attained at least a bachelor's degree, and the population density of the area. In total, three different town or metropolitan areas (urban, suburban, and rural) were included [19].

#### *2.4. Covariates*

In addition to age and sex, area of residence was incorporated into the study design as a demographic variable. Comorbidities considered in this study included hypertension, DM, hypertension, COPD and asthma, liver cirrhosis, and neoplasms, which were defined by major diagnosis codes with at least three months' long-term prescriptions before the index date with relevant ICD-9-CM codes mentioned above.

#### *2.5. Statistical Analysis*

The distribution of the study population, which was based on demographic and disease history data such as sex, age, diabetes, chronic kidney, pulmonary, and liver diseases, living place, social economic status, cardiovascular, cerebrovascular diseases, and mortality was analyzed. Besides A case control study to evaluate time-to-event outcomes was applied. To evaluate differences among different study groups, the chi-square test was used for categorical variables, and the Student's *t* test was used for continuous variables. The control group was matched for age and sex using propensity scoring methods. Major underlying diseases and living locations were selected from the univariate and multi-variate Cox proportional hazard regression models with backward eliminations to estimate hazard ratios of mortality among affected adults with a history of cardiovascular and cerebrovascular events. All tests of significance were 2-tailed, and a *p* value of 0.05 or less was considered statistically significant. Robust Cox models using robust sandwich variance estimators were applied considering clustering within matched sets. Moreover, stratified Cox models using stratification of the propensity scores were also adopted to treat the matched sets. The former approach resulted in an unbiased estimation of marginal hazard ratios that were compared with a biased estimation of marginal hazard ratios resulting from the latter approach. Each comparison value and its 95% confidence interval (95% CI) were also analyzed. All data management and statistical analyses were performed with SAS 9.4 software (SAS Institute, Cary, NC, USA). All statistical tests were 2-sided, and *p* values less than 0.05 were considered statistically significant.

#### *2.6. Ethical Issues*

All provisions of the study were performed in accordance with the principles of the Declaration of Helsinki and the Declaration of Taipei [20]. Patients' personal information was encrypted to protect their privacy, and the electronic databases were decoded for research; therefore, the requirement for informed consent was waived by the institutional review board (IRB). This study was approved by the IRB of the study hospital (IRB No: A-ER-103-176).

#### **3. Results**

A total of 715,244 adult patient files were identified; 199,991 patients (28%) were in the elderly group. Female patients were predominant (398,819, 56%). The study diagram is shown in the Appendix A Figure A1. The demographic characteristics of the study population between different severity of affected areas are presented in Table 1. Compared to the adults living in the moderately affected area, those in the severely affected area before Typhoon Morakot were older, lived more in rural areas, had a higher rate of multiple underlying diseases (CCI ≥ 1: 45% vs. 41%) and cardiovascular disease history (18% vs. 15%). Besides, the adults living in the severely affected area also had higher rates of underlying chronic diseases, such as DM (16% vs. 14%), hypertension (33% vs. 26%), heart disease (12% vs. 10%), COPD and asthma (6% vs. 4%).


**Table 1.** Demographic and clinical characteristics of adults living between different severity of affected areas before Typhoon Morakot.

\* Standard deviation; † Charlson Comorbidity Index. DM: Diabetes mellitus; CKD: Chronic kidney disease; COPD: Chronic obstructive pulmonary diseases.

Figure 1 shows the number of monthly visits for acute ischemic heart disease and cerebrovascular events by affected adults before and after Typhoon Morakot, i.e., from 2008 to 2010. There was a markedly increased peak in visits in the same month Typhoon Morakot occurred in both the severely and moderately affected areas and in both elderly and non-elderly adults. Compared to that for acute ischemic heart disease, the difference in the number of monthly visits for acute cerebrovascular diseases was more prominent in the moderately affected areas, particularly in the elderly group.

The demographic data before and after matching are summarized in Table 2. After matching for age and sex, the proportions of adults living in the severely affected area with a history of cerebrovascular diseases (44% vs. 42%, *p* < 0.001), hypertension (62% vs. 58%, *p* < 0.0001), hypertension (62% vs. 58%, *p* < 0.0001), COPD and asthma (12% vs. 11%, *p* = 0.0001) was higher than those in the moderately affected area. Adults living in the severely affected area had a higher proportion of living in rural area (71% vs. 32%, *p* < 0.0001) and low-income status than those living in the moderately affected area (monthly income < 750 USD: 73% vs. 60%, *p* < 0.0001). However, there was no significant difference in the CCI score between adults living in the moderately and severely affected areas.

**Figure 1.** Monthly visits for acute cardiovascular and cerebrovascular events by affected adults before and after Typhoon Morakot. (**A**): Ischemic Heart Diseases; (**B**): Acute Cerebrovascular Diseases. Vertical solid lines: Index Month (August/2009); vertical dotted lines: August/2008, August/2010 and August/2011.



\* Standard deviation; † Charlson Comorbidity Index. DM: Diabetes mellitus; CKD: Chronic kidney disease; COPD: Chronic obstructive pulmonary diseases.

A mortality analysis of adults with cardiovascular and cerebrovascular disease histories was performed. Robust Cox models using robust sandwich variance estimators for clustering within matched sets were used to analyze survival after Typhoon Morakot in affected adults with cardiovascular and cerebrovascular disease histories in the moderately and severely affected areas. Before Typhoon Morakot, the robust sandwich estimation for three-year survival between moderately and severely affected areas revealed a significant difference in the survival rate of affected adults with a cardiovascular events history (*p* = 0.0047) but did not reveal a significant difference in the survival rate of affected adults with a cerebrovascular event history.

Hospitalizations after Typhoon Morakot in affected adults with cardiovascular and cerebrovascular histories before Typhoon Morakot were analyzed (Table 3). Compared to the hospitalization rate in affected adults in the moderately affected area, a slightly increased hospitalization rate was noted among affected adults living in the severely affected area after Typhoon Morakot in terms of acute ischemic heart diseases, acute cerebrovascular diseases, heart failure, infection-related diseases, traumaand injury-related diseases (*p* < 0.0001).


**Table 3.** Hospitalizations after Typhoon Morakot in affected adults with preexisting histories of cardiovascular and cerebrovascular diseases before Typhoon Morakot.

\* *p* < 0.05.

To determine the important risk factors for mortality among affected adults with cardiovascular and cerebrovascular disease histories and to eliminate confounding factors, a multivariate stratified Cox proportional hazards regression model was estimated (Table 4). Among affected adults with a cerebrovascular disease history, patients with diabetes (adjusted HR: 1.40, 95% CI: 1.29–1.52, *p* < 0.0001), CKD (adjusted HR: 2.05, 95% CI: 1.81–2.33), COPD and asthma (adjusted HR: 1.89, 95% CI: 1.72–2.08, *p* < 0.0001), liver cirrhosis (adjusted HR: 2.25 95% CI: 1.79–2.79, *p* < 0.0001) and neoplasms (adjusted HR: 1.51, 95% CI: 1.33–1.71, *p* < 0.0001) had significantly increased mortality rates. Additionally, among affected adults with acute ischemic heart disease histories, patients with diabetes (adjusted HR: 1.55, 95% CI: 1.43–1.68, *p* < 0.0001), CKD (adjusted HR: 2.40, 95% CI: 2.15–2.69), COPD and asthma (adjusted HR: 1.69, 95% CI: 1.54–1.86, *p* < 0.0001), liver cirrhosis (adjusted HR: 2.76, 95% CI: 2.29–3.32, *p* < 0.0001) and neoplasms (adjusted HR: 1.87, 95% CI: 1.67–2.09, *p* < 0.0001) had higher mortality.

**Table 4.** Hazard ratio of all-cause mortality in affected adults with previous histories of cerebrovascular and ischemic heart diseases before Typhoon Morakot.


† Enrolled variables in Cox model for cerebrovascular disease history and cerebrovascular disease history: Location, DM, Hypertension, CKD, COPD and asthma, liver cirrhosis, Neoplasm. DM: Diabetes mellitus; CKD: Chronic kidney disease, COPD: Chronic obstructive pulmonary diseases. \* *p* < 0.05.

#### **4. Discussion**

This study suggested that affected adults with acute ischemic heart diseases and acute cerebrovascular disease histories before Typhoon Morakot had an increased numbers of visits healthcare facilities for acute ischemic heart disease and acute cerebrovascular diseases two months after Typhoon Morakot in both the moderately and severely affected areas. The survival analysis suggested that affected adults with acute ischemic heart disease history before Typhoon Morakot living in the severely affected area had significantly higher mortality than those living in the moderately affected area. Compared with those of affected adults in the moderately affected area, slightly more hospitalizations were noted among affected adults living in the severely affected area after Typhoon Morakot. Increased mortality rates were observed in affected adults with cerebrovascular disease and ischemic heart disease histories who had comorbidities such as diabetes, CKD, COPD and asthma, liver cirrhosis, and neoplasms.

Increases in the numbers of hospital visits for acute ischemic heart disease and stroke following natural disasters have been reported. A medical records analysis after the Great Hanshin-Awaji Earthquake in 1995 found that acute coronary syndrome and stroke rates rapidly increased and then decreased within 7 weeks, similar to rates after the Great East Japan Earthquake in 2011 [21,22]. After Hurricane Sandy in 2012, researchers detected increases in hospitalizations and deaths due to myocardial infarction and stroke during the 2 weeks following the hurricane compared to the same 2 weeks from 5 years previously [23]. In Taiwan, the rate of hospitalization due to acute myocardial infarction increased during the 6 weeks after the Ji-Ji Earthquake in 1999, and a significantly higher number of patients were hospitalized with acute myocardial infarction during that period than during the same 6-week period in the previous year [24]. Our study revealed similar results. Increased numbers of visits for acute cardiovascular and cerebrovascular events were observed in the first two months after Typhoon Morakot. The survival analysis also indicated a significant increase in mortality in the two years following the typhoon in those with previous acute ischemic heart disease who had lived in the severely affected area. Typhoon Morakot has had significantly short- and long-term health impacts on affected adults with acute ischemic heart diseases and cerebrovascular disease histories, especially those with chronic comorbidities such as DM, CKD or end-stage renal disease (ESRD), COPD, liver cirrhosis and neoplasms.

The abrupt increase in cardiovascular diseases (CVDs) in the acute phase was likely due to psychosocial and posttraumatic stress caused by the disaster and inadequate response after the disaster, especially in the severely affected areas [25–27]. Early morbidity from some cardiovascular events was likely predominantly attributable to the psychological stress of the event2 and corresponding physiologic derangements; psychosocial factors such as missed medications, changed diet, poor living conditions, and the stress of disorders following a disaster have been noted as well [26,28,29]. Psychosocial stressors and lack of medication and a proper support system likely play a key role. After Hurricane Katrina, multiple reports indicated high rates of psychosocial stress and posttraumatic stress [30]. Survivors had to cope with stressors such as searching for food and shelter, relocating, crowding, financial hardship, and navigating social services. Events such as hurricanes and other natural disasters can cause chronic and acute mental stress, which can trigger cardiovascular events [31]. Traditional cardiac risk factors account for only half of the incidence of CVDs, with most of the remaining risk explained by psychosocial factors [31]. Our study revealed similar results and indicated a short-term increase in the numbers of visits for acute ischemic heart and cerebrovascular diseases by affected adults with cardiovascular and cerebrovascular disease histories after Typhoon Morakot. After matching, increased two-year mortality was observed in severely affected adults with an acute ischemic heart disease history, highlighting the importance of the care of high-risk populations with cardiovascular diseases, especially those residing in severely affected rural areas. However, our study also suggested that affected populations with more co-morbidities (i.e., liver cirrhosis, CKD, COPD, asthma and neoplasms other than cardiovascular disease) have higher all-cause mortality hazards. The increased all-cause mortality occurred when patients with multimorbidity and pre-existing cardiovascular

diseases such as stroke and ischemic heart diseases had increased risk of complications from either underlying diseases or comorbidities [32–37].

Health disparities are differences in the health statuses of specific populations and the general population. Health disparities are usually defined as "a particular type of health difference that is closely linked with social, economic, and/or environmental disadvantages" [8,38]. Although medical teams and acute social and financial relief programs are provided to affected populations in affected areas, affected populations with chronic comorbidities in severely affected areas still have an increased risk of mortality. Specialty and subspecialty healthcare services are usually less likely to be available in rural areas, and rural areas are less likely to offer specialized and highly sophisticated or high-intensity care than suburban or urban areas. Reliable transportation to healthcare facilities might also be a barrier for rural residents due to long distances, poor road conditions, and the limited availability of public transportation options in rural areas. These exacerbate problems for rural patients seeking specialized care who are required to travel significant distances for treatment [39]. Following a disaster, medical infrastructure usually becomes overwhelmed with acute injury and illness patients [40,41]. If chronic diseases controlled, preexisting chronic health problems can quickly become acute in nature, and these diseases have been linked to increased mortality in vulnerable populations in the wake of a disaster [40]. Chronic disease within the context of a disaster might have a bidirectional effect, whereby initial acute disorders may advance to long-term illnesses if insufficiently treated. This effect creates a "secondary surge" in required medical treatment long after the event and amplifies health disparities among medically underserved populations. The secondary surge of chronic diseases after a disaster coupled with inherent healthcare disparities, such as those commonly found in medically deprived rural areas, makes access to routine healthcare very difficult during the recovery phase [7]. Our data revealed similar results. Severely affected areas were more likely to comprise households with low socioeconomic levels. Patients with underlying comorbidities had higher mortality rates and morbidities after Typhoon Morakot that those without underlying comorbidities. Affected adults with an acute ischemic heart disease history before Typhoon Morakot living in severely affected areas had substantially increased hospitalization and mortality rates, highlighting the importance of medical accessibility. Most of the acute medical teams departed the severely affected areas approximately 6 months after Typhoon Morakot. Although relocation villages with permanent houses and healthcare centers were established within one year after Typhoon Morakot in the severely affected area, the accessibility of public transportation from these relocation villages to advanced cardiovascular care centers was limited. Accessibility limitations have great negative impacts on affected adults with cardiovascular risks, resulting in negative health outcomes.

The effects of spatial disparities after a disaster could be improved by the adoption of an improved universal healthcare system. Previous studies have suggested that universal healthcare systems substantially reduce socioeconomic inequalities in primary care access and quality but lead to only modest reductions in disparities in healthcare outcomes [42–44]. The Sendai framework for disaster risk reduction 2015–2030 aims to achieve substantial reductions in disaster risks and losses by enhancing the resilience of national health systems; corresponding efforts include strengthening the development and implementation of inclusive policies and social safety-net mechanisms and ensuring access to basic healthcare services, with the ultimate aim of eradicating poverty [45]. High-quality and accessible universal healthcare systems could maintain and support affected populations after disasters and are regarded as an important aspect of community resilience [46]. In order to be well-prepared for the foreseeable catastrophic natural events caused by the climate change, after Typhoon Morakot, a series of laws have been enacted including the most important "Spatial Planning Act 2016" [47] (state of the new climate-adaptive and environmental-oriented land use policy). In 2018, the Ministry of the Interior announced the implementation of the National Spatial Plan in accordance with the law and further emphasized on prioritizing the conservation of environmentally sensitive areas as well as important public facilities including medical services in rural and coastal areas. In addition, the United Nations Sustainable Development Goals (SDGs) [48] (United Nations, 2015) form the basis of spatial

plans at all levels to ensure the health inequality and urban-rural gaps are taken into account in future land use policies. At the same time, more comprehensive affected area response plans including early evacuation of vulnerable groups and more medical and mental resources deployments before and after the heavy rainfall in susceptible areas were implemented to decrease the vulnerabilities and increase the resilience of the communities [49].

This study has several strengths. First, the large sample of population-based data covering almost the entire population affected by this disaster provided a large enough sample size for matching. Second, in addition to the use of traditional analyses, stratified Cox models and robust Cox models using robust sandwich variance estimators considering clustering within matched sets were employed. The latter approach resulted in unbiased estimations of marginal hazard ratios that could be compared with biased estimations of marginal hazard ratios resulting from the former approach.

Some limitations also need to be mentioned. The limitations of this study include the fact that the clinical data were collected from a registry; thus, completely verifying the data was impossible. Additionally, the registry does not include non-affected adults as reference group and provide detailed cardiovascular risk information regarding smoking habits, alcohol consumption, body mass index, physical activity, and family history, which are potential confounding factors in the analysis of chronic diseases and populations affected by disasters. The estimates of mortality among patients in this study were not verified with death certificates; however, the missing for more than 6 months is closely related to death and the proportion of disqualifications as an insurance applicant is negligible given the fact that foreigners only constitute around 2% of all insured individuals of Taiwan's NHI from the insurance service database. The misclassification rate of the estimated mortality method was <2.37%, and the method we used has been adapted from some previous studies [50,51]. Sex and age were chosen to match in these two groups to minimize the effects of comorbidities in the elderly. Although the underlying diseases between the two groups were still different, the CCI between the two groups after matching was similar and revealed similar comorbidities between groups. Hypertension is known as an important risk factor for cardiovascular death during the acute phase of a disaster. In most patients, the increases in clinic blood pressure and self-measured blood pressure are transient, and the blood pressure levels return to the pre-disaster baseline levels within 4 weeks. The blood pressure levels should be monitored and the dose of antihypertensive medication should be reconsidered every 2 weeks during the disaster situation [52]. An inverse relationship between diastolic pressure and adverse cardiac ischemic events (i.e., the lower the diastolic pressure the greater the risk of coronary heart disease and adverse outcomes) has been observed in numerous studies. This effect is even more pronounced in patients with underlying coronary artery disease (CAD) [53]. Previous studies also indicated uncontrolled hypertension increased risk of all-cause and cardiovascular disease mortality but no significant differences were identified between normotensives, and treated and controlled hypertensives [54]. Non-adherence to antihypertensive medication increased the risk of all adverse health outcomes, including all-cause mortality and hospitalization for cardiovascular diseases such as myocardial infarction, heart failure and stroke [55,56]. The hypertension-mortality risk would be significant reduced if good adherence to anti-hypertension agents is achieved [57,58]. Our study suggested that hypertension did not increase all-cause mortality within two years among affected adults with an ischemic heart disease history after Typhoon Morakot; patients with hypertension have lower mortality hazard might be because patients marked as hypertension received regular anti-hypertensive agent treatment with relative good adherence and blood pressure levels. Further longer follow-up studies should be considered to observe the effects. The monthly visits and mortality rates after this disaster might be underestimated if affected adults were not enrolled in the national healthcare system or did not have medical records during the study period. Furthermore, the areas covered in the spatial-temporal analysis were based on the primary medical facilities that were most commonly used by the residents of the affected areas after the disaster. It was difficult to locate all the affected individuals because some patients visiting these medical facilities might not have been residing in the same area they did before the disaster. The database did not include the unaffected population

as the reference group in evaluating the baseline status of mental health before and after Typhoon Morakot. The effects of Typhoon Morakot are difficult to evaluate precisely. Moreover, classifications of township development and the proportions of elderly individuals in the populations were based on the 2005 Taiwan Social Change Survey (Round 5), which primarily considered socioeconomic change and township development in Taiwan; thus, these results may not apply to other spatiotemporal analyses of the relationships between socioeconomic factors and diseases. However, data from the national census, such as the proportion of elderly individuals in the population and household income, were acquired to achieve optimal validity. Finally, nondifferential misclassification of diseases in the registry at baseline might lead to bias toward the null.

In conclusion, the health status of patients with histories of acute cerebrovascular and cardiovascular diseases before Typhoon Morakot was negative effected in both the short term and long term after Typhoon Morakot. The increase in mortality was predominant among the elderly population and was associated with comorbidities and living in the severely affected area. High-quality and accessible universal healthcare systems are important to maintain and support affected populations after disasters. Further long-term spatial and socioeconomic analyses of the health of affected populations with chronic diseases are warranted to build community resilience with optimal deployment, beneficial healthcare strategies and a collaborative framework.

**Author Contributions:** Conceived and designed the study: H.-I.S., C.-M.C. Provided study materials and assembled the data: H.-I.S., T.-Y.C., C.-M.C., Y.-F.T. Analysed and interpreted the data: H.-I.S., C.-M.C., Y.-T.H., T.-C.S. and J.-D.W. Wrote the paper: H.-I.S., C.-M.C., T.-C.S. and Y.-T.H. Final approval of the manuscript: H.-I.S. and C.-M.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** The funders played no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. No additional external funding was received for this study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**

**Figure A1.** Flow diagram for patients enrolled in this study.


**Table A1.** Affected Townships of Typhoon Morakot.

\* All townships were severely affected.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

International Journal of *Environmental Research and Public Health*

### *Article* **The Mortality Risk and Socioeconomic Vulnerability Associated with High and Low Temperature in Hong Kong**

**Sida Liu 1,2, Emily Yang Ying Chan 1,2,\* , William Bernard Goggins <sup>2</sup> and Zhe Huang 1,2**


Received: 31 July 2020; Accepted: 30 September 2020; Published: 7 October 2020

**Abstract:** (1) Background: The adverse health effect associated with extreme temperature has been extensively reported in the current literature. Some also found that temperature effect may vary among the population with different socioeconomic status (SES), but found inconsistent results. Previous studies on the socioeconomic vulnerability of temperature effect were mainly achieved by multi-city or country analysis, but the large heterogeneity between cities may introduce additional bias to the estimation. The linkage between death registry and census in Hong Kong allows us to perform a city-wide analysis in which the study population shares virtually the same cultural, lifestyle and policy environment. This study aims to examine and compare the high and low temperature on morality in Hong Kong, a city with a subtropical climate and address a key research question of whether the extreme high and low temperature disproportionally affects population with lower SES. (2) Methods: Poisson-generalized additive models and distributed-lagged nonlinear models were used to examine the association between daily mortality and daily mean temperature between 2007–2015 with other meteorological and confounding factors controlled. Death registry was linked with small area census and area-level median household income was used as the proxy for socioeconomic status. (3) Results: 362,957 deaths during the study period were included in the analysis. The minimum mortality temperature was found to be 28.9 ◦C (82nd percentile). With a subtropical climate, the low temperature has a stronger effect than the high temperature on non-accidental, cardiovascular, respiratory and cancer deaths in Hong Kong. The hot effect was more pronounced in the first few days, while cold effect tended to last up to three weeks. Significant heat effect was only observed in the lower SES groups, whilst the extreme low temperature was associated with significantly higher mortality risk across all SES groups. The older population were susceptible to extreme temperature, especially for cold. (4) Conclusions: This study raised the concern of cold-related health impact in the subtropical region. Compared with high temperature, low temperature may be considered a universal hazard to the entire population in Hong Kong rather than only disproportionally affecting people with lower SES. Future public health policy should reconsider the strategy at both individual and community levels to reduce temperature-related mortality.

**Keywords:** H-EDRM; climate change; extreme temperature; climate change; socioeconomic vulnerability; health disparities

#### **1. Introduction**

The adverse effect of extreme temperature on human health has been intensively reported during the last decade [1,2]. Epidemiological evidence shows that both high and low temperature were associated with increased mortality and morbidity in western, Asian and African countries [3–5]. With the effect of climate change, extreme weather events are expected to be more frequent and intensive in the future, which may impose a significant health burden in the years to come [6]. Previous studies have also identified that temperature may pose different health impact to population with different characteristics, some risk factors related to the vulnerability of extreme temperature include pre-existing diseases, housing, behavior as well as socioeconomic status (SES) [7–9].

The association between health inequities and socioeconomic disparities have been well established [10–12]. A number of research revealed that such disparities were also existed for health effect induced by temperature [13,14]. Current research has shown inconsistent results on whether socioeconomic disadvantages were associated with higher temperature-related health risk, especially for cold weather [7,8,14,15]. Previous studies of such kind were mainly conducted by city or country-level comparison [8,16,17]. This may be problematic since other confounding factors such as lifestyle, behavior, temperature-related policy and acclimation are difficult to properly controlled. An intra-city analysis has advantages to minimize the effect of other confounding factors. This was rarely conducted in previously due to methodological challenges that SES data are often absent in retrospective health-related data in most countries.

In Hong Kong, the mortality registry can be lined with area-level census data, which provide an opportunity to perform such city-wide analysis in which the study population shares virtually the same cultural, lifestyle and policy environment. Hong Kong is a developed metropolis with one of the largest income disparities in the world. The Gini coefficient was reported to be 0.539 for original and 0.473 for after social benefit transfer and it continues growing in the last two decades [18]. The large income disparity gap provides a good example with which to investigate the modifying effect of SES on temperature. Many mortality and morbidity related to extreme temperature are predictable and preventable [19]. As highlighted in UNISDR Sendai Framework and WHO health emergency disaster risk management (Health-EDRM), understanding the health risk and vulnerable population are priorities to reduce the exposure and adverse health consequences that are associated with disaster and environmental hazards [20]. In this study, we aimed to (1) examine and compare morality risk associated with high and low temperature in the subtropical city and (2) to address a key research question of whether the extreme high and low temperature disproportionally affects population with lower SES.

#### **2. Materials and Methods**

#### *2.1. Environmental Data*

Meteorological data including mean daily temperature, relative humidity and atmospheric pressure from 1 January 2007 to 31 December 2015 were obtained from Hong Kong Observatory, which is located at the center of the urban area. To adjust for the potential confounding effect of air pollution, daily mean air pollutant data including nitrogen dioxide (NO2), Sulphur dioxide (SO2), Ozone (O3) and particulate matter less than 2.5 µm aerodynamic diameter (PM2.5) of the same period were collected from the Hong Kong Environmental Protection Department. The daily air pollution variables were calculated by taking an average of ten general stations.

#### *2.2. Mortality Data*

The corresponding daily non-accidental mortality data were collected from the Hong Kong Census and Statistics Department for analysis. The data also contain the cause of death, gender and age for stratified analysis. Cause of death was also analyzed which classified according to International Classification of Diseases, Tenth Revision (ICD-10) non-accidental (A00-T99, Z00-Z99), cardiovascular (I00-I99) and respiratory (J00-J98) diseases and cancer (C00-D48).

#### *2.3. Socioeconomic Status Data*

Several indictors have been chosen to reflect socioeconomic status (SES) in previous studies, including education, income, occupation, and race/ethnicity [7,14,21–23]. To further capture the multidimensional feature of SES, some studies used composite indices composed of multiple indicators to represent an individual's socioeconomic advantage [24–26]. However, many of those indices were not officially adopted by local government, and aggregating SES may introduce additional bias when interpreting. In this study, we used a signal measure indicator, the median monthly household income (MHI) of a deceased person's residential area, as a proxy to reflect the one's socioeconomic status (SES). Income is the most widely used SES indicator, which directly and indirectly affects access to goods and services for health. In the 2011 Hong Kong census, the territory of Hong Kong was divided into 289 small areas, which are also known as Tertiary Planning Units (TPUs). To minimize issues regarding privacy, TPUs with a population less than 1000 were merged into the nearby TPUs to form 209 TPU clusters. One TPU was intentionally excluded due to the entire population within the area being prisoners, and therefore 208 TPU clusters were included in the final analysis. All TPU clusters were then grouped into four levels using the quartering method according to its MHI (52 TPU clusters in each group, where 1 indicates the group with highest SES and 4 is the lowest SES group), to evaluate whether there were differences in the effect of temperature on all non-accidental and cause-specific mortality (Figure 1). Each death record was linked with an SES group code according to his/her residential area by the TPU code.

**Figure 1.** The geographical distribution of socioeconomic status (SES) groups by median monthly household income of tertiary planning units (TPUs) in Hong Kong, where One = highest SES, shown in green and Four = lowest SES, shown in red).

#### *2.4. Statistical Analysis*

Distributed lag linear and non-linear models (DLNM) were performed to determine the association between daily mortality and ambient temperature for each SES group. As daily mortality counts generally follow an overdispersed Poisson distribution, a distributed lag model with a quasi-Poisson regression was used to evaluate the health effect of heat and cold while adjusting for the temperature at different lag days. The model used the following formula:

$$\begin{aligned} \text{Log[E(Y\_t)]} &= \mathbf{a} + \mathbf{c} \mathbf{b} (\text{Temp}\_t) + \text{ns} (\text{Humidity}\_t, \text{df} = 4) \\ &+ \text{ns} (\text{Pressure}, \text{ df} = 4) + \text{ns} (\text{O}\_3, \text{ df} = 4) \\ &+ \text{ns} (\text{SO}\_2, \text{ df} = 4) + \text{ns} (\text{NO}\_2, \text{ df} = 4) + \text{ns} (\text{PM}\_{25}, \text{ df} = 4) + \text{ns} (\text{Time}, \text{ df} = 9 \times 7) + \text{Down}\_t + \text{Holidity}\_t \end{aligned}$$

where α is the intercept; cb is the crossbasis matrix built up with *dlnm*() package in R; ns(): smoothing function using a natural cubic spline. A quadratic B-spline with three internal knots placed at the 10th, 75th and 90th percentiles of temperature was used to model the exposure–response relationship and a natural cubic spline was fitted with three internal knots placed equally spaced in the log scale to model the lag–response association [27]. 4 degrees of freedom (df) were used for potentially confounding metalogical and air pollutant variables. The model uses a maximum lag period of 21 days to capture the long delay of effects of cold and to exclude deaths that were advanced by only a few days (harvesting effect) [28,29]. The long-term and seasonality effect was controlled using 7 df per year. Day of the week (Dow) and local public holidays were also adjusted in the model. The Akaike Information Criterion for quasi-Poisson (Q-AIC) values were used for the choice of the degree of freedom.

The relationship between temperature and mortality was summarized as the exposure–response curve of relative risk (RR) accumulated across all lags. Minimum mortality temperature (MMT), with its 95% confidence interval (95% CI), was calculated. RR for total death and each SES group were estimated. Subgroup analyses for gender, age group (0–65 years and 65 years above) and cause-specific death (cardiovascular, respiratory and cancer) were performed. The results show that the MMTs for each subgroup analysis were close to 27 ◦C, so that was chosen as the fixed reference temperature to compare the temperature effect on each SES groups. Modelling choices were tested by a sensitivity analysis using different degrees of freedom for temperature and long-term trends, and length of maximum lag period. All analyses were performed using R project for statistical computing (version 3.6) packages *dlnm* and *mgcv* [30,31].

#### **3. Results**

#### *3.1. Data Description*

Hong Kong is a subtropical city with hot summer and mild winter. The mean and median daily mean temperature during the study period were 23.5 ◦C and 24.7 ◦C, with a range between 8.4 and 32.4 ◦C, respectively (Table 1). As a coastal city, the relative humidity is relatively high, with a median of 79%. Overall, 362,957 non-accidental deaths were included in the analysis after excluding cases without a valid TPU code (1.34%). There were more deaths in the lower SES group than higher SES in Hong Kong, with daily mortality ranging from 0 to 26 cases in the highest SES group and 14 to 192 cases in the lowest one (Table 2). Figure 1 shows the distribution of the TPU cluster by SES groups, of which the highest cluster (Deep Water bay, HK\$178,000 or US\$22,960/month) has reported 20 times higher median household income compared to the lowest (Tai O, HK\$8000 or US\$1031/month), indicating a large SES disparity among districts in Hong Kong.


**Table 1.** Descriptive statistics of metalogical and air pollutant variables.


**Table 2.** Descriptive statistics of daily mortality count in the study period.

#### *3.2. Main Findings*

As shown in Figure 2, the cumulative effect (lag 0–21) of cold posed a stronger effect on mortality when compared to heat with the reference temperature of 28.9 ◦C—the temperature with minimum mortality (MMT). When an extremely low temperature occurs (12.9 ◦C, 2.5th percentile of daily mean temperature), the mortality rate is almost 1.5 times higher compared to 27 ◦C. No significant overall heat was identified in the lag period of 0 to 21 days. For the lag–response effect, the hot effect tended to be more pronounced in the first 3 days with no harvesting effect. Meanwhile, the cold effect may last up to 3 weeks after the low temperature occurs (Figure 3). The cumulative cold effect in relative risk for lag 0–21 day was 1.463 (95% CI: 1.362, 1.571) and the heat effect for lag 0–3 days was 1.035 (95% CI: 1.011, 1.059) (Table 3). No overall clear harvesting effect in mortality was observed for both heat and cold effect.

For cause-specific temperature on mortality (Figure 4), the high temperature was found to be associated with significantly higher respiratory death, particularly among groups with the lower SES, whereas heat effect was not significant for cardiovascular deaths. On the other hand, the cold impact was significant for both cardiovascular and respiratory death and the effect was found to be generally higher in all SES groups. The results also suggested that the heat effect for respiratory death was higher than cardiovascular, whereas the cold effects were stronger for cardiovascular death than for respiratory death. The effect of cold temperature on cancer mortality was significant, although the effect size was much smaller than other causes. The cold effect on cancer deaths was found to be higher in lower SES groups, whereas the more affluent groups were not significantly affected. High temperature was found to be associated with increased mortality risk for cancer deaths.

Furthermore, the high temperature was associated with higher mortality risk among females than males, but no clear gender difference was observed for cold effect (Figure 5). Compared with the younger population, older individuals were generally associated with higher temperature-related mortality, especially for low temperature. Significantly higher cold-related mortality risk was observed in all SES groups, but heat-related mortality risk was only identified in the lowest SES group.

To compare the effect of high and low temperature among SES groups, the RR for each SES group were estimated (Figure 4). For non-accidental death, heat effect was found to be only significant in lower SES groups (Group Three: 1.039 (95% CI: 1.001, 1.078), Group Four: 1.040 (95% CI: 1.005, 1.076), while the cold effect was found to be significant across all SES groups (Figure 4).

**Figure 2.** The cumulative effect (lag 0–21 days) of daily mean temperature on non-accidental mortality during 2007 and 2015 with the distribution of daily mean temperature at the bottom (number of days). The dashed in the middle indicates the minimum mortality temperature (28.9 ◦C) the two vertical lines on the sides indicates the 2.5th (12.9 ◦C) and 97.5th (30.3 ◦C) percentile of the daily temperature during study period.

**Figure 3.** The Lag–response effect distribution during the 21 days lag period after exposed to a low (12.9 ◦C, 2.5th percentile) and high (30.3 ◦C, 97.5th percentile) temperature compared to reference temperature (27 ◦C).

**Figure 4.** The cold (12.9 ◦C, 2.5th percentile at lag 0–21) and heat effect (30.3 ◦C, 97.5th percentile at lag 0–3) compared to 27 ◦C by cause of death and SES groups (where One = the highest SES, Four = lowest SES).

**Figure 5.** The cold (12.9 ◦C, 2.5th at lag 0–21) and heat effect (30.3 ◦C, 97.5th at lag 0–3) on non-accidental deaths compared to 27 ◦C by age, gender and SES groups where One = the highest SES, Four = the lowest SES.


**Table 3.** The relative risk (RR) for cold (12.9 ◦C, 2.5th percentile at lag 0–21) and heat effect (30.3 ◦C, 97.5th percentile at lag 0–3) compared to 27 ◦C with 95% confidence interval.

\* Statistically significant at α ≤ 0.05.

#### **4. Discussion**

This study is the first study to investigate the SES-related disparities on the mortality risk associated with the high and low temperature in Hong Kong. A J-shape temperature–mortality association was identified for this subtropical urban metropolitan city. Results are consistent with previous studies, suggesting that the cold effect is stronger than heat effect in most populations across many countries [27,32,33], and in subtropical cities such as São Paulo [34], Guangzhou [29] and Taipei [27]. The finding provides supporting evidence to the hypothesis that population living in a warmer climate are more adapted to cope with high temperatures, and more susceptible to cold weather.

The minimum mortality temperature in Hong Kong was found to be 28.9 ◦C, at 82% percentile of the daily mean temperature in the study period, which was generally consistent with the finding from other studies conducted on population with similar climate [21]. One study examined the relationship between temperature and mortality in 66 Chinese communities found that the MMT was 27.4 ◦C in southern China, which is the highest compared to other regions in China [35]. Country-wide and worldwide studies found that the MMTs are usually higher in warmer regions [27,33,35,36]. Moreover, the MMT identified in this study was slightly higher than the mortality threshold (28.2 ◦C) reported in a previous local study [21] which used the data from 1998–2006. One assumption could be that the population are adopting the increasing temperature with the effect of climate change [3,37,38], leading to a slightly higher tolerance of hot weather. With the effect of climate change, some studies anticipated that heat-related mortality will increase and eventually compensate for the reduction in cold-related death after 2050 [39]. Other studies suggested that despite the fact that the cold burden will be reduced by the relative effect, heat-related death will remain high across the entire 21st century [40], and the net effect may be inconsistent and subject to local context [41,42], indicating that it is still too early to neglect the health impact of cold weather in the coming decades.

The climate in subtropical regions typically has a very hot summer and less harsh winter. Residents are usually acclimated to a high temperature, which was considered as an important reason why an overall insignificant effect of heat was observed in this study [29,43]. In Hong Kong, the air condition was commonly installed in almost all indoor areas and public transports. The actual exposure of the population could be substantially reduced. However, the study found that the 3-day cumulative heat effect was significantly higher among areas with lower SES, which was also reported in some local and international studies [21,44,45]. A possible reason may be due to the characteristics of housing. The buildings in low-income areas tend to have a higher proportion of old buildings with poor ventilation and insulation. Hong Kong is a city with the highest housing price in the world, and individuals who suffer from poverty usually live in subdivided or temporary dwellings [46]. Those dwellings are often small with poor ventilation, in which heat can be easily trapped inside.

This study found that the cold effect was significant across all SES groups for non-accidental, cardiovascular, respiratory and cancer death. The result also shows a counterintuitive pattern that communities with higher SES were associated with higher cold-related mortality risk, which despite

the differences was not statistically significant. Although this kind of analysis was rarely conducted in previous studies, some reported similar results [15]. In Hong Kong, communities with lower SES tends to have higher living density and stronger urban heat island effect [47]. This may suggest that the urban heat island effect may have potential benefit against cold weather in highly urbanized populations, which has been reported recently elsewhere [48]. The universal impact of low temperature on mortality may also be due to the low prevalence of central heating. Despite the high air condition coverage in Hong Kong, most air conditioning devices do not have heating functions and buildings were not designed to restore heat.

A stronger cold effect was identified on cardiovascular than respiratory mortality, which was consistent with previous local and international studies [34,49]. Previous physiological studies suggested that exposure to low temperature may cause elevated blood pressure [50,51], blood viscosity [52], plasma cholesterol and the tendency of blood clot formation in the vessels [53]. Changes in those risk factors may subsequently increase the risk of cardiovascular death. When a high-temperature event occurs, the risk of respiratory death increases as the SES decreases and only individuals living in the lowest SES group were significantly associated with higher risk. A similar pattern was also generally observed for cardiovascular death, despite the effect for all SES groups not being statistically significant. Some earlier local and international studies indicated that the area with lower SES was disproportionally affected by a higher concentration of air pollutants [54,55] and may pose both short and long term adverse effects on health [56,57]. A US study also suggested that in a large city, the exposure of NO<sup>2</sup> concentration is significantly higher for individuals with lower household income [54]. A local study showed that areas with a high level of social deprivation were associated with higher exposure to PM2.5 [24]. Furthermore, the effect on air pollution on mortality may also be modified by that temperature. Some studies found that an adverse effect of particulate matter on mortality is stronger under hotter weather in Chinese and European populations [58,59]. The disparity further supports that individuals living in a lower SES community had a higher relative risk after exposure to hot weather, especially for respiratory deaths. This study also identified that cold temperature was associated with significantly higher cancer mortality risk. Several recent studies suggested that cold temperature could be an independent risk factor for cancer [60]. Cold exposure may increase metabolic stress, may contribute to tumorigenesis and higher cancer deaths [61]. A study used data from 166 countries and found a positive association between cancer mortality rate and serum average total cholesterol, which could act as a mediator of cancer development [62]. Unlike other causes, a significant cold effect was only observed in groups with lower SES. higher deprivation and lower-income have been linked with higher cancer incidence and mortality rate due to inequalities on lifestyle, environmental factors and access to services [63,64]. However, the cold effect on cancer and the role of SES have been underreported in current literature, which may be a potential gap for future research.

Significantly higher cold-related mortality risk was found for both males and males, and gender difference was found to be minimum for cold effect. However, females were found to be more susceptible to high temperature. Such a pattern has also been found in a local study [21] and studies elsewhere [8,65]. Older age is a well-known factor associated with higher susceptibility to temperature-related mortality risk [8,29,66,67], and this study found no exception, especially for low temperatures. No clear intra-SES group heterogeneity was identified in both age groups for low temperature effect. However, older persons in the lower SES group were significantly associated with higher heat-related mortality risk, whilst high temperature only posed a very minimum effect to their counterparts in better-off groups. A UK study found that lower SES was associated with a higher uptake of protective measures when experiencing hot weather [68]. A local study indicated that low-income individuals did not have the same level of protective measures, and that some may face financial constraints and still have to work outdoors under hot weather [69]. However, a recent local study found that higher income and education level were not associated with a higher prevalence of protective behavior during cold weather, and vulnerable groups such as the older population

commonly underestimated their health risk [70], which partially explained the universal cold effect across all SES groups.

In the UK and Spain, cold weather has been recognized as a major public health concern at the national level, even though the weather during the winter in the two countries is generally milder compared to many other European countries. Nevertheless, the health impact of cold weather has not received adequate attention in Hong Kong, and only limited public services are made available when low temperatures occur [71]. Future policy should consider establishing a holistic strategy to enhance the protective measures at both individual and community levels and reduce the mortality and morbidity associated with low temperature.

A limitation of this study, which is shared by many other researchers of this kind, is the selection of SES measures. The small area level SES may not directly reflect the SES at the individual level. Future multilevel studies with both small areas and individual measures should be conducted, to further understand the effect of SES on the association between temperature and adverse health outcome. Moreover, the actual temperature exposure may vary from person to person, and outdoor ambient temperature may not reflect the exposure if a person spends more of the time in the indoor environment. Some behavioral and physiological factors such as lifestyle and pre-existing diseases may not necessarily be associated with SES, but have also proven to be related to personal exposure and the outcome of extreme temperatures.

#### **5. Conclusions**

Despite the fact that Hong Kong has a milder winter climate, the overall cold effect was found to be much stronger and last for a longer period than the heat effect on mortality in Hong Kong. The subpopulation with lower SES is more vulnerable to high temperature, but low temperature universally affects the entire population across all SES groups in Hong Kong. A higher cold effect was observed in groups with the highest SES, despite the different was not statically significant. The heat island effect is a risk factor for heat-related deaths, but it may provide a protective effect during cold weather. This study raises the concern of the health effect of low temperature in subtropical regions, which should be recognized as an important public health issue at territory level. Strategies on mitigating the cold effect should not only focus on populations with lower SES; additional public services should be provided for different socioeconomic classes in Hong Kong.

**Author Contributions:** Conceptualization, S.L., E.Y.Y.C.; methodology, S.L. and W.B.G.; data analysis, S.L., Z.H., W.B.G.; writing—original draft preparation, S.L.; writing—review and editing, E.Y.Y.C., W.B.G., Z.H.; supervision, E.Y.Y.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors would like to thank the Hong Kong Census and Statistics Department, the Hong Kong Observatory, and the Environmental Protection Department in Hong Kong for providing data for this study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*International Journal of Environmental Research and Public Health* Editorial Office E-mail: ijerph@mdpi.com www.mdpi.com/journal/ijerph

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com ISBN 978-3-0365-1749-0