Bulletin of the World Health Organization

Hospital admissions as a function of temperature, other weather phenomena and pollution levels in an urban setting in China

Emily YY Chan a, William B Goggins b, Janice SK Yue a & Poyi Lee a

a. Collaborating Centre for Oxford University and CUHK for Disaster and Medical Humanitarian Response, 3/F, School of Public Health and Primary Care, Prince of Wales Hospital, Shatin, NT, Hong Kong Special Administrative Region (SAR), China.
b. JC School of Public Health and Primary Care, The Chinese University of Hong Kong, Hong Kong SAR, China.

Correspondence to Emily YY Chan (e-mail: emily.chan@cuhk.edu.hk).

(Submitted: 17 September 2012 – Revised version received: 18 April 2013 – Accepted: 23 April 2013 – Published online: 31 May 2013.)

Bulletin of the World Health Organization 2013;91:576-584. doi: http://dx.doi.org/10.2471/BLT.12.113035 [PDF]

Introduction

According to the Intergovernmental Panel on Climate Change, global climate change poses major health threats and shifting temperature patterns in the 21st century are expected to substantially affect the burden of disease worldwide.19 Although there is evidence of an association between outdoor temperature and mortality rates, the interplay between the two remains only partially understood in Asia because of a lack of databases with comprehensive, comparable data for most urban communities in low- and middle-income settings.10,11 Evidence from European and American cities suggests that when outdoor temperatures are unusually high, there is a rise in hospital admissions10,1215 for respiratory ailments, renal diseases,12 and infectious diseases (both vector-borne and foodborne) and cerebrovascular accidents, including subarachnoid haemorrhage13 and transient ischaemic attacks.1622 The risk of admission appears to be higher among the critically ill,23 the very young, the very old, people of low socioeconomic status,24 people with pre-existing medical conditions,12,16 health-care workers, pregnant women25 and people who are institutionalized, live alone or have poor mobility.24

Hong Kong Special Administrative Region (SAR), China, has recorded one of the world’s highest average increases in ambient temperature.26 A study of the relationship between outdoor temperature and mortality in the city has shown a cumulative increase in mortality of 1.8% for every increase of 1 °C above 28.2 °C in mean daily outdoor temperature.26 According to another study on help-seeking behaviour in the city, elevated temperatures are associated with an increased volume of calls to an emergency hotline, particularly from women and elderly people.27 In light of these findings, the main objectives of this study were to examine whether rates of injury and disease fluctuate in relation to outdoor temperature and other environmental parameters in Hong Kong SAR by looking at hospital admissions, and to identify high-risk population subgroups. The ultimate aim is to facilitate policy to mitigate the negative health effects of changing climate patterns.

Methods

We conducted a retrospective ecological study using data on outdoor temperature, pollution levels and routine hospital admissions in Hong Kong SAR from 1 January 1998 to 31 December 2009. Meteorological variables, including mean daily temperature (MDT), mean daily relative humidity (MDRH), mean daily wind speed (MDWS), daily total global solar radiation (DTGSR) and total daily rainfall (TDR), were obtained from the Hong Kong SAR Observatory. Mean daily pollution levels were based on hourly concentrations of nitrogen dioxide (NO2), sulfur dioxide (SO2), ozone (O3) and particulate matter measuring ≤ 10 µm. These data were obtained from the 11 general environmental data collection stations belonging to the Environmental Protection Department.

The Hospital Authority in SAR keeps a citywide hospital database that covers all admissions to public hospitals, which represent 83% of all admissions. From this database we extracted data on routine hospital admissions by age, sex, district of residence, name of hospital, admission and discharge dates, cause of hospitalization, total length of hospital stay and discharge status. We analysed data on admissions for the following causes, as coded in the International statistical classification of diseases and related health problems, ninth revision:28 unintentional injuries (from external causes, codes 800–957; from accidents, codes E800–E949); intentional injuries (intentional self-harm, codes E950–E959; assaults, codes E960-E969); infectious diseases, codes 001–139, 320–323 and 614–616; renal failure, codes 584–586; diabetes, codes 249–250; cirrhosis, codes 571.5, 571.6, 571.8 and 571.9; cancer, codes 140–239; circulatory diseases, codes 390–459; and respiratory diseases, codes 460–519. All analyses were conducted by sex and age group (< 15 years, 15–59 years, 60–74 years and ≥ 75 years).

Statistical modelling

We developed generalized additive (Poisson) models to examine the association between changes in MDT and daily number of hospital admissions attributable to the causes described in the preceding section. Analyses were conducted separately for the hot season (June to September) and the cold season (November to March) to allow for the lag effect of temperature on hospitalization patterns. We used R version 2.10.1 (The R Foundation for Statistical Computing, Vienna, Austria) for the analyses. First we used R’s distributed lag nonlinear model (dlnm) package to determine how lagged effects should be modelled, and then proceeded to fit distributed lag models. We subsequently fit the generalized additive models with nonlinear smooth terms for MDT with a lag from day 0 (same day) to day k, where k was determined from the results of the distributed lag models. In all models, we controlled for time trend by means of a variable representing the day of the study (with values from 1 to 4383), and we controlled for season by means of another variable representing the day of the year (with values from 1 to 365). We allowed each of these variables a maximum of 4 degrees of freedom.

Results

From 1998 to 2009, temperatures ranged from 22.2 °C to 31.8 °C during the hot season; the average MDT was 28.4 °C and the average median daily temperature was 28.6 °C. During the cold season, temperatures ranged from 8.2 °C to 26.9 °C; the average mean and median daily temperatures were 18.7 °C and 19.0 °C, respectively. Table 1 shows the number of routine hospital admissions for various causes during the cold season (November to March) and the hot season (June to September). During the study period, there were 7 869 661 hospital admissions, and 33.1% of them took place during the hot season. Of the causes of admission selected for analysis, the most common one was respiratory diseases, which accounted for 17.1% of the admissions. This was followed by admissions for cancer (16.1%), circulatory diseases (15.1%), unintentional injuries (10.4%), infectious diseases (5.1%) and intentional injuries (0.3%).

Hot season

In the initial distributed lag models, the effects of high temperature on admissions peaked over a lag period of 3 to 6 days (lags 3–6) but persisted until the 10th day (lag 10) during the hot season. A smooth term with a maximum representing the MDT with a lag of 0 to 10 days (MDT(0–10)) was used in the subsequent generalized additive model, along with smooth terms for MDRH with a lag of 0 to 10 days (MDRH(0–10)), average DTGSR with the same lag period (DTGSR(0–10)) and MDWS with the same lag period (MDWS(0–10)). Same-day rainfall (square-root-transformed) was also included in all models based on the hypothesis that heavy rain would deter people from going to hospital. All terms were initially allowed a maximum of 5 degrees of freedom, with actual degrees of freedom determined by cross-validation using R’s mixed GAM computation vehicle (mgcv) package. In the final model, MDT(0–10) showed a statistically significant positive association with daily admissions that was roughly linear above 29 °C (Fig. 1). The association between MDRH10 and hospitalizations was complex (Fig. 2), but it was positive and approximately linear between humidity values of about 73% and 84% (i.e. the 20th and 80th percentiles of the variable’s distribution). Mean DTGSR(0–10) and MDWS(0–10) were positively and linearly associated with the number of daily admissions, although the associations were not statistically significant (P = 0.23 and 0.49, respectively); mean NO2, lagged by 0–4 days, was positively associated with the number of hospitalizations, with the association being linear over most of the range of this variable. The final model hockey stick terms created for temperature were: THOT = MDT(0-10) – 29if MDT(0–10) ≥ 29 °C, 0 otherwise:TCOLD = 29 – MDT(0–10)if MDT(0–10) < 29 °C, 0 otherwise.

Fig. 1. Relative risk of hospitalization as a function of mean daily temperature lagged by 0–10 days (MDT(0–10)), Hong Kong Special Administrative Region, China, 1998–2009
Fig. 1. Relative risk of hospitalization as a function of mean daily temperature lagged by 0–10 days (MDT<sub>(0–10)</sub>), Hong Kong Special Administrative Region, China, 1998–2009
Fig. 2. Relative risk of hospitalization as a function of mean daily relative humidity lagged by 0–10 days (MDRH(0–10)), Hong Kong Special Administrative Region, China, 1998–2009
Fig. 2. Relative risk of hospitalization as a function of mean daily relative humidity lagged by 0–10 days (MDRH<sub>(0–10)</sub>), Hong Kong Special Administrative Region, China, 1998–2009

Linear terms were used for DTGSR(0–10), MDWS(0–10), and NO2 and a smooth term with a maximum of 5 degrees of freedom was adopted to account for MDRH(0–10). In this final model, a significant association (P = 0.003) was found between MDT and daily hospital admissions at temperatures above 29 °C. For every rise of 1 °C above this threshold, there was a 4.5% (95% confidence interval, CI: 2.1–7.0) increase in admissions. MDRH(0–10) remained nonlinearly and significantly (P < 0.001) associated with admissions, in a pattern similar to that in Fig. 2. The number of admissions showed a significant positive association with mean NO2 concentration, a non-significant negative association (P = 0.18) with MDWS(0–10) and a non-significant positive association (P = 0.64) with DTGSR(0–10).

In subgroup analyses of the causes of hospitalization, a rise in admissions for respiratory diseases, infectious diseases and unintentional injuries was noted at temperatures above 28.5 °C. There was no significant rise in admissions due to other causes, including cancer, circulatory diseases, diabetes mellitus, renal failure or cirrhosis, at these temperatures.

For respiratory diseases, the initial model showed an increase in admissions at an MDT10 above 28.5 °C; a similar association was noted between admissions and mean MDRH(0–10). In the subsequent model, a significant increase in admissions for respiratory diseases was found at temperatures above 28.5 °C (P < 0.001). For every rise of 1 °C above this threshold, admissions rose by 7.6% (95% CI: 5.1–10.2). Mean MDRH(0–10) remained nonlinearly and significantly (P < 0.001) associated with the number of admissions, in a pattern similar to that in Fig. 2. For every increase of 10 mg per m3 in mean O3 concentration lagged by 0–4 days, admissions for respiratory diseases rose by 1.6% (95% CI: 0.7–2.4; P = 0.0002); the association between mean NO2 concentration and admissions became non-significant after controlling for O3. Mean DTGSR(0–10) was significantly (P = 0.003) associated with admissions; for every increase of 5 milliJoules per m2 in mean DTGSR(0–10) (approximately equal to the interquartile range for this variable), admissions for respiratory diseases rose by 3.2% (95% CI: 1.2– 5.2).

For infectious diseases, the initial model showed a significant association (P = 0.01) between MDT(0–10) and the number of daily admissions at MDT(0–10) values above 28.5 °C. Mean MDRH(0–10) also showed a significant (P = 0.0003) S-shaped association with admissions at MDRH(0–10) values between 73% and 84%. This pattern was also noted for admissions due to other causes. Mean DTGSR(0–10) was not significantly associated with admissions for infectious diseases. The subsequent model showed a significant association (P < 0.001) between admissions for infectious diseases and MDT(0–10) at temperatures above 28.5 °C; for every rise of 1 °C above this threshold, there was a 4.5% (95% CI: 2.2–6.9) increase in admissions. Admissions for Salmonella infections were particularly sensitive to the effect of hot temperatures; they rose by 11.4% (P < 0.001) on the same day and over the following 14 days for every rise of 1 °C in MDT(0–14) within the range of temperatures observed in the hot season (22.2–31.8 °C). MDWS(0–14) was also significantly (P = 0.006) and negatively associated with admissions for infectious diseases, whereas mean MDRH(0–14) remained significantly (P < 0.001) and positively associated with admissions, in a pattern similar to than in Fig. 3.

Fig. 3. Relative risk of hospitalization for any cause at 32 °C versus 29 °C (lag 0–10 days) during the hot season,a Hong Kong Special Administrative Region, China, 1998–2009
Fig. 3. Relative risk of hospitalization for any cause at 32 °C versus 29 °C (lag 0–10 days) during the hot season,<sup>a</sup> Hong Kong Special Administrative Region, China, 1998–2009
a June to September.

Fig. 3 shows that during the hot season, admissions due to all causes peaked 3 to 6 days after a hot day. Fig. 4 shows that admissions for respiratory diseases peaked immediately and remained higher but declined slowly over the next 4 days. Fig. 5 indicates that admissions for infectious diseases peaked about 3 to 5 days after a hot day.

Fig. 4. Relative risk of hospitalization for a respiratory disease at 32 °C versus 29 °C (lag 0–10 days) during the hot season,a Hong Kong Special Administrative Region, China, 1998–2009
Fig. 4. Relative risk of hospitalization for a respiratory disease at 32 °C versus 29 °C (lag 0–10 days) during the hot season,<sup>a</sup> Hong Kong Special Administrative Region, China, 1998–2009
a June to September.
Fig. 5. Relative risk of hospitalization for an infectious disease at 32 °C versus 29 °C (lag 0–10 days) during the hot season,a Hong Kong Special Administrative Region, 1998–2009
Fig. 5. Relative risk of hospitalization for an infectious disease at 32 °C versus 29 °C (lag 0–10 days) during the hot season,<sup>a</sup> Hong Kong Special Administrative Region, 1998–2009
a June to September.

Our findings also indicate a significant (P = 0.006) increase of 1.9% (95% CI: 0.6–3.2) in admissions for unintentional injuries for every rise of 1 °C in MDT above 28.5 °C. The number of admissions due to intentional injuries (self-harm or harm to others) was not significantly associated with a rise in MDT above a certain threshold.

We conducted additional analyses by sex and age group for admissions due to respiratory and infectious diseases (Table 2). No significant differences were observed between the sexes. For respiratory disease admissions, the increase observed during hot weather was greater among children (0–14 years) and young adults (15–59 years) than among people aged 60–74 years or people 75 years of age or older. For infectious disease hospitalizations, the increase in admissions observed during hot weather was greater among people 75 years of age or older than among younger people.

Cold season

In the initial distributed lag models, cold temperatures, lagged by 0–2 days, showed a negative association with admissions, but the association became positive from lag 3 onwards, peaked from lags 5 to 17 and remained positive to lag 21 (Fig. 6). In subsequent analyses we adopted the MDT lagged by 0 to 21 days (MDT(0–21)), along with averages of the other meteorological variables, lagged by 0 to 21 days. For air pollutants, the distributed lag models showed an association between admissions and mean NO2 only on the same day (lag 0) and the previous day (lag 1). Other pollutants, however, showed no significant association with admissions after controlling for mean NO2. Therefore, in the model we used mean NO2, lagged by 0 and 1 days (NO2(0–1)), and we dropped other pollutants. In subsequent analyses in which smooth terms were used for these variables, the associations between the number of admissions and MDT(0–21) (Fig. 7) and mean NO2(0–1) were linear over most of the variable ranges, and the associations between admissions and mean MDRH(0–21) and MDWS(0–21) were weak and nonlinear but significant. The model showed no significant association between admissions and mean DTGSR(0–21). Therefore, in the final models we used linear terms for MDT(0–21) and mean NO2(0–1) and smooth terms for mean MDRH(0–21) and MDWS(0–21), and we dropped DTGSR(0–21) from the models.

Fig. 6. Relative risk of hospitalization for any cause at 10 °C versus 25 °C (lag 0–21 days) during the cold season,a Hong Kong Special Administrative Region, 1998–2009
Fig. 6. Relative risk of hospitalization for any cause at 10 °C versus 25 °C (lag 0–21 days) during the cold season,<sup>a</sup> Hong Kong Special Administrative Region, 1998–2009
a November to March.
Fig. 7. Relative risk of hospitalization as a function of mean daily temperature during the cold seasona lagged by 0–21 days (MDT(0–21)), Hong Kong Special Administrative Region, 1998–2009
Fig. 7. Relative risk of hospitalization as a function of mean daily temperature during the cold season<sup>a</sup> lagged by 0–21 days (MDT<sub>(0–21)</sub>), Hong Kong Special Administrative Region, 1998–2009
a November to March.

In the cold season model, colder temperatures were positively and significantly (P < 0.0001) associated with admissions across the temperature range observed in the cold season (8.2 to 26.9 °C). For every drop of 1 °C below the top value in this range (26.9 °C) there was a 1.4% (95% CI: 0.9–1.8) increase in admissions. For every increase of 10 mg per m3 in NO2(0–1) there was a 1.2% (95% CI: 0.8–1.5; P = 0.0015) increase in admissions.

In subgroup analyses by cause of admission, the models with linear terms for mean NO2(0–1) and MDT(0–21), and without other pollutants or DTGSR(0–21), were also appropriate for admissions for infectious diseases. For every 1 °C drop in MDT(0–21) from 26.9 °C there was a 1.0% (95% CI: 0.4–1.6; P = 0.002) increase in admissions. For every increase of 10 µg per m3 in NO2(0–1) there was a 0.9% (95% CI: 0.4–1.5) increase in admissions.

For respiratory disease admissions, linear terms were used for MDT(0–21), mean NO2(0–1) and mean O3, lagged by 0–1 days (O3(0–1)), and smooth terms were used for MDWS(0–21) and MDRH(0–21). In this model, every 1 °C drop below 26.2 °C in MDT(0–21) was accompanied by a 4.3% (95% CI: 3.6–4.7) increase in admissions. Every increase of 10 µg per m3 in mean NO2(0–1) was accompanied by a 0.5% (95% CI: 0.03–1.0; P = 0.035) increase in admissions, and every increase of 10 µg per m3 in O3(0–1) was associated with a 2.2% (95% CI: 1.7–2.7; P < 0.0001) increase in admissions. MDWS(0–21) also showed a highly significant (P < 0.0001) association with admissions for respiratory diseases. A sharp decrease in admissions was noted at MDWS(0–21) values of 10 to 20 kilometres per hour (km/h), followed by a slight increase at values between 20 and 24 km/h (which roughly corresponds to the interquartile range for wind speed) and by another sharp decrease at values between 24 and 40 km/h (Fig. 8).

Fig. 8. Relative risk of hospitalization for a respiratory disease as a function of mean daily wind speed during the cold seasona lagged by 0–21 days (MDWS(0–21)), Hong Kong Special Administrative Region, 1998–2009
Fig. 8. Relative risk of hospitalization for a respiratory disease as a function of mean daily wind speed during the cold season<sup>a</sup> lagged by 0–21 days (MDWS<sub>(0–21)</sub>), Hong Kong Special Administrative Region, 1998–2009
a November to March.

For cardiovascular disease admissions, linear terms were appropriate for MDT(0–21), MDRH(0–21) and mean NO2(0–1), but MDWS(0–21) and DTGSR(0–21) were dropped. In this model, every decrease of 1 °C was accompanied, on average, by a 2.1% (95% CI: 1.6–2.5; P < 0.0001) increase in hospitalizations. For every increase of 10 µg per m3 in NO2(0–1) there was, on average, a 1.6% (95% CI: 1.1–2.0; P < 0.0001) increase in admissions. For every rise of 10% in MDRH(0–21), hospital admissions rose by 1.6% (95% CI: 0.3–2.9; P = 0.014). There was no significant association between MDT(0–21) and hospital admissions for renal disease or cancer; however, increases in mean NO2(0–1) were accompanied by significant increases in admissions for these diseases.

We conducted subgroup analyses by age and sex for hospital admissions due to respiratory, infectious and cardiovascular diseases (Table 3). Admissions for respiratory diseases were highly sensitive to cold temperatures among children and people 75 years of age or older, and admissions for infectious and cardiovascular diseases were the most sensitive to cold temperatures among people aged 75 years or older. No strong effect modification by sex was noted.

The number of daily hospitalizations due to accidental causes was not associated with same-day MDT during the cold season, but it showed a significant positive association with MDWS and mean NO2 concentrations (both with lag 0) and a negative association with MDRH. A higher MDT (lag 0) showed a significant positive association with the number of hospitalizations for intentional injuries (assaults and attempted suicides); every increase of 1 °C above the lower end of the temperature range observed during the cold season (8.2 °C) was accompanied by a 2.4% (95% CI: 1.6–3.1); P < 0.0001) increase in admissions.

Discussion

In this study, the high temperature thresholds above which morbidity, as reflected in hospital admissions, increases were 0.3 to 0.8 °C higher (28.5–29.0 °C) than the threshold above which mortality was found to increase in our previous study (28.2 °C).26 Overall, hospitalizations increased by 4.5% for every 1 °C increase in mean daily temperature above 29.0 °C; they also rose by 1.4% for each 1 °C decrease over the range of temperatures (8.2–26.9 °C) observed in the cold season. Thus, in Hong Kong SAR, a subtropical city, elevated temperatures affect morbidity to a greater degree than colder temperatures. However, since the effect of cold temperatures operates over a wider temperature range than that of hot temperatures, over the course of a year excess morbidity during cold temperatures exceeds excess morbidity during hot temperatures.

In subgroup analyses, admissions for respiratory and infectious diseases and accidents increased during both the hot and the cold season, but during the cold season cardiovascular disease admissions also increased. We found that sensitivity to high temperatures varies by age, an effect that has been seen in other studies. For example, in London, during hot weather the risk of having a respiratory condition is especially high among people who are older than 75 years, and the risk of having a respiratory or renal condition is also higher among children less than 5 years old.12,29 In our study, children (aged less than 15 years) were more vulnerable than adults to being admitted to hospital for respiratory conditions during high and low temperatures, but people 75 years of age or older showed a higher risk of being admitted for respiratory conditions only during lower temperatures.

MDRH was found to be strongly associated with the number of admissions for causes other than injuries during the hot season. This was true overall and in most subgroups, especially when humidity hovered between about 72% and 85%. In Hong Kong SAR, MDRH rarely drops below 72% in the summer, and values above 85% usually occur only on cooler days, when the temperature drops below our heat thresholds of 28.5–29 °C. Therefore, our results suggest that higher MDRH exacerbates the effects of high temperatures on health. The decline in admissions when MDRH rises above 85% suggests that high humidity may have a beneficial effect when temperatures cool. Higher MDWS was generally accompanied by fewer admissions for infectious diseases during the hot season and by fewer admissions for respiratory diseases during the cold season. This was an unexpected finding, since high winds can exacerbate the physiological effects of cold temperatures. One explanation may be that higher winds mitigate the physiological effects of high temperatures by facilitating the body’s natural cooling processes. Another possibility is that stronger winds help to clear pollutants from high concentration areas such as roadsides. Thus, for a given measured pollutant level, actual exposure may be less on days with high winds. In a recent study, wind speeds on hot days (mean temperature above 29 °C) were found to have a significant positive association with daily mortality, and the association was stronger in areas with a high urban heat island index.30 This merits further study.

This study has the limitations that are inherent to statistical models. First, we conducted a retrospective ecological study and the ecologic fallacy cannot be ruled out. In addition, we assumed that the increase in hospital admissions resulted solely from an increase in disease occurrence and that all patients sought care from a particular hospital system. In reality, several system-specific factors could have biased the data in the hospital admission records, among them patterns of clinical diagnosis on admission, admission policies and clinical auditing and reporting practices. Although the Hospital Authority covers 83% of all hospital services in Hong Kong SAR and the system provides good clinical services to all patients, regardless of their socioeconomic status or demographic characteristics, a small fraction of patients in Hong Kong SAR probably still prefer to go to private hospitals because waiting times are shorter and they can choose the physician. The retrospective nature of our data also failed to provide the information needed to control for factors that influence the risk of falling ill during hot weather. These include, for example, medical conditions leading to poor temperature regulation,31 ethnic group,24 alcoholism, living on a higher floor32 and activities such as working and exercising outdoors.31 In addition, there is scientific controversy surrounding the accuracy of temperature predictions, which makes it difficult to assess the impact of temperature on health. Our retrospective database was quite comprehensive and spanned more than 10 years, but because of the way in which the data were collected and recorded, it did not allow for fine geographic analyses that might have shed more light on how urban planning and design might affect the interplay between outdoor temperature and morbidity.

This study’s greatest strength lies in the use of meteorological and pollution data covering more than a decade to examine hospital admissions as a function of MDT and pollution levels in a Chinese city. Our finding that hospital admissions are affected by temperature corroborates findings from Taiwan, China,10 and South Australia, where ambulance transport of assault victims has been noted to increase by 13% during heat waves.11 Our findings will contribute to a better understanding of how temperature can affect health in a densely-populated Asian urban setting and will prove useful in the planning of clinical services and of emergency contingency response during periods of prolonged heat. Measures to increase public awareness, educate the public about self-protection and coordinate multidisciplinary and intersectoral responses will be required.


Acknowledgements

The authors thank the Hong Kong SAR Observatory, the Hospital Authority and the Environmental Protection Department for providing the data sets for this study.

Funding:

This work was supported by the Health and Health Services Research Fund of the Food and Health Bureau of the government of Hong Kong SAR [Ref: 08090261].

Competing interests:

None declared.

References

Share