Emerg Med J 29:60-64 doi:10.1136/emj.2010.096701
  • Prehospital care

Weather inference and daily demand for emergency ambulance services

  1. P C Lai
  1. Department of Geography, The University of Hong Kong, Hong Kong SAR, PR China
  1. Correspondence to H T Wong, Department of Geography, The University of Hong Kong, Pokfulam Road, Hong Kong SAR, PR China; frankwong{at}
  • Accepted 29 September 2010
  • Published Online First 28 October 2010


Aim To examine weather effects on the daily demand for ambulance services in Hong Kong.

Methods Over 6 million cases of emergency attendance from May 2006 through April 2009 (3 years) were obtained from the Hospital Authority in Hong Kong. These cases were further stratified by age, triage levels, hospital admission status, comprehensive social security assistance (CSSA) recipients and gender. The stratification was used to correlate against weather factors to assess the dependency of these variables and their effects on the daily number of ambulance calls. Adjusted-R2 values obtained from the regression analysis were used as a measure for evaluating predictability.

Results The adjusted-R2 of emergency cases by age groups showed proportional correlation with weather factors, which was more significant in older patients (0.76, p<0.01) than young patients (0.10, p<0.05). Furthermore, patients with more severe conditions were shown to have a higher adjusted-R2 (0.63, p<0.05 for critical as opposed to 0 for non-urgent patients). Weather effects were also found more significant in women (0.50, p<0.01) and CSSA recipients (0.54, p<0.01) when compared against their corresponding reference groups (respectively men at 0.46, p<0.01 and non-CSSA recipients at 0.45, p<0.01). Moreover, average temperature appeared to be a major weather effect.

Conclusions The presence of strong weather effects among different target groups indicates possibility for the development of a short-term forecast system of daily ambulance demand using weather variables. The availability of such a forecast system would render more effective deployment of the ambulance services to meet the unexpected increase in service demands.


Biometeorologists have claimed that weather exerts long and short-term effects on our bodies. Bad weather (heavy rain and extreme temperatures) have been blamed for many road accidents.1 2 Death rates have also reportedly increased with worsening weather conditions of extreme heat or cold.3 4 People of different traits (age, gender, general state of health, place of residence, occupation, etc) react differently to weather conditions. It is noteworthy that the old, young and patients with chronic illness suffer more whereas women are generally more sensitive than men to weather changes.3 4 Some patients with rheumatism or arthritis, for example, can notice changes in the atmospheric pressure causing pains in their joints.5 Older people are most susceptible to high temperatures while many young men drown each year trying to cool off with a swim. The risk of falling indoors due to hypothermia among older women is higher because of inadequate heating in cold weather.6

Two important global trends are happening—global warming and ageing population. Global warming and noticeably warmer urban temperatures due to heat island effects have been widely documented.7 Ageing population in both developed and developing countries has also drawn worldwide attention.8 As Hong Kong is a densely built city without serious planning consideration on street air ventilation, her urban heat island effect is widespread.9 At the same time, the demographic trend of Hong Kong's population shows a shrinking labour force and ageing population.10 The inevitable increase in the frequency of adverse weather occurrences and the growing ageing population will ultimately impose a heavier demand for emergency medical assistance. More prudent planning and better forecast are necessary to manage the anticipated increase in usage demand.

Our study aims to study weather effects on emergency ambulance demand of Hong Kong. Located in the southern part of China bordering the Guangdong Province, Hong Kong is a hilly area with many outlying islands. It has a sub-tropical climate tending towards moderately warm temperatures for nearly half of the year.11 However, it is not uncommon for the temperatures to plunge below 10°C or exceed 32°C in a relatively short time.

Yan attempted to investigate seasonal effects of Hong Kong weather on mortality from 1980 to 1994.12 Her results showed a significant negative relationship between minimum temperatures and mortality and a significant positive association between cloud cover and mortality. Although weather and mortality had no significant association in the age groups of 0–24 and 25–44, a strong and significant association was discovered in the age group of 65 and above (adjusted-R2=0.49). Furthermore, we previously demonstrated that common weather factors of Hong Kong had a significant correlation with monthly totals of ambulance calls.13 We found that the lowest temperatures in cold months were negatively associated with the daily average number of ambulance calls (adjusted-R2=0.38), but the average amounts of cloud and highest temperatures in hot months were positively associated with the daily average number of ambulance calls (adjusted-R2=0.34). Moreover, in that study we also demonstrated seasonal differences in which the daily average ambulance call in both summer and winter seasons was about 3% higher than that in the remaining months (p=0.00).

Although some research has studied the relationship between ambulance demand or mortality and weather factors of Hong Kong, the use of monthly weather data may not be of sufficient sensitivity to extract true weather effects. In this regard, our research aims at scrutinising daily weather statistics and their effects on the daily ambulance demand.

Data and methods

Ambulance demand and metrological statistics

Over 6 million records (over a 3-year period from May 2006 to April 2009) of emergency attendance at accident and emergency departments of all hospitals managed by the Hospital Authority (HA) in Hong Kong were obtained for this study. Each record is anonymised to contain an arbitrarily assigned but unique patient record number, age, gender, triage level, date, ambulance brought-in indicator and an indicator on whether the individual is a recipient of the Hong Kong Comprehensive Social Security Assistance (CSSA) scheme.

Daily statistics of common weather variables from May 2006 to April 2009 were obtained from the website of the Hong Kong Observatory.14 The weather variables selected include average temperature, relative humidity, pressure, percentage of cloud coverage, rain amount and wind speed. In addition to the above weather factors, we also computed numerical variants of these weather variables to embrace the possible non-linear nature of weather effects. The numerical variants include ‘average temperature square’, ‘difference of average temperature’, ‘sum of the difference of average temperature’ and ‘relative humidity times average temperature’.

The ‘average temperature square’ was meant to capture the quadratic nature of seasonal weather effects fluctuating between very hot and very cold days (as exhibited by the U shape relationship between average temperature and ambulance demand). The ‘difference of average temperature’ and the ‘sum of the difference of average temperature’ were intended to underscore consecutive days that registered a large temperature change. Finally, the ‘relative humidity times average temperature’ was used to represent the combined or interaction effects between temperature and relative humidity.


The research employed multiple regression analysis in two stages. Stage 1 of the analysis explored the underlying trend of ambulance data and simple relations with weather variables. We conducted lag effect analysis to test time lag effects of weather variables on the population by various age groups. Stage 2 made use of the forward variable selection to establish a series of regression models based on weather factors and by different target groups. We adopted the adjusted-R2 measure to assess the performance of our multiple regression models and to facilitate comparison of our results against those of other studies.12 13 15 The adjusted-R2 value for each model was computed to ascertain the predictive power of the independent variables on the dependent variable in which a larger adjusted-R2 value means a higher proportion of the explained variance. Independent variables rendered as significant through simultaneous testing and modelling will be adopted in the penultimate regression model.

Data analysis and results

Data pattern and pre-processing

The time-series plot of the number of daily ambulance cases (figure 1) shows a pattern of random fluctuation. There were many data points with extreme values. Moreover, the series seems to follow a linearly increasing trend.

Figure 1

A time-series plot of the number of daily ambulance cases (May 2006 to April 2009).

Three pre-processing steps are necessary to eliminate outliers that represent noise in our data analysis. First, data normalisation to remove the trend effect was achieved by multiplying a scaling factor to data of the first two years to standardise the total number of ambulance calls and render them the same as that of the third year. Second, data records occurring on public holidays of Hong Kong were identified. Holidays that consistently registered extremely low or high numbers of ambulance calls every year during our 3-year study period were removed from the series. The first day of the Chinese New Year, for example, had an extremely low number of ambulance cases every year because the Chinese believes that bad luck or ill health shall prevail throughout the year if one was attended at a hospital on the day. Although Hong Kong had experienced sudden demand for ambulance services arising from mass gathering events (festive celebrations, concerts, sporting events, etc) and epidemics,16 such events did not occur during the data period and was thus irrelevant for this research. Finally, the data series applied the technique of three-day moving average to smooth out short-term fluctuations that bring into focus longer-term trends or cycles. These three pre-processing steps helped clean the data series to allow for better visualisation and comparison of seasonal fluctuations on the demand for ambulance services.

Lag effect analysis

We made an attempt to examine lag effects of weather variables on ambulance attendance because lag effects of weather on mortality had been reported in previous research.17–19 Carder,19 for example, demonstrated that lag effects of mortality induced by cardiovascular and respiratory diseases could take as long as 3 weeks but the lag effects for all causes of death ranged between 1 to 6 days. Here, our quadratic regression models were constructed by using average temperature as the independent variable and considering lag effects of 1 to 5 days.

Table 1 shows the results of lag effect analysis of 1 to 5 lag days. The two most important target groups (all age groups and age group of 65 and above) were chosen for the analysis. Our results showed that 4 to 5 lag days produced the highest adjusted-R2 (0.470 on 5 lag days for all age groups and 0.714 on 4 lag days for 65+ age group). This result is consistent with that obtained by Braga and Leung17 18 who also showed a few days of lag effect. Our results further indicated that the older persons group had a shorter and more significant reaction time lag on weather effects compared to the population at large. A 4-day time lag was thus adopted for subsequent analysis to cater to the older population. The quadratic regression model in figure 2 showed that average temperature alone was a sufficiently good predictor of the number of daily ambulance cases for patients aged 65 and above (adjusted-R2=0.714 on 4 lag days).

Table 1

Lag effect analysis of two target age groups

Figure 2

A quadratic regression model of daily ambulance cases for patients aged 65 and above (using 4-day time-lag average temperature as a predictor).

Weather effects on daily number of ambulance cases by different target groups

Multiple regressions of time series data by different target groups were approached using the forward variable selection and a lag effect of 4 days. Adjusted-R2 values were employed as a measure to assess the predictive power of weather factors (the independent variables) on the daily number of ambulance cases (the dependent variable) by various target groups (age, triage level, hospital admission status, CSSA standing and gender).

To be absolutely certain that the respective causal relationships as reflected through our regression models do exist between weather factors and daily ambulance calls, the concept of clinical significance was adopted in this research.20 Given the large number of potential independent variables, some independent variables of statistical significance might not be meaningful in the clinical sense. To be clinically significant, the product of two times the SD and regression coefficient of an independent variable must be 1% larger than the mean of the dependent variable. This criterion was used to discriminate independent variables of clinical significance for inclusion in the penultimate regression model for further analysis. Moreover, the maximum variance inflation factor of all variables, except temperature and humidity related variables already verified pertinent in previous research,21 was capped at five, which was lower than the suggested cut-off point of 10.22


The ambulance records were aggregated by age groups: children (<15), young adults (15–34), middle age (35–64) and older people (≥65). The adjusted-R2 values of weather effects by different age groups based on 4-day time-lag data are summarised in table 2. Age was found to be directly proportional to the adjusted-R2 in which 76% of ambulance cases among older people could be explained by weather factors. Children, on the contrary, were indifferent to weather variables. Interestingly, regression coefficients of the ‘sum of average temperature difference’ and ‘pressure’ became statistically significant for the middle age group and very significant for older people. This observation is in line with the fact that the older population tends to have difficulties coping with sudden changes in the atmospheric temperature12 and pressure.23 It was also evident that ‘average temperature’ remained the dominant weather factor in the regression models.

Table 2

Results§ of regression analysis by various age groups (4-day time-lag data)

Triage level

Patients transported by emergency ambulance vehicles must be sorted into one of five triage categories as far as possible and according to patients' clinical conditions: critical (level 1), emergency (level 2), urgent (level 3), semi-urgent (level 4) or non-urgent treatment (level 5). It is reasonable to expect that weather implications on different triage groups tend to be either psychological (for non-urgent cases) or physical (for critical cases) since non-urgent cases are normally without serious illness. Table 3 shows that patients of triage levels 1 to 3 had a relatively higher adjusted-R2, (0.63 for levels 1 and 3, and 0.51 for level 2) although the slightly lower adjusted-R2 for triage level 2 was unexpected. The less severe cases at levels 4 and 5 yielded very low adjusted-R2 values (0.22 and 0 respectively). Furthermore, the ‘relative humidity’ factor and its interaction effect with average temperature were found insignificant for the more severe patients while ‘average temperature’ emerged again as the dominant weather factor in the regression models.

Table 3

Results of regression analysis by triage level (4-day time-lag data)

Hospital admission status

The number of hospital admission is an important concern given that bed capacity is often fixed. Consequent to the our earlier finding that patients of more severe conditions by triage level were highly affected by weather factors, a large adjusted-R2 value of 0.71 registered by the hospital admission status (table 4) confirmed the need for more hospital beds in certain weather conditions. Our regression analysis also showed that the adjusted-R2 of 0.71 was three times higher than that of non-hospital admission at 0.22. Moreover, ‘average temperature’ seemed to have more impact on hospital admission status.

Table 4

Results of regression analysis by select target groups (4-day time-lag data)


Past research has indicated that women were more sensitive than men to weather changes.3 4 Our results based on gender stratification showed that both genders reacted in almost similar manners to the same set of weather variables (table 4). Females were shown to have a slightly higher adjusted-R2 against their male counterpart (0.50 and 0.46 respectively). Our result was consistent with Yan's findings.12

Comprehensive Social Security Assistance recipients

CSSA recipients of Hong Kong are able-bodied adults whose individual asset is below HK$22 500 and who have passed the income test.24 Given that the median monthly domestic household income of Hong Kong residents in 2006 was HK$17 250,25 the CSSA recipients could qualify effectively as the lower income groups or the working class. This group of individuals would likely have difficulties protecting themselves from adverse weather conditions and thus are more vulnerable. The adjusted-R2 of 0.54 (which was 0.09 higher than non-recipients; table 4) also supported our speculation.


Our research establishes a few significant causal relationships between weather factors and the daily total of ambulance cases by different target groups. Moreover, we also evaluate the predictability of daily number of ambulance calls derived from various multiple regression models regressed on weather factors and by target groups.

We conclude that weather factors can reasonably predict the demand for emergency ambulance services by the following target groups: older people, more severe patients, hospital admitted cases and CSSA recipients. Our finding about older people is consistent with that observed by Yan12 who showed detectably stronger weather-mortality association for the older persons group. The corollary of insignificant association between weather and ambulance demand for children was also in line with a previous study by Attia and Edward26 who reported negligible weather effects on the number of visits to a paediatric emergency department. We detected a close association between adverse weather conditions and severity of illnesses/injuries which agreed with an earlier study by Tai et al15 who showed that low temperatures were linked with more severe triage sort. Although gender did not yield significant differences against weather factors, CSSA recipients representing lower income groups were found more exposed to weather effects. This finding confirms speculations of previous research which could not be substantiated at the time without a suitable income-related indicator.

Limitation and conclusion

An obvious limitation of this research concerns data completeness. Our ambulance data reflect only cases transported to and attended at accident and emergency departments of HA hospitals as reported and recorded in HA information systems. Moreover, false alarm ambulance cases and patients pronounced dead before arrival at a hospital were not accounted for in this data set. These demands are important since significant resources are consumed. More representative data could be derived from call records of the Fire Services Department, which is the managing authority of ambulance services. We also caution that the temperature relationships derived from our study may not apply in other geographical settings and populations. For example, people residing in the equatorial and polar regions may find the so-called ‘extreme’ temperatures in Hong Kong just right for them.

The research has significant practical and policy implications. A better understanding of the relationships between weather characteristics and patterns of ambulance run as we have derived from this research will enable more ‘intelligent’ forecasts of the short-term demand for emergency ambulance services based on specific weather conditions. Better prediction is the key to formulating logistic strategies for efficient deployment of ambulance services in rapid response time to meet demands.

We intend to extend our study to include spatial element by examining the distribution of ambulance depots to assess capacity against demand by geographic locations. Better ambulance logistics can be achieved by prioritising demand hot spots and recognising resource deficient areas. Changing supply to match demand based on reliable predictions means that potential victims can be served in a swift and efficient manner on the one hand, whereas unnecessary operational expenses (such as acquisitions of additional ambulances and manpower), can be reduced on the other.

The inferred relations between weather factors and ambulance demand, as demonstrated in our study, signal their potential roles in the long-term projection of ambulance need for Hong Kong. Even though the present demand for ambulance services was attributed mostly to extreme cold weather, as evident from figure 2, hot weather expects to exert more significant impacts given the global trend of rising temperatures. Along with the ageing population and their greater susceptibility to health problems in severe weather events, our ability to make long term ambulance demand forecasting has important implications on the sustainability, quality and equity of services.


We are grateful to the following government departments of the Hong Kong Special Administrative Region for access to data records used in the present study: Census and Statistics, Hong Kong Observatory and Hospital Authority.


  • Funding We are grateful for funding support from the University of Hong Kong GRF Incentive Award 2008-09 for the acquisition of research data.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.


Relevant Article

Free sample
This recent issue is free to all users to allow everyone the opportunity to see the full scope and typical content of EMJ.
View free sample issue >>

Don't forget to sign up for content alerts so you keep up to date with all the articles as they are published.