Comparison between our study cities
Our results concerning cool temperature effects are quite similar for Hong Kong and Taipei, two cities with generally similar cool season climates. While the lagged effects of cool temperatures lasted somewhat longer in Hong Kong, 20 days, than Taipei, 15 days, the estimated cumulative increases in mortality for the two cities were quite close, 45% per 10°C drop for Hong Kong vs. 36% per 10°C drop for Taipei. The subgroup analyses also gave similar results for the two cities, with both showing greater effects for non-cancer causes of deaths and generally greater temperature effects for older age groups, although the mortality increase for the 85+ group was considerably smaller for Taipei than for Hong Kong. The estimated pollutant effects were also similar, with O3 being the major pollutant associated with cold season mortality, although the effect size and the duration of lagged effects for Hong Kong were considerably larger. One major difference in the results was the fact that lower RH% was associated with significantly increased mortality in Hong Kong, but significantly decreased mortality in Taipei. This difference is surprising as the distribution of humidity in the cool season is similar between the two cities. A possible partial explanation is that while temperature and humidity are correlated in Hong Kong in the cool season (r = .28) there is no correlation between them in Taipei. Thus the fact that low humidity tends to occur with low temperature in Hong Kong may contribute to its observed association with higher mortality. Differing composition of causes of natural deaths between the two cities may also be a partial explanation. The stratified analyses indicate that the strongest association between higher humidity and higher mortality in Taipei is for stroke deaths. Stroke mortality in Hong Kong was also associated with higher humidity and stroke deaths made up a greater proportion of Taipei natural deaths (10.5%) than Hong Kong deaths (7.1%). Finally random variation may have also played a role in this difference. It should also be noted that the RH associations with mortality in both cities were weak compared to those for the temperature-mortality association.
Comparison of the O3 association between Hong Kong and Taipei is complicated by the difference in measurement units as the conversion between ppb and μg/m3 depends on air temperature and pressure, but the conversion factor is roughly 1 ppb = 2 μg/m3 under average conditions in the cool season in the two cities. Therefore the observed mortality increase of 3.0% per 10 ppb increase in O3 observed for Taipei is considerably weaker than the 4.0% mortality increase per 10 μg/m3 rises in O3 observed for Hong Kong. The mortality increases corresponding to increases in O3 from the 25th% ile – 75th% ile are 3.0% for Taipei and 9.4% for Hong Kong. While sensitivity analyses did show that the estimated ozone effect for Hong Kong was sensitive to the degrees of freedom used for trend variable the conclusion of a strong and significant effect of higher ozone levels on increasing mortality remained unchanged.
Comparison with other studies
Comparisons with other studies is hampered somewhat by differences in methodology, including differences in meteorological variables considered, statistical analysis methods, lags examined, confounder control, and whether whole-year or season-specific analyses were performed. The ISOTHURM study , which looked at temperature effects in several cities in developing countries worldwide, used lags 0–13 mean temperature to evaluate cold weather effects. Their methodology differed from ours in that they consider year round data. One of their studied cities, Monterrey, Mexico, has a sub-tropical climate. The estimated increase in mortality for 1°C drop in lag 0–13 mean temperature was 4.70% below a threshold of 17°C , a stronger increase but also a lower threshold than what we observed for Hong Kong and Taipei. Their results for Bangkok, Thailand, which has a tropical climate, showed a somewhat stronger effect size, 4.09% rise in mortality per 1°C temperature decrease, and a high threshold for cold effects, 29°C . Both Hong Kong and Taiwan are more economically developed than either Mexico or Thailand, however it is unclear to what extent this would affect cold weather-mortality associations as many homes in our study locations also do not have central heating. A study from the U.S. included Tampa, a sub-tropical city, and Miami, a tropical city, and the mortality increases per 1°C drop below the ‘minimum mortality temperature’ (MMT) for these cities (around 27°C) were 1.0% and 1.3%, respectively . This is a considerably weaker effect than we observed, however, the authors only considered lags up to 3 days, which may be too little to capture all of the cold effect. A study from Europe  used the average of lags 0–15 minimum apparent temperature to estimate cold weather effects on mortality during the cold season, October-March. This study did not include any cities with sub-tropical or tropical climates but did find a stronger effect for the warmer ‘Mediterranean’ cities, 1.62% natural mortality rise per 1°C temperature drop, than for ‘North-Central’ cities, 1.16% . Similar to our study they found that cold weather was more strongly associated with deaths among older people, and for deaths due to respiratory or cardiovascular disease . A study from a sub-tropical area of rural Bangladesh which used the average of lags 0–13 mean temperature estimated a 3.2% increase in natural mortality per 1°C drop in temperature across the temperature range observed, with stronger associations for cardiovascular and respiratory mortality and mortality among children and the elderly . This is quite close to our estimates for Hong Kong and Taipei despite a very different mortality profile for this area, nearly one-third of the deaths occurred among children 0–14 years of age .
Many of the previous studies which examined the association between specific cold waves and mortality did not adjust for the effects of individual cold days. Studies from the Netherlands , Czech Republic  and Russia , all found significant excess mortality during cold spells but since the short-term effects of daily temperature were not adjusted it is unsure whether or not some of the excess represented the additional effect of having several cold days consecutively. A study using data from 3 cities in Guangdong province, China examined the short-term effect of a severe 2008 cold wave, which also affected Hong Kong, and found that mortality was increased by 60% in two of the cities during and immediately after the cold wave compared to similar periods for 2006–7 and 2009 . This study also did not adjust for the general effect of cold temperatures. A recently published study  using data from 1994–2007 for four Taiwanese cities, including Taipei, found no additional effect of cold waves on mortality once the effects of individual cold temperature days, with lags up to 30 days, were accounted for. A study using data from 1987–2000 for 99 U.S. cities examined the additional effect of cold waves after adjustment for the cold temperature mortality response . This study found no additional cold wave effect and in fact found a small but statistically significant decrease in deaths for the ‘coldest’ cold waves, those with temperatures below the 1st percentile . Our study found an additional cold wave effect on raising mortality after controlling for the general effect of cold temperatures which was significant in Hong Kong and borderline significant in Taipei. Our finding implies that there is evidence that the effect of cold days is greater if they occur in sequence. One possible explanation for the difference between our finding and those from the Taiwanese and U.S. studies is differences in the definition of cold waves. The Taiwanese study, while considering several temperature thresholds and minimum number of consecutive cold days when defining cold waves , did not consider the potential lagged effects of cold waves, i.e. the possible persistence of increased mortality after the cold wave has finished. Given that substantial and prolonged lagged effects exist for cold temperatures in general it seems reasonable to assume that they would also exist for cold waves. The U.S. study did consider the possible lagged effects of cold waves, up to 7 days , but used a looser definition of cold waves, 2 consecutive days with cold temperature, vs. 5 for our study. In addition our cold wave indicator did not take a value of 1 until the 5th consecutive day of cold temperatures. We feel that this is appropriate since the fact that the first few days of cold weather eventually became part of a cold spell would not retroactively affect mortality on those days. Also most of the cities included in the U.S. study had considerably colder climates than our study cities, and thus would be expected to have weaker cold temperature, and possibly cold wave, effects on mortality.
Ozone levels have been found to have short-term associations with mortality in many studies. A meta-analysis of the results of 43 studies found that most observed a positive association between ozone and mortality , although considerable heterogeneity was observed. Their pooled estimate was for a 1.6% increase in natural mortality for each 20 ppb increase in 24-hour mean ozone . Our estimated ozone effects were considerably stronger, especially for Hong Kong.
Potential scientific explanations for the associations observed in this study
Cold stress has been found to result in blood pressure rises [21–23], cardiac hypertrophy , increased blood viscosity , and increased platelet counts [21, 23]. A prospective study looking at the association between air temperature and risk factors for ischaemic heart disease concluded that the most important effects of cold temperature were on the haemostatic system, including increases in fibrinogen, and α2 macroglobulin . A French study found that outdoor temperature and blood pressure were strongly correlated in the elderly . The elderly also have a higher prevalence of co-morbidities which would make them more sensitive to cold effects.
One reason cold days occurring in sequence may have a greater effect than when they occur in isolation may have to with the fact that many homes and workplaces in the study cities do not have central heating. Individual cold days occurring in the midst of several relatively warm days may not lower the temperature of indoor spaces in these cities very much. However several cold days in a row would result in indoor spaces becoming quite cold in the absence of a heating source. Thus the actual exposure to cold temperatures could be greatly increased by having the cold days occur in succession. This would be particularly true for subpopulations that are not particularly active, such as the elderly or those with chronic diseases.
Our findings regarding the association between relative humidity and mortality were inconsistent across cities and between different causes of death. Lower relative humidity has been found to be more favorable for transmission of the influenza virus  and this could lead to increases in mortality. The observed protective effect of higher solar radiation in reducing cold season mortality in Hong Kong may reflect the potential effects of sunshine to warm buildings and outdoor urban areas.
The observed ozone-mortality associations for our study were especially strong for respiratory and cardiovascular mortality, particularly for Hong Kong. There is evidence that exposure to ozone causes decreases in lung function due to reductions in inspiration capacity caused by sensitization of bronchial C-fibers and small airway narrowing . A recent controlled-human-exposure study  found that ozone exposure was associated with increases in vascular inflammation markers and pulmonary inflammation, decreases in lung function and changes in markers of fibrinolysis and markers that affect autonomic heart rate control.
Our study results indicate that cold temperatures result in substantially elevated short-term mortality rates in these sub-tropical Asian cities. Although cold weather cannot be prevented steps can be taken to reduce the exposure of vulnerable groups to cold temperatures. Promoting further awareness of the adverse health effects of cold weather among the public could help persuade individuals to take steps to reduce their exposure, including dressing more warmly, and keeping their homes and offices warmer. Measures to reduce excess mortality and morbidity have been found to be effective for hot weather [28–31] and should be for cold weather, e.g. issuing cold weather warnings, provision of warm public places and suitable clothing, particularly for people at high risk such as the elderly and those with chronic diseases. Since temperatures can be forecast a few days in advance a warning system could be set in which people belonging to high risk groups could be warned in advance when weather conditions with the potential to cause adverse health effects are coming .