Relation Between Temperature And Mortality In Shanghai Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Although the relation between temperature and mortality has been studied for many decades public health threat due to elevated temperature did not really become recognized as an issue with considerable magnitude until the late 1980s. To date, there is coherent evidence that air pollution has short-term effects on mortality.

Increased atmospheric concentration of greenhouse gases, of which CO2 is the most important, may increase the average temperature by 2-5.5 degrees (Celsius) by the end of this century (IPCC, 2001). Behind this average, with so-called "urban heat island effect", people living in urban area are especially vulnerable, not only because urban areas typically have higher heat indexes (combinations of temperature and humidity), but also because generally the urban population is more aged which naturally deteriorate the capability of people to do adjustment with un-nice weather (Basu et al. 2002).

A close reading of the IPCC 2007 report indicates that climate change will affect all countries and in general undermine the sustainability of the livelihoods of millions, but the worst impact will fall on developing countries, partly because of their geographical location, partly because of weak coping capacities, and partly due to more vulnerable social, institutional and physical infrastructure. This conclusion could indirectly be supported by the fact that to date there has been very little epidemiological research on the health impact of thermal stress on urban population of developing countries (Gouveia et al. 2003).

In this paper, we apply a time series modeling framework to specify and estimate a model for the relationship between temperature and mortality in the city of Shanghai¼Œa warm temperate zone city in the Asian Pacific rim, conditional on long-term trends in demographic characteristics, social & economical development, seasonal effect, and environmental condition.

As far as we know, the analysis in this paper is the first attempt to undertake a long term temperature-air pollution-mortality relationship study in China. Even in Asian Pacific region, the study based on daily mortality count is still quite rare. One in Beijing, China based on 1-year daily data, two in Australia, two in Korea and one in Hong Kong.

The daily meteorological data in Shanghai for 1956-2001 is extracted from the dataset of Daily Surface Climate Normal of International Exchanging Stations of China prepared by China Meteorological Administration. The daily death counts for all non-accidental causes (International Classification of Diseases, Revision 9[ICD-9] <800) in 14 sampled years (1956, 1961, 1963, 1966, 1968, 1971, 1976, 1981, 1986, 1990, 1991, 1996, 2000 and 2001) in 7 sampled districts (Changni, Hongkou, Luwang, Nanshi, Xuhui, Yangpu and Zhabei) in the city of Shanghai is extracted from the dataset digitalized by Vita Statistics Department, Shanghai Center of Disease Control and Prevention. The daily air pollution data for 1998-2006 is directly from the website of Shanghai Environmental Protection Bureau (

The remainder of the paper is structured as follows. In section 2, the 45 years daily temperature data is intensively analyzed with the aim that we try to summarize the seriousness of the climate change threat. Section 3 some summary statistics of mortality outcome, air pollution levels, and meteorological measures by season. In section 4, the General Additive Models (GAMs) is presented and we present the estimates results. Exposure-response relationships regarding with three key air pollutants in warm and cold seasons were also intensively examined using GAMs modeling. In section 5 comes the conclusion.


The daily meteorological parameters include daily mean, maximum and minimum temperature, relative humidity, barometric pressure and wind speed. We consider these as meteorological proxy representing whole Shanghai temperature level.

The analytical approach followed standard time series methods that have been developed for financial econometrics.

Suppose the time series for temperature can be decomposed into a macroscopic component and a microscopic component. The macroscopic component is described through a trend and seasonality, whereas the microscopic component is picked up by a noise. We assume the trend, the seasonality and the noise take an additive form.


whereis the temperature in period,is the time trend in period, andis the noise in period.

First, we estimate the trend part by a moving average filter over a complete cycle so that the effect of the seasonality is averaged out. This method is approved by Ngai Hang Chan (2002). In our data, we set the length of the filter at 6. Then a 13 point moving average is used to represent one year length effect. That is



After estimating, filter the trend out from the data. Next we apply a moving average method (Brockwell and Davis, 1991) to estimate the seasonal part from the residual. We assume that each month has a seasonal effect, that is, and satisfies.


Then,gives us the deseasonalized time series that include the trend and noise only.

To get the pure noise, we re-estimate the trendfrom the deseasonalized time seriesby applying the same filter as in formula (2).gives us the pure residual.

With the model set up above, we estimate the trend, seasonality and noise in time series for temperature to test whether the temperature has the intendancy (combination of global warming and the local urban heat island effect) to increase in Shanghai in the past half century and more importantly how much the increase is.

In figure 1, the top plot is for the raw temperature data from year 1956-2001. The second plot is the time series after deseasonalized. In which we suspect that it has an increasing tendency with the time. We examine this tendency further in the next paragraph. The third plot gives us the seasonal cycle that is estimated by moving average method. The last plot is the residuals after deseasonalized and detrending, from which, we could recognize a decreasing variance associated with time. We come to this tendency later too.

Figure 1 Deseasonalized and detrend time series for temperature in Shanghai

Based on the increasing tendency in the second figure in Figure 1, the deseasonalized time series is regressed on the time. Figure 2 shows the fitted regression line and table 1 gives the estimates. From 1956 to 2001, the temperature increases almost 1 Celsius degree in the linear prediction. Coefficient for timeis positive and significant at 1% level.

Figure 2 Time trend fitted for deseasonalized temperature data in Shanghai

Table 1 Time trend test for deseasonalized data


Standard Error




Coefficient for



Next, let us examine the decreasing tendency in variance of residual. We start to analyze this tendency from three dimensions. The first dimension is the difference between extreme high temperature and extreme low temperature for each year. The second is the extreme high temperature and the third is extreme low temperature. This first figure in Figure 3 shows that the difference between highest and lowest temperature becomes smaller in the recent past. The next two figures show that both the extreme high and extreme low temperature in each year tends to increase and the extreme low temperature increases at a higher speed, which gives the answer to why the variance of the residual in recent past year decreases.

Figure 3 Analysis for temperature changing from 1956 to 2001

All these tendencies in figure 3 are supported by regression results in table 2. All estimates in table 2 have the expected signs and significant at 1% level.

Table 2 Time trend test for temperature difference, extreme high and extreme low

Model 1, Temperature difference between Extreme high and Extreme low as Dependant variable


Standard Error




Coefficient for



Model 2, Extreme high temperature as Dependant variable


Standard Error




Coefficient for



Model 3, Extreme low temperature as Dependant variable


Standard Error




Coefficient for



From above analysis, we get the evidence that the temperature increased in the past several decades which supports the prediction as well as the intuitive social feeling of the weather is getting warming. Furthermore, the yearly extreme high and extreme low temperature tends to increase and especially the extreme low temperature increases at a higher speed. This tendency has important effects on the mortality not only in the sense that they are efficient in promoting mortality, but also in the sense that unusual high temperature and unusual low temperature have different effects on specific cause mortality which lead to different public health intervention strategies. Finally, the decreasing difference between extreme high and low temperature tells us a new weather pattern comes with the feature of unclear seasonality.


All pollutant levels are unacceptable high in Shanghai. Although referring with Chinese standard, Shanghai is regarded as advanced city in environmental protection comparing with Beijing, Guangzhou, and other big cities. From the period of 1998 to 2007 (3149 days), for PM10 (24-hr) is about 29.1% days, for NO2(24-hr) is about 50% days, and for SO2(24-hr) is about 10% days in Shanghai have reached the World Health Organization (WHO) air quality guidelines ( The situations become further worse in cool season, the proportions decrease to 23.9%, 39.2% and 6.8% respectively. If we consider the proportion of days with all three contaminants (PM10, NO2, SO2) reaching WHO guidelines, the figure is about 5%, and if the daily data of O3 were public available, with inclusion of O3, it is quite plausible that the proportion will converge to zero.

Table 3 Guideline values and the key health effects


Guideline values (µg/m3)

Key health effects


Value Averaging time


50 24-hour

20 Annual

Mortality, morbidity, hospitalization, work-affected days, increased use of medication.


200 1-hour

40 Annual

Apparent contribution to morbidity and mortality, especially in susceptible subgroups, including young children, asthmatics and those with chronic inflammatory airway disease.


20 24-hour

500 1-minute

Daily mortality, hospital admissions and emergency room attendances for respiratory and cardiovascular disease, increases in respiratory symptoms and decreases in lung function.


100 8-hour

Increased daily mortality, respiratory and cardiovascular disease; decreases in lung function; increases in hospitalizations, and in respiratory illnesses such as cough, phlegm and wheeze.

Table 4 Summary statistics of mortality, air pollution, and meteorological measurement






Mortality counts

(*for year 1996, 2000, 2001)

Nonaccident (ICD:<800)















Air pollution concentrations (μg/m3)¼ˆ*from 1998.5.28 to 2007.1.15¼‰














































Meteorological measurement

(*from 1956-2001)

Daily Average Temperature(0.1℃)















Daily Average Humidity (%)















Note: Warm season covers the period from October to March; cold season covers the period from April to September.

For the past 20 years, Shanghai has experienced profound industrial structure adjustments to transform a manufactory-based economy to a service-based one which has significant positive effects on air quality, especially, helps a lot with the decrease of sulfur content fuel consumption. The problem is like SO2, PM10 are some kind classic trans-boundary pollutants, without efficient regional environmental coordination, Shanghai's effort could be greatly offset by the free-rider behavior from the surrounding cities. Meanwhile as other Asian metropolitan cities, for the past 20 years, Shanghai also has experienced unprecedented traffic boom due to strong demand for automobiles from household. Therefore for the contaminants related with automobiles, it is quite likely NO2 and O3 are still on the track of robust increasing, or at least not going to decrease as easily and quickly as PM10 and SO2 do, although both Chinese central government and local Shanghai government has committed with the reduction of pollution emission.

The analytical approach for air pollution trend followed the same times series methods that have been deployed for temperature studies in the former part.

In figure 4, the upper three plots are for the raw air pollution data from year 1998-2007, the lower three plots are the time series air pollution data after deseasonalized. In which we suspect that PM10 takes quadratic form in time and had achieved its peak value in 2003 and currently is on the stage of decreasing path; SO2 takes cubic form in time and has unclear sigh in the long term; for NO2, we could recognize a significant decreasing trend which converges to some constant level at least in short term.

Figure 4 Original & deseasonalized time series data for air pollution level in Shanghai

We use a set of linear regression models to fit the deseasonalized time series pollution data based on the assumed model specifications we observe from the figure 4.

Table 5 gives the estimates for three fitted regression. All coefficients for timeare positive and significant.

Figure 5 Time trend fitted for deseasonalized pollution data in Shanghai

Table 5 Time trend test for three main air pollution contaminants in Shanghai

Model 1, Daily PM10 value as Dependant variable


Standard Error




Coefficient for



Coefficient for2



Model 2, Daily SO2 value as Dependant variable


Standard Error




Coefficient for



Coefficient for2



Coefficient for3



Model 3, Daily NO2 value as Dependant variable


Standard Error




Coefficient for



With the statistics models above, we could conclude with carefulness that Shanghai is now on the transition stage evolving from a most polluted city to a clean one, although some progress looks quite promising, while some a little bit weak. Comparing other Asian big cities, like Hongkong (Wong et al. 2001) and Seoul (Lee et al. 1999), Shanghai is about 10 years left behind in environmental protection. We start from model 1 in table 5, as we expected, from 2003 the slope of PM10 path became negative with average annually decrease rate 5µg/m3 which means with the given trend by the end of 2009 the daily expected value of PM10 in Shanghai would reach WHO guidance. From model 2 in table 5, there is no clear sign how SO2 will be going on especially in the long term. Since SO2 mainly has adverse effect on respiratory and cardiovascular disease, combing the population ageing, we would give more detailed analysis for this combination effect in section 4. Model 3 in table 5 shows a cheering trend NO2 is moving to the intercept value 39.92 which is exactly the value recommended by WHO.


The death counts per day were matched to the daily-averaged levels of temperature, humidity, PM10, SO2, and NO2 on the same day.

We used a generalized additive models (GAMs) with nonparametric smoothing functions (splines) to describe nonlinear relations between temperature, air pollution and mortality (TM) in Shanghai. We specify a GAMs especially designed to take into account long-term trends in demographic characteristics, social & economic development and seasonal effect, similarly to the analytical framework in Curriero (2002), as well as routine temperature and air pollution variables. We further estimate the TM by era, by specific communities to see the variation in time and in the inside-city.

In order to test whether there are independent effects of a single pollutant to account for a health outcome in the city of Shanghai, we again used a Poisson regression with daily mortality counts as the dependent variable. It is also highly interesting to test whether there are thresholds and linear (non-linear) relationships of TM. Since a similar research has been done in Hong Kong (Wong et al. 2001); therefore it is possible we could do some comparison analysis.

It is widely believed that the local air pollution policy should be based on local exposure-response relationship. In this regards, this research do significant contribution to the understanding of the health effects of air pollutants in the city of Shanghai.

The basic idea behind the GAMS is the function of TM is smooth but not necessarily linear which has self evidence advantage in its flexibility and efficiency in explaining the nonlinear relation between mortality and temperature and other independent variables, like pollution.

We consider using calendar time to serve as a catch-all variable to represent the unobserved events like long-term population structure change, improvement in individual nutrition, access to air-conditioner, trends in public health practice, etc.

Consider temperature, humidity and air-pressure have not only immediate effect on mortality but also lagged ones. We included same-day temperature/ humidity/ air pressure,, and adjusted lagged weather variables including preceding 3 days average temperature/humidity/air pressure ,,and adjusted preceding 4-10 days average temperature/humidity/air pressure ,,(the lag period have an exponentially increasing size.). The reason we use adjusted weather variables rather than original weather variables is to avoid the serious multi-collinearity between the original weather variables. Assume the lagged weather and the unlagged weather variables satisfy the following relation.


whereis the lagged weather variable,is a vector of unlagged weather variable, andis a white noise. For instance, consider the temperature, When,, ¼›whereas when¼Œthe adjusted weather variable is.

The formal GAMs model has the form:

Log expected mortalityt = (5)

Where represents a smooth relative risk function with degree of freedom, and represents calendar time. Here we consider every year has 2 df (we also try average of 4, 6 and 12 df respectively, per year, but the results are not interesting), for warm/cold seasons to adjust the seasonal confounders.

Regression models were fitted with the MGCV package in R.

Figure 6 reports how TM functions evolve steadily with time in the city of Shanghai.

Figure 7 reports the significant variation of TM function in different community in the city of Shanghai.

Figure 6 TM Relative Risk Function for Shanghai, 1956-2001.

Figure 7 TM log risk function for Yangpu, Changni and Xuhui

We extend the GAMs modeling (5) to analyze the relationship between the daily death count and the daily means of concentrations of PM10, NO2 and SO2. The formal extended GAMs model has the form:

Log expected mortalityt =


The main results of model (6) are summarized in figure 8, figure 9 and figure 10.

Figure 8 log risk function for PM10 in Shanghai (2000-2001)

Figure 9 log risk function for daily NO2 in Shanghai (2000-2001)

Figure 10 log risk function for daily SO2 in Shanghai (2000-2001)

Next we try to test whether there are independent effects of a single pollutant to account for a health outcome in the city of Shanghai. Follow the standard strategy (Wong et al. 2001), we start from obtaining a core model for each of the mortality outcomes (noaccident death, cardiovascular death, and Respiratory death) for all ages, nonparametric smoothing by means of loess function terms for trend on seasonality (warm season and cold season), temperature, and humidity were fitted as independent variables. To perform the stratified analyses, we first obtained expected mortality counts from the core model for all seasons. Poisson regression for the mortality outcomes was then fitted on pollutant concentrations to obtain the log relative risk estimatewith offset on separately for warm (April - September) and cold seasons (Other months). Offset is a computation procedure to treatas a reference value and does not proceed to estimate a parameter for it in the Poisson regression . Exposure - response curves in warm and cold seasons were again examined using Generalized Additive Modeling framework. That is, the Poisson regression is transformed into (whereis the smoothing function that are represented using penalized regression splines with smoothing parameters selected by Generalized Cross Validation. And the figures below are the relationship between pollutants concentrationsand the estimated log risk of mortality.

The figure 11, figure 12 and figure 13 show the seasonal exposure-response relationship for each pollutant for the three main mortality outcome.

Figure11 Smoothed plots of SO2 against mortality risk in log scale (deviated from overall mean)

Figure 12 Smoothed plots of NO2 against mortality risk in log scale (deviated from overall mean)

Figure 13 Smoothed plots of PM10 against mortality risk in log scale (deviated from overall mean)


5.1 The variation of V-shaped TM with time

Based on the half-century daily mortality data in the city of Shanghai, it is quite clear that the effect of temperature on mortality varied with time, even though the general V-shaped relation keeps constant in the sense that mortality risk is positively correlated with both the coldest temperatures and the hottest temperatures. The most striking thing is the threshold temperature (the temperature corresponding with the lowest mortality risk), which is around 23 Celsius, also keeps constant.

Even though from 1990s the home/office air conditioner is widely available in the city of Shanghai and quickly became saturation by the end of 1990s. We could see the TM's variation in time in two regards. Firstly from 1970s, the original V-shaped temperature-mortality relations had slowly changed into a reversed-J-shaped one with a flatting tail at warmer temperature, implying little increase in mortality risk for the hottest days. Secondly as time goes on, the TM's slopes (both the slopes in the coldest days and in the hottest days) significantly decrease with more progress in the coldest days from 1970s to 1990s. We could observe more disparity between the TM 1971-1986 curve and the TM 1990-2001curve in the cold season than in the warm season. Shanghainese benefit a lot from the introduction of air conditioner, the increase of personal awareness about the adverse health effect from the cold weather, the improvement of environment, etc, maybe the trend of consistent getting warming also should be included, even though detailed data on these events were not available, meanwhile these progress also happened simultaneously, therefore technically it is impossible to quantitatively differentiate these effects' own contribution.

5.2 The variation of TM with communities

We use three districts, Xuhui (XH represents rich and advanced communities in Shanghai), Changni(CN represents average communities in Shanghai), and Yangpu (YP represents less developed communities in Shanghai) to show the TM's variation among communities. We originally assume that the curve of TM function in advanced communities should be more flat than that of less advanced communities. Some empirical findings are consistent with our original hypothesis, but some do not. In cold season, the richer communities seem to be more experienced with adjusting the coldest temperature than poor communities which is consistent with our hypothesis; but in warm season, the poorer communities seem to be more "clever" with adjusting the hottest temperature than rich communities which turn out to be un-consistent with our hypothesis.

Since the information of community specific social-economical level is not available at current stage, we here just provide one possible explanation for this "puzzle" which actually partly based on the empirical finding from the first plot in Figure 7. It is highly possible that population ageing is much more significant accelerator than other so-called social-economical factors in promoting the mortality risk in the sense that being younger is more helpful than being rich in adjusting with un-nice temperature as well as air pollution. Xuhui is the most aged district in Shanghai and therefore people in Xuhui district have the most likelihood of being old and being fragile comparing with the people in the less aged districts, Changni and Yangpu. Even though Xuhui is the richest district in Shanghai and people living in Xuhui district generally are well educated and averagely have the best access with health service, air conditioner, inner heating system and related health knowledge. Based on the second plot in figure 7, it seems the precautious behavior (like using air conditioner, better living space, suitable clothing, etc) are more efficient in adjusting cold weather in winter season, but less useful in the hottest days.

5.3 The variation of mortality pollution function with communities

It is quite confusing with the inconsistent air pollution reaction functions by different districts. The city-pooled pollution index maybe is misleading in the sense that the pollution index published by the government may not relevant with specific district; meanwhile the air pollution level by district is not available. Even though we still could specify some consistent pattern based on the plots above. For the districts Zhabei and Yangpu, these two districts are traditional Shanghai heavy industry area; therefore Zhabei, Yangpu is tend to have the highest PM10 level comparing with the downtown districts, Xuhui, Luwang and Nanshi. From figure 8, Zhabei and Yangpu has the most sensitive reaction function with PM10. Again the age structure highly matters the reaction function with air pollution with same logic we discussed in former paragraph. Xuhui, Luwang, these most aged wealthy districts is the district most vulnerable.

5.4 The relationship between pollutant concentration and mortality stratified by season

The most striking finding was that SO2 had almost identical/linear effects on all three mortality outcomes during the cold season. According to our limited literature review, this finding is unique. In summer season, there is no clear exposure-response relationship were observed when SO2 <50(in Hong Kong, the number is 30), but there were some linear exposure-response relationship above that concentration for respiratory mortality and noaccident mortality. Unlike Hong Kong, in Shanghai it seems there was no exposure-response relationship between cardiovascular mortality and the concentration of SO2 for warm season.

During the warm season, we observed again the linear exposure-response relationship for cardiovascular and respiratory mortality for NO2, no exposure-response relationship for noaccident mortality for NO2. During the cold season there were some strong linear and nonlinear positive exposure-response relationships for the three mortality outcomes throughout the concentration levels. Again there is no threshold for NO2 in exposure-response relationship with three mortality outcomes during cold season.

For PM10, during warm season, there seems existing some threshold 125for cardiovascular mortality. Below this level, there was no exposure-response relationship. During warm season, there is negative exposure-response relationship for noaccident mortality for PM10, and some very strong linear positive exposure-response relationship for respiratory mortality for PM10. During cold season, there were again some almost linear positive exposure-response relationships for all three outcomes.

At very high concentrations, the risks of mortality could be reduced possibly because vulnerable subjects may have died before the concentration had reached the maximum levels. (Wong et al. 2001)

There are significant differences in the pollution level between the cold season and the warm season that could be easily recognized via checking the density distribution of X-coordinate in the figure 11 to 13, which could partly explain the significant disparity of exposure-response relationships in seasons. Also table 4 reported the detailed information about the seasonal variation in pollution level.

Comparing with the former research finding in Hong Kong, in Hong Kong, in warm season, except for respiratory mortality, no strong effects of particulate pollutants were observed, it is also reported that Hong Kong's case is quite unique comparing with other Western cities. This unique may partly due to Hong Kong's typical weather conditions in the warm season, it is reported heavy rain, rain storm and typhoons significant alleviate the actual exposure-response relationships between air pollution and mortality.