Article Text

Download PDFPDF

The effect of sandstorms and air pollution on cause-specific hospital admissions in Taipei, Taiwan
  1. M L Bell1,
  2. J K Levy2,
  3. Z Lin3
  1. 1
    Yale University, School of Forestry and Environmental Studies, New Haven, Connecticut, USA
  2. 2
    Western Washington University, Department of Environmental Studies, Bellingham, Washington, USA
  3. 3
    Chinese Academy of Sciences, Institute of Atmospheric Physics, Beijing, China
  1. Dr M L Bell, 205 Prospect Street, Yale University, New Haven, CT 06511, USA; michelle.bell{at}yale.edu

Abstract

Objective: Relatively little research exists focusing on the impact of air pollution on hospital admissions in Asia compared to the extensive work conducted in the USA and Europe. The issue is of particular importance because of the frequency, intensity and health effects of Asian sandstorms. This work investigates the relation between cause-specific hospital admissions and sandstorms and air pollution in Taipei, Taiwan’s capital.

Methods: Time-series analyses of asthma, pneumonia, ischaemic heart disease and cerebrovascular disease hospital admissions were performed for Taipei. An eight-year time period (1995–2002) was considered for various indicators of sandstorms and the pollutants NO2, CO, ozone, SO2, PM10, and PM2.5. Pollution effects based on single-day lags of 0, 1, 2 and 3 days were explored, along with the average of the same day and previous three days (L03).

Results: The risk of ischaemic heart disease admissions was associated with several sandstorm metrics, including indicators of high PM10 levels in the Taipei area, indicators of high PM10 at a monitor designed to measure background pollution, the PM coarse fraction, and the ratio of PM10 to PM2.5. However, the lag structure of effect was not consistent across sandstorm indicators. Hospital admissions for this disease were 16–21% higher on sandstorm days compared to other days. This cause was also associated with transportation-related pollutants, NO2, CO and PM2.5. Asthma admissions rose 4.48% (95% CI 0.71% to 8.38%) per 28 μg/m3 increase in L03 PM10 levels and 7.60% (95% CI 2.87% to 12.54%) per 10 ppb increase in L03 ozone. Cerebrovascular disease admissions were associated with PM10 and CO, both at lag 3 days. SO2 exhibited no relation with admissions.

Conclusions: Risk of hospital admissions in Taipei may be increased by air pollution and sandstorms. Additional research is needed to clarify the lag structure and magnitude of such effects.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Associations between ambient air pollution and hospital admission rates are well documented in North America and Europe1 2 with fewer studies in Asia. It is vital to further our understanding of air pollution and health relations in Asia, as a number of factors limit the reliability of results extrapolated from Western studies, including differences in healthcare systems, baseline health status, exposure patterns, air pollutant mixture, and population characteristics.3 Air pollution in Asia is of particular interest given the growing populations, increased economic activity, rise in vehicular traffic, as well as the increasing intensity and occurrence of dust storms originating in Mongolia and China.

During a severe Asian dust storm, particulate matter can be up to 36 times higher than during non-dust periods, whereas concentrations of gaseous pollutants such as ozone (O3) and sulfur dioxide (SO2) may be lower due to increased wind speeds and lowered light and temperature for O3 formation.4 5 The chemical composition of particles during dust storm events may differ from particulate matter at other time periods and locations, and thereby have dissimilar health impacts. A vast literature links airborne particle exposure to an array of adverse health effects including increased risk of premature mortality, hospital admissions, and respiratory symptoms.6 While epidemiological studies on the human health impacts of particles have implicated a variety of size fractions, recent research demonstrates that the observed epidemiological associations between PM2.5 (particulate matter with an aerodynamic diameter ⩽2.5 microns) and hospital admissions vary by region and season.1 Such spatial and temporal differences in the health effects estimates of particles could result from variation in chemical composition.

Studies of sandstorm particles in Taiwan and elsewhere in Asia identified higher concentrations of crustal elements (for example, Ca, Mg, Al, and Fe) than for non-storm particles.4 7 Consistent with those results, measurements from dust storms in central Taiwan exhibited increased concentrations of coarse particles (PM10-2.5),7 8 and a Beijing dust storm was reported to contain higher concentrations of particles >2 microns than were present in non-dust periods.5 To what extent is the health of the residents of Taipei affected by the chemical composition and size distribution of particles from sandstorm events? This question is critically important because particles from Asian sandstorms include a higher proportion of coarse particles compared to typical urban particle mixtures found in the cities of both developing and industrialised nations. In addition to these factors, the extrapolation of concentration-response functions from one study region to another is hindered by differences in the underlying population.3 Therefore, air pollution and health studies in East Asia address significant research questions and can help policy makers reach timelier and more effective air quality and health management decisions.

This study focuses on Taipei, the capital of Taiwan, whose greater metropolitan area includes a population of about six million, with 2.5 million in the city itself. Growing transportation networks and urbanisation have increased vehicular pollution. Motor vehicles account for over 95% of carbon monoxide (CO) emissions and 75% of nitrogen oxide (NOx) emissions, as well as substantial contributions to hydrocarbons, a precursor to O3.9 The valley topography encourages the accumulation of air pollution.

Under certain weather conditions, sand from dust storms originating in Mongolia and China are carried eastward to Taiwan by cold pressure systems. This long-range transport of air pollution influences air quality, contributing about 30 μg/m3 to particulate matter with an aerodynamic diameter less than 10 microns (PM10) in Taiwan and up to 87% for some episodes for Taipei.10 11 In fact, even minor Asian dust storms can influence Taipei’s particulate levels.11

As the intensity and frequency of Chinese sandstorms increase,12 their effects on the health of Taiwanese people may gradually worsen. A multi-criteria analysis of environmental quality in Taipei identified air quality and noise as the environmental issues most important to the public, outranking land-use conflicts, drinking water, solid waste, open space and water pollution.13 Air quality was particularly important to residents of downtown areas.

In order to better understand how air pollution affects health in Taipei, a study of air pollution and cause-specific hospital admissions was conducted for an eight-year period considering six ambient air pollutants and sandstorms, using four indicators of sandstorm events. To the best of our knowledge, this is the first study to examine the effect of air pollution on pneumonia or ischaemic heart disease hospital admissions in Taiwan. Previous studies of the impact of air pollution and sandstorms on hospital admissions in Taiwan are summarised in the Discussion section.

METHODS

Eight years of daily data from 1 January 1995 to 31 December 2002 are used for this study of Taipei. Eighteen air monitoring stations in Taipei City and County were in operation during the study period by the Taiwan Environmental Protection Agency (TEPA). The considered pollutants were PM2.5, PM10, nitrogen dioxide (NO2), SO2, CO and O3. The monitoring station instrumentation included β-gauges for PM10 and PM2.5, ultraviolet (UV) absorption for O3, UV fluorescence for SO2, chemiluminescence for NO2, and non-disperse infrared absorption for CO. Weather data, originally collected by the Central Weather Bureau, included temperature and dew point temperature. Twenty four hour averages were used for daily values of weather and pollution variables. Only two stations monitored PM2.5.

Daily weather values were based on an average of 16 weather monitors throughout the Taipei basin. For generating daily pollution estimates, five monitors were excluded because they were designed to measure transportation or background concentrations. The remaining 13 monitors were employed to measure air quality. Of these 13 monitors, five were in Taipei City and eight in Taipei County. For PM2.5, data from two monitors were averaged to generate daily averages. For the remaining pollutants, we estimated exposure to air pollution in several ways: (1) exposure over the entire Taipei area, based on the 13 air quality monitors; (2) exposure over Taipei City only, based on five monitors; and (3) exposure based on the subset of monitors for which all pairs of monitors had correlations of ⩾0.75 for a given pollutant. For each of these exposure metrics, daily values across monitors were averaged to generate an overall daily value, including only monitor values with ⩾16 h of measurements for a given day.

Sandstorm exposure was modelled using several approaches. First we applied Boolean indicators for the following: (1) PM10 levels >115 μg/m3 based on the Taipei area; and (2) PM10 >100 μg/m3 at the Yangmin monitoring station. While the use of an indicator variable for high PM10 levels is an imperfect marker of sandstorms, the measure does provide an indication of the extremely high particulate levels that may arise due to sandstorm events. The Yangmin station, located in Taipei City, is designed to monitor background levels; hence its PM10 measurements are less likely to be affected by local pollution sources. It follows that high particulate levels at this station are more likely to represent sandstorm events. Similar approaches, including the use of high PM10 levels at background monitoring stations as an indicator of sandstorms, have been applied previously.1418 Because sandstorm events contain a high number of particles in the coarse fraction, we considered coarse particles (PM10-2.5) and the ratio of PM10/PM2.5.

The health effects outcome was the number of hospital admissions for a given day at National Taiwan University Hospital for two cardiovascular causes (ischaemic heart disease and cerebrovascular disease) and two respiratory causes (asthma and pneumonia). Admissions for ischaemic heart disease included International Classification of Disease Ninth Edition-Clinical Modification (ICD-9-CM) codes 410, 411 and 414. Cerebrovascular disease included ICD-9-CM codes 430 to 437. Asthma was coded as ICD-9-CM 493 and pneumonia as ICD-9-CM486. Records were originally collected through the National Health Insurance Program, which insures over 96% of the Taiwanese population.19

A time-series analysis was performed to investigate the relation between the day-to-day variation in air pollution in Taipei and cause-specific hospital admissions using the following model:

ln(Yt) = α+βCit-l+θDOWt+ns(time,dftime)+ns(At, dfA)

where Yt is the expected number of hospital admissions on day t; Cit-l is the concentration of pollutant i on day t at a lag of l days (for example, l = 0 for same day exposure, l = 1 reflects the previous day’s air pollution); DOWt are indicator variables relating to the day of the week; ns(time,dftime) is a natural cubic spline of time with dftime degrees of freedom (dftime  = 7 per year); ns(At,dfA) is a natural cubic spline of apparent temperature (A) on day t with dfA degrees of freedom (dfA  = 6).

Day of the week was included as a potential confounder as both health effects and air pollution can follow weekly patterns. Long-term trends and seasonality were addressed with splines of a variable representing time. Adjustment by weather was incorporated through apparent temperature, which reflects overall temperature discomfort, accounting for temperature and humidity.20 In the preceding model Cit-l represents a single day’s exposure at various lags. In addition, a model reflecting multiple days of exposure was applied, using the average of exposure over the same day and the previous three days.

This type of time series modelling has been successfully applied to investigate the effects of air pollutants on hospital admissions among other health outcomes.1 The initial choice of degrees of freedom for the functions of temporal trend and weather were based on previous research.1 Sensitivity analyses were performed to ascertain the robustness of results to variation in the degrees of freedom for long-term trends. Initial results used 7 degrees of freedom per year for the natural cubic spline of time. In the sensitivity analysis, 3, 5, 9 and 11 degrees of freedom were applied. Previous research demonstrated that results using this type of model are generally robust to various measures of control for weather and season.1 Due to the different number of monitors for the PM metrics (13 for PM10, 2 for PM2.5), daily values for community-level PM2.5 estimates exceeded PM10 levels for a small number of days. Sensitivity analysis was conducted omitting these observations.

RESULTS

Over the study period there were 6909 hospital admissions for ischaemic heart disease, 11 466 for cerebrovascular disease, 10 996 for pneumonia, and 10 231 for asthma. Admissions peaked in November for asthma, January for pneumonia, and March for the other causes, and were lowest in August for asthma and cerebrovascular disease and in October for pneumonia and ischaemic heart disease. The respiratory causes followed a stronger seasonal pattern than cardiovascular causes. A summary of weather, pollution and health data by season is provided in table 1. No events occurred for 15.0%, 2.8%, 5.4% and 6.8% of days for the study period for ischaemic heart disease, cerebrovascular disease, asthma and pneumonia, respectively. No days were missing weather data for the study period. Table 1 also compares air pollution levels to World Health Organization (WHO) air quality guidelines.21 Taipei experienced high air pollution levels, exceeding WHO guidelines for PM10, PM2.5 and NO2 for all years.

Table 1 Summary of hospital admissions, weather, and pollution in Taipei, 1995 to 2002, and comparison with World Health Organization guidelines

PM2.5 measurements began on 12 April 1997. From that period through 31 December 2002, PM2.5 data were missing for 1.3% of the days. Ozone levels rose over the study period, exhibiting a 31.4% increase from 1995 to 2002. Other pollutants exhibited a downward trend, especially SO2, which declined 58.2%, likely in response to Taipei air quality measures including lower sulfur content for fuel oil in 1996 and diesel oil in 1998.22 Air quality trends over the study period are displayed in figure 1.

Figure 1 Per cent change in pollutant levels in Taipei, 1995 to 2002. Note: the baseline year is 1995 for all pollutants except PM2.5, for which the baseline year is 1997. PM2.5 data are not available for 1995 to 1996. The PM2.5 1997 data reflect the period from 12 April to 31 December 1997.

Concentrations of air pollutants can co-vary due to similar sources, atmospheric conditions, seasonal patterns and related pathways of formation such as through photochemistry. By definition, PM2.5 is included in PM10, so the levels of these pollutants can co-vary depending on the concentration of coarse PM. Many of the pollutant concentrations show similar patterns, with the strongest relation between NO2 and CO at a correlation coefficient of 0.87, based on the Taipei area-wide data. Other notable associations between pollutant levels were for PM10 with PM2.5 (correlation 0.61), SO2 (0.63), NO2 (0.64) and CO (0.62). SO2 was correlated with NO2 and CO at 0.63 and 0.68, respectively.

Table 2 shows results from the single-lag (that is, one day) and cumulative exposure models for the per cent increase in cause-specific hospital admissions per a given increase (near IQR) in pollution. For all pollutants other than PM2.5, results are shown for three exposure metrics: (1) exposure over the entire Taipei area; (2) exposure for Taipei City only; and (3) exposure based on the subset of monitors for which all pairs of monitors had correlations of ⩾0.75 for a given pollutant, as described in the Methods section. Table 2 also provides the average correlation among pairs of monitors for each pollutant and exposure metric. Note that under all three exposure measurements, correlations among monitor values were highest for PM10 compared with other pollutants, indicating that estimates from this pollutant are less likely to suffer misclassification of exposure due to spatial heterogeneity in pollutant levels. Conversely, SO2 generally had the lowest correlation among monitor pairs. The different exposure metrics identified similar associations between air pollution and hospital admissions. Estimates for exposures based on the city area generally had smaller confidence intervals (that is, smaller standard error of the regression coefficient) than did estimates based on the entire Taipei area.

Table 2 Per cent increase in cause-specific hospital admissions per specific increase in pollutant, using single lag and cumulative lag models, with various exposure metrics

Statistically significant relations were observed for ischaemic heart disease admissions and same day pollution for NO2 and CO and L03 exposure for PM2.5. Cerebrovascular hospital admissions were associated with PM10, NO2 and CO at lag three days. Asthma admissions were associated with L03 PM10 and O3 at lags of one or two days or L03. No statistically significant associations were observed for pneumonia and any pollutant. SO2 did not exhibit any statistically significant relations with any cause of admissions.

Sandstorm days were identified for 2.1% of days for the indicator based on PM10 levels >115 μg/m3 in the Taipei area, and for 1.6% of days for the indicator using PM10 >100 μg/m3 at the Yangmin background monitoring station. The correlation between these two indicators was 0.64. Table 3 shows the relation between the various sandstorm indicators and cause-specific hospital admissions. No sandstorm metric demonstrated associations with cerebrovascular, asthma or pneumonia hospital admissions. However, all sandstorm measures exhibited a relation with ischaemic heart disease admissions. The lag structure of this association was not consistent across various metrics.

Table 3 Per cent increase in cause-specific hospital admissions for various sandstorm indicators

Several sensitivity analyses were conducted. Analysis of the relation between air pollution or sandstorms and cause-specific hospital admissions for all lags and outcomes considered was repeated for exposure metrics that rely on community-level PM10 or PM2.5 values (PM10 in the Taipei area, PM10 in Taipei City only, PM10 based on highly correlated monitors, PM2.5, coarse PM, sandstorm indicator of PM10 >115 μg/m3, and PM10/PM2.5), by omitting the small number of days where community-level PM2.5 exceeded PM10 values. Effect estimates were similar to those of the original analysis (correlation 0.998).

The original results used 7 degrees of freedom per year for the natural cubic spline of temporal trend. We also applied 3, 5, 9 and 11 degrees of freedom per year for the Taipei-wide exposure metrics for PM10, NO2, SO2, O3 and CO for all lags and outcomes. Results were robust to more or less aggressive adjustment for long-term trends and seasonality.

DISCUSSION

Our results indicate a link between ambient air pollution and sandstorm and hospital admissions in Taipei. In particular, higher rates of ischaemic heart disease admissions were linked with CO, NO2 and PM2.5; all pollutants related to transportation sources, as well as several indicators of sandstorms, such as coarse particles. Asthma admissions were associated with PM10 and O3. Cerebrovascular admissions were associated with PM10 and CO, and displayed an implied relation with NO2. Results were consistent across multiple exposure metrics for air pollution. However, results for various sandstorm indicators implicated different lag structures, so further research is needed to clarify the lag structure and magnitude of such effects.

A limited number of other studies also examined the effects of sandstorms or air pollution on hospital admissions in Taiwan. Table 4 summarises these previous studies and includes one study on emergency room visits. Results from this study are provided for comparison. A previous study found higher asthma hospital admissions with PM10, SO2, NO2, CO and O3, whereas we observed effects for PM10 and O3 only. Possible explanations for differences in the results are the study designs, timeframe and city, as different regions may have divergent pollutant mixtures.

Table 4 Summary of air pollution or sandstorms and hospital admissions studies in Taiwan

No other studies of pneumonia hospital admissions and air pollution in Taiwan were identified in a literature review. This research is also unique because it is the only study to observe statistically significant relations with total cerebrovascular disease hospital admissions, which were identified for PM10 and CO. Other work found relations with specific forms of the disease and for emergency room visits for cerebrovascular disease. Two previous works found higher hospital admissions for cardiovascular causes in Taipei and one in Kaohsiung. We observed a relation between NO2, CO and PM2.5 with ischaemic heart disease, a subset of the cardiovascular outcomes modelled in other studies.

Other work has examined the relation between air pollution and hospital admissions and related events in China, finding associations between CO, SO2, NOx or PM10 and emergency room visits for colds, pneumonia and bronchitis;30 total suspended particles (TSP) and unscheduled non-surgical out-patient visits;31 SO2 and paediatric out-patient and surgery visits;31 TSP and SO2 and hospital admissions;32 SO2 and TSP and non-surgical out-patient visits;33 SO2 and TSP and hospital out-patient and emergency visits;34 and TSP and SO2 and chronic obstructive pulmonary disease hospital admissions.35

Previous research has investigated the impacts of sandstorms and coarse particles. Several studies based on Taipei found indication that risk of adverse health events is higher after Asian dust storm events, however results were not statistically significant.14 16 17 Such results include analysis of clinical visits for allergic rhinitis from 1997 to 200118 and respiratory and circulatory disease mortality from 1995 to 2000.15 In Korea, dust events were linked with respiratory symptoms of people with bronchial asthma,36 and weakly associated with mortality.37 Statistically significant relations were not observed between PM10-2.5 and mortality in China38 or dust events and hospital admissions in British Columbia.39 Studies that have examined PM2.5 and coarse PM together indicate an adverse effect of coarse PM on health, but stronger effects for PM2.5.40 Previous research also supports the theory that coarse PM contributes to cardiovascular hospital admissions.40

Our study uses one of the longest study periods applied to research in air pollution and hospital admissions in Taipei. It is also one of the first to investigate the impacts of air pollution on ischaemic heart disease and pneumonia hospital admissions in Taiwan. Limitations of our methodology include the use of ambient monitors as a surrogate for actual exposure. This method, while commonly employed, does not include differences in daily activity patterns, sub-community heterogeneity in concentrations (which may differ by pollutant) or occupational exposure. Advantages of this approach are the use of existing measurements and the ability to thereby study a larger population over an extended period of time. Misclassification of exposure may differ by pollutant, as indicated by the variation in correlations of pollutants between monitors. The limited number of monitors measuring PM2.5 hinders our ability to investigate this pollutant and related measures such as coarse PM. While we used a variety of metrics to gauge sandstorm levels, these definitions are imperfect and may not fully capture the presence of sandstorm events. Further, the sandstorm indicators applied in this research are related to high PM10 levels, including at a background monitoring station, and coarse particles, but do not assure that the particulate pollution is in fact the result of sandstorm events. Future research on sandstorms and health could include air quality modelling such as trajectory analysis or source apportionment techniques to confirm that particles on days identified as sandstorm days originate from sandstorm sources.

The outcome variable was the number of hospital admissions rather than the hospital admissions rate; therefore changes in populations over time could potentially affect our results. However, an investigation that covered about 72% of our study period found no significant trends in population rates for Taipei residents older than 50 years.27 Because the concentrations of air pollutants can co-vary due to similar sources and atmospheric conditions, it is not possible to distinguish among the effects of highly correlated pollutants, such as NO2 and CO. A single pollutant may act as a surrogate for another pollutant or a pollutant mixture from the same source or formation pathway. Therefore, results may be indicative of the pollutant mixture or a related pollutant, rather than the individual pollutant modelled.

Results from this study provide evidence that air pollution in Taipei may increase risk of hospital admissions for a variety of causes, and that sandstorm events in particular may adversely impact health. Additional research is needed to better understand physiological mechanisms, to further identify the relative toxicity of various pollutants in Taiwan, and to measure the extent to which the health consequences of dust storm particles and other pollutants in this area differ from the health impacts of air pollution in other regions of the world. Such efforts would better enable decision makers to effectively design air pollution control strategies. The use of multi-city air pollution studies for Taiwan, such as have been conducted in Europe and North America, would minimise sample size concerns and potentially lower uncertainty surrounding effect estimates.

Main messages

  • The presence of sandstorms and coarse particles may be associated with increased risk of hospital admissions.

  • Exposure to ambient air pollution may be associated with increased risk of respiratory and cardiovascular hospital admissions in Taiwan’s capital and other regions with similar air pollution characteristics.

Policy implications

  • Air pollution and sandstorms appear to present a substantial risk to public health in Asia; this information should influence the development of air pollution control strategies and warning systems for sandstorms.

  • Given the growing transportation systems and the high levels of air pollution currently experienced in this and other parts of Asia, in-country evidence on related health impacts is critically important for decision makers.

REFERENCES

Footnotes

  • Competing interests: None declared.