# Impact of the 1990 Hong Kong legislation for restriction on sulfur content in fuel.Res Rep Health Eff Inst. 2012 AugRR

### INTRODUCTION

After the implementation of a regulation restricting sulfur to 0.5% by weight in fuel on July 1, 1990, in Hong Kong, sulfur dioxide (SO2*) levels fell by 45% on average and as much as 80% in the most polluted districts (Hedley et al. 2002). In addition, a reduction of respiratory symptoms and an improvement in bronchial hyperresponsiveness in children were observed (Peters et al. 1996; Wong et al. 1998). A recent time-series study (Hedley et al. 2002) found an immediate reduction in mortality during the cool season at six months after the intervention, followed by an increase in cool-season mortality in the second and third years, suggesting that the reduction in pollution was associated with a delay in mortality. Proportional changes in mortality trends between the 5-year periods before and after the intervention were measured as relative risks and used to assess gains in life expectancy using the life table method (Hedley et al. 2002). To further explore the relation between changes in pollution-related mortality before and after the intervention, our study had three objectives: (1) to evaluate the short-term effects on mortality of changes in the pollutant mix after the Hong Kong sulfur intervention, particularly with changes in the particulate matter (PM) chemical species; (2) to improve the methodology for assessment of the health impact in terms of changes in life expectancy using linear regression models; and (3) to develop an approach for analyzing changes in life expectancy from Poisson regression models. A fourth overarching objective was to determine the relation between short- and long-term benefits due to an improvement in air quality.

### METHODS

For an assessment of the short-term effects on mortality due to changes in the pollutant mix, we developed Poisson regression Core Models with natural spline smoothers to control for long-term and seasonal confounding variations in the mortality counts and with covariates to adjust for temperature (T) and relative humidity (RH). We assessed the adequacy of the Core Models by evaluating the results against the Akaike Information Criterion, which stipulates that, at a minimum, partial autocorrelation plots should be between -0.1 and 0.1, and by examining the residual plots to make sure they were free from patterns. We assessed the effects for gaseous pollutants (NO2, SO2, and O3), PM with an aerodynamic diameter < or = 10 microm (PM10), and its chemical species (aluminum [Al], iron [Fe], manganese [Mn], nickel [Ni], vanadium [V], lead [Pb], and zinc [Zn]) using the Core Models, which were developed for the periods 5 years (or 2 years in the case of the sensitivity analysis) before and 5 years after the intervention, as well as in the10-year (or 7-year in the case of the sensitivity analysis) period pre- and post-intervention. We also included an indicator to separate the pre- and post-intervention periods, as well as the product of the indicator with an air pollution concentration variable. The health outcomes were mortality for all natural causes and for cardiovascular and respiratory causes, at all ages and in the 65 years or older age group. To assess the short- and long-term effects, we developed two methods: one using linear regression models reflecting the age-standardized mortality rate D(j) at day j, divided by a reference D(ref); and the other using Poisson regression models with daily mortality counts as the outcome variables. We also used both models to evaluate the relation between outcome variables and daily air pollution concentrations in the current day up to all previous days in the past 3 to 4 years. In the linear regression approach, we adjusted the data for temperature and relative humidity. We then removed season as a potential confounder, or deseasonalized them, by calculating a standard seasonal mortality rate profile, normalized to an annual average of unity, and dividing the mortality rates by this profile. Finally, to correct for long-term trends, we calculated a reference mortality rate D(ref)(j) as a moving average of the corrected and deseasonalized D(j) over the observation window. Then we regressed the outcome variable D(j)/D(ref) on an entire exposure sequence {c(i)} with lags up to 4 years in order to obtain impact coefficient f(i) from the regression model shown below: deltaD(j)/D (ref) = i(max)sigma f(i) c(j - i)(i = 0). The change in life expectancy (LE) for a change of units (deltac) in the concentration of pollutants on T(day)--representing the short interval (i.e., a day)--was calculated from the following equation (deltaL(pop) = average loss in life expectancy of an entire population): deltaL(pop) = -deltac T(day) infinity sigma (j = 0) infinity sigma f(i) (i = 0). In the Poisson regression approach, we fitted a distributed-lag model for exposure to previous days of up to 4 years in order to obtain the cumulative lag effect sigma beta(i). We fit the linear regression model of log(LE*/LE) = gamma(SMR - 1) + alpha to estimate the parameter gamma by gamma, where LE* and LE are life expectancy for an exposed and an unexposed population, respectively, and SMR represents the standardized mortality ratio. The life expectancy change per Ac increase in concentration is LE {exp[gamma delta c(sigma beta(i))]-1}.

### RESULTS

In our assessment of the changes in pollutant levels, the mean levels of SO2, Ni, and V showed a statistically significant decline, particularly in industrial areas. Ni and V showed the greatest impact on mortality, especially for respiratory diseases in the 5-year pre-intervention period for both the all-ages and 65+ groups among all chemical species. There were decreases in excess risks associated with Ni and V after the intervention, but they were nonsignificant. Using the linear regression approach, with a window of 1095 days (3 years), the losses in life expectancy with a 10-microg/m3 increase in concentrations, using two methods of estimation (one with adjustment for temperature and RH before the regression against pollutants, the other with adjustment for temperature and RH within the regression against pollutants), were 19.2 days (95% CI, 12.5 to 25.9) and 31.4 days (95% CI, 25.6 to 37.2) for PM10; and 19.7 days (95% CI, 15.2 to 24.2) and 12.8 days (95% CI, 8.9 to 16.8) for SO2. The losses in life expectancy in the current study were smaller than the ones implied by Elliott and colleagues (2007) and Pope and colleagues (2002) as expected since the observation window in our study was only 3 years whereas these other studies had windows of 16 years. In particular, the coefficients used by Elliott and colleagues (2007) for windows of 12 and 16 years were non-zero, which suggests that our window of at most 3 years cannot capture the full life expectancy loss and the effects were most likely underestimated. Using the Poisson regression approach, with a window of 1461 days (4 years), we found that a 10-microg/m3 increase in concentration of PM10 was associated with a change in life expectancy of -69 days (95% CI, -140 to 1) and a change of -133 days (95% CI, -172 to -94) for the same increase in SO2. The effect estimates varied as expected according to most variations in the sensitivity analysis model, specifically in terms of the Core Model definition, exposure windows, constraint of the lag effect pattern, and adjustment for smoking prevalence or socioeconomic status.

### CONCLUSIONS

Our results on the excess risks of mortality showed exposure to chemical species to be a health hazard. However, the statistical power was not sufficient to detect the differences between the pre- and post-intervention periods in Hong Kong due to the data limitations (specifically, the chemical species data were available only once every 6 days, and data were not available from some monitoring stations). Further work is needed to develop methods for maximizing the information from the data in order to assess any changes in effects due to the intervention. With complete daily air pollution and mortality data over a long period, time-series analysis methods can be applied to assess the short- and long-term effects of air pollution, in terms of changes in life expectancy. Further work is warranted to assess the duration and pattern of the health effects from an air pollution pulse (i.e., an episode of a rapid rise in air pollution) so as to determine an appropriate length and constraint on the distributed-lag assessment model.

### MeSH

### Pub Type(s)

### Language

### PubMed ID

### Citation

*Research Report (Health Effects Institute),*2012, pp. 5-91.

*Res Rep Health Eff Inst*. 2012.

*Research Report (Health Effects Institute)*, (170), 5-91.

*Res Rep Health Eff Inst.*2012;(170)5-91. PubMed PMID: 23316618.