Article Text

The smoke-free legislation in Hong Kong: its impact on mortality
  1. Thuan-Quoc Thach,
  2. Sarah M McGhee,
  3. Jason C So,
  4. June Chau,
  5. Eric K P Chan,
  6. Chit-Ming Wong,
  7. Anthony J Hedley
  1. School of Public Health, The University of Hong Kong, Hong Kong, Hong Kong
  1. Correspondence to Dr Thuan-Quoc Thach, School of Public Health, The University of Hong Kong, 5/F William MW Mong Block, 21 Sassoon Road, Pokfulam, Hong Kong 852, Hong Kong; thach{at}


Objectives To examine trends in deaths for conditions associated with secondhand smoke exposure over the years prior to and following the implementation of a smoke-free policy in Hong Kong.

Design Time-series study.

Setting Death registration data from Hong Kong Special Administrative Region (SAR) Government Census and Statistics Department.

Participants All deaths registered from 1 January 2001 to 31 December 2011.

Main outcome measures Deaths for conditions associated with passive smoking include cardiovascular disease (CVD), respiratory disease and other causes.

Results There was a decline in the annual proportional change for ischaemic heart disease (IHD), acute myocardial infarction (AMI) and CVD mortality in the year after the intervention for all ages and those aged 65 years or older. There were also clear declines in the cool season peaks for these three conditions in the first postintervention year. There was a further drop in the cool season peak for AMI among all ages in the year after the exemptions ceased. No declines in annual proportional change or changes in seasonal peaks of mortality were found for any of the control conditions.

Conclusions The findings in this study add to the evidence base, as summarised in the Surgeon General's report, extending the impact of effective smoke-free legislation to those aged 65 years or older and to cerebrovascular events in younger age groups. They also reinforced the need for comprehensive, enforced and effective smoke-free laws if the full extent of the health gains are to be achieved.

  • Smoking Caused Disease
  • Public policy
  • Secondhand smoke

Statistics from

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.


On the basis of the evidence on adverse health and economic effects of passive smoking, in January 2007, Hong Kong implemented legislation to protect against secondhand smoke exposure (SHS) in indoor workplaces and in public spaces.1 While the Hong Kong ordinance mandated no smoking in the indoor areas of restaurants, workplaces and other public spaces, it allowed exemptions until July 2009 to premises that predominantly served alcohol and that met other requirements.2 By June 2009, 1346 premises, including 1004 bars, 112 massage establishments and bathhouses, 87 clubs, 79 nightclubs and 64 mahjong-tin kau clubs had obtained an exemption.3

Many other jurisdictions have implemented smoke-free laws and these have been associated with reductions in hospitalisation principally for cardiac disease but also sometimes for cerebrovascular and respiratory diseases (RDs).4 In addition to numerous studies on the impact of smoke-free legislations on hospitalisation, some studies have demonstrated a reduction in mortality after implementation of the smoke-free legislation. These include a reduction in acute myocardial infarction (AMI) deaths in Spain5 and Massachusetts, USA,6 out-of hospital deaths for acute coronary events in Rome, Italy,7 and sudden circulatory arrest in the Netherlands.8 In the Spanish study, Villalbí et al5 found that there were declines in deaths due to AMI of 9% for men and 8.7% for women in the first year after the smoke-free legislation in Spain came into force. The decline was sharp and especially marked in the age group above 64 years. In Massachusetts, after a comprehensive smoke-free workplace law came into effect, the AMI mortality rates for residents aged 35 years or older in cities and towns with no prior local smoking ban decreased by 1.6% in the first 12 months and by 18.6% after the first 12 months.6 The overall adjusted effect was 7.4%. In Rome, out-of hospital deaths and hospitalisations for acute coronary events both decreased after 1 year of a smoking ban in all indoor public places, with relative risks of 0.85 and 0.84 for the age groups 35–64 and 65–74 years, respectively.7 In the Netherlands, a 6.8% reduction was noted in the number of sudden circulatory arrest cases after 1 year of legislation to make workplaces, but not the hospitality sector, smoke free.8 No further decrease was found after the extension of the ban to the hospitality sector around 4 years later. One study, which claimed to find little effect on AMI deaths in the first year after adoption of statewide smoke-free laws in six US states, was supported by grants from tobacco manufacturers.9 The latest US Surgeon General’s report compiled all the recent evidence and concluded that there was sufficient evidence to support a causal relationship between effective smoke-free legislation and reduction in coronary events in those aged under 65 years.10 It found suggestive but not sufficient evidence to infer a causal relationship with cerebrovascular and other heart disease outcomes. We set out to examine whether we could discern any impact of the Hong Kong legislation on the pattern of mortality in spite of the partial nature of the initial legislation. We wished to see whether there would be an impact on the short-term mortality rates; however, it is difficult to isolate the effects of changes in community-wide exposures such as SHS when using routine data. There is a lot of variation in mortality data and underlying trends have to be accounted for. In particular, mortality data in Hong Kong show a clear seasonal peak in the cool months, in common with many developed countries. We therefore used an approach that had identified the impact of a specific air pollution intervention on mortality in Hong Kong in the 1990’s.11 We compared the annual proportional change in mortality before and after the introduction of the legislation on 1 January 2007 as well as the magnitude of the seasonal changes in mortality over the same periods.


Mortality data

Data on all deaths registered from 1 January 2001 to 31 December 2011 were extracted from records of the Hong Kong Census and Statistics Department (CSD). The dataset included age, sex, date of death and the underlying cause of deaths according to the International Classification of Disease, Revision 10 (ICD-10). The underlying cause of death categories used as outcome measures in our study were ischaemic heart disease (IHD) (ICD-10, I20-25), AMI (I21-22), cerebrovascular disease (CBD) (I60-69), cardiovascular disease (CVD) (I00-I99) and RD (J00-98). Lung cancer (C33-34) was used as a quasi-control because it would not be expected to change over the time period of the study. Control conditions were injury, poisoning and external causes (S00-T98), cancer excluding lung cancer (C00-C32, C35-D48), and other causes (A00-A09, C00-C32, C35-D48, D50-D77, E00-E07, F00-F19, G00-G09, K00-K93, M00-M25, N00-N99). The number of deaths for each condition was stratified into four age groups: 15–34, 35–64, 65 years or older, and all ages.

Statistical analysis

We assessed the impact of smoke-free legislation on mortality using three approaches. First, we calculated the annual proportional change in mortality for pre- and post-legislation periods separately. Second, to test for any significant difference between these changes, we analysed annual relative changes. Third, amplitudes of seasonal variations in mortality were computed separately for pre-legislation and post-legislation periods.

First approach: annual proportional change for pre-legislation and post-legislation periods

We used Poisson regression to model weekly rates of mortality, taking into account seasonal trends, time trends, mean temperature (in Celsius) and mean relative humidity (in per cent).12 A quadratic term was included in the Poisson model to account for possible non-linear trend of disease. The F test showed this term to be significant and was included in the model. We fitted the regression model with an offset13 of log(γ), with γ representing the population of each age group to adjust for the change in deaths due to variation in size of population. The quantity 100(exp(52×β1)−1), where β1 is the estimated log relative risk associated with the variable for long-term trends in the model, represented the average annual proportional change in mortality (in per cent). The analyses were carried out separately for the periods before (2001–2006) and after (2007–2011) the implementation of the smoke-free legislation.

Second approach: annual relative change for combined pre-intervention and post-intervention periods

To estimate the annual relative change in trends for the combined 11-year period across the implementation period of the legislation, a dummy variable intervention was created by defining the pre-legislation period as 0 and the post-legislation period as 1. Whether or not the legislation has an impact on mortality rates was assessed by an intervention-by-trend interaction term. The annual relative change was quantified as 100(exp(52 × β2)–1), where β2 is the estimated log relative risk associated with the interaction term, which reflects the relative change in the mortality rates from pre-legislation to post-legislation periods (in per cent). Each age group was analysed separately.

Third approach: seasonal variations in mortality

To estimate seasonal variations in mortality, we first fitted a Poisson regression model of mortality adjusted for trend, intervention and intervention-by-trend interaction terms, and obtained weekly expected number of deaths (δ) assuming no seasonal variations were present. Second, a ratio of observed mortality to its expected value was fitted using cosine and sine terms with one cycle per year, in the form Embedded Image. The coefficients α and β associated with the trigonometric terms were used to compute the amplitude (λ) of the seasonal variations: Embedded Image. This quantity can be interpreted as the percentage increase in mortality from mean to the seasonal peak, where 2λ is the percentage increase from the warm (April–September) to the cool (October–March) season.14 We calculated the seasonal variation (λ) before and after the legislation, stratified by causes of death and by age groups. We examined λ in the pre-intervention and post-intervention periods by comparing the average in the year in the baseline period to the values in the individual years after the intervention as well as examining the seasonal trend in the post-intervention period. We also carried out a separate analysis using a moving average of 2 years for the pre-intervention and post-intervention periods to generate a more stable trend of mortality.

Sensitivity analyses

We conducted several sensitivity analyses to ascertain the robustness of the estimates. Mortality trends were adjusted for air pollutants (nitrogen dioxide: NO2, ozone: O3, respirable particulates PM10, and sulfur dioxide: SO2) and smoking prevalence. The estimates in the main analyses had not been adjusted for these variables because of the possibility of over-adjustment in the younger age groups and increased error due to multiple missing values for the confounders. For example, information on smoking prevalence was available once every 2–3 years, so data in intervening years had to be interpolated. All sensitivity analyses were carried out for all ages only to maximise the statistical power.


Annual proportional and relative changes

Before the intervention, annual proportional change in mortality had been increasing for IHD, AMI, CBD, CVD and RD among all ages, and for IHD, AMI, CVD and RD among older ages. No clear pattern was found for those aged 35–64 years over the pre-intervention period (appendix 1). For the control conditions, mortality for lung cancer, and injury and poisoning, was increasing among all ages; it was also increasing for injury and poisoning among older ages, but for other causes was decreasing in those aged 35–64 years before intervention. After the legislation, there was a significant decline in mortality due to IHD, AMI, CVD and RD among all ages by 9.3%, 12.6%, 5.7% and 5.4%, respectively (table 1). There was a similar decline for those aged 65 years or older for IHD, AMI, CVD and RD. There was also a reduction in deaths from CBD among those aged 15–34 and 35–64 years, but no significant reduction among all ages. There was no significant post-intervention difference in the annual proportional change for lung cancer or other control conditions except for a significant increase in mortality from cancer excluding lung cancer for those aged 15–34 and 35–64 years, and for deaths from other causes for all three age groups.

Table 1

Annual proportional change for pre-smoke-free and post-smoke-free legislation years and relative change with 95% CI in mortality

Seasonal variations

In the first year after the intervention (2007), the cool season increase in deaths from IHD among all ages declined from 20.7% to 11.0%; AMI declined from 18.6% to 11.1% and CVD declined from 19.3% to 13.2% (table 2 and appendix 1). In the second year (2008), IHD and AMI showed a rebound in seasonal amplitude to their pre-intervention levels. A decline was also observed in year 4 (2010), the year after the lifting of exemptions in July 2009, for AMI compared to its amplitude in year 3. Similar year 1 declines were found for those aged 65 years or older.

Table 2

Seasonal variation in mortality 2001–2011, examined year by year for the post-intervention period relative to the pre-intervention period

For the control conditions, there were no significant changes from the expected seasonal patterns in the years after the intervention for any age group. In the alternative analysis, the 2-year moving average model yielded a significant decline in cool season deaths from IHD and CVD among all ages, and among those aged 65 or older in the first interval of the post-intervention period, with the cool season peak for IHD declining from 23.7% in 2005/2006 to 16.7% in 2007/2008, and that for CVD declining from 20.9% to 14.6% (data not shown). As in the main seasonal model, the cool season peak gradually returned to the level seen before the intervention. No significant change from the expected pattern was found for other outcome conditions, although they all showed a similar trend to the main model with the change in AMI deaths (21.4–16.8%) just failing to be significant.

Sensitivity analysis

Adjusting for air pollutant concentrations resulted in only minor changes for each cause of death. The annual proportional change in mortality for IHD declined by 5.8–8.9%, AMI by 10.6–12.3%, CVD by 3.8–5.4% and RD by 2.8–5.1%, for all ages (table 3). After adjusting for smoking prevalence, the reductions in mortality for IHD, CVD and RD were 12.6%, 6.2% and 10.4%, respectively, while the reductions in AMI and CBD were not significant. These estimates were consistent with the main analyses and did not alter the conclusions.

Table 3

Annual proportional change pre-smoke-free and post-smoke-free legislation and relative change in mortality, adjusted for levels of specific air pollutants or smoking prevalence


We developed several models, including annual proportional change and seasonal variation models with two variants, to test whether we could detect an effect of the partial ban on smoking in the hospitality industry and other public areas on mortality at the population level. We adjusted for air pollutant levels and smoking prevalence in sensitivity analyses.

Overall, the results obtained from different models and sensitivity analyses were consistent.

Among all ages and those aged 65 years or older, we found declines in the annual proportional change for IHD, AMI and CVD mortality in the year after the intervention. There were also clear declines in the cool season peak for these three conditions in the first post-intervention year. For AMI, there was a further drop in the cool season peak among all ages in the year after the exemptions ceased for some of the bars. Adjusting for declines in smoking prevalence removed some of the impact on the AMI mortality trend, which lost statistical significance. A decline in the annual proportional change for RD was found among all ages and those aged 65 years or older, but no change in the seasonal peak. The annual proportional change in CBD mortality in the younger age groups (15–64 years) declined after the intervention, but there was no change among those 65 years or older and no change in their seasonal peak. There were no declines in annual proportional change or changes in seasonal peaks of mortality for any of the control conditions. This finding is an important outcome, and supports the assumption of a causal relationship between the public health legislation and the prevention of cardiopulmonary injury over a relatively short timescale.

There were some other tobacco control policies implemented at the same time the smoke-free law was introduced. For example, the tobacco tax in Hong Kong was raised by 50% in 2009, 2 years after the smoke-free law took effect. The effect of this measure would likely be reflected on the changes in smoking prevalence. In the sensitivity analysis, the effects estimates attenuated for some of the health outcomes but did not alter the results in the main analysis.

The conclusions that can be drawn from the mortality analysis are that there is evidence of a decline in the average annual mortality of 12.6% from AMI, 9.3% from IHD, 5.7% from CVD and of 5.4% from RD, after the intervention. In 2011, this translated conservatively into about 1055 deaths avoided per year (578 CVD and 477 RD). These effects were mainly for those aged 65 years or older. Some of the effect on AMI mortality may have been contributed by declines in smoking prevalence, as the annual proportional change for AMI was not significant after adjusting for smoking prevalence. There is possible evidence of an effect on CBD mortality, particularly for those under 65 years of age. These findings are plausible and consistent with the currently established global biomedical and epidemiological evidence that acute and chronic exposures to SHS both increase the risk of serious damage to the integrity of blood vessels and other tissues in the heart, lungs and central nervous system. Other study outcomes were consistent with the main analysis after adjusting for smoking prevalence in the sensitivity analysis. The data on annual proportional change indicate that the absolute risks of passive smoking-associated deaths appear to have been reduced by the legislation. The change in the seasonal pattern of mortality, with a marked reduction in the cool season amplitude for cardiopulmonary causes of death in the early post-intervention period, indicates that deaths that would have occurred at this time were delayed or prevented altogether. The later, apparent post-intervention rise in cool season amplitude indicates a possible mortality displacement pattern. Some of the delayed deaths may have occurred at this time, and others in subsequent years. The stability of the annual proportional change measure indicates that the reduced absolute risk persists at the population level. It is important that the re-establishment of the seasonal pattern, at a lower absolute risk, is not interpreted as a reversal of the post-intervention harm reduction. This same phenomenon was observed in Hong Kong in 1990–1995 when public health legislation was enacted to reduce the sulfur content of fuel.11 We included lung cancer deaths in the analysis because lung cancer is strongly related to both active and passive smoking, although the long latent period and slow progression of the disease mean that we would be unlikely to observe changes in deaths from lung cancer in these short-term timescales. We found no significant change in the annual proportional change in mortality from lung cancer.

The US Surgeon General’s analysis of the findings from studies on the impact of smoke-free interventions around the world concluded that there was an impact of effective smoke-free laws on coronary events in younger people, but they did not find evidence of an effect on older people.10 In their analysis, they did not distinguish between fatal and non-fatal events, nor between mortality and admissions. In this new analysis, we are able to supplement and strengthen the Surgeon General’s analysis with our data showing declines in mortality from IHD, AMI and CVD rubrics contributed mainly by those aged 65 years or older. The Surgeon General's report attributed the lack of effect on older people to the possibly lower exposure of older people to bars and similar venues, and also a potentially smaller relative risk associated with SHS exposure in older people. Also, in Hong Kong, older people are not likely to frequent bars but many of them do spend several hours in local tea houses and mahjong parlours. Eating out with the entire family is also a common activity in Hong Kong. Before the smoke-free laws, these venues tended to have high levels of SHS to which patrons and staff were continuously exposed. It is possible that this relatively higher exposure of older people in Hong Kong led to our findings of an impact of the smoke-free law on mortality of those aged 65 years or older. A cross-sectional study on exemption of licensed catering premises from smoke-free law in Hong Kong suggested that the indoor fine particulate (PM2.5) concentrations could be four times as high in smoking premises as in non-smoking premises, and higher urine cotinine measures were found in catering staffs.15 The results also provided evidence on the impact of the smoke-free law.

The lack of a significant effect on some of the outcomes in younger people in our models may have two explanations. First, in Hong Kong, younger people do not spend as much time in bars as their counterparts do in the countries from which the Surgeon General’s report data were derived. Second, the legislation in Hong Kong initially allowed exemptions for some bars, so the younger people who visited these bars did not experience much reduction in exposure until mid-2009, when the legislation came fully into force. In fact, those who frequented the bars with exemptions from the legislation between January 2007 and July 2009 may have been exposed to even higher levels of SHS due to smokers congregating in these venues. However, we did note a significant decline in CBD mortality among younger people. We also found a decline in deaths from RD in older people, which adds to the Surgeon General's findings of suggestive evidence of an impact on other outcomes.


The findings presented in this study added to the evidence base, as summarised in the Surgeon General’s report, extending the impact of effective smoke-free legislation to those aged 65 years or older and to CBD events in younger age groups. They also reinforce the urgent need for comprehensive and effective smoke-free laws without exemptions if the full extent of the health gains are to be achieved.

What this paper adds

  • Multidisciplinary independent, scientific, peer-reviewed reports, including systematic reviews, from around the world, have demonstrated the harmful and lethal effects from secondhand smoke exposure. A recent review concluded that public health legislation to create smoke-free environments does protect public health and has resulted in decreased cardiac events in those aged under 65 years.

  • This study adds, to the global knowledge base, evidence of reductions in mortality risks among older people for cerebrovascular, respiratory and heart disease, following smoke-free legislation. These reductions in mortality risk translate conservatively to about 1000 fewer deaths per year in Hong Kong.


The authors would like to thank the Health and Health Services Research Fund for the funding of the original study, and the Census and Statistics Department for the provision of death registration data in Hong Kong, and the Environmental Protection Department for the provision of data on air pollutant concentrations in Hong Kong.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Deceased 19 December 2014

  • Contributors TQT was responsible for the study design, advice for the data analysis, writing and critical review. SMG was responsible for the study design, data interpretation, writing and critical review. JCS was responsible for data collection, data analysis, data interpretation and writing. JC was responsible for data collection and writing; EKPC was responsible for statistical advice. CMW was responsible for the study design and advice for the data analysis. AJH was responsible for research questions, study design, data interpretation, writing and critical review.

  • Funding This study was funded by the Tobacco Control Office, Department of Health, the Hong Kong Special Administrative Region.

  • Competing interests None declared.

  • Ethics approval The study obtained ethics approval from the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (HKU/HA HKW IRB), IRB reference number UW 13-507.

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