|Home | About | Journals | Submit | Contact Us | Français|
The estimated effects of recent pubic and workplace smoking restriction laws suggest that they produce significant declines in community rates of heart attack. The consistency of these declines with existing estimates of the relative risk of heart attack in individuals attributable to passive smoking exposure is poorly understood. The objective is to determine the consistency of estimates of reductions in community rates heart attacks from smoking restriction laws with estimates of the relative risk of heart disease in individuals exposed to passive smoking.
Meta-analysis of existing estimates of declines in community rates were compared to a mathematical model of the relationship between individual risk and community rates. Outcome measure is the ratio of community rates of acute myocardial infarction: after, divided by before, a smoking restriction law. There is a significant drop in the rate of acute myocardial infarction hospital admissions associated with implementation of strong smokefree legislation. The primary reason for heterogeneity in results of different studies is the duration of follow-up period after adoption of the law. The pooled random effects estimate of the rate of acute myocardial infarction hospitalization 12 months following implementation of the law is 0.83 (0.80, 0.87) and this benefit grows with time. This drop in admissions is consistent with a range of plausible individual risk and exposure scenarios.
Passage of strong smokefree legislation produces rapid and substantial benefits in terms of reduced acute myocardial infarctions and these benefits grow with time.
Passive smoking increases the risk of coronary heart disease and acute myocardial infarction (AMI) in non-smokers,1–3 which has lead to adoption of many laws making public areas and workplaces 100% smokefree. The risk of AMI falls rapidly following smoking cessation4 and the effects of secondhand smoke (SHS) on many biological mediators that lead to heart disease occur rapidly and are nearly as large as smoking.5–10 It is reasonable to expect that implementing strong smokefree laws would lead to a reduction in AMI. Several such laws in North America and Europe have been studied as natural experiments to estimate the reduction in community AMI risk,11–24 with reductions ranging from 11% (Italy12 and Ireland15) to 40% (Montana22). These reductions seemed large compared to early estimates of an approximately 30% increase in risk of heart disease associated with chronic SHS exposure.5 25 These estimates may underestimate the actual risks associated with passive smoking because of a downward bias due to exposure misclassification; Whincup et al26 reported substantially higher risks of major coronary heart disease when stratifying risk based on levels of cotinine (a biomarker for SHS) than self-report, which should give a more accurate risk estimate because of more accurate SHS exposure assessment. We update previously published meta-analyses27 28 (that only considered the first 4 and 8 studies, respectively) by adding more studies and accounting for the length of follow-up. The present analysis demonstrates that the community risk reduction associated with smokefree laws grows with time and are consistent with a wide range of actual observed individual risk and exposure scenarios.
This study uses a simple random effects meta-analysis and a meta-regression to estimate the reduction in the community rate of heart attack as a function of time since smokefree laws went into effect. Simulation estimates of the reduction in community rates of heart attack calculated as a function of individual AMI risk due to exposure to passive smoking produces estimates consistent with the observed reductions in community risk.
The first paper reporting a drop in acute myocardial infarctions following implementation of a comprehensive smokefree law (in Helena, MT) was published in 2004,22 We used PubMed, Science Citation Index and Social Science Citation Index to locate 9 papers published since then. In addition, we identified 3 studies (Ireland, Massachusetts and France) which were reported at meetings. One city (Pueblo, CO) was used to publish two reports, one on results through the first 18 months after the law went into effect13 and one after 36 months11 and data from Piedmont, Italy was used as part of one study 2 months after the law went into effect24 and another of Piedmont alone after 6 months.12 Even through there was some overlap in the data used in Pueblo (reuse of the same baseline data) and Italy (Piedmont represented 25% of the total cases in the 2 month study), we treated these four studies as independent observations. Table S-1 in the Online Data Supplement provides detailed descriptions of all the studies and available data on changes in SHS exposure.
The studies were combined using a random effects meta-analysis. The expected relative reduction in community rates of AMI as a function of follow-up period was estimated using
See Table S-1 in the Online Data Supplement for detailed descriptions of all the studies located.
Calculations were done using Stata 9.0 procedures metan, metareg and metabias.29
The simulation estimates used a model of the community rates of AMI as a function of the distribution of individual relative risks due to current active and passive smoking exposure, the prevalence of active and passive smoking, quitting smoking due a smoking restriction law and effect of lower SHS exposure among people who continue to have some exposure after the law goes into effect.
The incidence rates of AMI in the community before (rb) and after (ra) the smoking law are
The prevalence of passive, pt,p, and current active smoking, pt,c, are population prevalences. The population prevalence of passive smoking exposure,pt,p, is lower than the corresponding prevalence in the non-smoking population, which is pt,p/(1–pt,c).
The parameter estimates are average values for one year before and after the law. The proportion of current active smokers quitting due to a smoking law, pq, and relative risk of current active smokers who quit due to the law are assumed to be the average values over six to eighteen months following the law.
The model assumes that only non-smokers are affected by passive smoking and current smokers are not affected by other smokers’ passive smoke. The AMI risk for recent quitters is determined only by their residual risk due to past current active smoking (and not current passive smoking), which underestimates AMI risk over the year following cessation. The base incidence rate of community AMI, i, is assumed constant before and after the law.
The parameters and their distributions are in Table 1.
We consider four distributions of individual relative risks.
The low estimate of 1.31 (95% CI 1.21, 1.41), from Barnoya and Glantz’5 meta-analysis of 29 long-term studies of the risk of AMI and other cardiovascular disease outcomes, are based (with one exception26) on self-reported SHS, which tend to underestimate total exposure. These studies likely underestimate individual risk associated with SHS because people considered in the unexposed reference (the denominator) in the risk estimate should have been in the exposed group (the numerator).
Whincup et al26 published the only longitudinal study where subjects were classified based on cotinine levels at baseline, with the reference group consisting of subjects with serum cotinine ≤ 0.7 ng/ml. This cutoff is high enough to include both unexposed nonsmokers and light passive smokers (This decision may have been necessary because, in England during the 1978–80 period, at baseline almost everyone had some detectable exposure to secondhand smoke). Not surprisingly, Whincup et al26 found higher risks than the studies based on self-report.
We considered three possible risks from this26 study. 1.49 (95% CI 1.03, 2.14) represents the AMI relative risk during the 20 year follow-up for the midrange of baseline serum cotinine levels (1.5–2.7 ng/ml). The other two estimates are 1.95 (95% CI 1.09, 3.48), representing the risk from comparing people with cotinine above and below 0.7 ng/ml followed for 5–9 years, and 3.73 (95% CI 1.32, 10.58) for people followed for 0–4 years. Because cotinine was only measured at baseline; the reliability of this baseline cotinine as a measure of actual exposure is probably most representative of actual exposure during the 0–4 year follow-up.
The community effects of reduced SHS exposure were estimated for three cases to describe the range of prevalence of passive smoking exposure before and after a smoking restriction law, each based on cotinine-validated population data. We considered a person “unexposed” if their level of serum cotinine was below the limit of detection (0.0530 31 or 0.1032 ng/ml). We estimated D, the reduction in SHS exposure among people who continued to be exposed as the ratio of geometric mean cotinine after the law to geometric mean cotinine before the law.
The first case (“Large Drop in Exposure”) is based on cotinine measured in 1999–2002 in the large US National Heath and Nutrition Examination Study (NHANES) among non-smokers 20 years of age and older,31 which represents the average experience of US communities with and without smoking laws. Forty-six percent of non-smokers living in jurisdictions without a law had detectable cotinine (≥0.05 ng/ml), compared to 13% living in jurisdictions covered by strong laws, a 33 percentage point difference in prevalence. The ratio of the geometric mean cotinine with laws compared to those without the law was 0.19.
The second case (“Moderate Drop in Exposure from Moderate Base”) is based on the New York State population with detectable cotitine (≥ 0.05 ng/mL) before and after its strong state smoking restriction law went into effect.30 The survey sample excluded New York City and Nassau County, which had already passed strong smoking restriction legislation. The sample did, however, include 3 counties (Dutchess, Suffolk, and Westchester) which had strong local smokefree laws in place, representing 27% of the sample (American Nonsmokers’ Rights Foundation, personal communication, January 27, 2009). Including these counties in the sample probably reduced the baseline level of cotinine exposure and change associated with implementation of the state law. The prevalence of non-smokers with detectable cotinine dropped from 68% to 52%, a 16 percentage point drop, comparing before and after the state law went into effect. The geometric mean cotinine following the law was 0.54 of that before the law.
The third case (“Moderate Drop in Exposure from High Base”) is based on results before and after the Scottish smokefree law went into effect.32 Before the law, 89% of nonsmokers had detectable (≥0.10 ng/ml) levels of cotinine, which fell to 72% after, a 17 percentage point drop. The high level of SHS exposure before the law was comparable to that observed in the US in 1988–91, before most smoking restriction legislation was enacted33 (No standard errors are available for Scottish exposure prevalence; assuming lack of correlation with the other parameters, the omission of the standard errors will reduce the standard error of the estimated community risk, therefore will underestimate the variance of RAMI.and produce conservative results). The geometric mean cotinine following the law was 0.61 of that before the law. The high level of exposed people following the law may reflect the fact that exposure to secondhand smoke at home remained high after the law.
The population prevalence of current smoking before a law, pb,c, was set to 23% (95% CI 21% to 25%) for all scenarios, the pooled random effects estimate from the latest available year for the places that enacted laws.34–36 The distribution of the proportion of active current smokers quitting within a year of the law, pq, was from a meta-analysis of the effects of smokefree workplaces.37 Distributions of the individual relative risks for current smokers, Rc, and those who quit due to the smokefree law, Rq, were estimated using gender specific rates for adults4 using never-smokers as the reference group.
The simulation estimates of the ratio of community rates of AMI after, divided by before, the law, RRAMI were calculated for 48 combinations of parameters: the passive smoking exposure scenarios (3 cases), individual risk for AMI associated with passive smoking (4 cases), presence or absence of a dose-response effect among people still exposed to SHS after the law took effect (2 cases) and presence or absence of smoking cessation due to the law (2 cases). The combinations of individual relative risk, dose-response and smoking cessation (16 cases) were termed ‘scenarios’ and grouped by the three passive smoking exposure parameters.
Crystal Ball Version 5.238 was used for the estimates using a Monte Carlo simulations with 20,000 trials.
Calculations and details on sources of the parameters are in the Online Data Supplement.
Several sensitivity analyses were used to determine the robustness of the results to changes in the sample and statistical method. These analyses include dropping the initial (18 month) estimate from Pueblo CO and the estimate for Piedmont, Italy from the Vaselli et al.24 study, so that all remaining observations were independent; adjustment for an insignificant trend in AMI baseline incidence, i for estimates Scotland21 and Piedmont, Italy,12 which had not accounted for possible secular trends in AMI;; and use of non-parametric tests for statistical significance of the slope parameter; and alternative random effects estimators.
Sensitivity analyses for the simulation included alternative formulas for dose-response in non-smokers who remain exposed to passive smoking before and after the smokefree law, alternative methods of pooling current smoking prevalence, and different age adjustments for average relative risk due to current smoking.
We used a random effects model for the meta-analyses because of significant heterogeneity in the estimates, which yielded a pooled RRAMI of 0.81 (95% CI: 0.78, 0.85) (Figure 1). A funnel plot and Begg’s test did not suggest publication bias (Figure 2).
Some of the heterogeneity could be due to differences in end points, confounding variables considered, analytical methods, changes in level of SHS exposure after the law, and duration of follow-up.
There was a significant relationship between the duration of study follow-up and RRAMI (Figure 3) with ln RRAMI falling by 0.0113/month (SE 0.002, P<.0005). The intercept was not significantly different from 0 (−0.046, SE .037 P= 0.242). Duration of follow-up accounted for 76% of the between-study variance. The meta-regression between ln RRAMI and duration of study (Figure 3) provided a good estimate of changes in risk over time for the observed period. We used the results of the meta-regression in Figure 3 to standardize all relative risks and associated upper and lower bounds of the 95% confidence intervals to the values that what would be observed at 12 months post implementation of each law with
where T duration of study follow-up in months. This adjustment substantially reduced the variability in the estimated values of RRAMI(12) (Figure 4). The random effects meta-analysis of the resulting values adjusted to 12 months yielded a pooled risk estimate of 0.83 (95% CI: 0.80, 0.87). While lower than before, there was still significant heterogeneity, probably reflecting the differences noted above (other than study duration). Geographic heterogeneity may be an important factor in remaining heterogeneity: after adjustment, the reductions in AMI risk appear slightly larger in the United States than Europe and Canada.
Eight of the 48 scenarios produced point estimates of RRAMI that were within the 95% confidence interval for the pooled risk estimate produced by the meta-analysis of the individual studies at 12 months, RRAMI(12) (0.80 to 0.87), and 28 of the 48 scenarios produced interval estimates (defined as the 95% CI) of RRAMI that overlapped the 95% confidence interval for the pooled estimate of individual risks (Figure 5). The scenarios that are most consistent with the pooled community risk ratio at one year, RRAMI, (12), are those with individual SHS relative risks of 1.95. The presence of a dose-response and effect on quitting were not required to obtain reasonable agreement with the observed population risk.
The sensitivity analyses did not produce noticeable changes in the met-analysis or simulation results.
After adjusting for variable length of follow-up the individual community studies yield remarkably consistent elevations of the community benefits of reduced AMIs following implementation of strong smokefree legislation (Figure 3), with about a 15% drop during the first year and continuing exponential declines, reaching about 36% in 3 years (the limit of currently available data). The fact that the intercept of the risk reduction with time has an intercept of zero is consistent with the results of the very large study of the effects of a strong smokefree law in New York State17 which tested for an immediate offset (change in intercept) and interaction between the presence of a law and time after the law went into effect. The New York study found no immediate shift in the AMI rate immediately after the law went into effect, but that the rate of AMI declined continuously with time after the law. Results from Bowling Green, OH18 and Pueblo, CO11 also showed that AMI rates fell with time.
Eight simulation scenarios have point estimates closest to the point estimates of the meta-analysis at one year (Figure 3) and are within their 95% confidence intervals. All 8 scenarios use an individual relative risk of 1.94 (from Whincup et al26 for 5 to 9 years follow-up). 6 of these scenarios assumed a dose-response relationship (lower SHS exposures among people still exposed after the law) except 1 using the Large Drop scenario. Four of the 8 scenarios included smokers quitting due to the law.
Richiardi et al39 published a mathematical model based on the assumption that the effect of tobacco smoke exposure was a large acute and short term increase in individual AMI risk (Relative Risk (RR) of 4.5 in the first hour of passive smoking) combined with a 50% reduction in SHS exposure was consistent with a reduction in community-level risk of 5–15%, less than has been actually observed. In contrast, we assume smaller risks that have been observed in long-term epidemiological studies. By accounting for the fact that the community risk reductions fall with time after the law (something not in Richiardi et al’s analysis) and adjusting the observed risks to those corresponding to 12 months of follow-up, our model provides results consistent with the full range of observed risks (Figure 4).
The largest shortcoming in the simulation is the lack of biomarker-based levels of general population exposure to SHS before and after these laws went into effect. Most of the available data relate to self-report levels of SHS exposure, which understates exposure,40 41 with biomarker data (cotinine) generally limited to restaurant and bar workers. The restaurant and bar workers studies are useful to document high compliance with the laws, but biomarker data for population samples would be more useful for evaluating health effects of the law in the entire population. It is important to publish not only the changes in arithmetic and geometric mean cotinine levels, but also results that can be used to estimate the fraction of the population with very low levels (below 0.05 ng/ml in serum) to estimate changes in prevalence of exposure. Changes in current smoking due to smoking laws can noticeably affect the decline in community rates and should also be measured.
All studies with adequate data were included in the meta-regression in order to avoid potential bias due to selection and adjustment of individual estimates. With one exception,32 (omitted from the meta-analysis because of small sample size) there are no reliable data on smoking status at the time of admission for subjects in the studies. It would be desirable to have cotinine measures on admitted patients, both to assess smoking status and level of SHS exposure. Such data would allow estimates of any differential effects of smokefree laws on nonsmokers and smokers and also estimates of the effects of the laws on cessation among people at risk of AMI. Collection and reporting such data would be useful in further refining the linkages level of SHS and risk of AMI.
We did not account for the effect of lower SHS exposure among people who quit smoking following the law or the fact that continuing smokers reduce consumption when smokefree policies go into effect;37 these omissions mean that our model probably underestimates the effect of the law.
The parameters in our model were treated as independently distributed even though some may covary, such as relative risks of current and all former smokers and correlation of prevalence of current smoking and exposure to passive smoking. The parameter distributions were taken from observational studies and may be subject to associated biases.
Analysis of reductions in community rates of AMI shows that the measured effects of smoking laws are consistent after accounting for follow-up period of different studies. Simulations estimates of expected reductions in community rates, based on existing estimates of individual relative risk of AMI due to exposure to passive smoking, are consistent with each other.
This analysis shows, using data from five countries, that passage of strong smokefree legislation produces rapid and substantial benefits in terms of reduced acute myocardial infarctions and that these benefits grow with time.
This work was funded by National Cancer Institute Grant CA-061021. The funding agency played no role in the design or conduct of the research, or preparation of the manuscript.
Journal Subject Codes:
 Acute coronary syndromes
 Acute myocardial infarction
 Health policy and outcome research
The authors have no competing for conflict of interest to declare.