A new study published online ahead of print yesterday in the journal Archives of Internal Medicine concludes that smoke-free restaurant and bar ordinances in Olmsted County, Minnesota produced a 34% decline in heart attacks and a 17% decline in sudden cardiac death. The authors assert that the decline was due to reduced secondhand smoke exposure.
(See: Hurt RD et al. Myocardial infarction and sudden cardiac death in Olmsted County, Minnesota, before and after smoke-free workplace laws. Arch Intern Med 2012; doi: 10.1001/2013.jamainternmed.46).
A smoke-free restaurant law went into effect in Olmsted County on January 1, 2002 and a smoke-free workplace and bar law went into effect on October 1, 2007. The investigators examined age-adjusted rates of myocardial infarction (heart attack) and sudden cardiac death during the 18 months before and after each ordinance.
The results were as follows: "Comparing
the 18 months before implementation of the smoke-free restaurant
ordinance with the 18 months after implementation of the smoke-free
workplace law, the incidence of MI declined by 33% (P < .001), from 150.8 to 100.7 per 100 000 population, and the incidence of sudden cardiac death declined by 17% (P = .13), from 109.1 to 92.0 per 100 000 population."
The paper concludes that these declines are attributable to the smoking bans and that the most likely factor influencing the decreased heart attack and sudden death rates is a reduction in secondhand smoke exposure because of the bar and restaurant smoking bans. According to a Reuters Health article: "The lead researcher on the work said that decline was likely due to less
secondhand smoke exposure in restaurants and bars, as smoke can trigger
heart problems due to its effects on arteries and blood clotting."
According to the news article, the lead study investigator stated: "There have
been lingering doubts among some people about whether or not this was a
real finding. We think we have produced the most definitive results that
anyone can produce related to smoke-free laws and heart attacks."
The Rest of the Story
There are two major problems with the study that lead me to question the validity of its conclusions.
1. There is no comparison group so we have no idea whether the observations in Olmsted County differ from other counties in Minnesota during the same time period.
The study methodology is flawed, in my view, because it doesn't include any comparison group. While the study demonstrates a decline in heart attacks in Olmsted County during the period 2006-2009, it is entirely possible that similar declines took place in other counties in Minnesota. In other words, from this study design, it is impossible to determine whether the observed decline in heart attacks in Olmsted County represents an effect of the smoke-free bar law or simply reflects a secular change in heart attacks that was occurring anyway during this time period.
It is important to note that during the same time period, heart attack rates were declining rapidly throughout the United States. Major advances in the surgical, medical, and pharmaceutical treatment of heart disease occurred during the study period and were associated with substantial declines in the incidence of heart attacks in may regions of the country.
Although the study does not report heart attack trends in the state of Minnesota as a whole with which we can compare the Olmsted County findings, such data are readily available from the Health Care Utilization Project (HCUP) State Inpatient Database. This database reports trends in hospital cases of myocardial infarction in all hospitals in Minnesota during the period 2001-2010.
Thus, we can compare the observed change in heart attacks in Olmsted County with changes in the entire state of Minnesota. The paper reports a 33% decline in heart attacks in Olmsted County from 2001 to 2009, and concludes that this 33% reduction is due to the smoking ban.
Now, it's time for the rest of the story.
According to the HCUP data, the number of heart attack diagnoses in all Minnesota hospitals declined from 10,626 in 2001 to 7,817 in 2009. This represents a 26% drop in heart attacks.
Thus, the actual finding of this paper is a 33% decline in heart attacks in Olmsted County under secular conditions that saw a 26% decline in heart attacks in the state of Minnesota during the same time period.
Thus, it is clear that the observed 33% reduction in heart attacks in Olmsted County during the study period is not attributable to the smoking ban.
Note that even if one looks only at hospital admissions for heart attacks in Minnesota, there was a 23% decline in these admissions between 2001 and 2006 alone (the data is not provided by HCUP past 2006). However, if one extrapolates to the year 2009 based on the secular rate of decline in the 2001-2006 period, the estimated number of hospital admissions for heart attacks in 2009 represents a drop of 34% from 2001.
In other words, in the absence of a smoking ban, heart attack admissions in Olmsted County would have been expected to drop by 34%. In the presence of the smoking ban, heart attack rates dropped by 33%.
Thus, it appears that the observed decline in heart attacks in Olmsted County is roughly representative of the overall trend in heart attacks in the state as a whole. In this light, the present study hardly supports a conclusion that the Olmsted County smoking bans produced a 33% decline in heart attacks.
If these are "definitive" results, I'd hate to see what "tentative" results look like.
2. The study authors present a heavily biased interpretation of the study findings, questioning the overall objectivity of the analysis.
The authors conclusion that the Olmsted County smoking bans led to a 17% decline in sudden cardiac death is based on their finding that from 2001 to 2009, the rate of sudden cardiac death dropped by 17%. Their conclusion is not based on the finding of a statistically significant drop in sudden cardiac death rates. In fact, they acknowledge that the observed 17% decline was not statistically significant and that they cannot conclude from the analysis that the point estimate is significantly different from zero.
Nevertheless, the authors ignore the lack of statistical significance, writing it off by arguing that had there been a higher sample size, the result would have been statistically significant. They write: "We observed a statistically non-significant decline in the incidence of SCD [sudden cardiac death], which may reflect the relatively smaller number of events in the SCD group. These findings suggest that SHS [secondhand smoke] exposure could be a risk factor for SCD."
OK, fine. Let's accept, for a moment, the authors' explanation that we should ignore the lack of statistical significance of this finding because the sample size was small. (I would argue that since the authors set an a priori significance level, this is not appropriate, but let's ignore that for the time being.)
Now, it turns out that the authors also found, but hide from the readers, that there was a 17% increase in sudden cardiac death rates associated with the implementation of the smoke-free restaurant law in Olmsted County. That's right - from 18 months prior to the smoke-free restaurant law to 18 months after the law, the rate of sudden cardiac death (as reported in Table 2) increased by 17% (the identical amount by which sudden cardiac deaths decreased over the entire study period).
This finding, like the 17% decline in sudden cardiac deaths, was not statistically significant. Curiously, however, the authors do not similarly argue that we can ignore the lack of statistical significance of this finding because of a small sample size.
It appears that statistical significance only matters when the finding in question is an "unfavorable" one. But when the finding is a "favorable" one, then statistical significance can be ignored.
In other words, it is quite clear that the investigators were "hoping" to find an effect of the smoke-free laws on sudden cardiac death and that - although I believe subconsciously - they slanted their interpretation of the data in order to favor such a finding.
I can understand this bias because as a lifelong supporter of smoke-free laws, I would love to see data showing that these laws have an immediate effect in reducing sudden death from heart attacks. But you can see how this kind of bias, which is natural, can affect one's interpretation of the data.
In this case, the investigators are presented with identical findings of a non-significant 17% change in sudden cardiac death rates. In the case where the change is a decline (which they favor), the significance does not matter. In the case where the change is an increase (which they disfavor), the significance absolutely matters and the finding can be completely thrown out as if it doesn't exist.
The rest of the story is that the conclusions of this study are invalid because there was no comparison group and one cannot rule out the possibility that secular declines in heart attacks were occurring in Minnesota during the study period anyway (independent of the Olmsted County smoking laws). In fact, when one examines heart attack incidence rates for the state as a whole, one finds that there were comparable rates of decline in heart attacks in Minnesota during the same time period, nullifying and rendering invalid the study's conclusion. Very clearly, the 33% decline in heart attacks in Olmsted County is not attributable to the smoking ban. A 26%-34% decline in heart attacks would have been expected anyway based on secular trends in the state of Minnesota.
More problematic is the revelation that the interpretation of the study data was biased in the direction of finding an effect of the smoking laws. While understandable because of the passion of advocates who have worked much of their careers to promote smoke-free laws, this bias does ultimately affect the integrity of the science being produced by the tobacco control movement, especially in this area of the smoking ban - heart disease relationship.