Thursday, April 09, 2009

New Study of National Heart Attack Admissions and Mortality Finds No Evidence of a Short-Term Effect of Smoking Bans


Most Definitive Study to Date Refutes Conclusions of Many Earlier Studies and Demonstrates Why These Studies Obtained Positive Findings


A new study by researchers from the RAND Corporation, Congressional Budget Office, University of Wisconsin, and Stanford University is the first to examine the relationship between smoking bans and heart attack admissions and mortality trends in the entire nation, using national data. All previous U.S. studies only examined one particular city. In contrast, this study examined data from the Nationwide Inpatient Survey (NIS), which is nationally representative and includes 20% of all non-federal hospital discharges in the United States. The study has been published as Working Paper 14789 of the National Bureau of Economic Research Working Paper Series.

Study citation: Shetty KD, DeLeire T, White C, Bhattacharya J. Changes in U.S. hospitalization and mortality rates following smoking bans. Working Paper 14790. Cambridge, MA: National Bureau of Economic Research, 2009. The study is copyright 2009 by Kanaka D. Shetty, Thomas DeLeire, Chapin White, and Jayanta Bhattacharya.

The key conclusions of the study are as follows:

1. "In contrast with smaller regional studies, we find that workplace bans are not associated with statistically significant short-term declines in mortality or hospital admissions for myocardial infarction or other diseases."

2. "An analysis simulating smaller studies using subsamples reveals that large short-term increases in myocardial infarction incidence following a workplace ban are as common as the large decreases reported in the published literature."

The study uses state and local workplace smoking ordinance data from the American Nonsmokers' Rights Foundation tobacco control database for the years 1990 through 2004 and national data on heart attack admissions and mortality from the National Inpatient Survey (1993-2004), as well as from the Multiple Cause of Death database (1989-2004). Using a fixed effects regression model, the authors analyze outcomes (heart attack admissions and mortality) before and after the implementation of all workplace smoking bans in the nation, as identified in the ANR database.

The regression coefficient of interest represents the change in heart attack admissions or mortality associated with the implementation of a smoking ban, while controlling for secular trends in the outcome variable as well as regional differences in outcomes and regional differences in population size, number of physicians, number of hospital beds, household income, and percent of the population in the labor force.

The study also simulates the results from the comparison of all possible combinations of regions in the U.S. by examining subsets of the data, where one region is an intervention unit and the other is a comparison or control unit. The authors are therefore able to simulate what the results would be for each of the 19,406 possible comparisons of intervention and control regions in the country.

The main study result is that the regression coefficients for the smoking ban variable is not statistically significant in either the heart attack admission or heart attack mortality model, indicating that the smoking bans had no effect on either heart attack admissions or heart attack mortality.

The study estimates that workplace smoking laws reduced heart attack mortality by a non-significant 1.5%, with a 95% confidence interval of -4.8% to +1.8%. The study estimates that workplace smoking laws reduced heart attack admissions among 18-64 year-old adults by a non-significant 4.2%, with a 95% confidence interval of -10.2% to +1.7%.

Most interestingly, the simulation of all possible comparison studies of local regions in the U.S. finds that just as many studies would find an increase in heart attacks associated with smoking bans as would find a decrease in heart attacks (see Figure 2). The mean difference in heart attack admissions among all studies was 0. The exact same result was found for heart attack mortality (see Figure 3).

The paper concludes: "We find no evidence that legislated U.S. smoking bans were associated with short‐term reductions in hospital admissions for acute myocardial infarction or other diseases in the elderly, children or working‐age adults." ...

"We also show that there is wide year‐to‐year variation in myocardial infarction death and admission rates even in large regions such as counties and hospital catchment areas. Comparisons of small samples (which represent subsamples of our data and are similar to the samples used in the previous published literature) might have led to atypical findings. It is also possible that comparisons showing increases in cardiovascular events after a smoking ban were not submitted for publication because the results were considered implausible. Hence, the true distribution from single regions would include both increases and decreases in events and a mean close to zero, while the published record would show only decreases in events. Thus, publication bias could plausibly explain why dramatic short‐term public health improvements were seen in prior studies of smoking bans."

The Rest of the Story

Without a doubt, this is the most definitive study yet conducted of the short-term effects of smoking bans on cardiovascular disease.

To give you an idea of the scope of this study compared to previous ones, the Helena study involved a total of 304 heart attack admissions in one community over a period of six months. This study examined a total of 217,023 heart attack admissions and 2.0 million heart attack deaths in 468 counties in all 50 states over an eight-year period.

This study fails to find any significant short-term effect of smoking bans on heart attack admissions or heart attack mortality, although a small effect cannot be ruled out. The study refutes the claims from previous studies that smoking bans result in a short-term reduction in heart attacks in the range of 20-40%, as many anti-smoking groups are asserting.

The most important finding of this study is that there are just as many smoking ban communities in which heart attack admissions and mortality have increased in comparison with control communities as there are smoking ban communities in which heart attacks have decreased relative to control communities. The mean difference was found to be zero.

Thus, the study not only fails to find a short-term effect of smoking bans on heart attacks, but it also explains the positive findings of previous studies. What appears to be going on is what is referred to as publication bias.

What this means is the following: if one wanted an unbiased estimate of the effect of smoking bans on heart attacks, one would ideally include all communities that have enacted a smoking ban. In reality, what has occurred is that there have been what essentially amount to anecdotal studies conducted in several communities. These few studies have been published in the literature. It is possible that similar studies were conducted that failed to find an effect and that these studies were therefore not published. It is also possible that the finding of positive results in the few communities studied was essentially a result of chance. There may be other reasons why certain communities were selected for study. For example, researchers may subconsciously have a feeling that heart attacks have decreased and may want to conduct research to confirm if this is the case or not.

For whatever reason, it is apparent that the sample of communities in which this issue has been studied represents a biased sample of all possible studies that could have been conducted. When one examines, systematically, data for all regions in which smoking bans have been enacted, one fails to find a significant effect of smoking bans on either heart attack admissions or heart attack mortality.

This doesn't meant that smoking bans will not eventually result in reductions in cardiovascular disease. That will take considerable time. It just means that the conclusions of anti-smoking researchers and groups that heart attacks drop immediately upon passage of smoking bans appear to be wrong.

When I first questioned the validity of the conclusions of the Helena and Pueblo studies, I was attacked by many of my colleagues, called a traitor, and expelled from list-serves because they did not want me to spread my dissenting opinion. Now, it has become quite clear that my skepticism was well-placed to begin with. The most definitive study to date has refuted the findings of these studies.

It is important for me to point out that I never took issue with the data presented in the individual studies. There were demonstrable declines in heart attack admissions in the intervention cities. What I questioned was the conclusion that these declines were attributable to the smoking ban, rather than to random variation in the data and/or secular changes in heart attacks, which we know are declining throughout the country, even in the absence of smoking bans. There have been considerable advances in surgical and pharmaceutical treatments for coronary heart disease and these may well be the major contributing factors to the observed decline in heart attack rates.

This new research demonstrates why tobacco control researchers and groups have to be very careful in drawing causal conclusions, and why it is better to uphold high standards of scientific rigor rather than to jump to premature conclusions that may later be shown to be incorrect. Tobacco control groups, and the movement as a whole, will lose some scientific credibility because of these new findings which do not support their conclusions. But it is too late to retract those conclusions because they have already been widely disseminated through the media.

It is far better to get it correct the first time. But that requires adherence to solid science. You cannot allow advocacy concerns and goals - no matter how noble they may be - to interfere with the process of objective scientific evaluation. That is exactly what has occurred in tobacco control.

No comments: