Thursday, April 02, 2009

Journal Club - 2nd April 2009

It was suggested in this week's journal club to post an overview of the discussion on the Geary blog. The hope is that this will encourage further discussion and participation.

This week's paper was:

Shetty, K.D., DeLeire, T., White, C. and Bhattacharya, J. (2009). "Changes in U.S. Hospitalization and Mortality Rates Following the Smoking Bans" NBER Working Paper, WP14790

This paper used nationwide data for the first time to analyse whether smoking bans had a short time effect on mortality and hospitalization rates. "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".

Their empirical strategy was as follows: they "compared trends in regions where smoking bans were implemented to those in control regions where smoking restrictions were not imposed". They used a fixed effects model, specification below:


OUTCOMEit represents the number of AMI deaths or hospitalisations in region i and time t. Bs is the indicator of interest which was tested for significance to test for a structural break. They also tested for admissions in other diseases across different age groups. (0-17, 18-64, 65+)

Their results found that that "workplace smoking restrictions are unrelated to changes in all-cause mortality or mortality due to other AMI in all age groups". These results are contrasted with the results found by previous research; (Sargent, Shepard et al. 2004; Bartecchi, Alsever et al. 2006; Cesaroni, Forastiere et al. 2008). The authors suggest possible explanations for this including publication bias and small sample comparisons may have led to atypical findings due to large year to year variation in myocardial infarction death and admission rates.

In their conclusion, the authors also highlighted the importance of further research into non-health related benefits of these bans to non-smokers, the impact on smoking rates as well as research into the long term health benefits of smoking bans.


Conversation from the journal club revealed different opinions on the paper. Some participants believed the paper was well-written and made an important contribution to the literature. Examples of the importance included policy makers who might incorrectly use the previously suggested reduction in myocardial infarction rates due to the smoking ban as evidence for the short term benefits of a smoking ban. The paper was also applauded for highlighting the potential pitfall of publication bias.

Other participants believed that this paper placed too much of an emphasis on myocardial infarcation as the primary short-term benefit of a smoking ban. The paper refers to a U.S. Department of Health and Human Services (2006) report on the health consequences of second hand smoke to back this position up. However, in this report very little discussion is given to myocardial infarction and there doesn't appear to be a clear cut relationship in the medical literature for this association. This report does conclude that there are causal relationships between second hand smoke exposure and sudden infant death syndrome, low birth rate and respiratory illnesses (which could be examined in the short term) as well as middle ear syndrome, asthma in children, wheeze illnesses, impaired child lung function, lung cancer and coronary heart disease.

I was of the second opinion. I also thought there were potential methodological issues including

1) Most workplaces had already implemented workplace bans prior to the legislation. This would result in a downward bias on the Bs estimate.

2) There is wide year to year variation in myocardial infarction rates. This could lead to imprecise and varying Bs coefficients which might lead you to fallaciously reject the null.

3) If there is a big lag between legislation and enforcement this could also result in a downward bias on the Bs estimate.

4) If you are using a fixed effects model where you are including a time trend, and a large proportion of the country brought in a smoking ban in the same year, then some of the impact of the smoking ban might be captured by a decrease in the average yearly rate resulting in a downward bias on the Bs estimate.

5) The authors state they tested for changes in admissions in other diseases. However these other diseases aren't specified. The medical literature suggests short term benefits such as reduction in sudden infant death syndrom and low birth rate, however using their age bands of 0-17, 18-64 and 65+ and only hospital admission data you would not capture these reductions.

Please leave comments or correct me on any of the above.


Colm Harmon said...

This is a progressive move and to be welcomed. I do think that if we do this - and we should - we should recognise that this is an open forum and we should let the authors of the papers know that we are studying their work in the journal clubs. I would welcome their input, and it would sharpen our insights when we know the authors are also aware and have a right of response. In fact, ideally we should set up a Journal Club forum page, let authors know we are discussing their work, and encourage debates - a sort of online seminar.

Liam Delaney said...

yes, very good idea to post things after the journal clubs. I would not pursue the score-out-of-ten rating.

Kevin Denny said...

I was thinking of getting a summer student to see could one replicate the study for Ireland.If anyone is interested in helping out let me know.