Hardening the Methods, a Comment on Fagerström & Furberg 2008

Fagerström & Furberg (2008) found that, across countries, smoking prevalence was associated inversely with ratings on the Fagerström Test for Nicotine Dependence (FTND) 1. They conclude that this finding supports the ‘hardening hypothesis’, which states that when smoking prevalence decreases, the less dependent smokers quit first and the remaining population of smokers consists of highly dependent smokers. There are a number of problems both with this paper and this hypothesis. This study is based upon a selected sample of 15 studies from 13 countries, conducted between 1992 and 2007, in samples that were not always representative of the general population. No systematic attempt was made to collect all relevant studies, and the 15-year interval was not taken into account in the analysis, even though smoking prevalence changed greatly in this interval. A sound study should be based upon a comprehensive literature review of population-based samples, not on an arbitrary selection of a few studies. In fact, the ‘hardening hypothesis’, that was first formulated 5 years ago 2, was never supported by data 3, 4. We recently tested the ‘hardening hypothesis’ using data from the Behavioral Risk Factor Surveillance System (BRFSS). Across the United States, there are sufficient variations in smoking prevalence (from 14% in Utah to 32% in Nevada) and in anti-smoking policies to allow for a test of this hypothesis. We found that, across states, a higher smoking prevalence was associated with a higher proportion of ‘heavy’ smokers (who smoked 21+ cigarettes/day), but with a lower proportion of daily smokers who tried to quit smoking for 1+ days in the previous year, and with lower levels of motivation to quit among smokers 5. We interpret these data as a rejection of the ‘hardening hypothesis’. A low smoking prevalence is usually the sign of effective policies and of an active anti-smoking climate, in which smokers have several incentives to smoke less and thus to be less dependent. In particular, smoking bans make it more difficult for smokers to smoke as much as they like, high taxes make cigarettes less affordable and social pressure elicits smokers to either reduce or quit smoking. So why did Fagerström & Furberg find such a strong correlation between prevalence and FTND? Looking at the scatterplot of their data, four outliers emerge: the three US studies and the study in Sweden. These four studies report high FNTD scores (median 4.5) and low smoking prevalence (median 19.5%), whereas in the other 11 studies conducted in Europe, there are lower FTND scores (median 3.1) and higher smoking prevalence (median 33%). Sweden is a special case because the prevalence of smoking is low among men, who use snus instead, and this forces the correlation. The prevalence of tobacco use (snus + smoking) in Sweden is, in fact, very similar to the smoking prevalence in the rest of Europe, where snus is prohibited 6. Data from more than three selected studies are needed to determine whether smokers in the United States really have higher FTND scores than European smokers. Within the other 11 European samples included in this paper, there is no correlation between smoking prevalence and FTND (r = −0.02, P = 0.9). It is true that heavy smokers may find it hard to quit even under a great deal of social pressure, in particular if they have psychiatric comorbidities. It is also quite possible that clinicians receive more highly dependent smokers now than they did previously, which led them to formulate the hardening hypothesis. However, this is not because all smokers in the population are more dependent, but probably because there is more social pressure upon heavy smokers to quit smoking and to seek help from clinicians. The ‘hardening hypothesis’ should probably be rejected. None.