Even the FDA Itself Concludes that Graphic Warning Labels Will Have Minimal Impact, and Perhaps No Impact, on Cigarette Smoking

In a little-known and little-publicized analysis, the FDA itself has concluded (in its Regulatory Impact Analysis) that the graphic warning labels will have very little impact, and perhaps no impact at all, on cigarette smoking. The analysis, to which the tobacco companies refer in their legal brief seeking an injunction against the warning label requirement, predicts that the graphic warning labels will only reduce smoking prevalence by 0.088 percentage points (less than 0.1 percentage points). Importantly, the 95% confidence interval for the effect includes zero, meaning that it is not statistically different from zero (see: Federal Register /Vol. 76, No. 120 /Wednesday, June 22, 2011).

In very clear terms, the Regulatory Impact Analysis emphasizes that its scientific evaluation concludes that the predicted effect of the graphic warning labels is not statistically different from zero: "FDA has had access to very small data sets, so our effectiveness estimates are in general not statistically distinguishable from zero; we therefore cannot reject, in a statistical sense, the possibility that the rule will not change the U.S. smoking rate. Therefore, the appropriate lower
bound on benefits is zero." (Federal Register 76 at 36776)

The Rest of the Story

What is even more striking than the fact that FDA has itself predicted a minimal impact of the graphic warning labels is the fact that its own analysis provides very flimsy evidence that the warning labels will have any impact. The analysis is based on a comparison of smoking rates in the U.S. and Canada before and after the implementation of graphic warning labels in Canada. The analysis compares smoking rates after accounting for the effect of cigarette tax changes, and attributes all unaccounted for differences to the Canadian warning labels. Given the huge variability in the data, examining the confidence intervals around the point estimate of the warning label effect is essential.

In this case, there was no observed significant effect of the graphic warning labels on smoking prevalence in Canada, once the tax increase was accounted for. Were this a scientific paper instead of a regulatory impact analysis, the paper would be forced to conclude that there was no significant effect of the graphic warning labels in Canada on cigarette smoking prevalence over an 8-year follow-up period.

Even more striking is the tremendous variability in the data. The analysis is sensitive to single outlying points, most importantly, the observed smoking prevalence change in Canada during the first year after the warning labels. This suggests that any minimal impact that was observed occurred immediately after the warning labels were introduced, due to their shock value, and that there was no effect whatsoever following that initial shock.

In fact, if you delete the year 2001 from the analysis, the observed difference in unexplained smoking rates between the U.S. and Canada is -0.23 percentage points, indicating that after the initial year of the warning labels, this intervention actually increased smoking prevalence in Canada by 0.23 percentage points.

Of course, an alternative explanation is that the widely fluctuating data are due simply to the large amount of random variation in the data, and that the overall analysis finds no statistically significant effect of the graphic warning labels on smoking prevalence.

Most significant, but shocking, of all, if the analysis is restricted to the period 1994-2008 (not including 2009), then the results show that the graphic warning labels in Canada increased smoking prevalence by 0.066 percentage points.

Thus, had the same analysis been conducted one year ago, before 2009 prevalence data were available, the very same report would have been forced to conclude that the proposed regulation is predicted to increase smoking rates and cost the U.S. billions of dollars.

This shows how flimsy the FDA's analysis is, and how scientifically shaky is its conclusion that the graphic warning labels will have a substantial negative effect on cigarette consumption.

The rest of the story is that the FDA's own analysis demonstrates no statistically significant impact of the graphic cigarette warning labels on smoking prevalence in Canada. This evidence - presented by the FDA itself - supports my earlier conclusion that the proposed graphic warning labels in the U.S. will likely have a minimal impact on cigarette smoking, and that this miminal effect will occur due to the immediate shock value of the warning change, not to any sustained effect of the warnings.

This analysis could also have implications for the FDA's ability to defend the graphic warning labels in court, since the FDA must show that the intervention will advance a substantial government interest, and if the FDA cannot provide evidence of a statistically significant effect of similar warnings in other countries on cigarette smoking, it makes it difficult to convince a court that the proposed regulation will advance the government's interest in reducing cigarette smoking rates.

As with so many other aspects of science that involve "unfavorable findings" for tobacco control, this is yet another one that you are not going to see disseminated by anti-smoking groups. Instead, you'll hear the rhetoric about how these new warning labels are going to "save countless lives." Unfortunately, the scientific evidence indicates otherwise: they will have a minimal effect and will not substantially lower cigarette company sales or profits.