Yesterday I mentioned a study published in
Pediatrics which claimed that childhood asthma rates fell after the English smoking ban was introduced in 2007. The study concludes:
The implementation of smoke-free legislation in England was associated with an immediate 8.9% reduction in hospitalizations for asthma along with a decrease of 3.4% per year.
The lead author is listed as Christopher Millett, an Australian social scientist who has worked in "obesity prevention" and assists Stanton Glantz's campaign to get smoking out of the movies. Glantz himself—founder of Americans for Nonsmokers' Rights and a veteran campaigner for smoking bans—is a co-author and the study should be seen as another entry in
Glantz's unenviable canon of junk science.
First, it is necessary to understand a few things about asthma. Although many people intuitively believe that there is a close relationship between smoking and asthma, this is not borne out by empirical evidence. As
the BMJ noted in 2005:
A broad consensus exists that in most Western countries the prevalence of asthma increased over the last four decades of the 20th century.
Rates in England and Wales are shown here:
This huge rise in asthma rates is totally inconsistent with the belief that smoking causes or exacerbates asthma, since we all know what happened to
smoking rates in the last four decades of the 20th century.
Asthma rates go up and down quite inexplicably. After rising at an astonishing rate for years, there was a large drop in childhood asthma cases in the UK from the early 1990s until 2001, since when rates have stayed fairly flat. The reasons for this are not known, as the BMJ article says:
While theories abound, we must admit that neither the rise nor the recent flattening or fall in the prevalence of asthma can be explained on the basis of current knowledge. Any single explanation would need to account for both the rise and fall of the prevalence of asthma. A substantial change in prevalence would require a large change in an important risk factor to which a large proportion of the population was exposed. Various explanations have been proposed to explain trends and geographical variations in asthma, including air pollution, tobacco smoke, aeroallergens, diet, and infections in early life, but none has been shown by epidemiological studies to fit the above requirements.
In Canada, the number of children with asthma rose in the 1990s, but the number of childhood asthma
attacks fell.
Childhood admissions for asthma in Australia fell in the 1990s but rose in the 2000s, at a time when smoking bans were introduced (see below). Adult admissions continued to fall, however.
In
America, rates of asthma—including childhood asthma—have been flat or rising slightly since 2001. And in the Netherlands there was a huge increase in childhood asthma prevalence in the 1980s and 1990s which ended around 1995.
As authorities have noted, "there has been no satisfactory explanation for this observation."
What does any of this tell us? Not much, except that it's futile to predict whether rates of asthma will rise or fall in the future since the causes of the condition are not well understood. Asthma rates vary enormously between countries and fluctuate greatly over time, and no one really knows why. It also shows that if smoking (let alone passive smoking) has any impact, it is not big enough to show up in aggregate data. However, as we shall see, Glantz's new study relies on the conceit that asthma rates can be forecast with precision and that the effect of passive smoking can be identified and isolated in the aggregate data.
Although there is no link between smoking and asthma rates at the population level, some epidemiological studies have found a correlation between childhood asthma prevalence and smoking households, just as there is a correlation between asthma prevalence and income (and, perhaps tellingly, also between smoking and income; income may therefore be
the third variable). But, as
Statistics Canada notes:
...there was no difference in the rate of past-year asthma attacks between those in smoking and non-smoking households.
Between 1994/1995 and 2000/2001, in smoking households, the prevalence of asthma among children increased, but past-year attacks among those with asthma decreased. Curiously, only children in non-smoking households experienced an increase in the prevalence of wheezing or whistling in the chest. The presence of other allergenic factors in the home (pets, for instance), which was not assessed in the NLSCY, may be related to the increase in asthmalike symptoms among children in non-smoking households.
So what about Britain in recent years? After dropping off in the 1990s, asthma rates have varied only slightly from year to year since 2000 and there has been neither a rise or decline in the medium-term (which is a source of frustration for
Asthma UK).
The graph below shows hospital admissions for asthma amongst 0-14 year olds in England (the group studied by Glantz and his team). It shows that the smoking ban had no effect on rates of asthma amongst this group (nor,
as I showed in my previous post, did it have any effect on rates amongst all age groups.) The data come from
HES England and cover all NHS hospitals in England. Note that these are financial years, but the smoking ban began in July (2007). The graph shows twelve years' data, whereas the
Pediatrics study shows only 8.5 years. Click to enlarge.
Between 2000/01 and 2006/07, the average number of childhood hospital admissions for asthma was 23,747. Between 2007/08 and 2011/12, the average number of admissions was 23,851. In other words, the rate remained essentially the same before and after the smoking ban. Evidence from
London below (red line) shows much the same picture for adults and there are similar statistics for England and the North West
here.
Whichever dataset you use, it is apparent that the rate of hospital admissions for asthma was higher in 2006/07 than in any other year in the series, albeit only slightly. This is crucial for Glantz/Millett's hypothesis because it allows them to portray the rate as
rising before the ban. No such claim can legitimately be made on the basis of a one-year peak (the data show several other peaks, for example in 2001/02 and 2008/09, which did not represent the start of a long-term rise).
Predicting the number of hospital admissions for asthma even
one year in advance is a mug's game. Nevertheless, having asserted that numbers were on the rise, they make a post hoc quasi-prediction that the rate would have continued to rise had the ban not been introduced.
Finally, having observed that the rate did
not rise in reality, they conclude that the ban resulted in there being 6,802 fewer hospital admissions for asthma than there otherwise would have been. Note that their methodology means that it is not necessary for the rate to decline
at all after the ban for them to claim the smoking ban reduced asthma cases. It is only necessary for the number of admissions in Glantz's hypothetical parallel universe to be higher than the rate recorded by the NHS.
We estimated the number of admissions prevented in the 3 years after implementation of the law by estimating the number of admissions if there were no legislation (the counterfactual) using the coefficients estimated in the model using all the data but setting the law dummy and time after the law variables to 0 for the whole time period, then adding the differences in number of admissions each month between the actual admissions and counterfactual estimate.
This is sheer junk science, using
a trick that we have seen used before for similar purposes. It's a heads-I-win, tails-you-lose fiddle. The only way the authors' hypothesis could be disproved was if asthma admission rates happened to soar up at an exceptional pace straight after the ban. Even that would probably not silence them since they would only say that, had it not been for the ban, the situation would have been still worse.
One only needs to look at the graph displayed in the study itself to see how fanciful it is to say,
as the BBC did, that there was "a sharp fall in the number of children admitted to hospital with severe asthma after smoke-free legislation was introduced in England."* This chart (shown below) shows Glantz's prediction of what would have happened without the smoking ban (red) versus the actual data (blue). The red vertical line represents the smoking ban. Do the rates to the right of the line look lower than those on the left to you?
 |
| Find the decline |
The "prediction" would have been more impressive if it had been made before the ban was introduced, rather than once all the data had been collected. Once they knew the outcome, the slightest adjustment could be used to show an effect from the ban.
How can you predict rates of a disease in a whole nation when you don't know what causes it and you don't know why the rate has risen and fallen in the past? You can't, and neither can Glantz. Instead he extrapolates from the ultra-short term to create unfalsifiable figures. That would be poor practice in any study, but when you don't have any variables and the long-term trend is downwards (see
below), it is absurd. But he has to do so because if the data were allowed to stand on their own, it would be quite obvious to any observer that the asthma rate has ticked along in the usual way, unencumbered by Glantz's beloved legislation.
Returning to Glanz's graph, if one ignores the arbitrary "counterfactual" and looks only at what actually happened (in blue), the story seems to be one of continuation and moderate random fluctuation. The number of admissions rise or fall by between 3-15% every year, meaning that the 8.9% decline claimed in this study is by no means exceptional or unusual.
There is a clear pattern in each year. Rates hit an annual low in August before hitting an annual high
in autumn. (When they say there was an "immediate" decline of 8.9%, please don't tell me they are comparing the June to August?! Surely no one would be that dishonest, but I have my suspicions.) You will notice that the high in 2006 lasted longer than usual—that's why the 2006/07 year was the peak year in the series. You will also notice that 2008 saw an unusually sharp rise in admissions—the authors make no comment about this.
Nor do they attempt to explain why, if the ban was responsible for the slightly lower rate of admissions in 2007/08, the benefits of the ban were not seen in subsequent years when the rate should have been equally low, if not lower. In fact, the press release that accompanied this study explicitly, and quite falsely, stated that a large initial decline in admissions was followed by further drops.
NHS statistics analysed by researchers at Imperial College London show a
12.3 per cent fall in admissions for childhood asthma in the first year after the law on smoking in enclosed public places and workplaces came into effect in July 2007. The researchers found that asthma admissions continued to fall in subsequent years, suggesting that the benefits of the legislation were sustained over time.
As bad as the study is, it does
not claim there was a 12.3% decline in the first year nor does it claim the rate fell further thereafter. Imperial College should be ashamed for putting out garbage that the press has regrettably, but understandably, swallowed.
Is there anything useful to be taken from this piece of research? I'm afraid not. Recall that Statistics Canada found "no difference in the rate of past-year asthma attacks between those in smoking and non-smoking households". That being the case, the idea that there would be an immediate decline in asthma admissions is most unlikely. Even if there was a plausible mechanism, the idea that smokers immediately banned smoking in their own homes in July 2007 as a result of the government banning smoking in pubs is highly dubious and Glantz provides no evidence for it. Instead, he relies on a circular argument, saying:
The reductions in asthma admissions identified here are consistent with findings that smoke-free legislation may be associated with reductions SHS exposure in children due to an increase in voluntary smoke-free policies in the home or while traveling in a car.
What we have here is a study that claims to find a correlation between asthma rates and passive smoke exposure over time despite no correlation having ever been found with
active smoking in the past (inverse correlations being much more common). It does not even attempt to show that children were "exposed" to less passive smoke after the ban was introduced; instead the authors infer from their own conclusion that they must have been. It uses a single year peak to wrongly imply a long-term rise which it then incorporates into a hypothetical model which uses no other variables.
As with previous efforts of this sort, the publicly available hospital admissions data speak for themselves. Trust the evidence of your own eyes, not manipulated graphs and speculative models.
There is a fundamental dishonesty in all this. Leaving aside the questionable assumptions and partisan biases of Glantz and his team when they were making their model, it remains
a model and should be presented as such. The model cannot be allowed to unseat recorded, demonstrable facts. History—such recent history—cannot be allowed to be rewritten. That, however, is what is happening.
I suggest to you, dear reader, that most people who read about the "sharp fall in the number of children admitted to hospital with severe asthma" would have assumed that the number of children admitted to hospital with severe asthma fell sharply in an
actual year, not in a hypothetical year which exists in the imagination of a man who
Dick Puddlecote has described, with some justification, as "the foremost anti-smoking crank on the planet". But then maybe I'm very old fashioned.
* Stanton Glantz has since
said: "The BBC did a nice story on the paper." You can always rely on them, Stan. Actually, the Beeb's report wasn't even the worst. Special mention should be given to the
Daily Mail, the
Telegraph and the
Guardian, all of whom took more than three-quarters of their reports directly from the shoddy press release. Their
churnalism.com scores are below...
