As we move from debating the statistical association to debating the cause of the association, there is an element of the Nirvana fallacy creeping in ie. if some of the evidence is not absolutely flawless, disregard all the evidence as if it were worthless.
Studies are never perfect and the first studies in any field will usually have the most imperfections. The quibbles about Doll/Hill 1950 made by the defence in this debate were made by the tobacco industry at the time. Some were valid, others less so. None of them came close to explaining how the smokers were 14 times more likely to get lung cancer. The main purpose of the criticism was to plant doubt in the public’s mind. By focusing on trivial issues, they hoped to divert attention from the main findings.
And they were trivial. I’ve lost count how many times Doll/Hill’s definition of an ever-smoker has cropped up in this debate as if it were a matter of supreme importance. ‘Why define a smoker as one who had smoked 100 cigarettes or more in their life?’ asks the defence? They didn’t. They were defining an ever-smoker ie. someone who had ever smoked, and they succeeded. From the way the defence go on about this issue, you would think there were millions of people who smoked 101 cigarettes when they were young and then gave up, never to smoke again. In reality, we know that’s not how it works. The vast majority of the subjects were classified as smokers would have been current smokers, while the rest (a small minority) would have been regular ex-smokers at some point before giving up. Either way, their history as smokers made them valid subjects to study.
But even if there was some misclassification here, the defence is not able to argue why it should favour the smoking theory. There is no reason to think that misclassification would be more common in the case group as amongst the controls. This is an important point because it applies to so many of the quibbles about methodology. Unless these supposed selection biases can be expected to affect one group more than the other, there is little point mentioning them. There is no reason to think that misclassification—if it existed—wouldn’t be evenly spread across both groups, and therefore would not affect the main finding. In any case, the main finding is so strong that it is totally implausible that poor methodology or confounding factors could explain it.
Doll/Hill 1950 has featured so heavily in this debate that anyone would think it was the only study to have ever found a link between smoking and lung cancer. The defence fails to acknowledge that whatever shortcomings the 1950 study had were addressed in subsequent studies, by Doll and many others. The question of smoker classification was addressed by separating current smokers, former smokers and nonsmokers and by asking detailed questions about how much they smoked. Recall bias was addressed by setting up a number of large cohort studies which found the same results. The inhalation finding was shown to be anomalous by later, larger studies. And biological tests on animals were set up to demonstrate that cigarettes were carcinogenic.
But, of course, nothing scientists around the world could do would ever be enough for the critics. Having criticised Doll/Hill for asking a vague question about smoking status, Rich criticises them for being too specific in their questions for the doctors study (1953). The perfectly reasonable question: “How much tobacco did you average per day during the past 20 years of your smoking?” is “guaranteed a wrong answer” because, apparently, no one can remember how much they smoked. Since most smokers smoke at a fairly steady rate, it seems fair to expect most of them to be able to say that they are/were 20-a-day, 30-a-day or 40-a-day smokers. If they smoked a pipe, they would know that they smoked around 10 pipes a day. It really doesn’t matter if someone who says they are on 30-a-day actually smokes 28-a-day or even 20-a-day. Doll and Hill wanted only to (a) establish that they were really smokers, and (b) get an idea of whether they were light, medium or heavy smokers to see if there was a dose-response relationship (there was). We also know that people tend to under-report how much they smoke, so any bias here will underestimate the risks from smoking, not the other way round. This criticism, like so many in this debate, expects an impossible level of accuracy that could only be achieved by having a video camera pointed at every single subject every day of their lives for 20 years. Anything short of that is considered to be ‘flawed’, irrespective of the fact that the supposed flaws are so incidental to the larger body of evidence that they are little more than distractions.
The issue of animal experiments also illustrates the impossible task scientists have in convincing people who do not want to be convinced. An early tobacco industry objection to the epidemiological evidence was to point out the lack of biological evidence that tobacco contained carcinogens. Such evidence soon arrived when tobacco tar painted onto mice caused tumours to develop. The industry then argued that this proved nothing since mice were different to humans and, in any case, no study had shown the actual smoke causing tumours (they were, of course, secretly getting the same results in their own laboratories). Auerbach then proceeded to get dogs to develop tumours by making them smoke through the use of tracheotomies. In public, the industry protested (the doses were too high, dogs were different to humans, it was cruel, blah, blah, blah) but they privately conceded that “the Auerbach work proves beyond all reasonable doubt that fresh whole cigarette smoke is carcinogenic to dog lungs and therefore it is highly likely that it is carcinogenic to human lungs”. A Philip Morris scientist described it as a “reliably conducted experiment”, but perhaps he was one of these ‘anti-smoking fanatics’ who orchestrated the whole scam because the defence think the experiment was “utter garbage” and that “beagles are prone to cancer.”
Having started off the debate by denying any such experiment even took place, the defence now complain that “one study finding a link does not prove anything”. But it did. It proved that tobacco smoke can cause malignant tumours in the lungs of dogs. 12 of the smoking dogs developed tumours but none of the controls did. In addition, the heavier smoking animals had more tumours and the dogs smoking unfiltered cigarettes had more tumours than those who smoked filtered cigarettes. But proof. If only we had proof!
Rich complains that people do not smoke like the beagles in this experiment. This is exactly what the industry said in public at the time. “There is no satisfactory animal model for smoking experiments,” said the Tobacco Institute. In other words, after years of saying that they would not accept the statistical evidence until it was supported by animal experiments, they now argued that no animal experiment could ever be conducted that would satisfy their requirements. Richard Kluger quite rightly called this “a transparent dodge” and we now know that the industry did not believe what they were saying themselves.
Rich also complains that “no adjustments were made for stress levels” as if such a thing would be possible or even relevant. The lack of tumours in the controls suggest that the only possible cause of the tumours was the smoking, not the conditions (by all accounts, the dogs quite enjoyed the smoking and would wag their tails!). The suggestion here is that the real cause of lung cancer is stress. (Presumably stress also caused the blood clots, pulmonary embolisms and heart disease that were found in the smoking dogs). But if stress is really to blame, there needs to be an explanation for why stress follows the smoking rate around the globe and closely parallels the geographical footprint of cigarette sales. That has never been forthcoming.
The defence is on stronger ground when they discuss possible confounding for cirrhosis of the liver. It seems very likely that the doubling of risk for smokers reported by the Surgeon General is due to alcohol consumption correlated with smoking status. It also seems likely that people who didn’t smoke in the 1960s were less likely to drink heavily. What Rich doesn’t mention is that the Surgeon General understood all this and concluded that “the data are not sufficient to support a direct or causal association.” And none of this refutes my contention that smoking was not a predominantly lower class activity in this era. Nor, indeed, was alcoholism. What the liver cirrhosis finding shows is that there can be confounding from other lifestyle factors. I’ve never disputed that. It does not show that smokers came from a lower socio-economic groups and there is not an obvious confounding factor that can help explain the larger risks found for lung cancer, heart disease and the rest.
In the end, it becomes obvious that there is no evidence that will satisfy the critics. Epidemiological studies are dismissed as mere statistics. Surveys are dismissed because human beings are prone to forgetfulness and lying. Animal experiments are dismissed for not relating to humans. Ecologic studies are dismissed for not showing causation.
Every study worth its salt will have a section in which strengths and weaknesses are discussed. All the sceptic needs to do is read what the researchers consider to be a limitation and then say “look, even the researchers admit that [insert ‘flaw’]...” Because studies are always imperfect, the sceptic calls for us to wait until a perfect study arrives, even though he knows it never will. The standard of proof the defence demands is simply beyond the scope of science. In truth, the best way to find out whether smoking increases the risk of lung cancer in human beings is to take large groups of them and see if smokers get lung cancer more than nonsmokers. That has been done around the world time after time, always with the same result. But that, of course, is dismissed as a mere statistical association and we are back to the start.
But while the defence expects impossibly high standards from everybody else, they seem happy to take any old theory that supports their case without asking the most basic questions of it. Let’s take the Semai tribe of Malaysia who supposedly start smoking as toddlers and are immune from lung cancer. The hunt for any evidence for this claim starts and ends with an unreferenced one paragraph letter that was published in the BMJ in 1977. The correspondent—a Dr GY Caldwell—provided no study, no citation and the miracle of the Semai never again appears in the scientific literature. No paper on the subject has ever been published but a Google search for George Yuille Caldwell shows him to be a prolific letter-writer on all sorts of topics. This anecdote from the BMJ letters page was revived by William T. Whitby (who the industry considered to be as “nutty as a fruitcake”) mentioned it in his book The Smoking Scare De-Bunked and Rich mentioned it in his book last year. That is the sum total of knowledge about the Semai tribe’s legendary resistance to lung cancer. I hope I don’t seem unduly cautious when I say that I’d rather wait for more (indeed any) research on this issue before we disregard 60 years of research.
Similarly, the doctors study was made up of (yes) doctors and, as Phil points out, therefore they were of a higher socio-economic class than most. But, again, so what? Phil speculates that they may have smoked more expensive brands of cigarettes. What is this supposed to tell us? Are more expensive brands deadly while cheaper brands are not?
He also speculates that heavy smokers have a ‘ferociously awful diet’. Again, that is possible but if a bad diet was the true cause of lung cancer how do we explain why lung cancer rates were low until the 20th century and then rose sharply just as diets and nutrition were improving? How do we explain why the nonsmoking wife of a smoker is not ten times more likely to get lung cancer despite eating the same meals? We can’t. Like stress, nuclear testing and diesel exhaust, diet is thrown into the mix as a confounding factor despite their failure to fit the rest of the facts, and despite the fact that none of the alternative theories—or even all of them combined—can explain the magnitude and consistency of the association between smoking and lung cancer.
A 1979 memo from Peter Lee, who was then working for British American Tobacco, explained all this rather well. TA73 was a Tobacco Institute publication called ‘The Continuing Controversy’:
“TA73 is so highly selective on what material is presented that one almost gets the false impression there is hardly any case to answer at all …TA73 does not appear to understand what causation is … Discussion of the role of other factors can be particularly misleading when no discussion is made of relative magnitudes of effects. For example, heavy smokers are observed to have 20 or more times the lung cancer rates of non-smokers. Sure, this does not prove smoking causes lung cancer, but that it does mean, and TA73 never considers this, is that for any other factor to explain this association, it must have at least as strong an association with lung cancer as the observed association for smoking (and be highly correlated with the smoking habit) ... TA73 seems ready to accept evidence implicating factors other than smoking in the aetiology of smoking associated disease without requiring the same stringent standards of proof that it requires to accept evidence implicating smoking. This is blatantly unscientific”
A few quick points:
On the American Cancer Study. It’s well known that CPS-I and CPS-II have a disproportionately large number of people from middle and upper incomes and that the subjects tended to be healthier than the general population. Once again, so what? The study still found the smokers to be dying of lung cancer at a much higher rate than nonsmokers. I’m struggling to see what Rich is getting at here, to be honest. He seems to be saying that there was no data on whether or not people smoked, but of course there was. Nor did it compare 80 year olds to 30 year olds.
- No. As I said in my previous post, it’s down to them breathing what is effectively secondhand smoke through their noses. What Auerbach did was make them breathe mainstream smoke directly. This is why it was a more natural (ie. realistic) experiment and was described as such by the Surgeon General.
- Because smokers get lung cancer 5 to 50 times more frequently and no one has come up with a plausible alternative suspect.
Arguing against received wisdom is fine. Rewriting history isn’t. All sources show that women started smoking cigarettes in a big way from the 1920s and by 1940 were smoking as much as men had done in 1900. There is some variation between countries (the graph below shows the USA), but the general trend is always the same. The reason there is a “convenient” relationship between smoking rates and lung cancer is because smoking causes lung cancer.
- What?!! Can we have a weblink for this outlandish claim?
Er, they were (see graph below).
The examples Rich uses to make this point all involve secondhand smoke, not undiluted smoke, or refer to the amount of a chemical produced by a certain machine. The relevant comparison would be measurements of normal indoor or outdoor air compared to that delivered by a cigarette. In every case, the amount of BaP, CO, toulene etc. delivered by the cigarette will be higher.
- Different races (and sexes) differ in their resistance or susceptibility to all sorts of diseases. There’s not space here to debate that point, but it is considered a given in medicine.
- I didn’t say they were “immune” and I didn’t say they had a high rate of lung cancer. I said that Japanese men had a much higher rate than Japanese women.
Frank asks where I got the figure of 75% for male smoking prevalence in 1950. I didn’t look it up, it was what I had remembered. I’ve since found a reference here which gives a figure of 80%.
“By 1950 lung cancer rates among men in the United Kingdom had already been rising steeply for many years, but the relevance of smoking was largely unsuspected. 2 7 At that time about 80% of men and 40% of women smoked (fig 1 and BMJ's website, table A). But few of the older smokers had smoked substantial numbers of cigarettes throughout their adult life, so even male lung cancer rates were still far from their maximum (except in younger men), and rates in women were much lower.”
I’m happy to accept the possibility that the rate could have been even higher in London, but common sense tells us that it wasn’t 98%, let alone the 99.7% seen amongst the lung cancer patients. Even in 1950 there were enough people with moral, religious or taste preferences to keep the rate well below that. And if the rate in London was below 98%, Doll’s findings would be even stronger. I do accept that the choice of control group was flawed, for reasons that Doll couldn’t have known about, but I don’t think this works against the overall finding. Quite the reverse.
On the subject of anti-smoking fanatics, it is certainly a theme of Velvet Glove, Iron Fist that they have always been with us, but it is also clear that there are peaks and troughs. The period 1930-50 was one of the biggest troughs. It is important to stress again that the prevailing bias was against the smoking theory, not towards it. Only a few years earlier, the pipe-smokers at the Medical Research Council had virtually kicked the truly fanatical Lennox Johnson out of the door for making the same claim Doll made in 1950. In fact, Johnson always resented Doll for NOT being anti-tobacco and for being a mere ‘researcher’.
And if you can divide 4.2% by 0.3% to get 13.5, I can divide 99.7% by 98% to get 1.017 - a 1.7% higher figure for the number of smokers in the lung cancer group than would be expected on the null hypothesis - and clearly nothing to be concerned about.”
Well, you could do that but it wouldn’t give you the relative risk. I’ll confess to having been a bit lazy and dividing 27 by 2 to get 13.5, because the total sample was the same for both groups, but I wasn’t far off. The odds ratio comes from the total smoking cases multiplied by the total non-smoking controls (647 x 27 = 17,469) divided by the smoking controls multiplied by the nonsmoking cases (622 x 2 = 1,244) to arrive at 14 (ie. 17,469/1,244 = 14.04).
On detection bias. I don’t dispute that there is some detection bias ie. smokers being more likely to be diagnosed as lung cancer sufferers but, again, this would only apply once it was widely believed that smoking caused lung cancer. It’s another reason why the 1950s and 60s studies are especially free from bias.
Not so. Smoking is associated with higher cancer mortality and higher mortality overall. This has been shown countless times.
It’s true that a 5-10% chance of getting lung cancer might not be viewed as significant in the colloquial sense of the word (ie. large). Obviously lots of people think it’s a chance worth taking (even if they still believe, deep down, that it won’t happen to them). It’s a value judgement, of course, but compared to your chance of getting most other very nasty diseases it’s really pretty high. But it’s your call, and mine. As you know I will defend people’s right to smoke to the, er, y’know...
A good and important question. The answer is that we just don’t know. It would seem to combine with other factors. One of them is genetics, probably. Age is certainly another. Some people would have developed lung cancer but died of something else before they got the chance. Presumably if everyone was able to live to 150, many more people would get it. The fact that not everyone gets lung cancer does not mean smoking is not a major risk factor. Human beings are different and very few risk factors kill everyone that comes into contact with them.
Because if we didn’t, you’d say something like this...
You can’t win! The truth is that there aren’t dozens and dozens every year, nor have the studies completely dried up. New studies appear occasionally. There are a host of studies that are not directly about smoking and disease but which do show the effects of smoking incidentally. Any study that involves smoking has to adjust for smoking since it is a major confounder.
No, it seems to have gone down from 90% to 85%, but these are very rough figures. Of course, the defence could always say that this is because other factors are causing more nonsmokers to get lung cancer, not because fewer people are smoking.