Chris Snowdon responds to Rich White and Frank Davis.
Frank points out that 98% of the (male) patients in the first Doll/Hill study were smokers. This is true, but we know that even in 1950, 98% was not the general smoking rate amongst men (it was around 75%). So how did Doll end up with such a disproportionately large group of smokers in his sample? He didn’t grab them off the street. They were in the sample because it was a study of hospital patients, half of whom had lung cancer. The fact that a random sample of seriously ill people had an unusually large number of smokers amongst them is not the best advertisement for smoking! Doll and Hill weren’t to know that smoking causes all sorts of other diseases that would land smokers in hospital, such as heart disease. Had they known, they might have picked healthy controls from the general population, which had a lower smoking rate and would, therefore, have strengthened their findings.
As it happened, they had no need to. Even with the very high rate of smoking amongst the patients, a clear difference was found. Only 2 of the 647 lung cancer patients were not smokers (0.3%), compared with 27 of the 647 non-cancer patients (4.2%). The lung cancer patients were 13.5 times more likely to be smokers.
Rich brings up the issue of inhalation, which has been doing the rounds since the arch-sceptic Ronald A. Fisher raised it in 1950. Doll and Hill asked patients whether or not they inhaled their cigarettes. 61.6% of the lung cancer patients said they did compared to 67.2% of the non-cancer patients. The difference between the two figures is actually pretty slight. The main thing it tells us is that there were a lot of liars in both groups, as the large number of supposed non-inhalers is difficult to believe. Subsequent (larger) studies in Canada and USA found the proportion of non-inhalers to be 11% and 6% respectively (Surgeon General’s Report, 1964; p. 91 – see Figure 1 below). Both of these studies found the non-inhalers to have the lowest risk. The Doll study is a slight outlier in this respect, although the difference in responses from the two groups is not great, and, as Doll and Hill say in the text of the study, it is not statistically significant once adjusted for confounding.
Fisher didn’t have to get the raw data to see any of this, by the way, Doll and Hill drew attention to it themselves on p. 744, which they are unlikely to have done if they were trying to trick anybody. This brings me nicely on to the question of reliability. It has already been noted that Doll was a smoker himself, as was Evarts Graham (whose study with Ernst Wynder predated Doll’s by a few months). As indeed were several members of the Surgeon General’s panel and many senior doctors. These were not people who were fanatical haters of tobacco and it is difficult to see what their motive might be if—as is being suggested—they twisted the evidence. Like all scientists who do newsworthy research they could be accused of careerism and publicity-seeking, but this is not sufficient either to explain a “scam” that has lasted sixty years, nor why they would choose to demonise their own pleasures.
Many researchers working in tobacco control today can be justly accused of egotism, rent-seeking, being in the pay of pharmaceutical companies, having a background in anti-smoking fanaticism and—not uncommonly—having a personal obsession with smokers and smoking legislation. The same was not true in the 1950s and 1960s. It is a sad reflection on the state of the academic field that, as a result of the biases that have grown up over the years, research written decades ago is more likely to be unbiased and reliable than studies published today.
Neither of the Doll and Hill studies were perfect, just as no epidemiological study is perfect, but the case against smoking hangs on much more than two articles. In 1964, the Surgeon General evaluated 7 prospective studies, including the massive American Cancer Society study of men in 25 US states. The number of people included in these studies totalled over 1,100,000 (Doll’s doctors study was the smallest of them). All of them found higher rates of mortality amongst smokers, higher rates of lung cancer amongst smokers and—with remarkable consistency—found that the risk increased with the amount smoked.
This was supported by 9 case-control studies (see Figure 2 below) which again found a consistent raising of risk and a consistent dose-response relationship. Lung cancer risk was elevated from 2.4 for light smokers to 34.1 for the heaviest smokers.
As has been pointed out, epidemiology can never prove causation, but the correlation is undeniable (even Fisher never denied that there was a correlation). Since the obvious explanation for this strong, dose-sensitive association is that chronic smoking raises the risk of lung cancer significantly, any defence of smoking as the causative factor must provide an alternative. That alternative must be closely correlated with smoking itself and must have a correlation with lung cancer that is at least as strong as the correlations found for smoking and lung cancer.
It is difficult to force animals to develop lung cancer because it’s difficult (a) to get animals to smoke, and (b) to have the necessary time-frame for lung cancer to develop (a major risk factor for all cancers is ageing). It takes several decades for human beings to smoke and it is therefore very hard to replicate the human experience of chronic, long-term smoking on animals. A bigger problem is that animals breathe through their nose, thereby filtering the smoke. Most experiments with animals are really showing the effect of secondhand smoke (albeit in unrealistically high doses) and, yes, most of them fail. However, Oscar Auerbach did manage it with beagles in 1970, by giving them tracheotomies. 12 of the 86 dogs did get lung cancer over an 18 month period, although Auerbach was criticised by animal rights campaigners and the experiment wasn’t repeated (and tobacco testing with animals was banned in the UK in 1997). Privately, the tobacco industry accepted the experiment as valid. A memo from a Philip Morris scientist read: “I would say that the experiment is a crude one but effective in that carcinoma in dogs has been produced.” A scientist at Gallagher wrote a memo saying: “We believe 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 would be carcinogenic to human lungs… the results of the research would appear to remove the controversy regarding the causation of the majority of human lung cancer.”
The geographical spread of lung cancer follows several decades behind the geographical spread of smoking. For the lung cancer epidemic to begin, you need to have a sizeable group of people who have been smoking for at least 20 years. Rich says that the “highest smoking rates tend to have the lowest lung cancer rates and vice versa” and uses Greece, Japan, China and Hungary as examples. In fact, Hungary has one of the highest rates of lung cancer mortality and Greece’s rate is medium to high (see these graphs). You can see that nearly every country has a higher rate of lung cancer for men than for women, and that in most cases the female rate is going up while the male rate is flattening off or falling. This is simply a reflection of what happened in the second half of the last century when smoking rates went down for men and up for women. Notice also the experience in Sweden, where snus use by men since the 1960s has helped the male smoking rate fall below that of women, resulting in Sweden being the only European country where men have a lower lung cancer mortality than women. Again, if smoking isn’t the primary cause, what other factor fits the specific experiences of each country?
Rates of lung cancer have always been somewhat lower in Asian countries due to an assumed biological resistance. Nevertheless, it is worth comparing the lung cancer incidence for Japanese men of 44.6 per 100,000 against the rate for Japanese women (few of whom smoke) of 13.3 per 100,000. It is a similar story in China (67.5 for men against 26.6 women). In India, where smoking is fairly uncommon amongst men and very rare amongst women, the incidence rates are just 12.1 and 3.8 respectively.
What we are looking for, then, is an unknown factor which is as strong or stronger than the statistical correlation between smoking and lung cancer, which causes lung cancer to rise and fall several decades after smoking rises and falls, which affects men first and then women and which is less common in countries where smoking is less common. Over to the defence…
PS. A few quick points in reply to several comments:
[Rich] MRFIT was primarily a study into cholesterol and heart disease. I’m not aware that they stratified the results to show smoking and lung cancer but I’m happy to be corrected on it.
[Rich] “People have been smoking for millennia.” Yes, but not cigarettes. It is the deep inhaling of cigarettes that poses the hazard. The Surgeon General found little or no elevated risk for pipe and cigar smokers. Really, this debate should be called ‘Does cigarette smoking cause lung cancer.’
[Anon]: “How does one explain that the type of lung cancer have changed over the last 3 decades from squamous cell cancer to adenocarcinoma cancer?” Nonsmokers have always been much more likely to get adenocarcinoma if they get lung cancer at all, and smokers have always had higher rates of other forms of lung cancer. Recent decades have seen more smokers getting adenocarcinoma and the reason you give in your comment is the most common explanation, ie. ‘light’ cigarettes allow (or require, even) the smoke to go deeper into the lungs. Old, unfiltered heavy cigarettes were not inhaled so deeply and so the tumour would form higher up the lung.
[Rich] “We know most cigarette smokers come from the lower classes, and those people tend to be die younger and suffer more disease than the richer in society also.” That is true today but not back then. In the first half of the century and beyond, cigarette smoking was not class-sensitive. Today, it is quite obvious when you see a study which purports to show that maternal smoking breeds criminality in unborn children that mothers who smoke during pregnancy are simply more likely to be from socio-economic classes where criminality is more common. This was not the case when the landmark studies were being conducted and it is another reason why older studies can be more reliable than modern studies.
[Frank]: Definition of a smoker. Doll and Hill were looking for ever-smokers and asking if they’d smoked 100 cigarettes in their life was as good a measure as any. Obviously they wanted to avoid people who bought one pack when they were 15, or had one cigarette every New Years’ Eve. Whether it was 50, 100 or 200 makes little difference. Other studies have found similar results after stratifying for current smokers, ex-smokers and never-smokers. Only counting current smokers would be a poorer system, especially with lung cancer patients who often, and understandably, give up smoking when they develop symptoms.
[Rich] “The quantities of the chemicals and toxins [in cigarettes] are tiny, less than we inhale in normal air.” This statement requires some support and explanation!