Confrontation at Hammersmith
A crucial public exchange of views between Philip Burch and Richard Peto took place on 6 May 1976 at Hammersmith Hospital Postgraduate Medical School in west London before an audience of senior doctors specialising in respiratory disease. Richard Doll should also have been there but was prevented by a misunderstanding about dates. We do not have the text of either man’s lecture, but we have a handout which Peto distributed to the audience and two very different accounts of the meeting by Burch and Peter Lee, who was also present. Lee thought that the confrontation was a clear victory for Peto, while Burch came away feeling that he had been subjected to a show trial and convinced of Peto’s intellectual fraudulence.
According to Lee:
All in all, I got the clear impression that PB did not come out of the discussion very well. RP showed that the causation hypothesis was consistent with all the observed facts… However, on the basis of Occam’s Razor, PB’s arguments did not come off very well, needing to invoke more and more assumptions to fit the data.
Peto spoke first, followed by Burch.
PB’s reply was for most of the time off the point. He went to painstaking lengths to churn out all his old graphs to prove the point that cigarette smoking could not explain the whole of the observed secular changes in lung cancer rates. As RP had already accepted that diagnostic changes were a very large determinant of those secular changes, PB’s effort was not really worthwhile.
Burch gave his version to Seltzer in a letter dated 10 May:
The Richard Nixon ‘dirty tricks’ department could have learned something from a study of the Richard Doll organization.
Peto had agreed to let me have a draft of his paper by mid-April on the understanding that my task would be to comment on it. Needless to say, nothing came along. That did not surprise me. They wanted to disadvantage me as much as possible. I arrived at Fletcher’s office at about 10.45 and we greeted one another cordially. He gave me Peto’s handout ‘Smoking and Lung Cancer’ (copy enclosed) and apologised on behalf of Peto who shortly turned up and joined our discussion. He is a young, short, shrill, angry man with abundant hair who looks at you sideways. I straightway accused him of abominable behaviour and needless to say he was full of insincere apologies. After a fairly rancorous discussion we went into the lecture theatre. During this discussion I asked him why ‘Trends in Mortality among British Doctors in Relation to Their Smoking Habits’ had been published by Doll and Pike without confidence limits. He thought this was done for ‘propaganda purposes’! Fletcher could hardly believe his ears!
Peto duly delivered his bit, angrily and with various misrepresentations. I delivered mine, correcting various errors in Peto’s presentation and pointing out conflicts between his model and the data. Fortunately, I kept cool and refrained from insults. Quite a number of faces in the audience looked sympathetic. We then had a slanging match for about 1/2 hour and Fletcher finally said a few words. He admitted that I had corrected some of his false beliefs…
Although I find Peto an execrable character (he is nevertheless clever in a slick, smart alec kind of way) I must confess I rather enjoyed the whole business. Their tricks were not wholly successful.
Burch and Lee cannot both have been right. To understand why Burch saw it as he did, the logic of his case and Peto’s, and also the structure of the meeting need discussion.
Burch had claimed to refute two predictions of the causal theory.
- 1. Lung cancer ought to strike earliest in the heaviest smokers and those who started youngest but it does not.
- 2. The observed rising rate of lung cancer was larger than predicted by the causal theory.
He then has to explain the rise in lung cancer.
- 3. Lung cancer was first under- and then over-diagnosed: the rise is an artefact.
Peto’s lecture is designed to answer this.
- 1. On a linear dose-response relationship, lung cancer would strike all groups of smokers at the same average age.
- 2. The rise in lung cancer is partly real and partly artefactual.
- 3. Divergent rates of lung cancer in men and women show that part of the rise is real.
That Burch had answers to this is known from his hand-written notes on Peto’s handout and letters to Seltzer and Fletcher.
- 1. The linear dose-response relationship predicts higher age-adjusted rates of lung cancer in old age than in middle age, but the reverse is observed.
- 2. Peto’s partial concession requires no refutation.
- 3. The argument from sex ratios assumes that misdiagnosis distorts male and female rates of lung cancer in exactly the same proportion, and this is unwarranted.
Peto began the proceedings. On the issue of dose-response relationships (1) he presented data from mouse-painting experiments apparently demonstrating a linear relationship in animals and data from the Hammond and Doctors studies apparently demonstrating a similar relationship in humans. On the issue of misdiagnosis he claimed to show from data on rates in men as against women that it was indeed a factor (2), but that it could only explain part of the rise in lung cancer, with the implication that smoking explains the rest (3).
Few of Peto’s facts were new, but his line of argument was novel. Previous versions of the case against cigarettes had downplayed or ignored misdiagnosis and none had invoked sex ratios, male-female differences, in this way. With apparent generosity, Peto conceded four “points of agreement” acknowledging some role for genetic factors and misdiagnosis, and rounded off with nine questions addressed to Burch personally.
Burch then spoke. It is clear from Lee’s report that he made the same points as in the Lancet and New Scientist, adding no new material. Lee will not have known, and neither will the cancer specialists who made up the bulk of the audience, that Burch had not had advance sight of Peto’s lecture, and that his own lecture did not address Peto’s points in detail because he was hearing them for the first time. Each man was supposed to speak for 25 minutes with a further 25 minutes allotted to discussion. If this was adhered to, Burch finished his lecture at 12.35, an hour when conference-goers tend to have their minds on lunch, with Peto’s nine questions to be answered off the cuff. He answered three of them (numbered 1, 2 and 8 in the handout).
Question 2 was a good choice, as it related to misdiagnosis, and Burch was able to point out a hidden assumption in Peto’s argument from sex ratios. But, Lee reported,
PB never answered question 4, which was quite an interesting one. PB has claimed that the mean age of heavy smoking lung cancer patients should be less than the mean age of light smoking ones, if smoking had caused the lung cancer and had quoted results to show this was not found. RP showed clearly that one would expect equality of mean age on the assumption that smoking caused lung cancer.
Question 4 was worded as follows.
If the incidence rate at each age is twice as big in heavy as in light smokers,
And the age distribution of heavy and of light smokers was similar,
Then mean age of lung cancer patients might be the same for heavy and light smokers.
Does Professor Burch agree with this?”
The argument here turns on three tables in Peto’s handout with which Peto attempted to show that the first of Burch’s claimed refutations of the causal theory fails.
Table 1 presents evidence for a linear dose-response relationship in experiments on mice.
Table 2 constructs a hypothetical scenario in which light and heavy smokers have virtually the same mean age of onset of lung cancer.
Table 3 presents real world data with a strong superficial resemblance to the hypothetical scenario. It is too strong to say, as Lee did, that equal mean ages are predicted by Peto’s model, but he seemed to show that they are compatible with it.
Peto constructed table 2 from a mathematical formula quantifying the excess risk of lung cancer supposed to be caused by smoking c cigarettes a day for d decades:
For instance, a 45 year old who has been smoking 20 a day since age 15 has a predicted excess risk of 20 ∙ (45 – 15)4 = 1620/million. Added to a background risk of 50/million observed in non-smokers, this gives a predicted death rate of 1670/million in that category of smoker. (Peto did not attempt to derive lung cancer rates in non-smokers mathematically.) 1670 / 50 is about 33: that is the risk ratio between smokers and non-smokers. Peto supplies the ever-increasing risk ratios for heavy smokers (33, 56, 75) and light smokers (17, 28, 38). Though he does not say so, the steeper rate of increase in heavy smokers implies that they have a marginally higher mean rate of onset, but the point is one he can reasonably ignore (see Burch on Doll for fuller discussion of this and other points at issue between Burch and Peto).
The full version of table 2 looks like this.
|Non-smokers||Heavy smokers||Standard smokers|
|Age first started||*||Age 15||Age 15||Age 25||Age 40|
|Incidence rate at age 45||50/m||1670/m||(33x)||860/m||(17x)||210/m||(4x)||51/m||(1x)|
|Incidence rate at age 60||150/m||8350/m||(56x)||4250/m||(28x)||1650/m||(11x)||310/m||(2x)|
|Incidence rate at age 75||350/m||26270/m||(75x)||13310/m||(38x)||6600/m||(919x)||760/m||(2x)|
Peto’s audience is invited to compare it with the actual data in table 3. Misleadingly, table 2 groups smoking rates into columns and death rates into rows, while table 3 does the opposite. Burch made handwritten annotations to table 3 in a copy of Peto’s handout which he sent to Carl Seltzer.
|Kahn’s data age 55-64||S/NS||Kahn’s data age 65-74||S/NS||Hammond’s data age 55-69||S/NS|
The columns headed S/NS (ratio of smokers/non-smokers) were added by Burch. They represent the risk ratios, which Peto predicted must diverge indefinitely with increasing age. In the real life data they do not diverge: rather (5.3 > 4.4 > 3.6 etc) they converge. Peto’s prediction was refuted by his own data.
In later years, Burch refused on principle to address any gathering also addressed by Richard Peto. There can be no doubt that Peto was negligent in withholding an advance copy of his paper, that his data tables are poorly formatted, and that it was unreasonable to expect answers to nine tricky questions in 25 minutes at the end of a morning session. Whether his actions were merely thoughtless or a deliberate attempt to discredit Burch must be a matter of opinion.