Abstract
As we have seen, over the years the tobacco industry spent some time and effort casting about for something other than tobacco products that could be blamed as the cause of lung cancer in smokers. In past postings we have seen that they proposed, for example, the "Carotene Hypothesis" (the theory that an excess of dietary carotene causes lung cancer), the "Reverse Hypothesis" (which proposed that lung cancer causes smoking), and the "Constitutional Hypothesis" (which proposed that something in a person's personality predisposes him or her to get lung cancer). The author of today's document brings up yet another theory closely related to the "Constitutional Hypothesis," called the "Unidentified Common Factor theory" (or U.C.F. for short).
The "Unidentified Common Factor" theory held that some unknown characteristic shared by a certain group of people predisposes them to both smoking and lung cancer.
The author states that
"...there are some eminent scientists who have suggested that there may well be a common factor which leads people to smoke and to develop lung cancer through a causal mechanism which does _not_ involve smoking...Thus the late Sir Ronald Fisher, who was one of the world's greatest geneticists, suggested that such a factor might exist in the type of genes a person inherited...Eysenck (who is a psychologist), Burn, (a pharmacologist) and Berkson (a medical statistician) have all suggested quite independently that heavy smokers have less resistance to lung cancer, perhaps because they live at an accellerated rate...[or] perhaps because they are less careful of their health, [or] perhaps because they are less biologically self-protective, and that some personal characteristic such as these causes the statistical association between smoking and lung cancer."
Yet in the same paper, the writer discusses the unliklihood of the U.C.F. Theory:
"It is not difficult to build up a hypothesis about an Unidentified Common Factor that fits all the known statistical facts about smokig and lung cancer. It can also be argued that many scinetific theories are developed simply by building up a hypothesis that fits all the observed facts. But is it inherently likely that an Unidentified Common Factor with all these very detailed characteristics really exists, waiting to be discovered as a result of some lucky experiment? Speaking in terms of probability...all I can say is that it is most improbable that such an Unidentified Common Factor exists. Finally, if such a Factor were to exist, it would only make the hypothesis that smoking causes lung cancer unnecessary: it would not prove that the smoking hypothesis was false."
The rest of this paper is also full of insights into how people inside the tobacco industry think about public health (or not). For example, near the beginning of the paper, the speaker takes an interesting viewpoint about the importance of the number of people dying from cigarette use:
"The vital statistics I would like you to bear in mind are 7, 57, 139 and 227...There is no glamour about these figures. They are the death rates per 100,000 per year from cancer of the lung of men who were nonsmokers (they are the 7), men who smoked 1-4 cigarettes daily (they are the 57), men who smoked 15-24 cigarettes daily (they are the 139) and men who smoked 25 or more cigarettes daily (they are the 227)...These four vital statistics are basically the reason why we are here to-night. They are the reason why the tobacco manufacturers in this country have spent over 25m to date on smoking and health reasearch... These vital statistics are really vital. They threaten the life of the tobacco industry in every country of the world."
Fields
- Notes
Thanks to Ron Davis for citing this document to Doc-Alert.
- Quotes
"...There are some eminent scientists who have suggested that there may well be a common factor which leads people to smoke and to develop lung cancer through a causal mechanism which does _not_ involve smoking...Thus the late Sir Ronald Fisher, who was one of the world's greatest geneticists, suggested that such a factor might exist in the type of genes a person inherited...Eysenck (who is a psychologist), Burn, (a pharmacologist) and Berkson (a medical statistician) have all suggested quite independently that heavy smokers have less resistance to lung cancer, perhaps because they live at an accellerated rate accordin to Eysenck, perhaps because they are less careful of their health according to Burn, or perhaps because they are less biologically self-protective according to Berkson, and that some personal characteristic such as these causes the statistical association between smoking and lung cancer."
It is necessary to emphasize that the common factor has only been suggested; it has never been positively identified, so we really ought to describe such theories as Unidentified Common Factor theories--U.C.F. theories for short. Moreover, if the hypothesis that smoking causes lung cancer and the U.C.F. hypothesis are both true, which was one of our alternatives, then it is still true that smoking causes lung cancer. The hypothesis that smoking causes lung cancer only becomes unnecessary if the U.C.F. hypothesis explains _completely_ all the features of the statistical evidence about the relationship between smoking and lung cancer...
- Company
- Philip Morris
- Author
- Unknown
- Recipient
- Unknown (appears to be text of a talk given to an audience related to the tobacco business).
RegionUnited States
United Kingdom
Named OrganizationBRITISH MEDICAL REGISTER; EATON; HOUSE OF COMMONS; MEDICAL RESEARCH COUNCIL; MEDICAL RESEARCH COUNCIL; MINISTERY OF HEALTH; PARLIAMENT; PLAYBOY; PRIVY COUNCIL; ROYAL COLLEGE OF PHYSICIANS; SEVENTH DAY ADVENTISTS; STANDING ADVISORY COMM ON CANCER &; STATIS
TypeSpeech
Presentation
Named PersonADLER
BERKSON
BURN
Doll, Sir William Richard, M.D. (Epidemiologist, Oxford U, Plaintiff's Expert)Dr. Doll is an eminent physician from the University of Oxford in the United Kingdom.
EYSENCK
FISHER,R
HENRY,O
Hill, Sir Austin Bradford, Ph.D. (Medical Statistician, U. of London, worked with Doll)In the September 1950 British Medical Journal, Richard Doll and Dr. A. Bradford Hill published preliminary report on smoking and lung cancer. They examined smoking rates for hospital patients with and without lung cancer. They did 1954 prospective studies of 40,000 physicians and concluded that heavy smokers were 24 times as likely to die of lung cancer (E. Whelan 1984).
PRATT
ROGERS,T
RUNYAN,D
SAMPSON
VONHARTMANN
JAMES,I
SubjectEffects—Smoking Behavior (Effects)
smoking history
smoking initiation
health belief
health effects
cancer
Document Images
Page 1: xdn74e00
3 4RIQTLY CONFIDENTIAL
..r .
f°j,iiY+l
_ 'U,-~-~.~~~
'f ~'
.
G 665
a`~.~
E;
7
,,,,::,
-~~L"i::c.
y~.
~~~M
;a-
.`n
:~~~~."'y
~ t'" Smoking and Health: the Present Position in the
U. K. , and How it came about'
I
, z:~
;The Present Position: the Main Evidence against Smoking
We all like~to think of ourselves as men who have the moral
strength to face facts, even if they are unpleasant facts. So I am
going to start by asking you to tace certain facts, certain vital
.::. . .-., .<'.:.;.t :.,.,.... :... " : _.. ~;',,, . .. . . . . .....- ...,'. .'.. . . . /i .
statistics. llhen.I mention the words "vital statistics", some of
my friends immediately think of figures like 38 - 24 - 36. The
~vital statistics I would like you to bear in_mind are 7, 57, 139
and 227.-There is no glamour about these figures. They are the
death rates per 100,000 per year from cancer of the lung of men
who were non-smokers (they are the 7), men who smoked l- 14
cigarettes daily (they are the 57), men who smoked 15 - 24 cigarettes
daily (they are the 139) and men who smoked 25 or more cigarettes
daily (they are the 227). These figures are all included in table 1
which you have before you.
.., . . . .... ,-..
Thes@ fonrnitalstatistics are basically the reason why we are
here to-night. They are the reason why the tobacco manufacturers in
this country have spent over £5m. to date on smoking and health
research through the Tobacco Research Council, and perhaps as much
~
again on their ownresearch. These vital statistics are really vital.
They threaten the life of the tobacco industry in every country of
,.the world.
,
g
p
p
;.~
cigar smokers, for men who smoked other products as well as cigarettes. ~
- since the figures I have quoted refer to smokers of cigarettes only
While the four figures which I have given you are the ones that
really matter
I have also included in table 1 fi
ures for
i
e and
- and for men who have given up cigarette smoking. Table 1 also
includes the corresponding figures for the death rates for two other
diseases - bronchitis and coronary thrombosis.
Since we are going to consider If these figures are really
valid, it is necessary to describe how these vital statistics were
arrived at. They were the results of an epidemiological survey.
- 1 -
Page 2: xdn74e00
..._.,t.. -
~-
.
;!',Epidemiology" simply means the science of diseases found in a large
. number of people. Epidemiology is a particular form of applied
the epidemiological studies of smoking and other diseases are
smoking and lung cancer, and they all reached similar conclusions.
smoking and disease have now been carried out in many country of
the world. There are well over 50 good quality epldemlological studies of
statistics. A very large number of epidemiological studies of
added in, the total runs into hundreds. But the total is not
important, One good, conclusive study is enough - un2ortunately.
These epidemiological studies of smoking and disease were of two
called "prospective" studies. The retrospective studies were the
kinds - what were called "retrospective" studies and what were
first to be carried out, partly because they are simpler and partly
investigated). In addition, an approximately equal number of
patients who already have lung cancer (or whatever disease is being
retrospective study is that the statisticians select a number of
because they give results immediately. The procedure in a
= patients is selected who have diseases which could not be attributed
to the factors that might have caused the disease being investigated.
Such people, for example, might have been involved in a traffic
members of both the disease and the control groups are then inter-
a basis of comparison by which the disease group is assessed. The
accident. This second group is known as the control group and forms
in order to see if any particular characteristic, such as that of
viewed and asked various questions. The answers given are analysed
`being a heavy cigarette smoker, is found substantially more often
among the disease group than among the control group. If positive
results of this nature are obtained, then there is at least a
all apparently free from disease, are interviewed and asked questions
statistical association between the disease and the characteristic
concerned. By terming it a'"statistical association" we suspend
,)udgement about the nature of the association: we mean only that
a real association of some kind exists between the disease and the
characteristic.
The second or prospective type of epidemiological study is
rather different. In prospective studies, a large number of people,
~2_
Page 3: xdn74e00
1
1:
Aliout the subjects which are being investigated. It is then
'necessary toywait a number of years, perhaps 10 for example. At
:.,;. . "
the end of this period, having kept track of all the people who
~'replied as far as possible, information is sought about the causes
of death of those who have died. The results are then analysed,
very much as in the retrospective type of study, in order to see
if those'who have died from, say, lung cancer possessed certain
-characteristics substantially more often than those who had died
of other diseases.
After the end of the war, the Medical Research Council had
become concerned about the rapid increase in deaths from lung cancer
in this country. In 1947 they planned a large scale investigation
to determine whether patients with lung cancer differed materially
from other persons in smoking habits or in exposure to atmQspheric
pollutfon. This was a retrospective type of study, and a preliminary" :
'report by the two statisticians directing the study (Dr. Richard ."
Doll and Sir Austin Bradford Hill) was published in 1950. The study
was based upon interviews carried out with patients in hospitals in
Greater London, Bristol, Cambridge, Leeds and Newcastle. Altogether, t'
- . . , . , . . ~y~
over 4,000 patients were interviewed, of whom Just under 1500 had
c ,`.r~
lung cancer. Doll and Hill concluded that "there is a real association
.between carcinoma of the lung and smoking" and "that smoking is a
factor, and an important factor, in the production of carcinoma of
the lung". This was the first time that the association between
. i:'..
smoking and lung cancer had been brought to public attention in
this country. Two years later, in 1952, the final report of this
enquiry was published. Doll and Hill claimed that the results
confirmed their._earlier conclusion " that the association between
smoking and carcinoma of the lung is real".
The retrospective type of study is open to certain criticisms.
The most obvious is that people who know or suspect that they have
lung cancer may not give true answers when questioned about their
smoking habits. I will refer to these criticisms and their validity
later. But in consequence of the possible defects of the retrospective
type of study, the Medical Research Council started a prospective
type of investigation in October 1951. Doll and Hill wrote to some
3
; r__.~ .
~~~'~~'
Page 4: xdn74e00
rJ
i Yy 'F"nrr't
~' .
1e
6i0;00Q men and women on the British Medical Register who were
_~~ .. ...--~
r.~-y.;,:.~
resident in the U.K, and asked them for certain broad information
41
ar~~: ~;;
. ,.:
_.;..~
.- . . . - . .. . . . .h.
'about their smoking habits. Over 40,000 sent replies that were
sufficiently complete to be used. By the end of 10 years about
5,000 of the 40,000 had died. Doll and Hill analysed the results
in great detail and the key figures of this study have been given
in table 1.
Assuming for the moment that the figures in table 1 are valid,
they lead to certain conclusions.
,; , _ _ .
First, lung cancer death rates show a very rapid rise as the
number of cigarettes smoked daily increases.. As you will see, the
ky
man smoking 25 or more cigarettes a day has over 30 times the chance
of the non-smoker of dieing of lung cancer. For men smoking 35 or
.more cigarettes a day, the figure is 45 times. The lung cancer
death rate of non-smokers is so low that smokers provide about 90%
of all male deaths from lung cancer. This, therefore, leaves very
little room for a major contribution to lung cancer by air pollution.
Further, there is no level of cigarette smoking that is free from
increased risk of lung cancer. Even the smokers of 1- 4 cigarettes
a day (the figures are not shown in table 1) have several times the
,, -
lung cancer death rate of non-smokers:
There is a broadly similar pattern of statistical association
between smoking and bronchitis, also shown in table 1, but the
bronchitis death rates are only about half the lung cancer rates.
It is necessary to remember, however, that perhaps half the men
,and women who suffer from chronic bronchitis are shown on their
death certificates as having died from some cause other than
bronchitis. To put it bluntly, bronchitis cripples many more than
it kills.
The last of the three important diseases statistically
associated with smoking is coronary thrombosis. The essence of
coronary thrombosis is that one of the coronary arteries of the
heart is blocked by a thrombus or clot, cutting off the blood
supply to part of the heart muscle. If it is an important artery,
the heart will stop pumping blood through the body and the person
will die.
.~"'"' ,~~~F
'e
47
t'!r'
X:.
a
ax;
r"%: ,~r;5a`:: ~'
Page 6: xdn74e00
Criticisms of the Statistical Evidence
,,. ~~i :,y,~ :
As I have said, the core of the present situation is contained
in table 1. Let us now see if we can show that these figures have
German philosopher called von Hartmann used the theory of probability
real validity.
All of you have heard it said many times that you can prove
anything by statistics. As if to illustrate this, a 19th century
to prove that there are 59.06 chances in 100 that God exists. I hope
least, statistics can never decide the nature of an association
being able to prove anything by statistics, it might be nearer the
truth to say that you can never prove anything by.statistics. At
the recording angel duly noted that .06. But in fact, so far from
between two statistically associatdd factors.
1 are invalid, many people, including myself, have squeezed the last
In seeking to prove that statistics of the type listed in table
.ounce of intellectual questioning from their brains. Fame certainly
awaited anyone who could conclusively refute the claim that smoking
Depending on how they are classified, I think I could make a
caused lung cancer.
list of perhaps 50 criticisms that have been made of the epidemiologi-
their diagnosis and this led them to over-state the amounts
(1) In some studies, patients with lung cancer knew or suspected
cal data. The following are just a few of the criticisms:
The control groups were often hospital patisnts. Many would be
off smoking and this led them to under-state the amounts they
they smoked.
No one can really know how much he smoked in the past and the
information which the interviewers recorded was little better
than guesses.
had previously smoked.
(4) Diagnosis of lung cancer by general practitioners is subject to
frequent errors.
(5) The average man smokes less than twice the number of cigarettes
per day smoked by the average women yet there are five times as
many male as female deaths from lung cancer in England and Wales.
,
Page 7: xdn74e00
.than men whereas lung cancer increased faster in
Doll and Hill failed to find a general association
in women.
inhaling and lung cancer, which they ought to have
. smoking really caused lung cancer.
(8) People who emigrated from Britain to South Africa or U.B.A.
smoked more in their new country, where cigarettes were
cheaper, and yet they had Lower lung cancer rates than men
who continued to live in Britain.
Americans smoke many more cigarettes per head than Britons but
I could go on for a long time listing the criticisms that have
have a much lower lung cancer rate.
been made, but you would find it very tiresome if I was to outline
and answer each criticism. I propose therefore to express only two
general points:
1. 'The first point is that, as I have mentioned, there have been
.cancer and smoking which were found in the retrospective studies
spective. Now, you can say that the associations between lung
two types of epidemiological studies - retrospective and pro-
were due to one or more of the following facts - that the
not often true), that the control patients were not an adequate I
patients or interviewers knew the diagnosis (although this was
-among light smokers in the disease group and/or among heavy
match for the disease group, that there was excessive bias
to special groups, and particularly to those who volunteered
for inclusion in the study. Many British doctors did not
participate in the Doll and Hill prospective study. This almost
prospective studies had certain imperfections. They were limited
not possibly have been present in the prospective studies and
so could not account for the association between smoking and
lung cancer found in the prospective studies. Of course, the
then you are forced to admit that, because of the very different
procedure adopted in the prospective studies, these biasses could
results of the retrospective studies to biasses of these kinds,
much they smoked, say, 10 years ago. If you attribute the
smokers in the control group, that people can't remember how
- 7 -
I
Page 8: xdn74e00
, ~'~f~.,A%`:~+
sy certainlymeans that the death rates found in the study
would not be, representative of the population as a whole,
and perhaps not even be representative of all doctors. But
this would not mean that the statistical association between
smoking and lung cancer found in the study was invalid, unless
it had happened that the non-volunteers were a group where non-
smoking and light smoking were highly associated with lung
cancer and heavy smoking was associated with freedom from
lung cancer. An inverse association of this natureso that
over participants and non-participants combined there was no
real association, is so improbable that it can be dismissed.
(2) The second point is this. Of course, there were inaccuracies
in reporting past smoking habits, in diagnosing causes of
death, and so on. But when these errors are largely random
or fortuitious, they do not invalidate the positive findings
of a study. What they do, in fact, is to reduce the apparent
magnitude of such associations as do exist. If errors of this
nature hadn't occurred, men smoking 25 or more cigarettes a
day might have been found to have not 30 times but perhaps 50
or 60 times the lung cancer death rate of non-smokers, and
those smoking 35 or more cigarettes a day not 45 times but
perhaps 70 or 80 times the rate of non-smokers.
I should perhaps add something about the associations between
inhaling and lung cancer. Doll and Hill, in their first study,
failed to find in association between inhaling and lung cancer. The
late Sir Ronald Fisher made great critical play with this particular
finding. In later studies, however, In several countries, including
Britain, an association between inhaling and lung cancer was found
at lower levels of cigarette smoking but not at higher levels. This
was a peculiar result, but I believe that it-was due to a failure
to adjust the figures, as should have been done, ior differences
in the average ages of the different groups. When this has been
done, a general association between inhaling and lung cancer has
been found, as in recent studies in U.S.A. and Canada.
Practically all statisticians who have given thought to the
matter - I can think of only two or three exceptions - accept that 7;.
.,~,,..,.
the statistical association
Page 9: xdn74e00
: ~r...:.
between smoking and lung cancer has been adequately established
as really existing. The same applies to the statistical
associations of smoking with bronchitis and coronary thrombosis. x"`;
This,of course, does not mean that smoking necessarily causes
lung cancer, bronchitis or coronary thrombosis. Statistics can
not decide the nature of an association, which is an'entirely :.,~`
°different matter. Naturally, the statisticians could be wrong
-in their assessment of the validity of the statistical
associations between smoking and lung cancer, bronchitisand
coronary thrombosis. Improbable things do happen. An eminei-t
'doctor told me recently that, at his hospital newspaper shop,
he had picked up a copy of Playboy magazine. As he said to
ments".
ri;
t
Page 10: xdn74e00
The Nature of the Statistical Associations between
Smoking and the Three Diseases
If you accept the view of almost all statisticians that the
~ statistical associations between smoking and lung cancer, bronchitis
and coronary thrombosis are real and not spurious ass6ciations, you
cannot duck the next set of questions. These are: what is the
nature of each of these assoications: does it mean that smoking
causes the associated disease? You have, in fact, only two pos-
sibilities, which are not mutpally exclusive. Both could be true,..
The two possibilities are these. Firstly, that smoking causes
the disease. Theoretically:there is also the possibility that the
disease causes smoking. It is, however, so difficult to believe
that a disease manifested at age 60 existed undiagnosed at age lQ
and caused a man to start smoking at 15, that I propose to dis-
regard this possibility. The second possibility therefore is that
smoking and the disease are associated because they each have a
real association with some third common factor.
Since the common factor concept is extremely important, I
.should like to give you a fairly clear example of it. There is a
real association between divorce and death from coronary thrombosis.
Men and women with a high divorce rate have a significantly high
death rate from coronary thrombosis. Death from coronary thrombosis
obviously does not lead to a divorce suit and it is unlikely that
divorce of itself causes formation of a thrombus; The most probable
explanation of the association between divorce and coronary throm-
bosis is that people of a certain type live in a way which causes
them to have a higher than average divorce rate and a higher than
average incidence of coronary thrombosis. The common factor is
the type of person.
Now in examining the nature of the statistical associations
between smoking and lung cancer, bronchitis and coronary thrombosis,
I think the oourse of the argument may be clearer if we deal with
the three diseases in the reverse order.
The figures for coronary thrombosis in table 1 have two
features. First, there is a tendency for the death rates to
increase from non-smokers (331) to light cigarette-only smokers
