Tobacco Documents Online

7 Reference List 1. Lee PN, Forey BA. Trends in cigarette consumption cannot fully explain trends in US lung cancer rates [Abstract]. Presented at the 15th Annual Conference on Applied Statistics in Ireland (CASI), Killarney, March 29th-31st 1995. Statistician 1996;45:448. 2. Lee PN, Forey BA. Trends in cigarette consumption cannot fully explain trends in British lung cancer rates. JEpidemiol Community Health 1998;52:82-92.

4 Gave up at aee Years given up Risk relative to never smoker at age 85 Ratio 85 0 21.29 1.00 80 5 16.56 0.78 75 10 12.80 0.60 70 15 9.85 0.47 65 20 7.58 0.35 60 25 5.87 0.28 As per the epidemiology, this showed that the risk declined sharply with increased time of giving up. Then I looked at how the relative risk varied with age for someone giving up at age 60 (start of follow up) Aee Years given up Risk relative to never smoker of same age Ratio 60 0 16.01 1.00 65 5 12.24 0.76 70 10 9.72 0.61 75 15 7.98 0.50 80 20 6.75 0.42 85 25 5.87 0.37 Again the risk declines sharply with increased time of given up. Finally I tried to approximately model the actual situation for the Veterans. I noted from Enstrom's Table I that between 1955 and 1967 about a third of the original current smokers had given up and that in the next nine years a similar proportion gave up. Based on this I decided to assume the population of current smokers at 1955 was distributed as follows

6 Year Relative risk Period Relative risk 1955 16.01 1955-60 16.38 1960 16.76 1960-65 16.75 1965 16.73 1965-70 16.42 1970 16.11 1970-75 15.57 1975 15.02 1975-80 14.28 1980 13.53 It can be seen that though there is a decline eventually it is only right at the end of the period and then much less than Enstrom predicts. Indeed, if we estimate the average relative risk for the whole period of 25 years, we come up with an estimate of 15.88, which is virtually the same as the figure of 16.01 for 1955, when the group were still all current smokers. I conclude that Enstrom's expectations are wrong and his analysis is not an appropriate way to demonstrate effects of ex-smoking. My analysis is over-simple because inter alia it does not take account of the fact that (i) numbers surviving will decline with time, so that my overall estimate of 15.88 is an underestimate, and (ii) those surviving to the end of the period will have smoked less than average, so that my estimate of 15.88 is an overestimate. However the general point is clear enough. Increased duration will increase relative risks in continuing smokers and this will largely counterbalance the decline to be expected from the increased proportion of ex-smokers. P N Lee 3.4.2000

1 REVIEW 1098 , ~ CONFIDENTIAL Subject ref 3b "Smoking cessation and mortality trends among two United States populations" J E Enstrom Journal of Clinical Epidemiology (1999), 52, 813-825 Epidemiological studies have clearly shown that the risk of lung cancer, relative to that in a continuing smoker of the same age, declines substantially in ex-smokers, the decline being greater the longer the period of cessation. In this paper, which relates to a similar paper reviewed earlier (Review 968) based on a different population, Enstrom argues that such comparisons might be biassed because the ex-smokers are self-selected and because, to be an ex-smoker who has given up for N years, one has had to survive those N years. Ideally one would test the effect of ex-smoking by a randomized intervention trial, but such evidence is very limited, the existing data lacking power to detect the order of magnitude of effect that the epidemiological data would suggest. Enstrom suggests an alternative approach based on data from large prospective studies (here the US Veterans Study and the NHANES Study). He limits attention to those subjects who, at the start of the follow-up period, reported being current smokers or having never smoked and investigates how the relative risk of lung cancer of these two groups varies over time. Given that (i) the never smokers will stay being never smokers (a very reasonable assumption as the cohorts he considers in his analysis were relatively old at the start of follow-up; 55-64 for the Veterans and 55-74 for NHANES), (ii) a very substantial proportion of the current smokers at start of follow-up will have given up over the follow-up period of 1954-1979 for the Veterans and 1971-1992 forNHANES (again a very reasonable assumption, supported by data shown in Tables 1 and 6), and (iii) ex-smokers have lower lung cancer risks than continuing smokers,

5 Year Breakdown 1955 100% current smokers 1960 83.3% current smokers (C), 16.7% ex-smokers for an average of2%: years (x2'h) 1965 66.7%C, 16.7% x 2'/2, 16.7% x 7'/z 1970 50.0%C, 16.7o/u x 2%z, 16.7% x 7Yz, 16.7% x 12'/z 1975 33.3%C, 16.7% x 2'/z, 16.7% x 7'/z, 16.7% x 12Y2, 16.7°/n x 17%z 1980 16.7%C, 16.7% x 2'/z, 16.7% x 7'/z, 16.7% x 12'/z, 16.7% x 17'/z, 16.7% x 22%z I computed the lung cancer risk from my model, relative to that of never smokers of the same age, for each year and sub-population. This gave Year 1955 160 1965 1970 1975 1980 Continuing smokers 16.01(6) 17.20(5) 18.32(4) 19.37(3) 20.35(2) 21.29 (1) Ex-smoker gave up . for 2% years 14.53 (1) 15.69 (1) 16.79 (1) 17.82 (1) 18.80 (1) 7% years 11.42 (1) 12.51 (1) 13.57 (1) 14,57 (1) 12'hyears 9.27(1) 10.26(1) 11.23(1) 17Y:years 7.75 (1) 8.64 (1) 22'h years 6.66 (1) [Bracketed numbers are relative population weights assumed (with total = 6 for any year).] It can be seen that, as the population ages, the original current smokers divide into subgroups with disparate risks, depending on the time given up. Using the weights to estimate the average risks at each year, or for each five-year period, we have

3 while the risk for an ex-smoker aged T starting at age S, smoking for a period D and then followed for a further period F (so that S+F+D = T) is given by RH = aiTK-' + d1L('i+F)K-' - FK-'1 + d2L(D+S)K'2 - SK''] + d,d2DK-' } Here V, and V2 are the transition probabilities for the first and penultimate stage while smoking, expressed relative to those while not smoking, with dl = V1-1 and d2 = V2-1 the excesses due to smoking. For the work described here I took d, = 4, d2 = 8 and S= 20 years. I first looked at the model's predictions for the continuing smoker/never smoker relative risk. As shown in the table below, this rose continuously with age and was 33% higher at age 85 (end of follow up period for the Veterans) than at age 60 (beginning - taking the 55-64 year olds as 60 year olds for convenience). Age Relative risk for continuing smoker/ never smoker of same aee Ratio 60 16.01 1.00 65 17.20 1.07 70 18.32 1.14 75 19.37 1.21 80 20.35 1.27 85 21.29 1.33 I next looked at the model's predictions for ex-smokers in two ways. First I looked at the risk at age 85 for those who gave up at various times

2 he expects to find (if the epidemiological data are valid) that the lung cancer relative risk for the two groups as originally defined will decline substantially with time. In fact he finds no such evidence in either study. Notably, for the much larger Veterans study conducted over a period where giving up smoking was common, the relative risks were as follows: Year Relative risk (95% CI) 1954-59 10.9 (7.7-15.4) 1960-69 11.6 (9.5-14.1) 1970-79 11.4 (9.4-13.8) Enstrom argues that the failure to see a decline in the relative risk is surprising and raises "an important issue about the value of smoking cessation." My major concern with his work is that he has failed to take into account the fact that, as the cohorts age, the duration of smoking of the continuing smokers increases so that their lung cancer risk, relative to never smokers, will also increase. Though the risk of the ex-smokers will be lower, it is not at all clear what the overall effect on the lung cancer rate of the total original group of current smokers at the start of the study will be. To gain some insight into this I carried out some approximate estimates using a multistage model with five stages in which (following previous work relating to lung cancer trends in the US and UK that Barbara Forey and I have carried out'• Z) we assumed that smoking affected the first and penultimate stage, the effect on the penultimate stage being twice that on the first stage. Formally, if T is age and the risk for a never smoker is given by aTK" RN = - the risk for a continuing smoker aged T starting at age S is given by RC = a{.SK_ I i- VI V21 a-.S)K I+ v2\TK-I-SR I-(T-S)K I) )