Tobacco Documents Online

Aee 4 VJig 35-39 0.174 40-44 0.145 45-49 0.140 50-54 0.135 55-59 0.111 60-64 0.096 65-69 0.080 70-74 0.062 75-79 0.035 80-84 0.022 The age-adjusted death rate is then calculated by summing the products of the age- specific weights and rates. However, and this is a major error, age-specific rates have been calculated as zero when there were no deaths even if there were no people at risk! In the final period (1995-97) for example, rates of zero are calculated for ages 30-34 up to 60-64 despite the fact that there could not possibly be any qualifying people as men and women aged 30 or more in 1959 would perforce be 65 or more in 1995. Clearly the rate is not zero for these age groups, but is not estimable, and the age-adjusted rate as so defined is also not estimable. It should also be noted that the weighting system is heavily weighted to results for the younger age groups (e.g. almost 60% of the total weight for ages 35-54) and that the contribution of rate estimates for these age groups to the overall 1960-1997 age-weighted estimate comes wholly from experience very early on in the 38 year period. A proper assessment of time trends cannot be obtained from such analysis. In order to provide valid comparison of rates over the whole following period, one has to restrict attention to an age group with adequate numbers of person-years of observation throughout. Ages 70-74 and 75-79 are clearly best for this purpose and yet Enstrom and Heath's weighting procedure gives very little weight to these ages.

2 smokers as at 1959 (test-group) and never smokers as at 1959 (control group). On the basis that those who were never smokers in 1959 would be very unlikely to start smoking, and on the basis that a very large proportion of current smokers in 1959 would have given up smoking by 1997, the authors argue that study of trends in relative mortality over the follow-up period provides a "natural" experimenY' to test for the effects of ex-smoking. Though they note that this "natural experiment" does not involved randomisation of subjects and is not nearly as rigorous as a randomised controlled trial, they argue that inferences from such a comparison should provide a better test of possible ex-smoking effects than the "conventional" analysis described above. Clearly Enstrom and Heath expect that if giving up smoking has a major effect on mortality, the mortality of the test and control group should converge with time. In fact, their results show that this is not the case. Comparing mortality in the first ten years of follow-up (1960-89) with that over the whole 38 year period (1960-97), they find the following. Current Smokers Never Smokers CurrentlNever 1960-69 1960-97 1960-69 1960-97 1960-69 1960-97 Men - all cancers 20.67 18.68 10.51 9.46 1.97 1.97 - lung cancer 1.558 1.728 0.127 0.133 12.3 13.0 Women - all cancers 9.54 10.14 6.95 6.44 1.37 1.57 - lung cancer 0.208 0.806 0.094 0.116 2.21 6.95 It can be seen that there has been no evidence that current/ever ratios of mortality have been decreasing between 1060-69 and 1960-97 and that for female lung cancer rates the reverse is true, with current/never ratios increasing dramatically, from 2.21 in 1960-69 to 6.95 in 1960- 97. The authors conclude that "their findings, alone with the results of randomized controlled trials, suggest that the impact ofcessation on mortality, particularly lung cancer mortality, among cigarette smokers as a whole is less than currently believed", although they do not clearly state what actually is currently believed.

5 If one does something more reliable and simply averages the overall death rates for the 70-74 and 75-79 ages and also for adjacent 5 year periods; one gets the following: 1960-69 1970-79 1980-89 1990-97 Men - current smokers 71.1 70.0 62.2 53.3 - never smokers 40.7 40.4 30.3 28.5 - ratio 1.75 1.73 2.05 1.87 Women - current smokers 31.5 41.4 38.2 35.2 - never smokers 26.0 26.9 21.3 17.1 - ratio 1.21 1.54 1.79 2.06 Here one can see that there is no evidence of a decline in the current/never smoker ratio for men and a clear increase for women. So, although their statistical procedures are quite wrong, the conclusions of a lack of decline in the current/never smoker death rate is correct. However, Failure to consider the increase in duration of smokine By far the most important weakness of the paper is that the authors have not considered at all the huge effect on death rates of change over time in the duration of smoking among continuing smokers. This will be most pronounced for women, where cigarette smoking did not really take off until World War II. For women aged 70 for example, the average age of starting to smoke would be around 40 in 1959, but around 20 in 1997. If lung cancer rates are proportional to the 4.5th power of duration, then the risk of lung cancer for women aged 70 who continue to smoke until that age would, all other things being equal, be (50/30)° 5= 10 times higher in 1997 than in 1959. The effect would be less marked for men but even so, since smoking by men mainly started after World War 1, male 70 year olds smoking for 50 or more years would be much commoner in 1997 than in 1959. The maj or problem with Enstrom and Heath's methods, which renders it totally valueless, is that the time trends they hope indicate effects of giving up smoking actually indicate

3 In considering these findings a number of points have to be borne in mind: Incomplete mortality - follow-un 5% ofinen and 3% of women were known to have died with the cause of death unknown and 6% of men and 1 I% of women were lost. This may have caused some bias in the analysis, though it is probably not major. Limited data on smoking habits since 1972 This is not a major weakness. It is abundantly clear that a very large proportion of the men and women who were current smokers in 1959 would have given up by 1997. Limited information on effects of ig ving up early The study was limited to men and women aged 30+ and the median age for current smokers is between 50 and 54. It is clear that the study gives little information on those who smoke for only a few years, giving up early. Strange comparison of time periods Comparison of the overlapping periods 1960-69 and 1960-97 is very odd. I would have expected to see analysis looking at trends in mortality over discrete 5 or 10 year time periods. In fact some data (only for overall mortality) by 5 year periods are shown in the Appendix and do appear to show evidence of a decline over time. Thus, the age adjusted death rates for age 35- 84 are 19.57, 21.47, 18.95, 17.38, 16.83, 14.18, 10.44 and 8.87 for successive periods 1960-64, 1965-69, .... 1990-94, 1995-97. However....... Totally inannropriate age-adjustment procedure The footnote of Appendix AI, gives age weights as follows

6 contrasting effects of increased duration of smoking by those who continue to smoke (suggesting a marked increase in death rates) and of increased smokers giving up (suggesting a marked decrease). Failure to consider time trends in other lung cancer risk factors Analyses I have carried out with Barbara Forey indicate very clearly that there is some factor other than smoking in the US which is leading to an increase in lung cancer ratio over time, contrasting to the situation in the UK where the reverse is true, with ratio declining faster than expected given trends in smoking. Another problem with the Enstrom and Heath methodology is that they assume that trends in factors other than smoking act to the same extent in current and never smokers. But this may well not be so. The major problem with the paper, however, is that it totally ignores trends in duration of smoking. P N Lee 7.10.99 P.S. After completing this review I became aware that the Journal of Clinical Epidemiology (2, 827-836) had also published a"rebuttaP' to the paper by Kuller, a response by Enstrom and a rejoinder to the response by Kuller. These are also attached for information. Although Kuller agrees with me that the Enstrom and Heath paper is of little value, I found his rebuttal unclear and missing some of the major problems.

LlI2 610 1 REVIEW 1068 CONFIDENTIAI:, Subje_ct ref 3b "Smoking cessation and mortality trends among 118,000 Californians, 1960-1997" 7 E Enstrom and C W Heath 7r Epidemiology (1999), 10, 500-512 In 1959, over a million men and women aged 30+ enrolled into the American Cancer Society Cancer Prevention Study I (CPS I), and detailed data on smoking habits and other variables were collected. While mortality follow-up forthe majority ofthe sample did not extend past 1972, for 119,000 men and women in California mortality follow-up extended until the end of 1997. Conventional analyses of the effect of ex-smoking have compared mortality risk of subjects classified by their smoking status at the start of follow-up. The results typically show that risk of overall mortality and of lung cancer is highest in current smokers and short-term quitters, lowest inevsmokers and intermediate in medium and long-term quitters with increasing time offgi-v-ing up associated with decreasing risk. Indeed, analysis of the Californian subjects relating mortality over the period 1960-69 to smoking habits as recorded at 1959 shows this pattern reasonably clearly, especially for men (see Table 3 of the paper). Enstrom and Heath argue that such analyses are biased by the fact that the population studies are constrained to have survived to the time the study began. Thus, subjects starting in 1959 that may have given up for 15 or more years are known to have survived at least 15 years from the time of giving up, and may well be unrepresentative of all ex-smokers who gave up on or before 1944. In an attempt to give a more reliable insight into the effect of ex-smoking, Enstrom and Heath employ an alternative approach. They look at age-adjusted time trends in overall mortality and in mortality from lung cancer in two sub-groups of the Californian men and women, current