Tobacco Documents Online

. 435 Table: Relative Risk Estimates for Cases vs. Controls (from Table 4, p. 483) Relative Risk 90% Confidence Limits Among Smokers, OC users 4.7 (1.3-17.6) Among Nonsmokers, OC users 12.8 (1.8-90.2) Among OC Users, Smokers 2.3 (0.8- 7.1) Among Non-OC Users, Smokers 7.6 (1.6-36.2) Because the sample sizes for cases are so very small and the confidence intervals are so wide, these results are of question- able reliability. I:. Studies which do purport to find a definite and increased risk of heart attack and/or stroke for women who smoke and use OC. A. Studies criticized in my previous testimony: (i) Jain (1977) uses the data in Mann et al. (1975) consisting of 63 women under age 45 who had survived an MI and a control group. Among the cases, there were three nonsmokers and 13 smokers who were using OC at the onset of the MI episode. These numbers are too unbalanced and too small to justify any reliable conclusions about the interrelationship of smoking and OC use on MI. Jain concedes in his paper that these mortality data are based on small numbers and may be subject to significant sampling errors; this 0 -a h ~ i

caveat should not be ignored. Further, his analysis is based on two unwarranted but convenient assump- tions which he justifies making ". . . because the relevant data . . . are not available" (p. 51). This is a non sequitur, and highly unscientific reasoning. (ii) Beral (1977) uses the Royal College of General Practi- tioners (RCGP) data but the numbers are still very _-sma11, especially the deaths for nonsmokers. She concedes that "These estimates are based on small numbers and are necessarily approximate. Without more data it is not possible to examine the inter- relationships of age, smoking, and duration of oral contraceptive use . . ." (p. 730). (iii) Jick et al. (1978b) report data on 26 women with acute but nonfatal MI and 59 controls and give rela- tive risk estimates for OC users. However, no MI subjects were nonsmokers who did not take oral contra- ceptives so relative risk estimates for OC users who smoke could not be obtained. (iv) Petitti and Wingerd (1978) give relative risk factors for subarachnoid hemorrhage (SAEi) for women who smoke and use OC. flowever, this analysis is based on extremely small numbers, a total of U women, which includes none who were nonsmokers and non-OC users, and only six who were smokers and OC users. Certainly .-I this is insufficient data ing an interrelationship. 8. Shapiro et al. (1979) report a stud- with MI and 1742 control premenopau< the following age-adjusted rate rati dence intervals of MI for recent OC P. 745): Nonsmokers 4.5 (1. Smoke 1-24/day 1.2 (0. Smoke > 25/day : 4.3 (2. The estimated rate ratio for heavy sm, for non-smokers and the confidence in~ completely includes the confidence lir Further, the rate ratio estimate for ¢ siderably smaller than either of those smokers, and the confidence interval fc includes the value 1.0, which ehows no risk. The age-adjusted rate-ratio estimates g (p. 746), on the other hand, give a vern estimate for women who are heavy smokers that the Table VI results were derived f Table V, but the results in Tables V and that I do not see how they could have be,

this is insufficient data to use as a base for claim- ing an interrelationship. !, Shapira et al. (1979) report a study of 234 premenopausal women with MI and 1742 control premenopausal women. The authors give the following age-adjusted rate ratio estimates and 952 confi- dence intervals of MI for recent OC users (from Table V, - p. 745): Nonsmokers 4.5 (1.4-14.1) Smoke 1-24/day 1.2 (0.3- 4.4) Smoke > 25/day, ,- 4.3 (2.2- 8.2) The estimated rate ratio for heavy smokers is about the same as for non-smokers and the confidence interval for nonsmokers completely includes the confidence limits for heavy smokers. Further, the rate ratio estimate for moderate smokers is con- siderably smaller than either of those for nonsmokers and heavy smokers, and the confidence interval for moderate smokers includes the value 1.0, which shows no significantly greater risk. The age-adjusted rate-ratio estimates given in Table VI (p. 746), on the other hand, give a very large rate-ratio estimate for women who are heavy smokers. The authors state that the Table VI results were derived from the data shown in Table V, but the results in Tables V and VI are so inconsistent that I do not see how they could have been obtained from the

438 same data base. Further, the confidence intervals in Table VI are extremely wide for all categories of OC use and similarly for non-OC users who are heavy smokers. The confidence limits for moderate smokers who use OC include the value 1.0, a result that is consistent with the Table V results, but this again implies no risk. My independent calculations of rate ratios from the data in Table V without adjusting for age, as shown in Appendix C, are more consistent with Table V ratios than are those given by the authors in Table VI. Perhaps more important here is the small frequency of women in each group. See Appendix C. These limited sample sizes alone could justify an argument that the results given by Shapiro et al. are not reliable. C. Jick et al. (1978a) extend an earlier case-control study of nonfatal MI and its relation to OC use and smoking to include a total of 83 case and 154 control subjects. The authors claim a strong positive association between MI and OC use, and between MI and smoking. My independent analyses of their data, shown in'6ppendix D, confirm these conclusions. However, these authors also state "In both groups there is an extremely strong correlation between smoking and MI. Of the 83 case patients Interviewed, 74 (89%) were current smokers. The corresponding figure for the 153 controls is 67 (44Z)" (p. 2,549). As this D. 439 statement shows, it is certain smokers is larger in the case but this fact has nothing to d percentages have no relevance and smoking because the data b. cases and controls. This pape: frequences of women wit respec! acteristics. The NIH Report (1981) states "t firm that OC users who smoke or at somewhat [emphasis added] gr effects, particularly circulato However, no specific data or re the degree of greater risk real E. Petitti et al. (1979) also use - (p. 1,152) shows the relative r- smoke and 2.8 for women who smok risk is lower), and the lower 9C who smoke and use OC is 0.8, les MI cases are based on a total of Other relevant results in this t of 5.7 for smokers and 21.9 for based on a•total of Il observati

439 statement shows, it is certainly true that the percentage of smokers is larger in the case group than in the control group, but this fact has nothing to do with correlation; in fact, the percentages have no relevance for the relationship between MI and smoking because the data bases were not random samples of cases and controls. This paper gives no data on the joint frequences of women wit respect to MI, OC use or smoking char- acteristics. D. The NIH Report (1981) states "the Walnut Creek data also con- firm that OC users who smoke or who are older than 30 years are at somewhat [emphasis added]'greater risk of serious side effects, particularly circulatory disorders" (p. 1,071). However, no specific data or results are given to clarify what the degree of greater risk really is estimated to be. E. petitti et al. (1979) also use the Walnut Creek data. Table 1 (p. 1,152) shows the relative risks for MI as 2.9 for women who smoke and 2.8 for women who smoke and use OC (note this relative risk is lower), and the lower 90% confidence limit for women who smoke and use OC is 0.8, less than 1.0; these results for MI cases are based on a total of 26 observations, however. Other relevant results in this table are relevant risks for SAH of 5.7 for smokers and 21.9 for women who smoke and use OC, based on a-total of 11 observations, and relative risks for

440 other stroke as 4.8 for smokers and 2.0 for women who smoke and use OC (note this relative risk is lower than the relative risk for smokers), based on a total of 23 observations. Table 3 (p. 1,152) gives incidence rates for the combined types of cardio- vascular disease for 3 women under age 45 who neither smoke nor use OC, and 8 women under age 45 who both smoke and use OC. All of the aforementioned results or relative risks are highly questionable because of the extremely small numbers of cases in each subgroup (especially that of women under age 45). In spite of this severe limitation on reliability, the authors' conclusion is that "smoking and OC use appear to act syner- gistically to increase the risk of subarachnoid hemorrhage, hemorrhagic stroke, and @fI" (p. 1,154). In my opinion, their data do not justify this conclusion at all. . g. Layde, Beral, and Ray (1981) use the RCGP data to study the re- lationship between smoking and OC use in regard to mortality from SAH and from various circulatory diseases. An independent analysis of their data in Tables IV and V (p. 543), shown in Appendix E, implies that the following statistical conclusions are appropriate: In the population of women aged 35-44, there is no association between OC use and smoking for those who died from SAH, nor for those who died from circulatory disease. 441 The sample sizes here are small (20 deaths from circulatory disease). T deaths from circulatory disease to c and excess risk estimates for Ever-u Table V (p. 543), separately for eac smoking. The individual numbers of subgroups are extremely small (see A authors do not use these actual numb risks and excess risks; rather they 100,000 women-years and this leads t, larger sample sizes. Moreover, the ; risk [of circulatory disease] for ev, smokers than among non-smokers for e: The authors fail to point out that t! of relative risk for nonsmokers is cc for smokers for women aged 35-44, anc over; and in fact the interval for nc interval for smokers in each case, wt difference between relative risks foc G. Rosenberg et al. (1980) study the ef: MI in the presence and absence of ott factors including cigarette smoking. estimates and 95% confidence interval other predisposing conditions are as p. 63):

441 The sample sizes here are small (20 deaths from SAH and 65 deaths from circulatory disease). The authors use the 65 deaths from circulatory disease to compute the relative risk and excess risk estimates for Ever-users vs. Controls in Table V (p. 543), separately for each subgroup of age and smoking. The individual numbers of cases for each of these subgroups are extremely small (see Appendix E, Table 14). The authors do not use these actual numbers to compute relative risks and excess risks; rather they use mortality rates per 100,000 women-years and this leads to an impression of much larger sample sizes. Moreover, the authors state "The relative risk [of circulatory disease] for ever-users was greater among smokers than among non-smokers for each age group" (pp. 543-544). The authors fail to point out that the 95% confidence interval of relative risk for nonsmokers is considerably wider than that for smokers for women aged 35-44, and also for women 45 and over; and in fact the interval for nonsmokers includes the interval for smokers in each case, which implies no significant difference between relative risks for smokers and for nonsmokers. G. Rosenberg et al. (1980) study the effect of 0C use on nonfatal MI in the presence and absence of other predisposing risk factors including cigarette smoking. Their relative risk estimates and 95% confidence intervals for KI for women without other predisposing conditions are as follows (From Table 4, p. 63):

443 Normotensive nonsmokers 2.8 (1.0-7.8) Normotensive smokers 1.1 (0.5-2.6) These risk estimates are for current OC users relative to women who had never used OC. The relative risk for normotensive smokers is smaller than that for normotensive nonsmokers, and the lower confidence limit is 0.5, which implies that the additional risk for smokers using OC is probably nonexistent. And yet the authors claim "The increase in risk attributable to the combined effect of current OC use, cigarette smoking and hypertension was considerably greater than what would be predicted from the sum of the sepa- rate effects of these factors" (p. 59). Each of the risk estimates given in Table 4(p. 63) is based on a very small number of cases, however. Only 7 cases were OC users and smokers without other predisposing conditions, and only 12 cases were neither smokers nor OC users. Further, the data base is married U.S. female registered nurses, which is hardly representative of all U.S. females. Appendix A Analysis of Date from Krueger et Table 1. Number of Smokers and OC Users Amor (from Table 13, p. 666) Non OC Users Nonsmokers 42 Smokers 101 Totals 143 ~ . - .0033, .90 < P < .95 :?e appropriate statistical conclusion from Table Jomen aged 15-44 who died from MI, the factors of ao association. Table 2. Number of Smokers and OC Users Amor Predisposing Conditions (from Table Non OC Users Nonsmokers 6 Smokers 31 Totals 37 2 ( - .016, P - .90 1 Table 3. Number of Smokers and OC Users Amor disposing Conditions (from Table 1z Non OC Users Nonsmokers 36 Smokers 70 Totals 106 2 X - .1038, .70 < P < .80 1 -19- 95-077 95-077 0-82--29

disposing Conditions (from Table 14, p. 666) Non OC Users OC Users I Totals Nonsmokers 42 6 Smokers 101 14 Totals 143 20 2 ~ ~ .0033, .90 < P < .95 1 443 Appendix A Analysis of Date from xrueger et al. (1980) Table 1. Number of Smokers and OC Users Among the 163 Cases (from Table 13, p. 666) Non OC Users OC Users I Totals 48 115 163 The appropriate statistical conclusion from Table 1 is that in the population of women aged 15-44 vho died from MI, the factors of Cigarette Use and OC Use show so association. Table 2. Number of Smokers and OC Users Among the 44 Cases Without Predisposing Conditions (from Table 14, p. 666) Non OC Users OC Users Nonsmokers 6 1 Smokers 31 6 Totals 37 2 X " .016, P - .90 1 Nonsmokers 36 5 Smokers 70 8 Totals 106 13 2 X - .1038, .70 < P < .80 1 Totals 7 37 44 Table 3. Number of Smokers and OC Users Among the 119 Cases with Pre- 9SA77 0-82--29 -19- 41- 78 119

The appropriate statistical conclusions from Tables 2 and 3 are that in the population of women aged 15-44 who died from M2, the factors of Cigarette Use and OC Use show no association, irrespective of whether there are predisposing medical conditions. The odds ratios associated with OC use and smoking as given in Table 13, p. 666, are simple proportions of cases versus controls in each smoking cate- gory relative to the same proportion for nonsmokers, non-OC users, calculated as follows: Nonsmokers, non-OC users Reference Category - 1.0 Nonsmokers, OC users 6/9 * 42/138 - 2.19 Smokers, non-OC users 101/154 t 42/138 - 2.15 Smokers, OC users 14/25 • 42/138 ~ 1.84 -20- 445 Appendix B ' Analysis of Data from Jick et al Table 4. Number of Cases and Controls Who Sm Control Strok Nonsmoker 33 7 Smoker 23 Totals 56 14 2 1 - . 36, .50 < P < .70 r. L Statistical Conclusion: There is no association bE stroke and smoking in these women. Table 5. Number of Cases and Controls Who Use Control Stroke Non-OC User 49 3 OC User 7 11 Totals 56 14 2 ~ 25.60, p 1 < .001 Statistical Conclusion: There is a significant ass of nonfatal stroke and OC use in these women. -21-