Tobacco Documents Online

425 STATEMENT OF JEAN D. GIBBONS My name is Jean Dickinson Gibbons. My current position is Professor of Statistics and Chairman of the Applied Statistics Program at the Graduate School of the University of Alabama. I am currently a Fellow of both the American Statistical Association and the International Statistical Institute and a member of the Committee on National Statistics of the National Academy of Sciences. I received the bachelor's and master's degrees in mathematics from Duke University and the Ph.D. degree in statistics from Virginia Polytechnic Institute and State University. My previous faculty appointments were at the University of Pennsylvania and the University of Cincinnati. I was a senior Fulbright-Hays scholar at the Indian Statistical Institute in 1973. I was Associate Editor of The American Statistician for eight years, currently act as editorial collaborator on many statistical journals, includ- ing The Journal of the American Statistical Association, Biometrics, and Technometrics, and serve as a reviewer for grant proposals to the National Science Foundation. I am a member of several professional societies and have served two terms on the Board of Directors of the American Statistical Association. My publications include four scholarly books on statistics and over 30 articles in refereed professional and learned journals in my field. I was named Outstanding Scholar in 1981 and Board of Visitors Research Professor in 1974 at the University of Alabama. My current curriculum vita is attached to this statement.

42' In February of 1978, 1 was asked to review the reported statistical studies that formed the basis for the Food and Drug Administration's (FDA) decision to include a boxed warning in the patient and physician labeling on oral contraceptive (OC) products which states that cigarette smoking increases the risk of serious cardiovascular side effects for OC users. In October of 1978, at the request of Representative L. H. Fountain, I testified as an expert witness at a House Subcommittee Hearing on the "Quality of Scientific Evidence in FDA Regulatory Decisions (The Adoption of an Antismoking Warning in Oral Contraceptive Pill Labeling)." My conclusion at that time was that the statistical evidence published in the literature about the interactive effects of smoking and oral contraceptives on risk of cardiovascular disease is quite weak because the sample sizes in'most studies are extremely small, the results may be subject to significant sampling errors, and the results in some studies are based on convenient but unfounded assumptions. The authors of these papers in many cases pointed out these limitations of their data and deficiencies in their analyses. At that time I suggested that the FDA should run a controlled experiment to obtain sufficient and relevant data on factors such as length of time of OC use; number of years and amount of smoking; genetic, environmental, and psychological characteristics; among others. My prepared testimony and the discussion following at that hearing are a part of the written record. Bill H.R. 4957 contains a finding that states "(5) women who take birth control pills and smoke are more likely to suffer a heart attack or stroke than women who don't smoke". This finding is similar to the wording on the patient insert warning on boxes of OC which was at issue in my previous -2- testimony. Therefore, when Representati• February of this year on this matter, I studies that have been published on this current study included over 20 papers, i: for the previous testimony and the 3 add and were mentioned during the questionin references for the papers appearing sinc to this statement. This statement includes a brie studies published since 1978. I have lo data, findings, and conclusions of the a pendent calculations to measure the stat and OC use in their data and to check th analyzed those papers in which the auth:, tncreased risk of heart attack or stroke well as those papers where the authors c increased risk of heart attack or stroke have not included in this written analys relevant to the finding stated in H.R. 4 9ignificance, or do not help clarify the My primary overall conclusion 3i11 H.R. 4957 is at present groundless vhich it is based is limited, weak, conc criticism by impartial experts. There that, while also subject to criticism, I 95-077

427 al )n ~ A ic ag .t ise 1, 9 in or. and .ould :tors MY rt of testimony. Therefore, when Representative Fountain again contacted me in February of this year on this matter, I examined and reviewed the statistical studies that have been published on this topic since my previous analysis. My current study included over 20 papers, in addition to the 13 papers examined for the previous testimony and the 3 additional papers that appeared in 1978 and were mentioned during the questioning at that hearing. A complete list of references for the papers appearing since 1978 that I have studied is attached to this statement. This statement includes a brief analysis of each of the relevant studies published since 1978. I have looked carefully and objectively at the data, findings, and conclusions of the authors, and also performed some inde- pendent calculations to measure the statistical relationship between smoking and OC use in their data and to check their conclusions. I have critically analyzed those papers in which the authors claimed there is no (or a slight) increased risk of heart attack or stroke for women who smoke and use OC, as well as those papers where the authors claimed that there is a definite and increased risk of heart attack or stroke. The papers which I reviewed but have not included in this written analysis are, in my opinion, either not relevant to the finding stated in B.R. 4957, or do not add any new results of significance, or do not help clarify the situation. My primary overall conclusion is that Congressional Finding (5) in Bill H.R. 4957 is at present groundless because the statistical evidence on which it is based is limited, weak, controversial, and subject to severe criticism by impartial experts. There are current studies in the literature that, while also subject to criticism, have come to the opposite conclusion 95-077 0-82-28

428 429 and indeed claim that there is no interactive effect of OC use and smoking on the occurrence of heart attack or stroke in women. The evidence for either conclusion is limited and weak and subject to serious scientific criticism. In my professional judgment, I believe that the Congress should not in good conscience find that "women who take birth control pills and smoke are more likely to suffer a heart attack or stroke than women who don't smoke" because the scientific evidence is inadequate. Most of the women who will be affected and influenced personally by this finding will not have the scientific back- ground to form their own conclusions and will interpret the finding as truth, when, in fact, it is at best a questionable opinion that has been neither confirmed nor denied by the data in the reported studies. I again strongly urge the Congress to recommend that a controlled study be carried out with a good data base so that the issue can be addressed properly. More research is urgently needed before Congress can make a finding of such public importance. In support of these.general conclusions, I have attached a brief eummary of my analyses of (I.) the group of reported studies that claim no increased significant risk of heart attack and/or stroke, and (II.) the group of reported studies that do claim an increased risk for women who smoke and use oral contraceptives. A complete list of references is attached. Addi- tional details are given in the Appendices. Analysis of Reported St, I. Studies which purport to find no (or sligh attack and/or stroke for women who smoke a A. Krueger et al. (1980) report a c of death from myocardial infarct for the period January 1974-June the five largest metropolitan ar on smoking habits and OC use wer died of MI and 326 control women sizes on which to base a conclus reasonably reliable. - _ A primary stated conclusion of t effect of OC use and smoking on recent studies in the U.S. of nc (p. 672). My independent stati: conclusion of no interactive ef: based simply on the reported nur among the cases. My three conc analysis are as follows: 1. In the population of from MI, the factors no statistical associ

® 429 Analysis of Reported Studies I. Studies which purport to find no (or slight) increased risk of heart attack and/or stroke for women who smoke and use OC. A. Krueger et al. (1980) report a collaborative case-control study of death from myocardial infarction (MI) in women aged 15-44 for the period January 1974-June 1975 in regions representing the five largest metropolitan areas in the United States. Data on smoking habits and OC use were reported for 163 women who died of MI and 326 control women; these are reasonable sample sizes on which to base a conclusion and these data appear reasonably reliable. A primary stated conclusion of the authors is "An interactive effect of OC use and smoking on risk of MI, as reported in recent studies in the U.S. of nonfatal MI, was not found . . ." (p. 672). My independent statistical analysis to verify their conclusion of no interactive effect, shown in Appendix A, is based simply on the reported numbers of smokers and OC users among the cases. My three conclusions from this independent analysis are as follows: 1. In the population of 163 women aged 15-44 who died from MI, the factors of cigarette use and OC use show no statistical association.

430 2. In the population of 44 women aged 15-44 who died from MI and had no predisposing conditions, the factors of cigarette use and OC use show no statis- tical association. 3. In the population of 119 women aged 15-44 who died from MI and had predisposing conditions, the factors of cigarette use and OC use show no statistical association. Another primary stated conclusion of the authors is "Smoking and OC use together appeared to be no more of a risk factor for fatal MI than either smoking or OC use separately, compared to a reference group of nonsmokers and non-OC users" (p. 667). The authors based this conclusion on the odds ratios reported in Table 13, p. 666, and I have verified these odds ratios in Appendix A by independent calculations. The odds ratios and 95% confidence limits (from Table 13, p. 666) are as follows: Nonsmokers, OC users 2.19 (0.60, 7.33) Smokers, non-OC users 2.15 (1.38, 3.39) Smokers, OC users 1.84 (0.81, 4.06) The fact that the confidence interval for smokers and OC users includes 1.0 is statistical evidence that there is not neces- sarily any increased risk of MI for women who smoke and use QC over those who do neither; the same conclusion applies to women who do not smoke and do use OC of these statements is .95. 1 smokers and OC users is smalle of the. groups (Nonsmokers, OC users) in fact implies the opp It is unfortunate that the pap the joint characteristics of c ceased controls so that the re who died from MI and women who also unfortunate that the data have no breakdown according tc use, and age category within t duration of OC use, duration c smoked, and age are important ing whether a relationship ex` B. Slone et al. (1981) report on of nonfatal MI with respect tc of current and past OC use, t: within 25-44 years, and four only data on rate-ratios prov smoking status snd OC use are p. 423):

431 who do not smoke and do use OC. The confidence level for each of these statements is .95. The fact that the odds ratio for smokers and OC users is smaller than the odds ratios for each of the. groups (Nonsmokers, OC users) and (Smokers, Non-OC users) in fact implies the opposite of a synergistic effect. It is unfortunate that the paper does not give information on the joint characteristics of cigarette and OC use among the de- ceased controls so that the results could be compared for women who died from MI and women who died from other causes. It is also unfortunate that the data given on both cases and controls have no breakdown according to amount of smoking, years of OC use, and age category within the 15-44 years. Surely the duration of OC use, duration of smoking, amount of cigarettes smoked, and age are important factors to consider in determin- ing whether a relationship exists. B. Slone et al. (1981) report on a case-control study of the rate of nonfatal MI with respect to the characteristics of duration of current and past OC use, three subcategories of age group within 25-44 years, and four categories of smoking status. The only data on rate-ratios provided in the paper that concern smoking status and OC use are as follows (from Table 7, p. 423):

432 Never Smoked 2.5 Ex-smokers 2.9 1-24 cigarettes/day 1.5 > 25 cigarettes/day 1.4 The authors conclude from these findings that "The rate-ratio estimates declined with increasing cigarette smoking, and the trend was statistically significant. This finding is at vari- ance with previously published observations on current use of oral contraceptives" (p. 423). HoweveY, the authors warn of possible bias in this study and recommend that the data be interpreted with caution. C. Maguire et al. (1979) extend the analysis of some previously reported data on a case-control study of four diagnostic cate- gories of thrombosis (including MI). The primary conclusion relevant here is that their analysis indicates "no strong evidence of modification in the relative risk associated with oral contraceptive use by age or smoking for any of the throm- bosis diagnoses considered. It is of interest, however, that in all groups except predisposed venous thrombosis the effect modifier coefficients were negative, suggesting a consistent pattern of decreasing estimated relative risk associated with pill use with both smoking and older age" (p. 193). 432 s.is.y et al. (1979) use vital :crt.lity from cardiovascular ,~%V pre-pill period as control :sers. (This is a re-examinat :)r a 1976 publication in Lanc eare included.) These author creasing mortality from cardio increased levels of pill use u 7~teir analyses fail to show a ;.ercent of women smoking and c: authors do point out, however, statistics as opposed to a cas, .I:ck et al. (1978c) report on c stroke in premenopausal women. 1a the report by the CGSS [Col'_ 3troke in Young Women], cigaret associated with stroke in healt independent analysis of their d that cigarette smoking is not a group of women. These authors indicate that oral contraceptiv stroke in healthy young women" of their data, also given In Ap indeed a positive association b

D. Belsey et al. (1979) use vital statistics from 21 countries on mortality from cardiovascular disease (CVD) for 1962-74, taking the pre-pill period as controls and post-pill period as OC users. (This is a re-examination of some data used by Beral for a 1976 publication in Lancet, but with an additional two years included.) These authors "find the conclusion of in- creasing mortality from cardiovascular disease associated with increased levels of pill use unsupported by the data" (p. 85). Their analyses fail to show a significant correlation between percent of women smoking and changes in CVD mortality. The authors do point out, however, the inadequacy of using vital statistics as opposed to a case-control study data base. E. Jick et al. (1978c) report on a case-control study of nonfatal stroke in premenopausal women. They conclude "In our study, as in the report by the CGSS [Collaborative Group for the Study of Stroke in Young Women], cigarette smoking was only weakly associated with stroke in healthy young women" (p. 59). My independent analysis of their data, given in Appendix B, shows that cigarette smoking is not associated with stroke in this group of women. These authors also state that their •"results indicate that oral contraceptives markedly increase the risk of stroke in healthy young women" (p. 59). My independent analysis of their data, also given in Appendix B, shows that there is indeed a positive association between OC use and incidence of

434 stroke at the .001 level of significance. It should be pointed out that the sample sizes for this study are quite small, a total of 56 control and 14 case subjects. F. Petitti et al. (1978b) use the Walnut Creek Data to study OC use, smoking, and other risk factors for venous thromboembolism and conclude "that OCa and smoking have independent effects in increasing the risk of the idiopathic form of the disease" (p. 484). Their conclusions are based on 17 cases without predisposing conditions grouped as follows: - Table: Number of Smokers and OC Users Among the 17 Cases (from Table 4, p. 483). Non-OC Users OC Users Totals Nonsmokers 2 4 6 Smokers 6 5 11 Totals 8 9 17 X . 0.7 .30<P<.50 Z • 1 Statistical Conclusion: Among women who have the disease and no predisposing conditions, the factors of cigarette use and OC use show no association. The authors also give relative risk estimates for cases versus controls, and the estimates for smokers and OC users are smaller than the estimates for women with only one of these factors. . 435 Table: Relative Risk Esti: Controls (from Tab Rc Among Smokers, OC users Among Nonsmokers, OC users Among OC Users, Smokers Among Non-OC Users, Smokers Because the sample sizes for cas confidence intervals are so wide able reliability. Studies which do purport to find a definit, attack and/or stroke for women who smoke aT A. Studies criticized in my previou; (i) Jain (1977) uses the di consisting of 63 women an MI and a control grc were three nonsmokers a OC at the onset of the too unbalanced and too conclusions about the i and OC use on MI. Jain these mortality data ar may be subject to signi-