Tobacco Documents Online

HI(C 0119014 v--' An Analysis of "Some Recent Findings Concerning Cigarette Smoking" E. C. Hammand, L. Carfinkel, H. Seidman, E. L. Lew by Marvin A. Kasteqbaun 1. In a press release dated September 14, 1976, the American Cancer Society announced "the first definitive evidence that smoking low tar and nicotine cigarettes results In lower" death rates. The "evidence" itself was to be presented that same day by B. Cuyler Hammond at a scientific meeting that was In progress at the Cold Spring Harbor Laboratory on Long Island, New York. That the reported results represent "definitive" evidence, or even acceptable evidence In the scientific sense is certainly debatable. The manuscript not only lacks citation or documentation, it also fails to pre- sent any raw numbers with which an independent investigator may verify or disclaim the stated assertions and conclusions. Indeed all the "important" nembgrs that appear in the tables have been masticated and predigested beyond the point of even partial recovery or salvage. Simple tabulations such as the "number of subjecte" and the "nember of deaths" hrve been "adjasted" by procedures which, though the authors suggest are commonly practiced by laboratory experimenters, manage, at the same time, to distort and to obfuscate the original observatious. One can only hope that when and if this "definitive evidence" to submitted for publication in a more scientific mediam, it vill be subjected to serious peer review by referees who are competent in the techniques of data analysis and statistical inference. 2. Slide BI - Total NeIDber+of Subjects at Start of Period. Marginal totals in this table indicate that, at the beginning of the period 1960-1966, there were 133.422 males and 136,590 females who smoked "cigarettes only" (See page 6). The.cor:esponding numbers presented by Hammond (1) in an earlier docuneat are 198,797 males and 181.302 females. Thus we bave an anexplained deficit of 65,375 males,and 44,712 females from the earlier report of these data. It ia possible, of course, that one-third of the males and one-fourth of the female "cigarette only" smokers failed to provide information about the brands of cigarettes they smohed. Nowever, no such explanation is provided in the current manuscript. For the period 1966-1972 there are totals of 76,079 males and 88,905 females in the "cigarette only" category. We are told (page 6) that the _ people represented in this period are those "who answered the 1963-1966 queetionnaira and ano..:eare-smohiag cigarettes daily." We are also told i (1) E. Cuyler Hammond (1966), "Smoking In Relation to Death Rates of One Million Men and Women", Boidemioloaical Approaches to the Study of Cancer and Other Chronic Diseases. National Cancer institute Monograph 19. Appendix tables 1 and 2b.

HKi4113t115 -2- (page 6) that 897,825 man and women were alive through September 30, 1965, "and of those still alive that time, 94.9x answered a relatively short repeat questionnaire." This would indicate C'xat over 852,000 people answered the 1965-1966 questionnaire. If so, why are there such dramatic drops in the nunbere of "cigarette only" smokers between the 1960-1966 and the 1966-1972 periods? Loss of Subjects between Perioda I and 11 i I ;8iah" T/N "Medimu" T/N "Lov" T/N Total 4 lost E lost S loet x lost 0 lost x loet f lost R lost Nale 33.906 53.8 14,909 27.1 8,528 55.5 57,343 43.0 Female 21,228 48.1 10,557 17.7 15,900 48.6 . 47,685 34.9 Could it be that 43% of males and 352 of females quit smoking cigarettes? Could it be that the largest percentage of such quitters mare males and females who smoked "high" and "low" TIN cigarettes? Why.would males and females who smoked "medium" T/N cigarettes tend not to quit smoking in the eame percentages as other "cigarette-only" smokers? f I These possible explanations make little sense in light of what the autaors tell us on page 7. "High" for 1966-1972 is defined as "high" In 1959-60 and "high" or "medium" in 1965-66. "Low" is defined in a similar. manner. One would tdisk that these definitions would tend to move more subject's into the "high" and "low" T/N classes for 1966-1972 at the expense of the "medium" T/N clape. Instead, the reverse appears to have taken place. In fact, there ie no equivalent definition for the "medivm" TIN smoker in the period 1966-1972. One acceptable explanation for some of the reduction In "cigarette- only" smokers is the impact that the 1964 Surgeon-General's report might heMe had nven In 1965-1966 when the ACS questionnaire was circulated. However, it does not seem likely that this event coold sccoust for a 432 drop in male smokere and a 35Z drop In female smokers. Another factor that would accoeat for some reduction In these numbers In the number of subjects who died during the first eix-year period. Earlier data (2) indicates that among "cigarette-only" smokers tn the original study, 11,636 males and 3,651 females had died by September 1963. Simple extrapolation of these nuoberu to a sia-year period would inflate them to about 17,200 and 5,200 respectively. Deaths might thus account for 302 of the lose in males, and 11x of the loae in females. (2) E. C. Hammond, 0.s cit. I

HKi0113016 -3- I 3. Slide B2 - Adjusted Nuober of Subjects. 8ere we have the first set of adjusted data. It should be made per- fectly clear at the outset that the "adiuetment" used ia generating these data may have little or nothing to do with the "adiustments" applied to generate data for the subsequent tables. The clues to this "adiustment" appear on page 7 of the manuscript, where the authors introduce the concept of "matched group analysis." The specific cine is in the first sentence of the last paragraph of this page where we are told, 'A group, so defined, was discarded if it did not contain at leaut one "lov" T/N, one "madium" TIN, and (our emphasis) one "high" T/® smoker.' To appreciate how this "adjustment" works, examine the numbers in Slide Bl, and note that the lowest numbers are in the group designated "lov" TIN. Assign each of these smokers to one of the many aubgroupe (unJefined incidentally) "in respect to age, race, number of cigarettes per day,..., occupational exposure to dust,..., education,..." Then assign esch of the "medium" and "high" TIN smokers to his respective subgroup, in a similar manner. By definition, (page 7, first sentence in the last paragraph), each subgroup must contain at least one smoker from each of the three T/N categories. Presem?ably, all subgroups failing to mest this requirement are omitted from the analysis--all subgroups and all smokers falling into such subgroups. Except for the "lov" TIN group, there is no way of gauging the number of subjects eliminated from the stvdy by this "adj ustment." For the • "low" Y/N group, comparison of slides 81 and B2 indicates a loss of 672 malee and 2527 females in the period 1960-1966, and a loss of 357 males and 1461 females in the 1966-1972 period. Qe are not told, nor can me ever determine from this manuscript,'hov maay "medium" and "high" T1N subjects vere eliminated at this stage of analysis. The amount of information omitted at this stage of the manu- script ie not insignificant. For example: a. Bov many age groupings were used, and vhat were they? Hon many groupings were used for "nanber of cigarettes emoked per day?" Hov many age categories for when they "began to smoke cigarettes?" Bov many residence, occupation, and education categories vere used? Hoa many other factors, not mentioned in the manuscript; I i vere used to classify the subjects? b. For malna, at least eight classificAtien factors (aga, race, etc.) are mentioned. If each factor had only tw levels (i.e. race: white, nonwhite; residence: rural, urban), then the number of subgroups for classifying males vould be 28 ot 256. There is teaeon to suspect that this number should be quadrupled to over 1000 subgroups, most of which would be discarded by the "occupancy" constraints described on page 7.

fi KC-u11ou1rj c. Which subgroups were discarded? Bow do the characteristics defined by these subgroups relate to (confound with) the smoking charac- teristics of the subjects who fall within them? Is a 65 year old urban male with a grade school education, working in a coal mine, who smokes 9"high" T/N cigarettes per day, likely to have been included in this study? d. How many "dead" subjects have been discarded from this stmiy as a resul; of this classification procedure? In what subgroups did they happen to fall? i I 4. Slide B3 - Adjusted Nueber of Deaths. All of the numbers in the body of this table have been subjected to ~ s second "adjustment" yhich is described in the first paragraph on page 8. This "adjastment" is designed eo simplify the arithmetic calculations of mortality ratios. It ie tantamount co finding a leaRt common denominator for the calculation of death rates (i.e. a denominator different from the usual "per 100,000"). In this case, it'e "per number of subjects in the group with the least nmober of subjects," namely the "low" T/N smokers. The most enigmatic neIDbere in this table are the adjusted ou®ber of deaths for the "lov" T/N groap. In theory, based on all the information provided in the manuscript, these nnabere should be integers. In,theory, no "adjustment" aho uld have been applied to the nmabers•in this cqlomn. The reasonable thing to have expected is that among the male and i•emale • subjects designated in slide B2, there were exact and integral (not "adiusted") avIDbers of deaths. One attempt was made to reconstruct the exact nsmber of deaths for the "low"' TIN group, as follow: Asaume that 1351.7 "adjusted" deaths occurred among 14,688 "adjusted" male eubje.ts in the period 1960-1966. Calculats the number of deaths that might have occurred among 15,360 such male subjects. The answer is 1351.7 a 15360 - 1413.5 14,688 Similar calculations for "low" T/N malee and females in both periods yield 801.3 for 759.4, 1142.2 for 1053.9, und 904.9 for 826.2. The obvious failure of this attempt at salvage sests in the fact that none of these reconstructed numbers are integers. What is more important however, is thas: this approach to reconstructing the actual nuober of deaths is erroneous in concept. If only 14,688 of the 15,360 "low" T/N smoking males in the 1960-1966 period are admissible to the stedy, by the rules laid down on page 7, then only those dead malen among these 14,688 subjects shonld be admissible to the study. And this atatber has to be an integer. i

HIK®0119O1b -5- The intepretation of all the numbers on Slide B3, as well as those on Slides B4-B8, depends, at the very least, on a resolution of this enigma. Yndeed, the same questions may be raised about the num- bera in Slides B9-B11. 5. Nith very little imagination, the introdoction, the motivation and' the general tone with which this paper was presented (tape recording) could be adapted, by a talented playwright, into a script depicting the opening remarks by the prosecution at the triall of Galileo in the early seventeenth century. The words and the thoughts are designed to rekindle the passions of partisan believers, referred to by Copernicus (3) as "drones among the bees," against any who would dare to raise questions about established dogma and beliefs. In such a climate, there is no need for citation or docwsentation. Indeed every statement may be made with maaimum authority, and each new complea concept may be introdoced with patroaiaing simplicity. Following are some of the many examples: Page I: "'_he epidemiological findings have been so similar in various atudies...--that data from any of them can be used to illastrate the major results." If we ignore some of the evidence, this statement may be true. "First, and most important, deashirates are higher in smokers than in non+smokera and increase with the degree of exposure to tobacco smoke." This "fact" has not generally been established. "...other diseases including peptic ulcers. The latter was most elegantly investigated in a clinical trial carried out by Sir Richard Doll and his associates." .pave_ehanp +resnlts been published? "This slide, based upon data fror one of the prospective studies,..." Which one? Page 2: "This appears due to the fact that the majority of pipe and cigar smokers do not inhale the smoke or..." This statement is merely a conjecture. "Among those few pipe and cigar smohers who con- sciously inhale the emoke, limg cancer death rates (as well as...) are as high as the death rates of cigarette smokers." Has this "fact" been prove44 "These two variables and the degree of inhalation are all interrelated." liow4 (3) Nicolas Copernicus (1543), Da Bevolutionibus Orbivm Ceelestium. Preface to His Holiness Paul ill, Pontifes Maximas.

-6- yK 0 0 1190l8 "This is because there were so few such mEa... that their death rates could not be reliably esti- mated." Are we being told that very few men who begin smoking at age 15, smoke 112 pack of cigarettes or less per day? Page 3: "This may have been due to•selective mortality." fihat does this mean4 Each of these statements ignoree contradictory evidence, or faila to provide a proper citatioo, or fails to give sufficient documentatl.on for an experiment that has not been reported in the published literat:ae, or fails to define a complex concept. 6. Slide A-4 - Age-Standardized Death Rates. Lung Cancer. i i Two points are to be noted in the text concerning this elide: a. ...some of the points on this chart, are very anscable statistically due to small numbera." This apparently did not inhibit the authors from displaying and distributing it. i b. "We are undertaking a further analysis taking both these exposure variables into consideration." The two exposure variables are the age at which a person begins to smoke and the number of cigarettes smofted per day,. Presumabxy this further analysis will provide points that are more stable statistically, in spite of the fact that the auo- bere of observations may not be increased. i 7. "Tar and Nicotine". Page 5: "(T6is speculation turned out to be incorrect when it was found that emokers who switch from high !1N to low T/N cigarettes do not usually increase the nuobera smoked per dayl)" The verification of this finding may be very i®portant. It does not appear to conform with the observations in recent years that per-cepita cooaumption of cigarettes in the U.S. is increasing, while the total number of smokers is decreasing. I Page 61 "...of those still alive at that time, 94.92 answered a relatively short repeat questionnaire." Ia the second paragraph on this page we find that 897,825 men and women were alive on September 30. 1965. Of these 852,036 (94.92) answered the abort questionnaire. ! s I

HH1o119'Dtio -7- Only 23 of the original 25 states are represented at this point. It is difficult to say how many subjects bad been "traced" through September 1971 and September 1972. Are these numbers (0.984)(897,825) - 883,460, and (0.928)(897,825) - 833,182, or (0.984)(852,036) - 838,403, andd (0.928)(852,036) - 790,689 7 "gigh" T/N is defined as 2.0-2.7 mg. nicotine and 25.8- 35.7 mg. "tar." "Low" TIN cigarettes are those that "deliver" umder 1.2 mg. nicotine and under 17.6 mg. "tar." "MediuID" TIN cigarettes would, by these defi- nitions, "deliver" 1.2-1.9 mg. nicocine and 17.6-25.7 mg. "tar." These definitions appear to ba rather arbitrary. *:o good reaaoa is given in the manuscript for the choice of these levels. One might suspect that they were established after the fact, i.e. after the numbers of subjects in the various response categories were observed. i 1 Page 7: "Natched groap.analysis was utilized." Is this a nae type of data analysis introduced by Hemmond for this atudy4 As with his previous studies, serious questions may be,raised. about the matching procedures used by Hammond for aqalyzing these ACS data. Page Bs "The logic of this procedure..." The statement is not necesaarily true, nor is the procedure described in this paragraph good laboratory practice. Hammond is trying to justify the fact that he has discarded il~orms.tion, by suggesting that every good "eaperimen- talist" does the setas thing. One would hope that spne good experimentalists will recognise this affront. 8. Concloaions - Page 10 and Transcript. Ue authors of this manuscript make the following aeserr.ionss ~ a. ~ ~ ~ b. , ~ tbe reductions In "tar" and nicotine In cigarette smoke did not make cigarettes "safe" for the men aad women In their study; "the serious riske inearred by smob.ing" are somewhat reduced for smokers of "low" T/N cigarettes, provided that the nmber of cigarettes smoked per day is not increased; I

-e- NK10119021 c. "the threat to the future health" of youmg people would be reduced if "high" ?1N cigarettes were removed fram the market; d, the aoe of "additives for flavor" may pose a serious health problem for smokers. The transcript suggests that these "additivee" are polycyclic aromatic hydrocarbons; a. both "additives" and cigarette smoke condensate should be tested for carcinogenicity before cigarettes are marketed. 9. Additional ObservatZons. a. The epidemiological technique of using "person-years" in the analysis'of death rates and mortality ratios has been abandoned in favor of "number of subjects." This suggests that the'current manuscript may have been assembled hurriedly in an effort, on the part of the ACS, to saxinise the PR impact. In this, the ACS has certainly achieved its goal. b. The saaneWered questions about the ACS etudy af 1,000,000 Americans that have been raised by such statisticians as Broenlee, MacDonald, Barkson and Sterling over the years are still unaneoered. In fact, these questions are serely brvshed aeide andiignored. In this matter one can only conclude that compntence has been supplanted by credibility.