Tobacco Documents Online

9 • The Dore Study of SmoWns and Mortality Among U.S. Veterans: Report on ptht and Ora-Half Years of Obserration HASotn A. SAaN! Bimaebiu Eesearoh Branch, Naticnal Heart bulitndc, Bceheadc, Maryland INITIATED by Dorn in 1954, the study of a group of U.S. vetarans was one of the fint three large prospective inves- tigationa undertaken to describe in detail the relationships between tobacco use and mortality experience. Two features made this study of special interest. One was the precise definition of the population, which permitted identification and follow-up of both respondents and non- 'respondents, and the other was its size. The Doll and Hill (6) study of British physicians, began in 1951, was based on a de6ned population of 59,600 men and women. The Hammond and Horn (1d) study of 188,000 men, begun in 1952, was large in scale, but lacimd Ibedement of population definition. The Dorn study was concerned with.a defiued population of over 293,000 holders of life insuraace policies. The study population and plan have been set forth in d.taII in two publications (7, 8). Some major features are summari+ed below. With the cooperation of the Veterans Administmqtion, policyholders of U.S. Government Life Insurance were selected for study. This insurance was available to persona who served in the Ermed forces of the United States from 1917 to 1940. Most of the policyholders were veterans of World War I; the remainder entered military service at a later date. The cohort comprised 293,658 persons who held active U.S. Government Life Insurance policies in December 1953. Beginning in January 1954, questionnaires on smolong habits (appec .d~a ~E) were mailed to these policyholders, with 198,834 (68%) respondini. Beginaiag in January 1957, a second questionnaire, essentially iden6caT to the first except for typography, was mailed to those not responding in 1954, which elicited ~ Nwew rmarew erH..nn. raeue 8wsh s.r,id, Us. Ds.em.eea a~.rn sdoo.eten..ee w.u.,.. ~TaY 1Ndr ~r m.d~ p~~y tlteu:h tb mepeRt:m ~Ld tllbdDeo o! t!~ vMAS ~dm~•"^`w"` ~1tr. r.m~r w. omlo ond D4. hec Lbtl at Wo CompuWles ane DW ~ 1eym4. N.tlena m.tl• Wrefgritp,~ +NPomtDYb'°'^"l"i^'N.OatlnD~urdnead,brm, nf7~ Ye.7anA.HaF ..v, xr. a,n n. Tmu..m )o. r,d zu.w..e at w. x.Mmu tmtlma et S.Irn cmroowm .od DW haaetlat BtmcC.in eoopuftle¢ w1Lh )Sn. Awne?.1YOtt ott0o N.ew.i8rtt Imtlmv Dlom.t4n 3meE. t=Lpd tW rr¢ehooY epopu ta bDehtlom la:e 1~~. pR eom..Yetly Soe.dma. ud tb. edl R'odoet. Hn. Dotty L. CzGr md Md.laphta S.I[nwC manood tho osmpki etrtst poadoxem npaYtlfntEoaa6ramo.t.cdYn. Dhq r.ohnl prortdodLeti SaMMandem1lodq lieedlegotmme.Lt7 rb. 1 0

w 2 49,361 additional replies, raising the response rate to 85 percent. AIl smolring classificatdons used in this report are taken directly from infor- mation supplied in these questionnaires. CHASACTEBISTICS OF THE POLICYHOLDERS Almost all policyholders were white malee. Less than ha8 of i percent were females, and only a negligible numberwere nonwhite males. Eighty- two percent wera white-collar or stilled workers, 7 percent were semi- skilled or umldlled workers, and 6 percent were farmers or farm laborers. Roughly comparable perceatages for U.S. white males aged 20 to 54 years as of the 1950 Census were 50, 35, and.13, respectively (19). Clearly, semieloDed and unelclled workers were underrepreeented in the study population; such selection is found in the e=perience of all insurance companies writing whole-life or endowment policies. ,Since the policyholders were drawn mainly from the middle and npper socioeconomic cLaeee, it could be anticipated that their death rate from all causee.during the period of observation would be lees than that for . the general white male population (SL, 2J). Mortality in the_ptudy population age 55 to 84 (respondents and nonmpondente) expreised as a proportion of the mortality for U.S. white males of like age compbsition was 0.75 for the interval July 1954 to June 1957, 0.73 (July 1957-Decem- ber 1960), and 0.67 (196142). An estimate of the number of death reports for 1981-82 not received until after the cutoff date for the present tabulations suggests that the true figure for the 1961-62 period is closer to 0.71. Pending the receipt of more complete data on deaths it seems reasonable to estimate the mortality of the population of insured veterans during the initia18J4 years of observation as about 0.73 of the correspond- iag U.S. rate. Judgments on the presence of a real and sustained im- provement in mortality over time among policyholders relative to the US. white male population should be deferred until sufficient time has elapsed to ensure that all delayed death notifications for the more recent years have been received. . FOLLOVO-UP Additional follow-up procedures over and above the normal Veterans Adminiatration routine were required which were carried out by the staff at the National Institutes of Health. Whenever a claim is filed for the 'psyment of a policy, a copy of the death certificate is routinely sent by the Veterans Adminietration to the NIB study office. Annually the Ot$ce of the Chief Actuary in the Veterans Administration also provides a deelc of punch cards for each policy terminated during the year, indicat- ing whether termination was due to death or other reasons. The cards for policies terminated by death serve as a checlc on the completeness NATIONAL CLNCIIt rNaT1TOTZ IlONOn86PH NO. 19

Il f- ~er for dy as .on m- &th an6 ' eer _ M" ans nd- im- the has -ent T2M DOEN STUDY OF eYOlrNO ItPD MORTALITY .8 of reporting of deaths. Records without death certificates are traced through appropriate VA offices and the missing data ultimately obtained. -Veterans may, of course, have more than one U.S. Oovezament Life Insurance policy and, when a notice of policy termination for reasons other than death is received, a check is made to determine whether the individual has any remaining active policies. If not, the record is traus- ferred to a specialfAe for "terminations" and periodically the Veterans Sdministration Index Section is asked to report whether these indi- viduals are still living. If not, a letter, requeeting a copy of the death certificate, is then written to the VA office having custody of the records. Termination of insurance means an automatic notification procedure for reporting death has been lost, but the VA has so many other pointa of contact with beneficiaries (including payments to defray funeral ex- peaeee) that mortality follow-up on "terminations" is considered to be quite sati.sfactory for inclusion in this study. Although very good, it would be unrealistic to assume that mortality reporting on "terminations" is as complete as for active policyholders. Therefore, it is of interest to investigate whether the termination rates are the same for smokers and nonsmokers and this point will be considered in a later eection. All mortality follow-up procedures apply identically to re.pondents and nonreepondente, so that the study findings can be related to the insurance policyholders as defined and need not be restricted to the subgroup who answered the smoking questionnaire. When a death certificate ie received, additional medical information including verification of statements on causes of death is requested from the certifying physician or the hospital where death occurred. The data reported here reflect the composite information available from the query and the original death certification. In about 6 percent of the deaths the query led to a change in assignment of the underlying cause, and in another 12 petoent information on contributory causes not mentioned in the original certification was added, though the underlying cause re- mained unchanged. The underlying reason and as many as two addi- tional contributory causes were routinely coded for all deaths. METHOD OF ANALYSIS -ans 3taff the t by the ides icat- srds mes+ The available data were summarized into the number of deaths, 4, and the number of pereon-years of observation, y„ at each single year of age from 35 to 84. Person-years were accumulated by attained age and • woi by a fixed age determined for each individual as of the start of the study. The ratio d,(y, provides an estimate of the average annual force of mortality at age z, µ, _+~ µ,dt. Person-yean are terminated in the . middle of the month of death and the values of d,Jy, could exceed unity at the oldest ages. The actuarial formula relating force of mortality and probabHity of death within the period z to z+ 1 can be written as STIIDT OT CLDiCEB AND OTHES CERONIC DmEAa13 .

i • 4 uMr . q. ~ 1-e1p-~+, µdt (ISJ , and the estimated annual probability of death at age x, q„ computed from t~he equation The q, values for individual yeus of age for the different emoking classes were calculated in this manner and then averaged into 10-year age groups 35 to 44, 45 to 54, 55 to 64, 65 to 74, and 75 to 84, with the 1960 divtribu- tion of the U.S. male population by single years, within the corresponding 10-year age group, used as weights. The study population is such that ages 55 to 84 represent about 98 percent of the deaths and about 85 per- cent of the person yeare. Therefore, several of the tables dealing with detailed smoking categories omit age-specific data for ages under 55. In addition to the age-specific probabilities, a mortality ratio of the number of observed to expected deaths was computed for each smoking category. The expected number of deaths for each smoking category was computed as the product of the person-years of observation at age z in that category by the force of mortality at age z for those who never smoked, summed over ages 35 to 84 inclusive. This is equivalent to stating bow - many would have died in a subgroup of smokers if the force of mortbi'ity observed among nonsmokers prevailed. The force of mortality, p,;:pen ' used rather than the probability of death, q„ to ensure that the e:pected number of deaths for nonsmokers would exactly equal the observed number. Throughout this report the terms "nonsmoker" or "never smoked" are to be understood asincluding persons who have never been regular smoke,s but who may have smoked occasionally. The mortality ratio is a relative indirect age-adjusted rate and as such is a function of the age structure of the smoking category being adjusted. For this reason it is technically incorrect, though the practical effect is often negligible, to compare directly the mortality ratio for smoking category A with that for another smoking category B, since the differing age structures of A and B are. not controlled in this comparison. Of course, all mortality ratios are comparable to the base experience for non- mnoketa which is defined to be 1.00. The mortality ratio is a convenient summary index, but wherever its use leads to a different inference than that derived from a direct ratio of age-specific rates, the latter is always to be preferred, subject only to the limitations of larger sampling errors associated with smaller subsets of the data. Because of the large number of deaths (6,932) and person-years of observation (443,856) available to estimate the force of mortality for nonsmokers, all references to sampling variability of mortality ratios in- corporate the simplifying assumption that the expected number of deaths was determined with so little error that sampling variability of the latter can be ignored. Given this assumption, the variability of any mortality ratio depends solely on the random sampling arror in the observed NATIONAL CINClB IH6ZTrOTD YONOOBLPS NO. 2e

!8) >m sea lpe 2u- ing iat -er- ith the ing ras :in ed, ,ow uty xa6 ted ~ ved Ter een uch ted. tis eing ring Of 1On- ieat hnn :ays rora a of for iin- athe ttter slity rved __ras aoa.N aTVns o) e11OZIIrc AND HOST.tL31T number of deaths. Unless otherarise apecified, all taeta of aignificance are at tSe 1 percent level. PRESENT REPORT The preaent report indudes all deaths lmown to us, ages 35 to 84 inclu- aive, occurriag from July 1954 through December 1962. Table 1 dis- tributes the 46,270 deaths and the 2,265,674 pe:son years observed during this period by age and respondent status. The second questionnaire to nonreapondente in 1954 was malled in January 1957 and for the next several months, while replies were being received, the probability of re- sponee was strongly associated with current health. In order not to exaggerate any differences in mortality between respondents and non- respondents, July 1957 was chosen as the date for transferring individuals from nonrespondent to 1957 respondent status. Thus, someone who answered the second questionnaire in March 1957 and died in May 1957 was treated entirely as a nonrespondent. A person who replied in March 1957 and died in August 1957 was counted as a nonrespondentfrom July ,_. 1954 through June 1957 and as a 1957 respondent thereafter, and his death =: . •counted in the 1957 respondent category. Tuca 1.=Dietribution of deaths and peno7~ observation by attained a'e' . and response etatue, ]d 19 . Respondents Total I 12" 1957 Deaths Non.epogdent. Number Percent of grand total . 35-84 46, 270 35,691 29,731 5,960 10,579 219 35-44. 559 389 302 87 170 30.4 45-84 ' 532 374 322 52 158 29.7 55-84 19,523 14,414 12 528 1, 686 5,109 2& 2 65-74 23.107 18,454 14,877 2,577 4,653 20.1 75-84 ' 2, 549 2, 960 1,702 358 489 19.2 Percent of death. witb in- oq b~~oa on m n ~ b e o Ic 8 ba i 4. 6 4.9 2.6 . Petwn-ycara 85-84 2,265, 674 1,801,119 1,547.905 253,313 44555 20.5 : aa-44 251,122 193,725 159,661 34.064 57,397 22.9 45-54 ' 37,985 70,787 60,099 10, 688 17,195 19.6 55-64 1,124, 385 888, 546 776,187 92, 359 255,839 22,8 65-74 765,033 637,082 625, 878 111,405 128,951 16.8 75-84 ~ 3& 249 30,979 26,182 4,707 5,270 14.5 Percent of penon-yeaa with Inadequate in(orma- tlon on .mohing babite. 4.6 4 9 2.7 tT4D: or CaNCLa 1ND OTHiCS cBaoNIC aIS7JSEe

M • 9 HNnN. OY,-T.W en~l,ul f-VM e-p~M 4-b~M w. „,..1 ~ ONr 0.h-TMa tr ~r. fiM Iw .w//q IIM O.h- T.M . YM e. 3/1q MAJOR RESULTS FROM INITIAL PERIOD (2K YEARS) _ COMPARED WITH EXTENDED PERIOD (83i YEARS) Dorn reported the findings from the initial 2!f years of observation (7, 8). Data from this study have also been extensively cited ia Smoking and Heafth (H). Te=t-figure 1 presents an ova-all comparison of mortality ratios for all causes of death by amokiag category for the two time periods. For each elaae of smokers (escept for one pipe-smoking category) the mor- tality ratios based on the total observation period were larger than those derived on first review and andysis. We will return to this point in some detail in a later section dealing with possible extent of selection biee, but some comments aeem appropriate here. There are two major differences between the study periods. Covarage of penone in the fuat 2y, yeais wee limited to those who answered the 1954 questionnaire. The 8)6-year period includes data for both 1954 and 1957 respondents. In the earlier reports, observation was terminated on persons who no longer held active insurance policies, but the data for the 8% years include both person-years and deaths eIperienced after termination dates. Despite these and other _ differences to be discussed under the heading Selection Bias, the corres- s i i A..~rr ~.ons ~ 1 I. .r.,M tl~al. Owr k h~ bw. w~ .M 7C {p l0-24 M>'1rWa t!-Y~OEIIS 0F CWIIRtn o~V ~ h1.1 1!M! M Na •.:d. enr 1 M . 1 i e a0 I.a. .r i:ar _, ( N 0 tb iiIIIILlIAI { 0 i:.i <.d : i N ~ O ~tY _~ i _ ~ U a1KITf U tA uTk t1 --- _ Tsza~aoaa. 1.-ComperLon of mortality ratias for aII eaew of dntb by smokfn8 eet.8otr-2)S- and 8ylrY+ar followap. NATlONA7. CA1(CZE rP8T1TOTf llONOOB'J!8 NO. 10

i i • aon KnD for For aor- iose ome but aces wa. Yeu rlier rtive ,es2e ther ROe- nOYing 40. ls -Tas aosN aTCa: 01 U2oswa .aID 1[osT.rsTr 7 pondence in mortality ratios based on the preliminary and extended obser- ntion perioda is striking and impressive. Thus, both sets of results show a strong positive gradient with number of cigarettes smoked, a negative gradient with age began smoking, lower mortality ratios for ez-cigarette smokers than for current smokers, and much lower mortality ratios for cigar and pipe amokers than for cigarette smokers. Tett-figure 2 contrasts the two periods with respect to mortality ratios for selected causes of death. They were chosen for iilustration from the disease with the strongest association with smoking history (lung oancer), - a disease with no risk gradient by amount smoked (cancer of inte®tiness and rectum), the disease with •the largest number of "excess" deaths associated with smoking history (coronery heart disease), and the disease with the smallest mortality ratio (paralysis agitans). The estimated ratios from the 8% yean tend to run somewhat higher, but the structure of relationships is again very consistent. All the original findings have been confirmed by more extended observation. Test-figuie 3 compares annual death ratee for the 2)4-year period with the annual probabilities of death calculated for the 8//ryear period. At the observed level of mortality experience, the two techniques for meaa- CWSF Of OFiTN !NO fYOYMT fYOR[0 I 2% ri.r I Q 1 i ' i ~ f /ar 0 , (Ki-li)) i <NI/Iq i 7 Lf0 r 1.1/ i r q-20/Iq iiiia ~~ i 1 i i Owr 20/a ~ na , L w I I w.fr J s m 20 I 1 2 3 eOeTLLITi ILLTW 0 Tsza•navsa 2-Comparison of mortility raNw for seketed ames of deatL (includes md.rlrtna or contributory au2e) among mrf.nt smskms of olprettem enlT-2yr and fyr7ear follow-up. a1Rn7 oF CANCLn AnD oT8£fi C88o2PIO an1L6i9 1a2-e71_

. 8 s.cnpi ai[w.r Lln wc illaM/ M ONNiN,I 0e, eS-i4 t3-71 amn wlu. r cq..m.. ~ !5-N tf-7i c.wr wiNn r riwN ary ff-N P-T4 wN/l wNN r rl/N m/, !f0-5. I ' f 1 I 1 1 o n m s w w urr o. Ipoo . e) 4NrM ViNA Nal k/Nr y, r/n N et Iitl,iMli.s r/1. YW 0.A ticb p/rYiY M 1244. Tsia•navas 3.-Comparison of annual death rate per 1,000 (2y4~year follos-up) with the annuai probability of death per 1,0D0 (8+year follow-up). All caeeeo of death by smoking category for agn bb-ed and 66-74. uring risk are essentially equivalent and any understatement in the prob- ability of death in relation to the death rate is unlikely to exceed 2 percent. The measures of mortality for cigarette smokers are almost identical for both periods, and no striking differenoev can be discerned for the cigar and pipe categories. The more extensive observations yield somewhat lower estimates of risk for nonsmokers, particularly at ages 65 to 74. Differences between time periods are small compared to differences among smoking classes. While the two sets of data lead to the same inferences, the longer observation period provides an opportunity for a more precise look at various relationships. The following sections present and discuss in more detail the'results for the 8K-year period. SMOffiNO CATEGORY Mortality ratioe by mloldng class are shown in table 2; supplementary datatl on number of deaths and ag"pecific probabilities of death are given in Appendix tables A and B. , Current smokers of cigarettes have mortality ratios directly related to the amount smoked. This statement holds true for smokers of cigarettes only and for all smokers who combine the uae of cigarettes with other ^ NATIONAL C"CL'R INSTITOTB YONOGRAPB NO. 19 LSN ~ 1 O e4yi.w NI , tlJ i NA ~ N.I

r-up) ea ot tary icen tlto ettes ither J. 19 T8i DOBN STUDY OF aMOSIHO AND MORTALITY 9 Tuts 2.-Mortality r.tioa* by smoking category, July 19S4-Dooember 1962 Et4mokes Smoking category Corrent smokers Btopped on dootorh otder pped 8tootha for raanons Ctµte~ 1 L 1 L 96 1.27 8 1-9/da L31 L 78 L08 f1o--89/d:y 2 0830 L 92 L 47 39+rdi 132 aai 58 y Clpr.i ~ tot,i " L 2.05 i3 s 0 6 1-9/day L45 L ls 10-201day L 97 28 21-891day am 2 6 is a 39+/day 2 32 7 L 60. y-tot.l t:!{at7 L80 ~ L04 L~ 9+1a+9 L 49 nlr totat plpe a 1 81 L q i~ d 93 1S 1-19 /day L10 L18 20.+lday L 20 Axet w.n eMm tln ~Dnree, eerr..a. forms of tobacco. Persons who smoked two packs of cigarettes per day or more had 2.3 times the mortality risk of nonsmokers. The gradient of risk with amount'smoked is slightly steeper now than first eetimated from 2}i years of observation. A gradient in mortality risk with amount smoked also appeared for cigar and pipe amokers. Both moderate (5-8 cigars or 5-19 pipes) and heavy(9 or more cigars or 20 or more pipes) smokers of cigars and pipes have mortality ratios significantly greater than 1.00. Those who cur- rently smoke only 4 cigars or pipes or less per day have mortality ratios ndt aignificantly different from nonsmokers. Thus, current users of cigarettes, cigars, or pipes experience excess mortality risks if they smoke more than an occasional cigarette or more than 4 cigars or pipes per day. In an effort to *ni++i+*+i>e a presumed artifact in the data for ez-smokers, mortality ratios were calculated separately for two groups of ei-emokers. Those who stopped on doctor's orders experience consistently higher ri.ks for all causes than those stopping for other reasons. By segregation of the former component (about 10% of all ez-emokets), analysis of the experience of ez-smokers can be partially 6roed from the distortion introduced because illness was the reason for stopping (14,16). Whereas it is difficntt to gauge the credence to be accorded a reported reason for an action, separation of these groups is a step in the right direction. About 5 percent of the records are coded "reason uDknown" and these have been induded with "other reasons." Unless otherwise specified, further STUDY ol cANCEa AidD oT8E8 cBHOYIC nISLSST.9 W

9 0 reference in this paper to ersmokers will be restricted to those who stopped for reasons other than doctor's orders. Among e:cigarette smokers mortality ratios in each subclaesification recede to an intarmediate position between the corresponding figure for current smokers and nonsmokers. In the process a gradient in risk by number of cigarettes formerly smoked is maintained. However, former cigar or pipe smokers have higher mortality ration than those who continue smoking. When Dorn fuat reported this rather curious finding (7) he atated that "many cigar and pipe smokers may have stopped smoking because of ill health, but it is not obvious why this should be true for cigar and pipe smokers but no6 for cigarette smokers." It is certainly not obvious why such an effect should persist among cigar and pipe smokers after excluding those who stopped smoking because of doctor's orders. In summary, for each form of tobacco use, mortality risk is directly rehted to amount smoked. The riaks for cigarette smokers greatly exceed those for cigar or pipe smokers and are lowar for those who have stopped than for those who continued smoking. A gradient of risk according to amount smoked is evident. AGE BEGAN AND NUMBEfi OF YE®RS SM08ED = Table 3 contains mortality ratios computed for current smokers of cigarettes cross-classified by amount smoked, age began, and number of yaas smoked: Discussion of these variables will be limited to cigarette smokers because of inaufficient data for analysis in the categories of pipe or cigar smokers subject to excess risk. Among those who began at age 20 or later, the relationship of years 'smoked to risk depends on the amount amoked. Those who smoke 1 to 9( cigarettes a day do not display a significantly higher risk than nonsmokers unti125 yesas or more have elapsed. Smokers of 10 to 20 cigarette a day eaperience a significantly greater risk after 15 yeats. Not shown in the table because it was based on only 48 deaths is a ratio of 1.66 for smokers of 21 to 39 cigarettes who started at age 20 or later and smoked for less than 15 year9. (Other calls omitted from table 3 are based on 16 deaths or ]ess.) This ratio of 1.66 is significantly greater than 1.00 and would suggest that persons who attain a rate of over a pack a day may be sub- ject to increased risk in less than 15 years. Among those who begsn moking before age 20, the results do not strongly indicate that mortality risks continue to rise with longer duration of exposure. Once a signif-- icantly higher risk is reached there is little evidence of further increases. One may reserve judgment as to whether this apparent plateau in risk correctly reflects the facts or whether the very great overlap among dura- tion categories has blurred beyond recognition an association with duration of exposure. The duration categories <15, 15 to 24, 25 to 34, and 35+ are those reported on the smoking questionnaire. Since the present report covela 99 years, the actual durations to which these labels now ' NITIONAL CLNCBS INaT1TOTt YONOO8LPH NO. 19