Tobacco Documents Online

Cancer Mortality Among a Representative Sample Nonsmokers in the United States During 1966-68' James E. Enstrom, Ph.D., 2 and Frank H, Godley, Ph.D. A~STRACT-.-Oata are presented on cancer and total mortality among a representative sample of nonsmokers and the total population 35--84 yee~ of age In the United States during lS66-68 that measured the influence of cigarette smoking on mortality r=,tu, inda~ndent of other health-related feelers. Of all U.S. white males, those who never smoked cigarettes have a total age-adjusted cancer death rate which is 37.% less than that of males a= a whole and 53.% less than that of tllose who currently smoke cigarettes. Corresgondingly. of all U.S. white females. thOSe who never smoked cigarettes have • total age-adjusted cancer death rate which is 15,% less than females as a whole and 33% less than that of those who currently smoke cigarette& The largest cancer rate reduction in the nonsmokers is concentrated in the respiratory system. Nonsmoker= have en ag~-adjusted total death rate which Is about 20% lest than the po~)ulat]on u a whole and about 43.% less than currant cigarette srnoker=. These and other result= and methodologi¢ Issues are discu=sed.--JNCI ~5: 1175-1183. 1980. Previous studies have established a strong, positive correlation betwmn cigarette smoking and mortality, with cancer mortality showing some o! the highest correlations. Beginning in I964. the Smoking and Health Report of the Surgeon General has summarized the evidence (1). Oniy one prtwious study in the United States has been based on representative national samples and that was limited to lung cancer deaths during 1958-59 (2, 3). The findings here axe based on a 1966-68 survey that measures among a representative U,S, sample the effect o[ smoking status on all causes of death inde- pendent of other health-related factors. The nonsmokers in this survey are meant to he used as a reference for comparisons with other nonsmoking populations who may he subject m additional health-related influences and selection biases. Our focus is on cancer mortality among nonsmokers and their risks ~'elative to those experienced by the total population. "Nonsmokers" here are defined as persons who smoked fewer than 5 packs o[ cigarettes (100 cigarettes) in their lifetimes. However. they include former and current cigar and'or pipe smokers. Unlike most other low-risk populations, nonsmokers comprise a substantial proportion o[ the adult popula- tion. They are not limited to a particular geographic region, ethnic group, or religion. Moreover, our use o[ national probability samples to examine the mortality experience o[ nonsmokers minimizes selection biases with respect to other environmental, socioeconomic. and life-style factors. MATERIALS AND METHODS Investigators who conducted most previous studies of dgarette smoking and mortality used a single-sample, prospective design. Mortality ratios were obtained from the relative proportions of the original sample who died within a specified follow-up period and then were weighted by survival times. However, serious disadvan- tages associated with representativeness and problems of follow-up are evident in the prospective studies (1). Nonresponse may have been as much as 32% overall and even greater among smokers. The bias in the mortality ratios of the respondents was estimated to be about a 0.2 or 0.3 overstatement because of lower mortality ratios of the nonrespondents. The low level of nonrespome in the present study (55) indicated that nonresponse was negligible as a source of bias. Our approach was to use a cross-sectlonal, two- sample design to estimate the relative risk o( non- smokers' mortality. This study was analogous to the one o! representative samples of 1958-59 lung cancer deaths among U.S. whites {2, 3). One sample was representative of U.S. deaths during 1966-68, and the other resembled the general U.S. population during the same period. The smoking characteristics of both samples were determined and then mortality rates were calculated as a funcdon o[ smoking status. Some limited results of the study, described in detail else- where,' have been published (4-6). Sources of data.---Data were from two sources. Esti- mates o[ observed deaths of nonsmokers 35-8,i years o[ age in the United States in 1966-68 were based on the NMS. a follow-back survey linked to a probability sample of 19.526 death registration records that in- A|S~VL~,TIO.~S VSEO: CP$=.Currem Population Survey(s}: ICD =In. ~erna¢ional Classification of Diseases: NCH$=Nadonai Career [or Health Statistics: NMS==Nadonal Mortality Survey: SM1~.=standard- ized mortality rauo(s~. = ~nted at the Workshop on Populations at Low Risk of Can- cer conducted at Snowbird. Utah. August 23-25. 1{)78, ~ S<hool of Public Health and Jonsson Comprehensive Cancer Center. University o| California. Los Angeles. Calif. 90024. ~ Division of Analysis, National Career for Health Statistics. ~700 East-West Hight~ay, Hvattsville. Md. 20782. • Godlev. FH. Ci~7~t~tte smoking, sc<ial factors. ~d mortality," New estimates from repre~.-=matise national sampi¢~. Unpublished Ph.D. dts~ertatton. L'mver~ttv el Mat?.-land. College Park. Md.. 1974. JNcL YOU ~5, NO. 5. NOVEMBER. Tl04891637

1176 Enstrom and GodMy eluded deaths of 11,318 white males and 5,636 femalesJ The underlying cause of each death was assigned in accordance with the Seventh Revision of the ICD Questionnaires were mailed to surviving family mem- bers and others named on the death certificate who provided smoking histories and social characteristics of the deceased. The second source of data is the CPS conducted by the U.S. Bureau of the Census in August 1967 (9). Smoking and socioeconomic iniormatlon comparable to that for the decedent was obtained by means oi household interviews for a probability sample of 60,920 adults 35--84 years of age, including 9_.5,266 white males and 29,308 white females.~ The technical details of each survey are given in the Appendix. Also included there are the age-sex distributions of the deaths and population-at-risk by smoking status and a sample calculation of the weighted number of deaths when the sample number of deaths is used. Inferences that can be drawn from the data in this study are conditioned by certain features of its design, sampling, and data-collection methods, as discussed in detail in the Appendix. In prospective studies, the decedents are a subset of the total study population: examples are the American Cancer Society and U.S. Veterans cohort studies in which a beginning sample is initially questioned regarding both current and past smoking habits (I0, II). The decedents of a prospec- tive study, who are in the original sample, match exactly with an individual in the study population. The cross-sectional design assumes statistical matching between the two independently drawn samples: de- cedents and population. This problem is common to conventional vital statistics rates and must be balanced with the expense, time, and follow-up problems of prospective studies. The value of a two-sample cross- sectional design was demonstrated in national studies of smoking and lung cancer mortality during 1958-59 (2, 3). Delinitions.--Detai]ed estimates of cigarette smoking history, by amount smoked and other relevant smoking variables were not made. Instead, persons were simply classified by lifetime history, o[ having "never smoked" or "ever smoked" cigarettes (at least 5 packs or I00 cigarettes). "Current" cigarette smokers are persons who have smoked at least 100 cigarettes in their life- times and have smoked cigarettes during the past 12 months. We coordinated smoking questions in the supplement to the CPS with those in the death follow.up survey in an effort to gain comparability. Unlike sizable proportions of the CPS sample who provided data on their smoking habits and other personal characteristics, such data were not awailable for the NMS death sample. Moreover, a mailed ques- tionnaire was used in the NMS, whereas a household interview was used by the CPS [or collection of data on : National Center [or Health Statistics. I.~o6-.~8 Nadonal Mortality Sur~'ey.. L'npubli~hed tethnical rmte~ at~d computer tape~. 19~6. the living population. Evidence from another sur~ey indicated that agreement between the smoking history reported for a decedent by. next o~ kin and by his own reports when alive was about 90%. Moreover, Haenszel et al. (2) noted that this type of bias is largest with respect to medium and heavy smokers but virtually nonexistent for nonsmokers. RESULTS Table 1 contains the average annual age-s!cxcific total death rates for 1966-68 for U.S. white males and females by smoking status: never smoked cigarettes, currently smoke cigarettes, and the total sample. The rotes are given in 10-year intervals for adults 35-84 ,vents old, with an overall rate age adjusted to the 1940 U.S. population. The 1940-U.S. population is the standard reference used by the NCHS (7). Included in table 1 are age-spedfic total death, rates ~or all U.S. whites based on the NMS and CPS and the average total death rates as published in the 1966--68 and 1970 Vital Statistics of the United States (8). The NMS-CPS rates generally agree with the 1966-68 U.S. rates to within ~%, except in the 75- to 84-year-old category., in which they are up to 10,% higher than the U.S. rates. (The CPS includes only the noninstitudonalized popu- lation as explained in the Appendix.) Agreement be- tween the sur~'ey rates and the U.S. vital statistics is good. Finally, it is evident that the nonsmokers in this survey have significantly lower rate~ than those of the total popuIation, which conforms with previous epi- demiologi¢ studies of nonsmokers (I, 10, I1). Non- smokers have an age-adjusted total death rate about .90% less than that of the population as a whole and about 4~% less than the rate for current cigarette smokers. In addition to the 1966-68 average annual age- specific cancer death rates for U.S. white males and females by smoking status, table 2 includes the rates for total cancer and for cancers of the lung, breast, and prostate, all of which are based on more than 100 sample deaths. The U.S. white males who never smoked cigarettes have a total age-adiusted cancer death rate 37.% le~s than that of U.S. white males as a whole and 53% less than that of white males who are currently cigarette smokers. With the rates for total cancer based on the N,NIS and CPS are those given in the 1966-68 and 1970 U.S. vital sradstics. The agree- ment between the survey and the vital statistics rates is within II% for those less than 65 years old. Over 65 years, the survey rates are 7--17% higher than those for vital statistics. These differences are greater than the corresponding differences for total mortality primarily because of statistical fluctuation in the sampling o~ cancer deaths in the NMS. The 1966.-68 weighted numbers o£ observed cancer deaths for U.S. whites who never smoked cigarettes for the sites listed ba,ed on at least 10 sample deaths in males and/or females are given in table 3. Also included are the expected numbers of cancer deaths jNcI. voL. 6~. xo. s. NOV~'.MBF~R T104891638

Cancer Among U.S. N,~nsmckers 1177 U.S. whims U.$. whims All U.S. whi~es who never wh~ ~ur- 19~6-68 i~65-~8 U.S. 1970 smok~ ~n~y smoke ~MS - ~1 35-44 216 442 352 342 344 45-54 570 1,153 93"/ 904 883 5,~64 1.686 2,878 2,352 2246 2~3 ~74 3.~ 6.572 4.9~ 4.9~ 4.810 7~ 9.5£8 I3.~ 10,749 9,918 " 10,~ To~: 1,425 2.~9 1,918 1,~2 1,827 Females 35-44 152 245 201 195 193 4,5-54 333 597 464 462 463 b5-64 771 L444 1,015 1,022 1,015 65-74 2.398 4.129 2.702 2.622 2.471 75-84 7,044 13,123 7,600 6.890 6,699 Total: 894 1,606 1.0,~3 1,017 989 • Death rates are based on the 19~6-68 NMS and August 1967 CPS. ' Ra~s are age zdjustod by the direct method to the 1940 U.S. population ~5-34 yr old, based on the death rotes among all the whites in the 1956-68 NMS sample. With this group as the reference population, the total cancer SMR for whites were 67 and 87% for male and female nonsmokers, n~- specdvely. Using 1970 U.S. whites as the reference population, as was done in a previous publication summarizing some of these NMS results (5), we de- termined, that the total cancer SMR were 70% [or white TABLE 2.--~7te 1968-@8 a~rage annual a~e-sp~c~ic cgncer death rates (deaths/lO0,O~O) among U.S. wkites by age. ~. and cigaret~ Age. yr 3,5-44 45-54 65-74 75-64 Total: 35-84~ Lung cancer Total cancer U.S. U.S. whites whites whites who U,S. whites U.S. whites ,, ,..,All U.S., whites who who All U.S. never who never who cur- 19~6-68 1966-68 U.S. 1970 U.S, never currently whiteF .smok~ smok~ smok~ ~oke ci~ss ei~re~s~ 2.3 (2) 17.5 (38) 12.9 (48) 0.7 (1) 32.0 (28) 56,8 (125) 492 (185) 50 50 3.5 (3) 81.1 (163) 59.3 (213) 2.1 (3) 80.7 (42) 211.2 (340) 172.4 (476) 170 172 33.4 (24) 270.4 (365) 177.8 (493) 15.2 (18) 306.0 (102) 695.7 (646) 527.5 (946) 479 498 63.8 (43) ,533.9 (298) 288,5 (499) 79.9 (89) 655.8 (204) L531.3 (~3) 1.079.7 (L041} 985 997 87.9 (36) 746.1 (122) 2672 (211) 244.4 (165) 1,360.8 (294) 2,224.3 (262) 1.548.0 (760) 1.472 1,593 19.9 (168) 179.8 (986) 103.3 (1.464) 24.0 (276) 232.8 (670) 497.4 (1,916) 368.6 (3.408) 343.9 3,54.9 Females 35-44 0.4 (1) 8.6 (16) 4.2 (19) £3.8 (46) 56.7 (110) 80.1 (134) 72.9 (288) 65 62 4,5-54 2.8 (6) 2-;.4 (43) 15.3 (65) 45.6 (46) 133.2 (141) 197.2 (179) 172.1 (377] 175 1,7 55-64 10.9 (24) 51.9 (46) £5.4 (8,5) 85.9 (45) 268.5 (166) 412.6 (141) 335.7 (360) 330 339 65--74 18.7 (37) 103.7 (31) 34.3 (82) 121.2 (28) 576.9 (182) 877.1 (66) 627.1 (274) 565 555 75-84 48.4 (55) 143.9 (9) 55.5 168) 143.0 (19) 921.8 (190) 1,303,9 (21) 999,2 (229) 868 904 Torah 7.5 (123) 39.4 (145) 17.5 (319) 59.4 (184) 220.9 (789) 3£8.8 (541) 260.6 (1.528) 244.5 246.2 " Unweight~l No. of sample deaths are given in pa~rnt~eses. * Des& rams ~ b~ on 1~8 NMS and Au~st 1967 CPS. In column 5. v~u~ a~ cancer in females. ~ ~ ~ ~ ~ hy ~e di~cc me~ to the 1940 U.S. population 3~ ~ for pr~s~zt.e ¢~ncer in ingles and br~as~ olcL JNCL VOL ~. NO, $. NOVEMBER 1~80 T!04891639

1178 Enstrom and Godley ICD 23.945,730 female nonsmokers (Seven~ Sample Sample PHmary aim Revision) Obs Exp# S~IR" Ohs Exp~ code No. No." No." Tom[ mM[~nt neop]ums I~205 106,478 670 157.798 67 237.2~1 789 271,~8 ~c~l ~qW ~d ph;~x 14~1~ ~8 ~6 4,403 62 1.702 ~ 3~48 ~s~ o~ ~d ~rimneum 15~159 45.~9 I~ ~,457 93 74.968 I80 87,5~ 85 Empha~s 150 h782 87 3~16 55 L601 2S 1,992 S~m~h 151 7.~0 17 10,65I 74 10.553 ~ 13.~5 ~lon 1~ 17,~7 ~ 13,5~ 133 30.~8 61 32,~6 94 ~mm 1~ 7,~ 18 6,463 117 8.117 18 10.573 77 ~r. ~llbl~der, ~d biliaw 15~156 3,~2 6 5~72 62 Ii,717 22 13,311 88 Panc~ 157 4.759 9 7,870 60 II,~ 23 13.983 ~spi~mw ~m 16~165 13,038 115 46,~1 28 I¢~8 131 21.71l 59 L~g, b~nch~ ~d ~chea 162-1~ 8,7~ 108 39,991 22 9,079 123 17.122 53 B~t 170 ~,~I 184 ~.659 94 Female ~ni~l sys~m 171-176 38,385 12~ 41,576 92 Ce~x 171 7.790 30 11,642 67 Co~us u~ # 172 1.126 6 1,937 ~t~, NOS 173-174 5.~8 16 6,316 Ova. ~lIopian mb~ ~d 175 21,563 ~ 19,393 III b~M li~en¢ Male geni~l s~m 177-179 13,784 278 16,563 P~ gland 177 13.~ 276 16,263 83 Urina~ ~mm 18~81 5,~1 I3 9,578 61 6,816 12 7,~6 86 Newo~ s~m 193 ~478 10 2,~9 95 6,830 24 5,543 123 L~phom~ 2~202 4.187 14 6.417 65 7.991 21 9,131 88 ~ukemia ~ 3,361 12 5~8 64 8,754 24 10,107 87 Total ~i~ant n~ 14~205 106,478 670 152,037' 70 237,261 789 ~0,5~I" pl~~ 9.525,41S male nonsmokers • Sample No. o! deaths from whMh weigh~:l No. of observed deaths is e~Ioulated with the use of pop,stratified ratio estimation pr~cedur~ (se~ text). • ~ Expected No. of d~.ths from ~neer ~s b~sed on 5-yr, ace-specific death ~tcs determined for U.S. whites (1966-68) from the total 1966--68 NMS and August 1967 CPS (s¢¢ t~xt). < SMR ~s the percentage defined a~ Ol~÷Exp x I00. ~ NOS=no¢ otherwise specified. • Expected No. of deaths from cancer is based on 5-yr, age-specific de~th ra~es for U.S. whites in 1970 (~). male and 95,% for female nonsmokers. For male non- smokers, the lowest SMR values occurred for the tradi- tional smoking-related cancer sites of the respiratory system, buccal cavity and pharynx, esophagus, pan- creas, and urinary, system, with a combined SMR of 39%, whereas the remaining cancer sites had a com- bined SMR of 91.%. However, the lowest SMR values for female nonsmoke~ occurred for the cancer sites o[ the lung and buccal cavity and pharynx, with a combined SMR of 5~%: the SMR values for most of the other sites were between 80 and 95,%. Because of the complex sampling and weighting procedures used in these data. confidence limi~ on an individual S,MR cannot he precisely calculated, although approximate limits can he determined with the use of the sample number of deaths, if one assumes they are subject to a Poisson distribution DISCUSSION The 1965-68 NMS represents the first and only at- tempt to date to determine death rates for all causes of death among a representative sample of the U.S. popula- tion classified according to smoking status. The basic findings in this paper show that during 1966-68, U.S. whites 35-8,t .years old who never smoked cigarettes have age-adjusted death rates from cancer and all causes substantially below those of the population as a whole and only about one-half the corresponding rates of those who currently smoke cigarettes. Phrased in other terms, whites who never smoked dgarettes have overall SMR which are substantially less than 100.%. Specific comparisons vary. by a few percentage points with different reference populations or with direct versus indirect standardization, but the basic patterns remain the same. Furthermore, these results indicate that most of the reduction in cancer deaths among male nonsmokers occurs among the traditional smok- ing-related cancer sites of the respiratory, system, buccal cavity and pharynx, esophagus, pancreas, and urinary system. The single greatest reduction in cancer deaths among female nonsmokers occurs in the respiratow system: however, small reductions also occur at many other sites. These results provide a methodologically sound data set on which to base comparisons with other popula- j.xcL VOL. .~. XO. 5. NOVK.'%II~ER I-~O TI04891640

Cancar Among U.$. Nonsmoker= 1179 tions at low risk to cancer. In other words, any nonsmoking population who has death mt~ similar to the I965--~8 NMS nonsmokers is probably at low risk solely because they do not smoke. These results help delineate the sF~.-cific influence of cigarette smoking on mortality rates from other health-related factors. REFERENCES (I) Advisory C~=nmittee to the Surgeon C~neral of the Public Health Service. Smoking and health. W~hington, D.C.: U.5. Govt Print Oft, I~-I [DHEW publication No. (PH5)IIOS]. (~) H.~ts~xt. W, LOVZL~.~D DB. SIItKL'~ MG. Lung-~ancer mortality • at reMted to residence and smoking histories. L White mal~. J Nail Can¢~ Inst I~6~;°2:947-II01. (J) H,,,L'~S~mL ~V, T.~c=~ K~. Lung-can¢~ mortality as related to residence and smoking hi,tone,. It. White females. J Nail Cancer Inst (4) Goat.~Y F. KXUtG~L DL C.i~,arette smoking and differential mortality. New estimates from representative nadorml samples. In: Health: United Stams. 1975. Rockville. Md.: National Center for Health Statistics. 1916:4~ (DHEW publi,"~tion No. O) E.'~sT~tOSt JE. Cancer anti total mortality among active Mormons. Cancer (6) --~.. Rising lung cancer mortality amon~ nonsmokers. JNCI (7) G~tovt, R.D. Hu~t~, AM. Vital ~tati=dcs rates in the United Stat~. I~.t0-1960. Washington. D.C.: U.S. Go,n Print Of L I~ [DHEW publication No. (8l National C~nter for Health Smtisdt:*. Vital statistics of the ~nited States, annttal reports for l~6& 196~. l~6& and 19~0. P, cckville. Md.: National C~nter for Health {9) ~. Cigarette smoking status--June 196~, Aueu~t I~I. and August 1~58. Monthly vital statistics report. Vol 18. No. (~uppl. December 5, 1969). Rockv/lle. for Health Statistics. 1~o9. UO) HA.'.t.~to,~ EC. Smoking in reladon to the death rates of one million men and women. Na¢l Cancrr (~I) KAI4.~ I-L-X. The Dora stt~d.v of smoking and mortality among U.S. tzterans: Report on eight and one-half year~ of ob~er- vat/on. Nad Cance~ Inst Mono~r (~,,0) U.$. Bureau o[ the Census. Estimates ol the population of the United States by a~e. color, and sex. Jui~ I. I~6Z Current population r, ports. ~er/es P.~5, No. 385. Washinsion. D.C,: U.$. Govt Print Off. l~. (~) KrraGAwA F-M. HAt:$t~t PM. Diffetent/al mortalit~ in the United States: A study o~ socioeconornk: epidemiolog¥. Cam~ridg,~: Harmtd Univ Press, 1973. (14.) F~,'~x~L MR. Inferenc~ from turveS, samples: An empirical in- vestigation. Ann Arbor, Mich.: Institute (or $oc/al R,e~earch, L'niv .Michigan. 1971. (IY) I~ttct4o~ H. R~60o.~ RH. Smokin~ habits of °.i.61.~ individuals in Te.'zas. J Natl Cancer Inst 1956:16:l~8"/-1304. APPENDIX National Mortality Survey The NMS is based on a systematic random sample of all deaths occurring and registered in the United StatesJ Death registration is essentially 100% complete and is performed on a state-of.~xcurrence basis with the use of the official State death certificate. The States generally follow the federally recommended model death certifi- cate. Date. place, and cause of death, age, sex. and other demographic characteristics are important legal facts noted on the dooament. One can extend the research potentials of d~ath certificates by using them as a sampling frame for surveys (such as the NMS) in which ad~tional information is collected through informants, funeral directors, physicians, and others named on the certificate. The scope of the NMS is all deaths of persons year, old that occurred in the United States during 19~6. 1967, or 19~8. The sample was stratified for month of death, geographic area, age, and selected causes 0[ death related to smoking. The total sample of 19,526 decedents between 35 and 84 years of age was drawn in two stages. I) A 10% sample is represented in the monthly "Current Mortality Survey" of death certificates submitted to the NCHS on microfilm by all States. the District of Columbia, and other death registration areas. 2) The strata of the 10% sample are subsampled; each 26th case was taken on the average. The overall sampling rate is therefore 1/o.60. The demographic and medical information coded from the sampled death certificate is supplemented by informa- tion from a mailed questionnaire on smoking habits, marital status, income, and other characteristics o£ the deceased. One-half of the respondents are spouses or ex-spouses o[ the deceased, one-fourth are either parent~ or offspring, one-eighth are siblings, and the remain- ing 10% are comprised of more distant relatives, friends, neighbors, and other associates. The overall response rate of 95.'.o in the NMS is comparable to that o~ the CPS. After follow-up queries of questionnaires with in- complete, inconsistent, or otherwise inadequate re- sponses, smoking information was still missing for about 8% of the sample deaths. For these cases, smok- ing history was assigned by computer processing of the records of the decedents. This procedure, in the absence of better information, assumes that the characteristics for which reported information is missing are like those of demographically similar individuals for whom the information in question was reported. This processing was performed by the NCHS, which also supplied the magnetic tape o{ 1966-68 NMS records. Each record combines the smoking and other NM$ questionnaire information with that ~rom the death certificate. In addition, each record contains an inflation factor (sample weight) that reflects the samp- ling fraction applicable to the decendent's demographic cause.of-death stratum. This sample weight is de- termined by a poststratified rado estimation procedure which forces the total number of deaths estimated from sampling to agr~ with the total number of deaths from which the sampling was done) The weight is a [unction of age, year of death, and cause of death and varies from about 10 m ~-t0. Appendix table i shows an example o~ how the total weighted number of deaths is calculated from the 184 sample deaths due to breast cancer among U.S. white females who never smoked cigarettes. The age-sex distributions of the total weighted deaths from all causes according to smoking status are given in Appendix table 2. The underlying cause o~ death (item =18 of the death JNCL VOL ~. NO. 5. NOVF-2*,II~E1~- I-e4~O T104891641

1180 Enstrom and Godley Uaweizhted No. d rumple Pes~tradfied r~tio esti- mation sample weight Age, ~ d~s by ~ by ~ 1~6 1967 1969 196~ 1968 I967 1968 ~ 18 15 13 46 81 82 ~ 17 ~ 17 4~ 163 162 166 ~ 17 I0 18 45 313 319 Total Averag'~ weigh*.ed Aus"ust 1967 annual Pro~rtional No. of dea~ rate distributes sample ~opulafisn- (dea~x/ of 1940 U.S. dea~s, zbrisk I00.0~1. ~pulad~n 196~8a 196~ 3.7S0 5.30L563 ~.8 0.34~40 7,~7 5,514,745 45.6 0.29~3 14.1~ 5,496,482 85.9 0.199205 17,619 4.~7,416 121.2 0.120142 11,952 2,7~5,516 14Z.0 0.04~30 55,051 23,945,7~ 59.4# 1.~0 • Weight No. of sample deaths |a each 10-yr age ga'oup is obtained by" multiplication: Unwelght~d No. of sample deaths in each dea~ year ~m~ ~ ~a~fi~ ~tio ~tima~on sample weight for ~e ~me d~ yeas add resul~ for the ~yr ~1.~ # A~j~ d~ ~ is ~leula~ by the di~t me~ ~ ~e 1940 U.S. ~p~lation 3~84 yr of age ~ the s~ndard. certificate) is determined from information provided by physicians and coded at the NCHS according to the rule~ of the Seventh Revision of the ICD (7, 8). Current Population Survey The CFS (9) is the source o[ estimates of the non- smoking and smoking populations. Smoking informa- tion was collected by the U.S. Bureau of the Census in its CPS of August 1967. Designed to be a representa- Eve sample el the Nation's noninsthutionalized popu- lation, the total consisted of about 50,000 households (residents of military, bases, home~ for the sick and aged, and other institutions were excluded). The over- all response rate was 95,%; of the 95,552 adults in the sample, those between the ages of $5 and 84 years numbered 60,92.0 and included 25.266 white males and 29,$08 white females..Based on this sample, the age-sex distributions of the noninstitutionalized population ac- cording to smoking status are shown in Appendix table $. Also shown for comparison is the CPS for July 1, 1967 (12), which represents the total resident popu- lation. The U.S. Bureau of the Census interviewers recorded inforriaadon about household and personal characteris- tics of those in the sample. Demographic and socio- economic information [or all persons in the household was reported by a single adult respondent. Smoking information v-as obtained from each adult responding for himselL if this was possible during the one interview. However, the smoking information for those not at home was also obtained from the person inter- viewed for the household. Inasmuch as men are more likely than women to he away from home when the interviews take place, only about one-third of the men, as compared with three-fourths of the women in the sample, reported on their smoking habits. Smoking information was missing for only about $,% of the sample. Estimates of population frequendes of nonsmokers by demographic characteristics were obtained from ,M'P~-NDIX 7aar.r. 2~E~t~mated 196~-68 U.S, total deatl~ of whites baaed on rel~'esentatit~ sample of death~ from th~ NMS by o~, sea. and c~zrette maoki~ql U.S. whi~ who never smok~ U.S. whi~ who cu~ently smoke 51ale Female Male Female 8~9 6.~ {81) 7.~0 {98} 28,181 (846) ~.087 {148} 4~ 8,581 (i05} 16~ (2~) 49.3~9 (~81 21.116 (262) 66,8~ (824} 4~9 13.784 (93) 20.697 (1~ 74.~ ~5~} ~,045 (221) 103.119 (732) 5~ 25~68 (170) 84.~ (~ 118.~6 (8~) 29.771 (273~ 169.694 (1.20~ 5~9 89.~3 ($~) 47.~ (179) ~46.~49 (774) 48.~6 (~6) 285.~9 (1.193~ 6~9 ~.~7 (~) 13Ll~ (8~) ~.8~ (691) 48.5~ (1~9) 327.~6 (I.~1) 7~74 I09.~6 (3~ 217.5~ (515) ~6 ~6~) 4L403 (~131 ~68.7~ 7~79 1~ (498) ~.0~ (708) 113.174 (482~ 89.244 (~6~ 377.1~ (1.~6) 8~ ~57.809 (4~ ~8.~69 (691) 70.~ (~0) 20.659 t53~ 309.~9 All U.S. whiten" Male Female 42,220 (521} 60.191 (398) 84.686 (a6~ 110,514 (448) 14L$~2 (~9) 195.314 (488) 280.660 (698) 349.9~$ (862) 337.4-°,3 17S7~ 1,625.460 (5.6~6) * T~.e first No. in eae~ c~ium~t is the weighted No. of deaths with the u~e o~ weight b~ced o= the l~St~tratified ratio estimation ~roeedureJ Nam~,ers in p~rs~ese~ a~e the unweigh~*d No. cf sample deaths ,.~ed in the NMS.' j.~. VOL dS. XO. ). NOVE.MBEK |~$3 T10489t642

Among U,S. N=nsm=kem 1181 U.S. whi~s wko never U.S, whites wP, o currently All U.$. ~ yr ~[~e ~,f~/e Female Female Male" Fete" Male" Fe~le" (July L (Au~c (JulT I. ~3~ 1.1~8.~1 2,5~.~2 2,779.~5 4~9 1,195~2 2,788.~4 2,894,~7 2~48,~5 5,0~,911 5,161,0~ 5,~4,324 5A64,C00 5~9 1,~9,~2 2,~8~8 2,0~,528 1.350,491 4,095~0 4.153.~ 4.454.802 4,489,~0 6~ ~,1~ 2.718,194 L559,~9 ~3,~3 3.392,~8 3.452,~ 3,~5,309 7~T4 883,178 £~.1~ 552.~3 241,868 2,~9,~7 2.067,0~ 2,650,9~ 2.T26,~0 8~ ~8.~8 951,496 114,~6 44,8~ 6~9.~2 800,~ 1,012.735 1.195.000 • Number= a~e from the Augamt 1£67 CPS of noninsfitudonslized U.S. populztion by eiczrette smoking status (9)." * Numbers ~re from the July 1. 196T CPS of tot~l resident TJ.S. population (I~). infladon facmm supplied by the U.S. Bureau of the Census in a magnetic rope of the sample person records. The inflation factors rake into account the sampling fraction and demographic characteristics of those in noninterviewed sample households. Missing demographic characteristic~ of those in interviewed households were computer assigned except for smoking information. "Not stated" or otherwise unknown smok- ing status is distributed statistically in proportion to the relative frequencies in the known smoking cats- gories. This procedure is followed, ~or lack of evidence of a smoking-related pattern of nonresponse, so the best estimate of population can be provided. Because it incorporates U.S. census control totals, the CPS is representative of geographic units, age, sex, and other demographic characteristics of the national population. Standard Death Rates Expected numbers of deaths in table ~ are the product of a set of standard death rates for 1966-68 (Appendix table 2) and the August 1967 CPS estimates of whites who never smoked dgarettes (Appendix table ~). The standard rates reflect the mortality experience of the total white population (irrespective of smoking history.). These rates are specific for 5-year age groups. sex, and the underlying causes of c~ncer death shown in table 3. We computed the standard rates by dividing the total of the NMS death estimates for 19~6, 1967, and 1968 by the CPS population estimate for August 1967. Thus both expected and observed deaths were derived from the same dam base of deaths. In other words, for each cause of death, the total number of expected deaths equals the total number of observed deaths for all white males and females. This procedure minimizes some biases that othemvise may be present in the SMR when the expected deaths are based on an arbitrarT., standard such as the 1970 U-S. white popu- lation. Sources of Error Various types of systematic errors likely to affect the estimates of mortality from smoking are discussed below. Also, the reliability of the estimated statistics in relation to sampling error and other sources of random variability is evaluated. 1) Matching between deaths and populations: The major source of bias affecting the results of this study arises from the cross.sectional two-sample design. The- oretically, the decedents are a subset of the total study population: ideally, the prospective design best ap- proximates this concept. The American Cancer Society and U.S. Veterans cohorts (10, 11) are prospective studies in which a beginning sample is initially questioned regarding current and past smoking habits. The cohort is followed longitudinally, and its decedents are recorded with respect to length of time in the study. Death rates are computed as follows: the total number of deaths divided by the total number of person-years lived during the study period by persons in various subgroups. These persons are classified by smoking habits and other characteristics. The decedents oi a prospective study, included in the original sample, match exactly with an individual in the study popu- lation. The present cross-sectional study, however, must rely on statistical matching between the two independently drawn samples: decedents and population. This prob- lem is common to conventional vital statistics rates G, 8) and must be balanced with the expense, time. and follow-up problems of prospective studies. Because mortality risk is measured by rate of deaths to population, the matching problem is a basic con- sideradon in the evaluation o[ errors from sampling, the data-collection process, and measurement techniques. 2~ The samples: A comparability problem exists in the extent to which the target populations differen- tially represent their universe. The universe of all JNC|. VOL. 65. NO. 5. NO%'F--%IBER I.~30 T[04891643

1182 Enstrom and Godley deaths occurring to persons 35-84 years old in the United States during 1966-68 is represented by offidal registrations of deaths occurring in the United States in any of the 3 years, irrespective of month of death or usual residence of the deceased. However, the popula- tion interviewed in the GPS consists of individuals who were not residents of institutions in the United States as of August 1967; this count was at about the midpoint and is believed to represent the 196~-68 population adequatdy. The CPS practice of replacing part of the sample every 4 months (rather than monthly) extends parts of the target population to cover a span of 7 months, May to November 1967. Conversely, exclusion of residents of institutions introduces a systematic bias that understates the popu- lation and thereby inflates death rates. A first con- sideration is also to exclude deaths of residents of insti- moons. The rationale for not doing so is as follows: a) In terms of death rates, the bias is really a funcdon of person-years of excluded instim0onal population. Fragmentaqt evidence suggests that many persons in the institutional population are there for only a fraction of a year.4 b) There is no sound basis for the estimation the person-years o[ excluded institutionalized popula- tion for the various socioeconomic groups classified by smoking status, c) The net effects of the bias are likely to be canceled in measures of differential mortality like the SMR. d) The bias is most likely to afIect only the extremely old, a small fracdon of the study population for whom differen0al mortality is greatly reduced and not of major interest. Only for persons 80-84 year~ age is the excluded population more than 5% o[ the total (Appendix table 3). 5) The respondents: Unlike sizable proportions the populadon sample who could provide information on their smoking habits and other personal characteris- tics, no such data were available from the death sample. Moreover, a mailed questionnaire was used in the death survey, whereas a household interview was conducted for collection of data on the living popula- tion, Other sources report substantial agreement tween the smoking habits reported for a decedent retrnspectively and his reports as obtained in a house- hold interview prior to death (2).4 With regard to social and demographic characteristics, evidence indi- cates agreement between the 1960 NMS responses and the matched 1960 census reports (13). The overall response rates co the NMS and the CPS were nearly equal at 95.%; these high levels eliminate any serious bias from differential nonresponse. There is no evidence on nonresponse differentials h.v strok- ing status from this investigation. In retrospective studies, in which nonresponse may he as much as 39.% o~rall and even more among smokers, the bias in the mortality ratio of the respondent has been estimated to be about a 0.2 or 0.3 overstatement because of lower mortality ratios of the nonrespondems (1). Again, the low level of nonresponse in the present stud.v f5%) in- dicates that nonresponse is negligible as a source of bias. ,t) .'V[easurement of cigarette smoking:. The o~era- 0onaI definitions of lifetime history concepts of never ~moking cigarettes are almost e.xacdy the same in the death survey {<5 packs of cigarettes] and popula- tion survey (<100 cigarettes). Copies of the actual NMS and CPS questionnaires are shown elsewhere." Ahhough the slight variations in these definitions do not appear to be a problem, a comparability problem does arise from the hct that those reporting for the decedent tend to understate cigarette consumption com- pared with what the decedent would have reported as his habits. Haenszel et al. (2) noted that this bias is largest with respect to medium and heavy smokers but virtually nonexistent for nonsmokers. 5) Sampling and data processing:. Complex sampling design, coding of reported information, construction o~ data giles, and production of the final data tables are all possible sources of systematic e~xor. The assump- tion is that these errors are negligible. Table 1 and Appendix table 3 show that the study estimates of death rates and population-at-risk are representative of the United States [or the study period. Table i illus- trates the good agreement between the total age-specific death rates as calculated in the survey and the 1966-68 U.S. vital statistics (8). The good agreement between the total age-specific U.S. white population as deter- mined by the August 1967 and the July 1, 1967, CPS (12), which was used to calculate the 1967 U.S. death rates (8), is depicted in Appendix table 3. 6) Reliability: In addition to systematic error, the estimates reported in this study contain some random error. This kind of variability is inherent in the sampling of deaths and population at a particular time. In more detailed analyses, random variation may be present in the indicators of mortality that are used for empirical tests of hypotheses. The 1964 Surgeon General's Report evaluated the stability of the SMR by a main assumption that deaths occurring within a given period are generated by a Poisson process (I). The random variability of SMR was small and de- pended on the number of deaths. The analysis was of ~6,000 deaths in 7 prospective studies. Inasmuch as this study is based on about 20,000 sample deaths, one can assume that variability in its SMR is a negligible problem for the major causes of death. 7) Sampling error:. In addition to variability over time, the SMR may contain some random variation due to sampling. Both death and population estimates are based, on complex, multistage sample designs. Another analysis of the 1967 CPS shows that simple random sampling standard errors understate the true standard errors of estimated population statistics (14). Standard te, ts of statistical significance are therefore inapplicable to this study. Hammond (10) used a frequency, of 10 deaths as a minimum ~or standards of reliability.. Although arbi- tral', the same rule of thumb (adopted in this study) primarily aifects deaths when classified by detailed underlying causes. The sample numbers of deaths from the 1966-~8 NMS are presented in tables 1-3. jNcI. t'OL ~. NO. 5. T104891644

$) Response error. At the level af the individual, reported informal/on may vary from one abservadon another. This variation may introduce error in mates of fixed characteristics like a histo~" oi e~er smoking. Kirchoff and Rigdon 115) compared the smoking habits of a group of Texans and reported Cancer Among U.S. N~:nsmokers 1183 different smoking habits f~r the same reference period when respondents ~'ere reinter~iewed. This difference partly reflects some memoD" lapse and some response error in smoking information. ~%'e assume that the extent to which this source of variation affects esti- mates of the pre,oent study is negligihIe but unknown. .~.'CCI. VQL 6~. .NO. ~. NO%'F-.%IBER 1~0 Tt04891645