higher for single than for married men. The size of the excess death rate for users of tranquilizers compared to men who do not use them is perhaps surprising {29.1 against 18.2 and 52.4 against 31.8). However, the tran- quilizers in question required a doctor’s prescription, so that some men in this group are presumably under medical attention for illness. The group of users is small, comprising only about 10 percent of those who answered this question. Death rates tend to decrease slightly as the educational level increases; this association may represent some facet of the association of death rates with socio-economic level. Degree of exercise displays an inter- esting association with mortality, the death rate declining steadily with additional degrees of exercise. In particular, the two “no exercise” groups show marked elevations in death rates. These groups, however, amount to only 2 percent of the respondents to this question. From the same data, Ipsen and Pfaelzer (14) made a further analysis of seven variables that appeared to be related to mortality, in order to see whether any of the variables had a stronger association with mortality than did cigarette smoking. They concluded that apart from previous serious disease, none of the other variables examined had as high a correlation with mortality as smoking of cigarettes. Further, the correlation of any of these other variables with cigarette smoking was too weak to reduce markedly the correlation of cigarette smoking with mortality after adjustment for the other variable. In the analyses above, smoking was matched against each variable sep- arately. In addition, Hammond (11) carried out a “matched pair” analysis, in which pairs of cigarette smokers and non-smokers were matched on height, education, religion, drinking habits, urban-rural residence and occupational exposure. The percentage who had died in the 22 months was 1.64 for smokers and 0.88 for non-smokers. These informative analyses are available, unfortunately, for only one of the studies. However, in order that the association of cigarette smoking with mortality should disappear when we adjust for another variable, the correlations of this variable with smoking and with the death rate must both be higher than the correlation between smoking and the death rate. Except for the breakdowns by longevity of parents and grandparents, the analyses throw little light, however, on the objection that a part of the differences in death rates may be constitutional, psychological or behavioral; ie. that regular cigarette smokers are the kind of men who would have higher death rates even if they did not smoke. Further discussion of this Point appears in the next section. MORTALITY BY CAUSE OF DEATH In all seven studies the underlying cause of death, as specified in the Inter- National Statistical Classification of Diseases, Injuries and Causes of Death, was abstracted from the death certificate. In the two American Cancer So- ciety studies, further confirmation of the cause of death, including histological evidence, was sought from the certifying physician for all cancer deaths; this 101 procedure was also followed in the British doctors’ study for all certificates in which lung cancer was mentioned as a direct or contributory cause. With these exceptions the data presented here represent the results of routine death certification. For current smokers of cigarettes the total mortality, after adjustment for differences in age composition, was found previously (Table 2) to be about 70 percent higher than that of non-smokers in these studies. The primary objective in this section is to examine whether this percentage increase ap- pears to apply about equally to all principal causes of death, or whether the relative increase is concentrated in certain specific causes or groups of causes. RESULTS FOR CIGARETTE SMOKERS For 24 causes of death, plus the “all other causes” category, Table 19 shows summary data over all seven studies.* In four of the studies the data are those for current smokers of cigarettes only, but in the two California studies and the 25-State study the cause-of-death breakdown was available only for all cigarette smokers including “cigarette and other” smokers and current and ex-smokers. For each listed cause, Table 19 shows the total numbers of expected and observed deaths of cigarette smokers summed over all seven studies, and TABLE 19.—Total numbers of expected and observed deaths and mortality ratios for smokers of cigarettes only * in seven prospective studies Mortality | Median |Non-smoker Underlying cause of death Expected | Observed ratio mortality deaths ratio Cancer of lung (162-3)..._......2--22-22._- 170.3 1, 833 10.8 11.7 12 Bronchitis and emphysema (502, 527.1) ?.__ 89.5 546 6.1 7.5 59 Cancer of larynx (161)......._.--...-..--_-- 14.0 75 5.4 5.8 8 Cancer of oral cavity (140-8). 37.0 152 4.1 3.9 a Cancer of esophagus (150)... -.___- 33.7 113 3.4 3.3 19 Stomach and duodenal ulcers (540- 105.1 204 2.8 5.0 67 Other circulatory diseases (451-468) ________ 254.0 649 2.6 2.3 170 Cirrhosis of liver (581).___.....-_--_--_---- 169. 2 379 2.2 2.1 98 Cancer of bladder (181)... 111.6 216 1.9 2.2 92 Coronary artery disease (420) 6, 430. 7 11, 177 1.7 L7 4,731 Other heart diseases (421-2, 430-4). ..___._ 526.0 868 17 1.5 398 Hypertensive heart disease (440-3) _______- 409. 2 631 15 15 334 General arteriosclerosis (450)_.....-.---...- 210.7 310 1,5 1.7 201 Cancer of kidney (180)....-.....--.--.----- 79.0 120 15 14 59 All other cancer 1, 061.4 1, 5% 1.4 14 742 Cancer of stomach (151)_ 2 413 L4 1.3 203 Influenza, pneumonia (480-493).__.____...- 303. 2 415 14 16 169 All other causes___-_..-__- 22.22 oe eee 1, 508. 7 1, 946 1.3 1.3 1, 036 Cerebral vascular lesions (330-4).-.-..---_- 1,461.8 1, 844 1.3 13 1, 069 Cancer of prostate (177) _....._....--..-2.-- 253. 0 318 1.3 1.0 198 Accidents, suicides, violence (800-199)... ._ 1, 063. 2 1,310 1.2 1,3 627 Nephritis (592-4)_--.-..-2.22-2-.22 2222. 156. 4 173 11 1.5 08 Rheumatic heart disease (400-416). ._.__.__ 290. 6 309 11 11 185 Cancer of rectum (154).__._-....--- 22. ee 207.8 213 1,0 0.9 150 Cancer of intestines (152-3) 422.6 395 0.9 0.9 307 All causes. ._.2- 20-22-02 22-2 15, 653. 9 26, 223 1.68 1,65 11, 168 ' Current cigarettes only for four studies: all cigarettes (current and ex-) for the two California studies and the study of men in 25 States. ? “Bronchitis and emphysema” includes ‘‘other bronchopulmonary diseases” for men in nine States and Canadian veterans. *The individual results for the seven studies are shown for reference purposes in Table 26. 102 the resulting mortality ratios, arranged in order of decreasing ratios. The combination of the results of the seven studies in this way is open to criticism, since it gives more weight to the larger studies than may be thought advis- able, and since the true mortality ratios for specific causes presumably differ somewhat from study to study. However, for some causes of death that are of particular interest the numbers of deaths are small in all studies, so that some procedure for combining the results is highly desirable. As an alternative measure of the combined mortality ratio, the median of the seven mortality ratios (obtained by arranging the seven ratios in increasing order and selecting the middle one) is also shown for each cause in Table 19. The median, of course, gives equal weight to small and large studies. Although there are some changes in the ordering of the causes when medians are used instead of the ratios of the combined deaths, the general pattern in Table 19 is the same for both criteria. Table 19 also presents the total numbers of non-smoker deaths on which the combined mortality ratios are based. Lung cancer shows the highest mortality ratio in every one of the seven studies, the combined ratio being 10.8. Other causes that exhibit sub- stantially higher mortality ratios than the ratio 1.68 for all causes of death in Table 19 are bronchitis and emphysema, cancer of the larynx, cancer of the ora] cavity and pharynx, cancer of the esophagus, stomach and duodenal ulcers, and a rather mixed category labeled “other circulatory diseases,” which includes aortic aneurysm, phlebitis of the lower extremities, and pulmonary embolism. For three of these causes—cancer of the larynx, oral cancer and cancer of the esophagus—the numbers of non-smoker deaths are small, so that the over-all mortality ratio cannot be regarded as accurately determined. The U.S. veterans’ study and the 25-State study provide an additional breakdown for two of the causes listed in Table 19. For the rubric 527.1 (emphysema without mention of bronchitis), these studies give mortality latios of 13.1 and 7.5, respectively. For ulcer of the stomach they give 5.1 and 4.3, whereas for ulcer of the duodenum their mortality ratios are 2.3 and 1.1. Bronchitis and emphysema also show a high rate, 12.5, in the British doctors’ study. There follows a list of 14-causes whose mortality ratios are not greatly different from the ratio of 1.68 for all causes in Table 19. These causes Tange from cirrhosis of the liver, with a ratio of 2.2, down to a ratio of 1.2 for the miscellaneous class which contains accidents, suicides and violent deaths. This group includes the leading cause of death, coronary artery disease, with a ratio of 1.7, cerebral vascular lesions with a ratio of 1.3, and the “all other causes” group with a ratio of 1.3. For each of these 14 causes the mortality ratio differs from unity, by the approximate statistical test of significance. Finally, there are four causes—nephritis, rheumatic heart disease, cancer of the rectum and cancer of the intestines—whose mortality ratios are close to unity. For smokers of cigarettes and other, the data from four studies agree in Beneral with the ordering of causes in Table 19, although the mortality Tatios for most causes are slightly lower than with smokers of cigarettes 103 only. These and the corresponding data for ex-cigarette smokers are shown in Table 20. Data on ex-cigarette smokers can be obtained from four studies. The causes of death with mortality ratios of 2.0 or higher are, in decreasing order, bronchitis and emphysema (7.6), cancer of the larynx (5.4), cance; of the lung (4.8), stomach and duodenal ulcers (3.1), oral cancer (2.0) and other circulatory diseases (2.0). , The group of 17 causes with mortality ratios below 2 in Table 19 requires discussion. If cancer of the bladder (mortality ratio 1.9) and coronary artery disease (mortality ratio 1.7) are omitted, since they receive detailed consideration elsewhere in this report, the numbers of expected and observed deaths for this group as a whole are as follows: Expected Observed Mortality Ratio 8,241.3 10,789 1.31 If we exclude from this total the four causes at the foot of Table 19, fo; which the mortality ratios are 1 and smaller, the corresponding totals become: Expected Observed Mortality Ratio 7,164.0 9,699 1.35 In either case the excess of observed over expected deaths is close to 2,500 or about 25 percent of the total excess in observed deaths in Table 19. Thus, although the mortality ratios for these groups are only moderately over 1, the group as a whole contributes substantially to the total number of excess ob. served deaths. The group consists mainly of a miscellaneous collection of chronic diseases. Several tentative explanations of this excess mortality ratio can be put for. ward. Part may be due to the sources of bias previously discussed. It was indicated in the section on “Non-Response Bias” that the bias arising from non-response might account for a mortality ratio of 1.3. Relatively high mortality ratios in certain causes of death that have not yet been examined individually may also be a contributor, although as these causes are likely to be rare, the contribution from this source can hardly be large. Part may be due to constitutional and genetic differences between cigarette smokers and non-smokers. Except for the breakdown mentioned previously by longevity of parents and grandparents in the men in 25 States study, there is no body of data available that provides a comparison of cigarette smokers and non-smokers on these factors as they affect longevity. But it is not un- reasonable to speculate that the kind of men who become regular cigarette smokers are, to a moderate degree, less inherently able to survive to a ripe old age than non-smokers. We know of no way to make a quantitative estimate of the difference in death rates that might be attributable to such constitu- tional and genetic factors. Studies reported in Chapters 14 and 15 indicate that some average differ- ences can be detected between smokers and non-smokers on behavioral, psychological and morphological characteristics. Nevertheless, the same com- parisons show considerable overlap between the individual men in a group of smokers and a group of non-smokers. For what they are worth, these com- 104 TaBLE 20.—Expected and observed deaths and mortality ratios for current smokers of cigarettes and other (three studies) 1 and for ex-cigarette smokers (four studies) * Cigarettes and other Ex-cigarette Underlying cause of death Number of deaths Number of deaths Mortality Mortality ratio ratio Expected | Observed Expected | Observed Cancer of lung (162-3)_..------ 60.9 510 8.4 30.4 145 4.8 Bronchitis and emphysema (502, 527.1) 3_.__.___._-------- 53. 2 191 3.6 17.4 133 7.6 Cancer of larynx (161) _. .---- 1.6 20 12.5 1.3 7 5.4 Cancer of oral cavity (140-8). . 111 42 3.8 5.9 12 2.0 Cancer of esophagus (150) __.- 18.1 57 4.4 5.4 6 11 Stomach and duodenal ulcers (840-1) --- 2. eee 23.0 99 4.3 13.0 40 3.1 Other circulatory diseases (451-468) __.-.....---------- 99.0 227 2.3 45. 8 93 2.0 Cirrhosis of liver (581) __.....-- 57.3 85 15 22.4 27 1.2 Cancer of bladder (181)._....-- 58. 2 73 1.3 29.8 31 1.0 Coronary artery disease (420) _ 2, 335. 0 3, 262 14 1, 245. 0 1, 731 14 Other heart diseases (421-2, ) 225. 9 321 14 124.1 178 14 144.4 174 1.2 93. 0 133 14 General arteriosclerosis (450)_- 106.8 146 L4 63.7 75 1.2 Cancer of kidney (180)___....- 25.0 37 1.5 13.9 25 1.8 All other cancer . ______- 272.9 339 1.2 199.3 239 1.2 Cancer of stomach (151) 101.0 139 1.4 51.4 66 1.3 Influenza, pn’ ur onia (480-493). 199. 2 153 0.8 55.1 55 1.0 All other causes ...........-. 769.3 790 1.0 308. 1 357 1.2 Cerebral vascular lesions (330- L ) a ee 634.0 605 1.0 300. 1 321 Lt Cancer of prostate (177) ..---- 97.1 118 1.2 52.0 57 1.1 Accidents, suicides, violence nee e een e nee ee eeee 287.1 316 11 169.6 159 0.9 Nephritis (592-4) _....--..--- 30.7 44 1.4 21.7 23 11 heumatic heart disease (400- 416)... oe eee 96.0 86 0.9 47.9 59 1.2 Cancer of rectum (154) _..._.- 89.7 64 0.7 43.3 38 0.9 Cancer of intestines (152-53)... 149.6 164 it 85.8 97 11 All causes._..-____--..-------- 5, 941.1 8, 062 1.4 3, 045. 6 4, 107 1.35 ' British doctors, U.S. veterans and Canadian veterans. ? British doctors, men in nine States, U.S. veterans, and Canadian veterans. +“Bronchitis and emphysema” includes ‘other bronchopulmonary diseases” for men in nine States and Canadian veterans. parisons suggest by analogy that the differences in death rates from constitu- tional or genetic factors may be moderate or small rather than large.* Fur- ther, it seems unlikely that constitutional or genetic differences between cigar and pipe smokers and between these groups and non-smokers can have any substantial effect on their death rates, since the over-all death rates of these three groups differ only slightly. Finally, part of the difference may represent a general debilitating effect of cigarette smoking in addition to marked effects on a few diseases. Pearl’s hypothesis that smoking increases the “rate of living” is of this type, though there are difficulties in making this hypothesis precise enough to be subject to medical investigation. Hammond (13) has suggested that the explana- tion might lie in the effect of cigarette smoking in decreasing the quantity of oxygen per unit volume of blood, but there are numerous medical objections to this hypothesis. This Committee has no information that would lead it to favor one or another of the possible explanations put forward above. *This question is discussed more fully in Chapter 9, p. 190. 105 Morrtatity Ratios FoR CIGARETTE SMOKERS BY AMOUNT SMoKgp For coronary artery disease and lung cancer, the mortality ratios are given by amount smoked in Tables 21 and 22 for current smokers of cigarettes only, In Table 21 an increasing trend with amount smoked appears in all five studies. The two California studies, in which the data are for all cigarette smokers (current and ex-smokers combined) show a less marked trend. TaBLE 21.—Mortality ratios for coronary artery disease for smokers of cigarettes only by amount smoked Number of packs per day British Men in 9 US. Canadian | Men in 25 doctors States veterans veterans States _— ————___. 1.2 1.3 1.7 13 1.9 18 1.7 24 2.1 L7 12.0 21 2.4 19} 25 1 More than one pack. TaBLE 22.—Lung cancer mortality ratios for current smokers of cigarettes only by amount smoked Number of packs per day British Men in U.S. Canadian doctors 9 States veterans veterans Wow nn ne eee nee 5.8 5,2 8.4 WA ieee ee eee ee eee nee 7.3 9.4 13.5 VQ lee 15.9 18.1 1b t Over 2______-_----- 2-22-22 eee 21.7 23.3 1 Over one pack. The trends in lung cancer mortality ratio with amount smoked are steep in all four studies. The two California studies also show marked trends for all cigarette smokers combined. For the six causes of death (other than lung cancer) that were pointed out in Table 19 as having unusually high mortality ratios, the numbers of deaths permit a breakdown only into two amounts smoked. The results from six studies are shown in Table 23. Data were not available from the TaBLE 23.—Expected and observed deaths and mortality ratios for current cigarette smokers, for selected causes of death, by amount smoked, in six studies : One pack or less More than one pack i I Causes of death ' Number of deaths Number of deaths i Mortality Mortality : ratio ratio Expected | Observed Expected | Observed Bronchitis and emphysema. __ 44.6 225 5.0 17.2 147 8.5 Cancer of larynx. .___._. 3.6 19 5.35 41 31 7.5 Cancer of oral cavity 16.8 53 3.2 14.8 60 41 Cancer of esophagus. ___..___- 13.2 40 3.0 9.7 48 4.9 Stomach and duodenal ulcers_ 32.5 110 3.4 31.2 91 2.9 Other circulatory......_.._--- 98.5 253 2.6 60. 4 175 2.9 Caneer of the bladder._______- 57.3 80 1.4 2.7 73 3.1 106 men in the 25-State study. Cancer of the bladder is included in Table 23 as background data for Chapter 9. , All causes except stomach and duodenal ulcers show some increase in the mortality ratio for the heavier smokers. The rate of increase cannot be regarded as accurately determined in view of the small numbers of deaths. CIGARS AND PIPES In view of the small numbers of deaths involved, the data for cigar and pipe smokers were combined in Table 24, which lists the total expected deaths, total observed deaths and mortality ratios from five studies (British doctors, U.S. Veterans, Canadian Veterans, and men in 9 and 25 States). Causes of death with relatively high mortality ratios are oral cancer (3.4), cancer of the esophagus (3.2), cancer of the larynx (2.8), cancer of the lung (1.7), cirrhosis of the liver (1.6), and stomach and duodenal ulcers (1.6). It should be noted that all these ratios are based on modest numbers of deaths. TaBLe 24.—Numbers of expected and observed deaths and mortality ratios for cigar and pipe smokers, in five studies + Number of deaths . Underlying cause of death Mortality ratio Expected | Observed Cancer of oral cavity (140-8). .-_.-----.----.------------------------ 13.5 46 3.4 Cancer of esophagus (150)... 10.2 33 3.2 Cancer of larynx (161)... 3.2 9 2.8 Cancer of lung (162-3). 65.2 113 L7 Cirrhosis of liver (581) _ _.---------- - 47.5 17 1.6 Stomach and duodenal uleers (540-1) - 35.2 56 1.6 Cancer of kidney (180)----.-------- - 30.8 39 1.3 Cancer of intestines (152-3) __..-- - 174.6 219 1.3 Other circulatory diseases (451-468) 7 89.1 105 1.2 Allother cancer... -..-------- - 396.7 456 11 Cancer of prostate (177). - 127.2 144 11 Cancer of stomach (151). - 116.8 132 11 Cancer of rectum (154)....-.---- - 78.2 88 1.1 Hypertensive heart disease (440-3) - - 194.5 218 11 Other heart diseases (421-2, 430-4)___ - 272.6 303 11 Bronchitis and emphysema (502, 527.1) - 33.7 37 L1 Cerebral vascular lesions (330-4) _ _ - 685.3 720 11 Coronary artery disease (420). 2 2,721.5 2, 842 1.0 All other causes_._...--.-------- - 612.9 8 10 Influenza and pneumonia (480-498) - - 93. 8 0.9 Accidents, suicides, violence (800-999) . - 347.1 318 0.9 Cancer of bladder (181)_. _-.-------- - 63.1 56 0.9 General arteriosclerosis (450) - 124.1 109 0.9 Nephritis (592-4)___.-.-.-------- - 63.6 55 0.9 Rheumatic heart disease (400-416)...........-.---------------------- 100, 5 69 0.7 All causes... 0-2-0. eee eee eee eee ee eee 6, 500.9 6,919 1.06 1 Includes British doctors, men in 9 States, U.S. veterans, Canadian veterans, and men in 25 States; cludes ex-smokers for men in 9 States; excludes pipe smokers for Canadian veterans. Separate breakdowns by cause of death for cigar-only smokers and for Pipe-only smokers are available in only three studies. The numbers of deaths are too few to throw any light on the question whether there are differences between cigar and pipe smokers in the causes of death for which mortality ratios are elevated. 107 Tue CONTRIBUTION OF DIFFERENT Causes To Excess MorTatiry Several of the reports previously published on these studies have included a table showing how the excess number of deaths of cigarette smokers over non-smokers is distributed among the principal causes of death. For each cause, the difference between the observed and the expected number of deaths for cigarette smokers is divided by the total excess for all causes, and multiplied by 100 to express the figures on a percentage basis. Table 25 presents these percentages for the seven studies for 13 groups of causes, A negative percentage, which occurs in a few places in the table, implies that for this cause the observed smoker deaths were smaller than the expected deaths. TaBLE 25.—Percentage of total number of excess deaths of cigarette smokers due to different causes * ; British | Men in U.S. |California|California|Canadian}| Men in Underlying cause doctors | 9 States | veterans oceupa- Legion | veterans | 25 States tiona Coronary artery disease ___.-... Other heart disease___--- a Cerebral vascular lesions. ------ Other circulatory diseases. Cancer of lung. -_-_.------------- Cancer of oral cavity, esopha- gus, larynx_. Other cancer. -- Bronchitis and s -- Influenza and pneumonia. ..--- Stomach and duodenal ulcers. - Cirrhosis of liver. __..-..----.-- Accidents, suicides, violence __. All other causes._.-.----------- All causes. ___._.--.------------ w on Seyyronan Bap Os ie & - cn Praor bo h e y _ “ Om wena -_ am | RONAN E BOm CNN CATR ONW COPrewme Swrrerrwes woewe CON DHMewHo onto o CoOouvrrrs.oosl Oro o cd Spoorvyego Sears ON WOW WOOD HN wR or CANN NAAN OS wrancnn SOPONMONN Sopom CORAMWOCUNNN Wewon SWC Mw we woo Sm mw wena an SOSNNNOSw - = _ S _ : — 7 3 _ 2 _ > 1 All cigarette smokers (current and ex-) for the two California and men in 25 States studies; current cigarette smokers only for the remainder. As previous writers have noted, all studies agree in showing coronary artery disease as the prime contributor to excess mortality, with lung cancer in second place. Other rubrics that show a substantial contribution in some studies, though not in all, are bronchitis and emphysema, cancers other than those of the mouth and lungs, and heart disease other than coronary. SUMMARY This report summarizes the results of the seven major prospective studies of the relative death rates of male smokers and non-smokers. ToTaL MORTALITY Cigarette Smokers The death rate for smokers of cigarettes only who were smoking at the time of entry is about 70 percent higher than that for non-smokers. 108 60T TABLE 26.—Numbers of expected and observed deaths for smokers of cigarettes only, and mortality ratios, each prospective study and all studies British doctors Men in 9 States U.S. veterans California occupational Cause of death Deaths Deaths Deaths Deaths Mortality Mortality Mortality ratio ratio ratio Expected| Observed Expected} Observed Expected; Observed Expected| Observed Cancer of lung 6.4 23.4 10.0 43.3 519 12.0 8.7 15.9 Bronchitis, emphysema_ 4.2 12.8 2.3 14.4 M41 9.8 2.6 4.3 Cancer of larynx. -._.. .0 1.3 13.1 2.4 14 5.8 Of Beetle Cancer of oral cavity .0 7.8 2.8 8.1 54 6.6 7.2 1.0 Cancer of esophagus .._ -----(¢ 3.3 2.7 6.6 5.2 33 6.4 5.5 .7 Stomach and duodenal ul ~ (540, 541) -0 12.2 5.0 21.5 67 3.1 23.1 5 Other circulatory diseases_ ~- (451-68) 17.2 19.7 2.7 66.4 228 3.4 11.5 1.6 Cirrhosis of liver. --..-.---------------- ee (581) .0 23.5 2.1 31.2 WW 3.6 14.7 4.0 Cancer of bladder _ -- (181) 13.9 9 17.2 2.4 31.4 55 1.8 2.2 6.0 Coronary artery disease __.....--..-...-.-. (420) 366. 9 1.5 927.7 1, 734 19 803.3 3, 037 1.7 273.9 2.0 Other heart diseases... ._. (421-2, 430-4) 78.8 1.5 72.5 108 15 122.2 244 2.0 23.8 1.0 Hypertensive heart disease__........_._. (440-3) 21.0 1.5 89.7 1.2 138.7 223 16 27.2 10 Genera) arteriosclerosis ____ --- (450) 21,2 1.0 9.1 2.0 97.0 163 1.7 .0 - Cancer of kidney... --- (180) .0 14.0 L5 23.1 34 1.5 .0 All other cancer...-...-----..-022 222 eee ee eee 81.7 9 132, 9 17 315.8 457 1.4 72.1 L5 Cancer of stomach_ ~~ (151) 28.3 11 33.7 2.3 41.5 9” 15 31.4 -8 Influenza, pneumonia_ (480-93) 47.0 .7 1.6 2.6 22.6 36 1.6 10.3 2.4 All other causes... -.....---..222-20-2--2--- eee 144.0 1.3 209. 5 1.3 354. 8 530 15 68.9 15 Cerebral vascular lesions - (330-4) 16L.1 1.2 208. 8 13 309. 1 467 1.5 42.2 L8 Cancer of prostate......__- ---(177) 29.0 5 32.4 1.6 53.7 106 2.0 8.6 5 Accidents, suicides, violence ~ (800-999) 89. 2 1.0 174.1 Ll 241.5 306 1.3 108. 4 15 Nephritis_--.........----- 8.1 2.1 43.3 -8 18.6 30 16 16.0 -6 Rheumatic heart disease 10.2 1.3 48.4 -9 67.4 77 Ll 22.9 L4 Cancer of rectum._ 4.2 3.6 29.8 8 68.7 62 a) 13.6 1.0 Cancer of intestines 26.1 Li 65. 6 25 121.2 152 13 23.7 9 All causes_____-.--2.--- 222202222202 1, 161.8 1. 44 2, 227.7 3, 781 1.70 4,043.1 7, 236 1.79 818.5 1.78 Ort TaBl.& 26.—Numbers of expected and observed deaths for smokers of cigarettes only, and mortality ratios, each prospective study and all studies—Continued California Legion Canadian veterans Men in 25 States Total, all studies Median Cause of death Deaths Deaths Deaths Deaths mortality Mortality Mortality Mortality} ratio ratio ratio ratio Expected] Observed Expected] Observed Expected} Observed Expected] Observed Cancer of lung......-.-------- (162-3) 19.9 4.9 27.1 11.7 41.5 9.6 170.3 1, 833 10.8 11.7 Bronchitis, emphysema - _ (502, 527.1) 3.6 8.4 36.5 4.6 15.4 7.6 89.5 546 6.1 7.5 Cancer of larynx._..__...-...--- 161 4.0 1.5 .0 6.3 3.7 14.0 75 5.4 5.8 Cancer of oral cavity 5.2 1.9 5.1 3.9 3.6 9.2 37.0 182 4.1 3.9 Cancer of esophagus_-..---..---- ( 1.8 5.1 6.8 3.3 8.4 2.4 33.7 113 3.4 3.3 Stomach and ducdenal ulcers 540, 541) 1.8 6.8 7.9 6.9 38.6 1.9 105. t 204 2.8 5.0 Other circulatory diseases --~ (451-68) 16.7 2.2 41.5 2.3 81.0 2.5 254.0 649 2.6 2.3 Cirrhosis of liver. 13.1 18 37.6 1.3 49.1 1.5 169. 2 379 2.2 21 Cancer of bladder--_--- 18 4.0 22.3 17 22.8 2.2 111.6 216 19 2.2 Coronary artery disease. _. 312.8 17 882. 5 1.8 1, 863. 6 17 6, 430.7 11, 177 17 17 Other heart diseases -_ (421-2, 430-4 13.1 26 2.0 75. 3 21 140.3 L4 526.0 868 17 1.5 Hypertensive heart disease. _. (440-3) 24.9 2 12 36. 2 1.6 71.5 2.2 409, 2 631 1.5 1.5 General arteriosclerosis (450) 39.1 2 5 14.7 3.3 29.6 1.2 210.7 310 1.5 17 Cancer of kidney.._._-....-- 8.3 6 .7 9.5 1.4 24.1 12 79.0 120 15 14 All other cancer. --.- 75.4 84 11 104.1 1.4 279.4 1.5 1, 061.4 1, 524 1.4 14 Cancer of stomach 20.5 25 12 41.2 1.9 68.6 1.3 285. 2 413 1.4 1.3 Influenza, pneumonia. ------ (480-93) 14.7 22 1.5 135.0 1.2 58.0 1.7 303. 2 415 1.4 16 All other causes. -__..--------.---.---- 39.1 O4 2.4 361.5 1.0 330.9 1.3 1, 508.7 1, 946 1.3 13 Cerebral vascular lesions- -- - - (330-4) 57.1 87 1.5 294.1 -9 380. 4 1.2 1, 461.8 1, 844 13 13 Cancer of prostate_.....--..---- (177) 22.1 19 9 32. 3 15 74.9 1.0 253. 0 318 1.3 1.0 Accidents, suicides, violence )) 45.0 62 1.4 101.3 17 303.7 11 1, 063. 2 1,310 12 1.3 Nephritis __.......----.------ (592-4) -0 3 |---------- 11.6 15 58.8 Vt 156. 4 173 11 Lb Rheumatic heart disease..._- (400-16) 14.2 18 1.3 48.1 .8 79.4 Li 290.6 309 11 Ll Cancer of rectum... .-.....------ (154 12.0 9 8 41.3 -6 38. 2 17 207.8 213 1.0 9 Cancer of intestines__..._..-.- (152-3) 33.2 13 a4 46.6 L4 106. 2 .8 422. 6 395 9 9 All causes........-.------------------ 799. 4 1, 264 1.58 | 2,420.1 1.65 | 4,183.3 1.63 | 15, 653.9 26, 223 1.68 1.66 The death rates increase with the amount smoked. For groups of men smoking less than 10, 10-19, 20-39, and 40 cigarettes and over per day, respectively, the death rates are about 40 percent, 70 percent, 90 percent and 120 percent higher than for non-smokers. The ratio of the death rates of smokers to that of non-smokers is highest at the earlier ages (40-50) represented in these studies, and declines with increasing age. The same effect appears to hold for the ratio of the death rate of heavy smokers to that of light smokers. In the studies that provided this information, the mortality ratio was substantially higher for men who started to smoke under age 20 than for men who started after age 25. In general, the mortality ratio was increased as the number of years of smoking increased, although the pattern of in- crease was irregular from study to study. In two studies which recorded the degree of inhalation, the mortality ratio for a given amount of smoking was greater for inhalers than for non-inhalers. Cigarette smokers who had stopped smoking prior to enrollment in the study had mortality ratios about 1.4 as against 1.7 for current cigarette smokers. Two studies reported the number of years since smoking was stopped. In these, the mortality ratio declined in general as the number of years of cessation increased. The mortality ratio of ex-cigarette smokers increased with the number of years of smoking and was higher for those who stopped after age 55 than for those who stopped at an earlier age. (These results were available in one study only.) Taken as a whole the seven studies offer a substantial breadth of sampling of the type of men and environmental exposures to be found in North America and Britain, although none of the groups studied was planned as a random sample of the U.S. male population. All the studies had death tates below those of the U.S. white male population in 1960. To some extent this is to be expected, since men in poor health were likely to be under-recruited in these studies. Only a minor part of these differences in death rates can be attributed to a failure to trace all deaths or to higher death rates among non-respondents in these studies. The data on smoking status and on amount smoked were subject to errors of measurement, particularly since smoking status was measured only once and some men presumably changed their status after entry into the study. For men designated as current smokers of cigarettes only, our judgment is that the net effect of such errors of measurement is to make the observed mortality ratios relative to non-smokers underestimates of the true mortality ratios. The studies suffered from a failure to obtain substantial portions of the study populations selected for investigation. For a non-response rate of 32 percent in the prospective studies, calculations based on the available information about the non-respondents indicate that reported mortality tatios lying between 1 and 2 might overestimate the corresponding figure for the complete study population by 0.2 or 0.3. In our judgment these biases can account for only a part of the elevation in mortality ratios found for cigarette smokers (see Appendix I). Tn three studies in which the data could be ‘subdivided by size of city, the mortality ratios differed little in the four sizes of communities studied. 714-429 0-649 111 In one study numerous other variables that might influence the death Tate, such as longevity of parents and grandparents, use of alcohol, occupational] exposure and educational level, were recorded. Adjustment for each of these variables individually produced little change in the mortality ratiog. Although similar information from other studies would have been wel. come, it is our judgment that the mortality ratios are unlikely to be explained by such environmental, social class, or ethnic differences between cigarette smokers and non-smokers. Except for the analyses reported above by longevity of parents and grand. parents and by previous serious disease, no direct information is available on whether there are basic constitutional differences between cigarette smoker, and non-smokers that would affect their longevity. As described elsewhere in this report, differences have been found between cigarette smokers and non-smokers on certain psychological and behavioral variables. However, even for these variables the distributions for cigarette smokers and non. smokers show considerable overlap. It seems a reasonable opinion that the same situation would apply to the constitutional hardiness of cigarette smokers and non-smokers, if it were possible to measure such a variable. This implies that constitutional differences, if they exist, are likely to express themselves in only a moderate difference in death rates. Cigar Smokers Death rates are about the same as those of non-smokers for men smoking less than five cigars daily. For men smoking five or more cigars daily, death rates were slightly higher (9 percent to 27 percent) than for non. smokers in the four studies that gave this information. There is some indi- cation that this higher death rate occurs primarily in men who have been smoking for more than 30 years and in men who stated they inhaled the smoke to some degree. Death rates for ex-cigar smokers were higher than those for current smokers in all four studies in which this comparison could be made. Pipe Smokers Death rates for current pipe smokers were little if at all higher than for non-smokers, even with men smoking 10 or more pipefuls per day and with men who had smoked pipes for more than 30 years. Ex-pipe smokers, on the other hand, showed higher death rates than both non-smokers and current smokers in four out of five studies. The epi- demiological studies on ex-cigar and ex-pipe smokers are inadequate to explain this puzzling phenomenon. According to Hammond and Horn (10) and Dorn (6) the explanation may be that a substantial number of cigar and pipe smokers stop smoking because of illness. Morratity BY Cause oF DEATH In the combined results from these seven studies, the mortality ratio of cigarette smokers was particularly high for a number of diseases: cancer of 112 the hing (10.8), bronchitis and emphysema (6.1), cancer of the larynx (5.4), oral cancer (4.1), cancer of the esophagus (3.4), stomach and duodenal ulcers (2.8), and the rubric, 451-468, “other circulatory diseases” (2.6). For coronary artery disease, the mortality ratio was 1.7. There is a further group of diseases, including some of the most important chronic diseases, for which the mortality ratio for cigarette smokers lay between 1.2 and 2. The explanation of the moderate elevations in mor- tality ratios in this large group of causes is not clear. Part may be due to the sources of bias previously mentioned or to some constitutional and genetic difference between cigarette smokers and non-smokers. There is the possibility that cigarette smoking has some general debilitating effect, although no medical evidence that clearly supports this hypothesis can be cited. The substantial number of possibly injurious agents in tobacco and its smoke also may explain the wide diversity in diseases associated with smoking. In all seven studies, coronary artery disease is the chief contributor to the excess number of deaths of cigarette smokers over non-smokers, with lung cancer uniformly in second place. For cigar and pipe smokers combined, the data suggest relatively high mortality ratios for cancers of the mouth, esophagus, larynx and lung, and for cirrhosis of the liver and stomach and duodenal ulcers. These ratios are, however, based on small numbers of deaths. APPENDIX I APPRAISAL OF PossIBLE BrasEs DuE to NON-RESPONSE The non-response rates in the prospective studies were approximately as follows: 15 percent for the California occupational study; 15 percent for the U.S. veterans’ study during the 3-year period 1957-1959 and 32 percent during the 3-year period 1954-1956: 32 percent for the British doctors’ study; and about 44. percent for the California Legion study and the Canadian veterans’ study. In forming a judgment about the size of the bias that may e due to non-response, we have concentrated on a non-response rate of 32 percent, since this represents roughly an average figure for these five studies, The objective is to estimate by how much the mortality ratio for the whole population might differ from that found in the respondents. The only useful information in any detail about the non-respondents comes from the U.S. veterans’ study. Table 27 shows data on death rates in 1958 and 1959 (16). For the present purpose the 1957 respondents will be regarded as a part of the 32 percent of non-respondents to the origina] questionnaire for whom we are fortunate to have some data. Table 27 indicates that the non-respondents in 1954 have higher death rates than respondents for both non-smokers and smokers. For non-smokers the ratio of the death rate of 1957 respondents to 1954 respondents was 1.35 in 113 TABLE 27.—Age-adjusted death rates (per 1,000 person-years) for 1954 respondents, 1957 respondents, and non-respondents in U.S. veterans study Proportion Death rates Groups in population 1958 1959 — Te, Non-smokers_____..-__ 0.17 13. 20 ] 1954 respondents... ---.-------.--+---+--++- {at smokers_.........- ‘51 19. 26 he Non-smokers_-_____._- 04 17. 96 i 97 1957 respondents. __.-.-...---.--.-.----.--- {RR smokers. ._._.._._. .13 22. 67 or Non-respondents.._.._.__._...-.----------- 7. | an 15 21.99 19. 1958 and 1.27 in 1959. For smokers the corresponding figures are 1.18 in 1958 and 1.14 in 1959. If the adjusted death rates in Table 27 are weighted by the proportions of men in the population, it is found that the over-all 1958 death rate for 1954 respondents was 17.77 as compared with 19.05 for the complete study popula. tion. The ratio 19.05/17.77 is 1.07, so that in 1958 the death rate for the study population was 7 percent higher than for the 1954 respondents. In 1959 the corresponding death rates were 17.46 for 1954 respondents and 18.31 for the complete population, the ratio being 1.05. These ratios agree with Doll’s judgment (4) that in the British doctors’ study the death rate in the complete population may exceed that in his 68 percent of respondents by from 5 percent to 10 percent. Comparison of the 1954 and 1957 respondents also suggests that the non- respondents in 1954 contain a higher proportion of smokers than the re. spondents. In the 1954 respondents, non-smokers contributed 183,094 person-years of experience during 1957-1959 as compared with 179,750 person-years for current smokers of cigarettes only, non-smokers represent- ing 50.6 percent of the total of the two groups. Among the 1957 respondents the corresponding figure was 46.8 percent. A further decline may have oc- curred in the non-respondents to the 1957 questionnaire. From these data the following assumptions were made in investigating the non-response bias as it affects the mortality ratio of current smokers of ciga- rettes only. 1. The proportions of the relevant groups in the complete population are as follows: Groups Non- Cigarette Total smokers smokers Non-respondents___.......-.--.----- 0.14 0.18 0. 32 Respondents___.....---------------- 34 34 - 68 Complete population. _____._- - 48 52 1.00 This assumes that in the 68 percent of respondents, non-smokers consti- tute 50 percent of non-smokers plus cigarette smokers, but in the non-re- spondents this figure has dropped to 44 percent. 114 2. The death rate in the complete population is 10 percent higher than in the respondents. 3. One further numerical relationship is needed in order to obtain con- crete results. For this, the computations were made under two different sets of assumptions. The more extreme (3a) is that cigarette smokers have no higher death rates among non-respondents than among respondents. The alternative (3b) is that the death rate of cigarette smokers was 10 percent higher among non-respondents than among respondents. Both sets of assumptions seem more extreme than the indications from the U.S. vet- erans’ study in which, as already noted, the smoker death rates were 18 percent and 14 percent higher among 1957 respondents than among 1954. respondents. For total mortality, the calculations of most interest are those for a mortality ratio of 1.7 among the respondents, since this is the average ratio found in the prospective studies for smokers of cigarettes only. For indi- vidual causes of death, however, the mortality ratios among respondents range from 1 to LO, so that calculations were made for a series of different mortality ratios among respondents. Table 28 illustrates the calculations made on assumptions (3a) and (3b) for a mortality ratio of 1.7 among respondents. TaBLE 28.—Illustration of calculation of non-response bias Assumption (3a) Assumption (3b) Mortality ratios Mortality ratios Non- | Cigarette Non- |Cigarette smokers | smokers smokers ; smokers Non-respondents_..-.- 4 (1. 865) 1.700 | 4 (1. 772) Non-respondents__-_..| # (1, 646) 1.870 | 3 (1. 772) Respondents_..._..__. 1. 000 1.700 | § (1.350) Respondents. _------- 1, 000 1.700 | 4 (1.350) Complete population | § (1. 252)] 6 (1. 700)] 2 (1. 485) Complete population_} § (1. 188)| ® (1. 759)] 2 (1. 485) MRL ? a 36) MLR. eee eee 7 (1. 48) The figures without parentheses in the mortality ratio tables represent the start of the computations. The indexes (#2 etc.) show the order in which other figures are computed. For assumption (3a): (1.350) 1 =[ (0,84) (1.000)-+ (0.34) (1.700) ]/ (0.68) (1.485) 2= (1.1) (1,350 (1.772) 3=[ (1.485) — (0.68) (1.350) ]/ (0.32) (1.865) 12 {1082 (1.772) — (0.18) (1.700)]/ (0.14) (1,252) §=[(0,14) (1.865) + (0.34) (1.000)}/(0.48) 700) $={ (0.18) (1.700)+ (0.34) (1.700) }/(0.62) (1.36) 7=1.700/1.252 Thus, the mortality ratio drops from 1.7 to 1.36 in the complete population under assumption (3a) and to 1.48 under assumption (3b). One conse- quence of assumption (3a) is that the mortality ratio of cigarette smokers among the non-respondents is less than 1. Table 29 shows the results obtained for a range of mortality ratios in the Tespondent population. For the high mortality ratios the assumptions may appear unduly extreme. For instance, under assumption (3a) with mortality ratio 10.0 in the respond- ents, the non-smoker death rate in the non-respondents has to be 3.6 times 115 that in the respondents, although the smoker death rates are assumed the same in respondents and non-respondents. It may be of interest to quote Berkson’s (1) example in the same form (Table 30). TABLE 29.—Mortality ratios in respondents and computed values for the complete population | In complete population Tn respondents (68 percent) Assump- Assump- tion (3a) tion (3b) GF 9 pot pt pe RSSSRRS Neprer RQUSSRR TaBLe 30.—Proportions and death rates for Berkson’s example Proportions Death rates Group Total Non- Smokers Total Non- Smokers smokers smokers Non-respondents._-....-..---- 0. 00494 0. 28360 0. 28854 60. 121 4, 217 5.174 Respondents__._._.....----__- . 19506 - 51640 . T1146 1. 553 2. 332 2 118 Total__..----.-.-..--.-- . 20000 . 80000 1. 00000 3. 000 3, 000 3.000 In their general direction, Berkson’s assumptions are similar to those made in this Appendix, but the differences in death rates between respondents and non-respondents were more extreme in his example. The death rate in the complete population (3.000) was 42 percent higher than the respondent death rate. The non-smoker death rate was over 38 times as high among non. respondents as among respondents (60.121/1.553), whereas among the smokers it was only 1.8 times as high. His calculations referred to the early years of a study, in which the effects of differential entry of ill persons among smokers and non-smokers are likely to be most marked. Further, as we in- terpret his writing, the example was intended as a warning against the type of subtle bias that can arise whenever a study has a high proportion of non- respondents, rather than a claim that this numerical estimate of the bias ac- tually applied to these studies. To summarize, the amounts of non-response in the prospective studies could have produced sizable biases in the estimated mortality ratios. Taking assumption 3b in Table 29, as representing fairly extreme conditions, it appears that a reported mortality ratio between 1 and 2 might overestimate by 0.3, a ratio of 5.0 by 1.0 and a ratio of 10.0 by 3.0. 116 APPENDIX II STABILITY OF Mortatiry Ratios In computing the mortality ratio of a group of smokers to a group of non- smokers, each group is subdivided into age-classes (usually 5-year). For the ith age-class let y,; denote the number of smoker deaths and x, the num- ber of non-smoker deaths. The “expected” number of smoker deaths in the ith class (expected on the assumption that smokers have the same age-specific death rates as non-smokers) is (Person-years for smokers in class i) (Person-years for non-smokers in class i) m= Aix (say) The estimated mortality ratio R is defined as =a R=" (1) SAX summed over the age-classes. In the interpretation of the values of R found in the seven studies, much weight has been given to the consistency of the values from one study to another, on the grounds that if the values of R for a particular cause of death are high in all seven studies, this evidence is more impressive than R values that are high in say, three studies but show no elevation in the remaining four studies. As a consequence, the question whether the value of R in an individual study is significantly above unity, in the technical sense of this term, becomes less important. Nevertheless, an answer to this question is occasionally useful in the analysis. Moreover, for some causes of death the total numbers of deaths, even when all seven studies are combined, are small enough so that a measure of the stability of the combined R is needed. Assumptions In attempting to get some idea of the stability of R without too much com- plexity, the following assumptions will be made. 1. The numbers of deaths y, and x, are distributed as Poisson variables. As Chiang (3) has shown, a more accurate assumption is to regard y,; and x, as binomial numbers of successes. But with causes of death for which the probability of dying in a 5-year age span is very small the Poisson assump- tion, which is slightly conservative, is reasonable. 2. The quantities \; can be regarded as known constants. This is not quite correct. Initially, the A; are the ratios of the numbers of smokers to non-smokers in the age-classes, which can reasonably be regarded as given. In subsequent-years, however, the numbers are depleted by deaths, and the number of deaths is’ a random variable. When death rates are small, how- ever, this assumption should introduce little error. 3. The variates y, and y; are uncorrelated. An error in the age assigned to a death, putting it in the wrong age-class, induces a negative correlation tween y, and y;. The existence of such errors should have no effect on 117 the variance ascribed to Sy; on the assumption of independence. The same remarks apply to the assumption that x, and x; are uncorrelated. 4. The variates x, and yi are uncorrelated. An error in assigning a death to the correct smoking category would induce a negative correlation between x; and y;. Such errors should of course not be allowed to happen, since they vitiate the comparison of the death rates that is the main point of the study, but occasional errors of this type may have occurred. With these assumptions the numerator Zy: of R follows a Poisson distri. bution. The denominator 3A,x, is a linear function of independent Poisson variates, and numerator and denominator are independent of one another, The exact distribution of a ratio of this type has not been worked out. Two approximate methods of obtaining confidence limits for the true mortality ratio R will be given. Confidence limits are_presented rather than the standard error of R because the distribution of R is skew when the numbers of deaths are moderate or sraall, so that the standard error is harder to interpret. The Binomial Approximation If the A; can be regarded as approximately constant (=A, say) then R becomes of the form y/Ax, where y and x are independent Poisson variates, Since Ax then represents the expected number of deaths of the smokers, the quantity 4 is estimated as the ratio of the expected number of smoker deaths to the number of non-smoker deaths. By a well-known result it follows that x/(y+x), the ratio of non-smoker deaths to smoker plus non-smoker deaths, is distributed as a binomial proportion with n=number of trials=y +x p=probability of success=1/(1+ AR) where R is the true mortality ratio. Confidence limits for R are found from those for p. Example. For the study of men in 25 States, the figures for lung cancer for cigar and pipe smokers are as follows: Se ee Non- Smokers smokers Observed | Observed Expected a, a [nna Number of deaths.....--- 16(x) 15{y) 9.71(Ax) Hence, A=9.71/16=0.607 and the binomial ratio is 16/31=0.516. Hald’s (9) table of the 95 percent two-tailed confidence limits of the binomial distribution gives 0.331 and 0.698 as the confidence limits for p. Those for R are given by the relation R=(1—p)/p This yields 0.7 and 3.3 as the 95 percent limits for R. Since the lower limit, 0.7, is less than unity, the estimated R, 1.5, is not significantly above unity. 118 Unfortunately the assumption that A, is constant is not true in these studies. For instance, in the study of men in 25 States A; has the value 3.85 for cigarette smokers aged 45-49 and declines steadily with increasing age to a value of 0.96 for men aged 75-79. For cigar and pipe smokers the fluctuation in y; with age is less drastic but is still noticeable. The Normal Approximation This approach avoids the assumption that the A, are constant, but makes other assumptions that are shaky with small numbers of deaths. If R is the true mortality ratio, the quantity y—Re where e=%A\x, is the expected number of smoker deaths. will follow a distribution that has mean zero. If pi, m,; denote the true means of y; and Xi. respectively, the variance of (y—Re) is 3 (pn; + R?A2m,) The basis of this approximation is to regard the quantity y—Re S(u, + R2\2m ) (2) A/S (yp; +R2A3m,) as normally distributed with zero mean, since yi and x; are regarded, as Previously, as independent Poisson variates. The 95 percent confidence limits for R are then obtained, by a standard device, by setting the absolute value of this quantity equal to 1.96 and solving the resulting quadratic equation for R. Since the #; and the m; are unknown, a further approximation is to substitute y as an estimate of Sw, and SAix; as an estimate of SA2m,. Example. For the example previously discussed the data are as follows: y= 15: e=9.71: 3A2x,=6.059 On squaring (2), the quadratic equation becomes (15—9,71R)?=3.84(15+6.059R?) The roots are found to be 0.7 and 3.4, in good agreement with the limits 0.7 and 3.3 given by the binomial approximation. This agreement is better than will usually be found with small numbers of deaths. The following are 4 comparisons of the confidence limits for cigarette smokers in the same study. Number of deaths 95 percent limits i Mortality i Cause of death Non- Cigarette smokers ratio smokers ! Binomial Norma] observed i Observed | Expected | paneer oftung.__.. 16 399 41, 20 9.7 (5.0,14.5)} (5.0, 21. 4) Taphysema Sees ee see 7 115 15. 31 7.5 (3. 5,18.1)] (4.0, 40.0) meer ofrectum.__. _....-.-| 16 64 38. 42 L7] (1.03.3) (1.0, 3.6) Uenza and pneumonia_____ 29 97 58. OL 17 (1.1, 2.6) (1.1, 2.9) The lower confidence limits agree well, but the upper limit runs higher for the normal approximation. For cigarette smokers the normal method is perhaps more accurate. The binomial method has some advantage in simplicity. 11. 12. 13. 14. 15. 16. 17. REFERENCES . Berkson, J. The statistical study of association between smoking and lung cancer. Proc Staff Meeting, Mayo Clin 30: 319-48, 1955, . Best, E. W. R., Josie, G. H., Walker, C. B. A Canadian study of mor. tality in relation to smoking habits, a preliminary report. Canad J Pub Health 52: 99-106, 1961. . Chiang, C. L. Standard error of the age-adjusted death rate. Vital statistics. U.S. Department of Health, Education and Welfare. Special Reports No 47, 275-85, 1961. - Doll, R. Personal communication to the Surgeon General’s Advisory Committee on Smoking and Health. . Doll, R., Hill, A. B. Lung cancer and other causes of death in relation tosmoking. Brit Med J 2: 1071-81, 1956. . Dorn, H. F. The mortality of smokers and non-smokers. Proc Soe Stat Sect Amer Stat Assn 34-71, 1958. - Dunn, J. E., Jr., Linden, G., Breslow, L. Lung cancer mortality ex. perience of men in certain occupations in California. Amer J Pub Health 50: 14.75~87, 1960. - Dunn, J. E., Jr., Buell, P., Breslow, L. California State Department of Public Health. Special report to the Surgeon General’s Advisory Committee on Smoking and Health. . Hald, A. Statistical tables and formulas. Wiley, New York, 1952. . Hammond, E. C., Horn, D. Smoking and death rates—report on forty- four months of follow-up on 187,783 men. Part I. Total mortality, Part II. Death rates by cause. JAMA 166: 1159-72, 1294-1308, 1958. Hammond, E. C. Special report to the Surgeon General’s Advisory Committee on Smoking and Health. Hammond, E. C. Special report to the Surgeon General’s Advisory Committee on Smoking and Health. Hammond, E. C. The effects of smoking. Sci Amer 207: 3-15, 1962. Ipsen, J., Pfaelzer, A. Special report to the Surgeon General’s Advisory Committee on Smoking and Health. Korteweg, R. The significance of selection in prospective investiga. tions into an association between smoking and lung cancer. Brit J Cancer 10: 282-91, 1956. Krueger, D. Personal communication to the Surgeon General’s Advis- ory Committee on Smoking and Health. Yates, F. Sampling methods for censuses and surveys. Griffin, Lon- don, (Section 9.4) , 1960. 120 Chapter 9 Cancer Contents CANCER MORBIDITY AND MORTALITY ....... Sources of Information . 2... 1... ee ee ee Sex Ratio ©. 2 2 we ee ee Geographic Variation . . 2... 1. we eee ee Urban-Rural Gradients... 2.2... 1 we ee ee Income Class. 2 1 1. ee ee et Occupation. ©... 6. ee ee Ethnic Group. . 2... 6. ee ee Trends. 2... 1 ee ee Age-Specific Mortality From Lung Cancer . . . 2... Effects of Changes in Lung Cancer Diagnosis on Time Trends. 2. ee ee ee CARCINOGENESIS ...............208. Fundamental Problems in Carcinogenesis in Relation to Induction of Neoplastic Changes in Man by Tobacco Smoke... 2. we ee Threshold 2... 1... 2 ew ee ee ee te es Carcinogenicity of Tobacco and Tobacco Smoke in Animals. . Skin. 2. ee ee Mechanism of the Carcinogenicity of Tobacco Smoke Condensate... 2. 1 wee ee ee ee Other Materials of Possible Importance in Carcinogen- icity. 2 ee he es Pesticides. ©. 2 2 6 ee ee Lactones . 6. 1 ee ee Radioactive Components... 2... . ee eee Summary. 2... 1 ee ee te et Carcinogenesis in Man. . 2. 2 2. 2 2 ee ee Polycyclic Aromatic Hydrocarbons .......... Industrial Products . 2... 1. ee ee ee Soot. 2... ee Mineral Oils . 2. 2... ee ee et ee Summary... 2.) 1 ee ee ee CANCER BY SITE ..............2.026. Lung Cancer . 2 1 2 1 ee ee ee Historical 2. 6 ee ee Retrospective Studies 2... 1. 2 ee ee Methodologic Variables. . 2... 6... 2 eee. Form of Tobacco Use... 2 2. 2 1. eee eee 122 14] 143 143 144 144 145 145 145 145 146 146 147 147 147 147 148 148 149 149 150 151 155 CANCER BY SITE—Continued Lung Cancer—Continued Retrospective Studies—Continued Amount Smoked ... 2... ...-..2..0..24-., Duration of Smoking. .. 2... 1... 7 ee Age Started Smoking .. 2... 2 2... eee Inhalation. - 2. 2. ee Histologic Type. . 2... 6. eee ee Relative Risk Ratios from Retrospective Studies. . . . Prospective Studies . 2... 2 2. ee ee ee Experimental Pulmonary Carcinogenesis . . ...... Attempts to Induce Lung Cancer with Tobacco and Tobacco Smoke... . 2 1. ee ee eee Summary. 2 6 6 ee ee Susceptibility of Lung of Laboratory Animals to Carcin- ogens 2 wee Polycyclic Aromatic Hydrocarbons ........ Viruses. 6 6 we Possible Industrial Carcinogens . . . . . 2... Summary... 2. ee ee ee Role of Genetic Factors in Cancer of the Lung. . . . . Summary. . 2... 2 ee ee ee Pathology-Morphology. . ........-50200084 Relationship of Smoking to Histopathological Changes in the Tracheobronchial Tree... . 1... ...~. Summary. .......... rr Conclusion 2. 2 1. 6 1 ee ee Typing of Lung Tumors . . 2... 2 2 ee ee Conclusions. 2. 2 6 6 ee Evaluation of the Association between Smoking and Lung Cancer. 2 6 6. ee ee ee Indirect Measure of the Association... . 2... . Direct Measure of the Association . ........ Establishment of Association . . 2... ee Causal Significance of the Association . ..... . The Consistency of the Association. . . ...... The Strength of the Association. . 2... 2... The Specificity of the Association . . ... 1.0... Temporal] Relationship of Associated Variables Coherence of Association. . . 2... 2... ee (1.) Rise in Lung Cancer Mortality ....... (2.) Sex Differential in Mortality ........ (3.) Urban-Rural Differences in Lung Cancer Mor- tality, 2... ee ee (4.) Secio-Economic Differentials in Lung Cancer Mortality ............0.. (5.) The Dose-Response Relationship. . . . .. . (6.) Localization of Cancer in Relation to Type of Smoking... 2... 6... eee ee ee Histopathologic Evidence. . 2... 1) ee ee ee Page 155 158 158 159 159 160 161 165 165 165 166 166 166 166 167 167 167 167 167 172 173 173 174 175 175 179 179 182 182 183 183 185 185 185 185 186 186 187 188 189 123 CANCER BY SITE—Continued Lung Cancer—Continued Constitutional Hypothesis . 2... 2... Genetic Considerations. . 2... ee Epidemiological Considerations .........2.~., (1.) Lung Cancer Mortality . 2. ....2.20202, (2.) Tobacco Tars. 2... .......0048, (3.) Pipe and Cigar Smoking .........2.2, (4.) Ex-Cigarette Smokers... .......2.., Other Etiologic Factors and Confounding Variables . . . (1.) Occupational Hazards .. 2... . 0... (2.) Urbanization, Industralization, and Air Pollution . (3.) Previous Respiratory Infections... ... . . (4.) Other Factors. 2.2... 0. ..20.2.04.4 Conclusions... 2... . ee OralCancer. 2 2 2... Kpidemiological Evidence... 2... .....0 4. Carcinogenesis... 2... 2 ee ee le Pathology ............2..00000008 Evaluation... 2... ee ee Conclusions. . 2 2... ee Laryngeal Cancer . 2... 2.2... 7 Epidemiological Evidence... 2... 2... 2 ee Retrospective Studies . 2.2... ......00.0. Prospective Studies . 2.2... . 2. ..004. Carcinogenesis «2 2... ee Pathology .............2....004. Evaluation of the Evidence... . 2... 0. ee Time Trends... 2... ee Sex Differential in Mortality ........2.202.2. Localization of Lesions... 2... ee Conclusion... 2... ee Esophageal Cancer . 2. 2... 2 eee te ee Epidemiological Evidence... 2... .....2... Retrospective Studies . 2... 2... Prospective Studies . 2. 2... 2... ee ee Carcinogenesis . 2... 1. ee ee Evaluation of Evidence .. 2... ......0.4., Conclusion... 2... ee ee Urinary Bladder Cancer. . 2... 2... ee Epidemiological Evidence... .........20.. Retrospective Studies . 2... 2... we Prospective Studies . 2... 2... 2... Carcinogenesis «2... 1. ee ee ee Evaluation of the Evidence. . 2... 2.0.2.0... Conclusion... 2... Epidemiological Evidence. . 2... 2... ....004. Retrospective Studies . 2... .......040. Prospective Studies . 2.2... 2... 2.1 124 CANCER BY SITE—Continued Stomach Cancer—Continued Carcinogenesis... 6) ee ee ee ee Evaluation of the Evidence. . .......+.--246-. Conclusion . 2 2 1 1 ee eee SUMMARIES AND CONCLUSIONS ...........- 3A. 3B. 10. Lung. 2 6 ce ee Oral Cancer. . 2 6 6 1 ee ee ee Larynx. 6 6 ee ee ee Esophagus . . 2. ee ee ee ee Urimary Bladder. «2 2 2 ee ee es Stomach . 2... 1 1 ee ee ee ee REFERENCES Figures . Mortality from cancer (all sites), U.S. Death Registration Area of 1900, 1900-1960. ©. 2... 2 2. ee ee ee . Age-adjusted mortality rates for cancer—all sites, in 17 countries, 1958-1959 2... 1 1 1 eee ee es Age-adjusted mortality rates for cancer of six sites in six selected countries—males . ... 2.2) ee ee ee Age-adjusted mortality rates for cancer of six sites in six selected countries—females . . . 2... - +e eee . Comparison of age-adjusted mortality rates by sex, United States, 1959-1961, with incidence rates from State regis- tries of New York and Connecticut. . ........ . Trends in age-adjusted mortality rates for cancer by sex— all sites and respiratory system in the United States, 1930-1960. 2 6. ee . Trends in age-adjusted mortality rates for selected cancer sites by sex in the United States, 1930-1960 ...... . Age-adjusted mortality rates for cancer of the lung and bronchus by birth cohort and age at death for males, United States, 1914, 1930-1932, 1939-1941, 1949-1950, 1959-1961. 6 6. ee ee . Age-adjusted mortality rates for cancer of the lung and bronchus by birth cohort and age at death for females, United States, 1914, 1930-1932, 1939-1941, 1949-1950, 1959-1961 . 2. 2 6 we ee ee ee te Crude male death rate for lung cancer in 1950 and per capita consumption of cigarettes in 1930 in various countries . . Percentage of persons who have never smoked, by sex and age, United States, 1955... 6. ee ee ee ee Page 228 228 229 229 229 233 233 234 234 235 235 128 129 130 131 132 136 137 138 139 176