The Prostate 29:249-260 (I 996) Prostate Cancer and Smoking: A Review of Case-Control and Cohort Studies L.H. Lumey Division of Epidemiology, American Health Foundation, New York BACKGROUND. The purpose of this study was to undertake a systematic analysis of the relation between smoking and prostate cancer. METHODS. All published case-control and cohort studies including relevant data on this topic were collected. The magnitude of potential sources of divergent results among casecontrol studies was estimated comparing current, former, and ever to never smokers, stratified by type of controls (hospital vs. population) and by race. Cohort studies were discussed individually. RESULTS. Neither a clinically nor a statistically sigruficant association between smoking and prostate cancer seems likely, but it cannot be ruled out entirely. The association between other purported risk factors and prostate cancer is not likely to be confounded by the incomplete measurement and control for smoking habits because smoking and prostate cancer are not related. CONCLUSIONS. For a valid assessment of other risk factors in prostate cancer patients the collection of complete smoking information is therefore less critical than in patients with other diseases. For a definitive assessment of the effect of smoking on prostate cancer incidence and mortality however, future studies need to focus on the collection of information on lifetime smoking habits, including tar exposure. 0 1996 Wiley-Liss, Inc. KEY WORDS: prostate cancer risk factors, cigarette smoking, dose response, cumulative tar exposure INTRODUCTION After lung cancer, prostate cancer is the second leading cause of cancer deaths in black and white men in the United States. In addition, the incidence and mortality of prostate cancer are increasing, with an estimated number of 244,000 newly diagnosed cases and over 40,000 deaths in 1995 [l]. Therefore, full attention needs to be given to the etiology and prevention of this type of cancer. The rates for clinically overt prostate cancer increase with age, a family history of the disease, and with living in Western and more developed countries. They are much higher in black compared to white men, and black men in the United States have the highest rates in the world [2-51. Incidence and mortality are low in Asia, but rise sigruficantly among immigrants to Western countries . The low rates in Asia compared to Northern Europe and the United States and the findings from immigrant studies suggest that environmental factors, especially differences 0 1996 Wiley-Liss, Inc. in dietary fat intake, could be an important determinant of prostate cancer disease. Indeed, most retrospective [7-141 and prospective [15-19] epidemiologic studies on dietary factors show a modest (30-5070) increase in risk for subjects with a high, compared to a low level of dietary fat intake. However, differences in fat intake between ethnic groups can explain only part of the differences in prostate cancer . The reported increase in risk could be an underestimation because of the intrinsic difficulties of dietary assessment of past andor present fat intake , and of the narrow range of fat intake  in the societies where these studies were carried out. On the other Received for publication January 26, 1996; accepted January 30, 1996. Address reprint requests to L.H. Lumey, American Health Foundation, Division of Epidemiology, 320 East 43rd Street, New York, NY 10017. ~~~ TABLE 1. Case-Control Studies: Prostate Cancer Prevalence and Smoking setting 1. Selected hospitals nationwide, American Health Foundation, New York City, 1%9-1984 2. Cancer registries, Atlanta, Detroit, New Jersey, 1986-1989 3. Local hospitals, Montreal, Canada, 1979-1985 4. Cancer registry, Alberta, Canada, 1981-1963 Reference Yu et al., 1988  Hayes et al.. 1994 1291 Cases" Controls' W: 989 B161 W 2,791 B 320 Hospital w Matching variables (stratum sizein years) Ever smokers among cased controls OR (95% CI)b (W)" Current smokers Former smokersc Age(5) Hospital Race Interview date(1) W 76/78 B 80/82 1.0 (A-1.2) 0.7 (.4-1.3) .9 (.7-1.1) 1.4 (.7-2.7) 1.2 (.a-1.7) 1.0 (.7-1.4) 1.2 (.9-1.6) 1.1 (.7-1.5) W: 502 B: 479 721 B:594 Population Age(5) Race Region W: 81/77 Siemiatycki et al., 1995 [MI 449 1,266 Hospital Age 87/80 Fincham et 382 625 Population Age m 6 .87 (.61-1.24) 362 358 685 679 Population Age51 57/58 1.28 (.76-2.17) (over 28.6 pack-years) al., 1990 B80/78 Comments No relation with amount (pack-years or CPD) or duration of smoking. No relation with amount smoked. Also 533 population controls, presumably not included for OR estimates. .63 (.61-1.13) No relation with amount (CPD) of smoking. No difference in risk between current and former smokers. Mixed ethniaty of mainly European origin. Relation with duration of smoking (pack-years) in aggressive tumors .99 (.76-1.29) only. 1.0 (.7-1.5) I311 5. Cancer registry and Slattery and pathology labs, West, 1993 Utah,1984-1985 (321 Elghany et al., 1990 Ever smokersC W1 6. Cancer registries, The Netherlands, 1988-1990 Van Der Gulden et al., 1994 1341 345 1346 Hospital WH) No matching, Cases 3 years older than controls. 95/90 7. Regional hospitals, Northern Italy, 1985-1991 Tavani et al., 1994 1351 Talamini et al., 1993 1361 281 599 76/78 271 685 Hospital Matching for hospital only. Cases 7 years older than controls. 8. Cancer registry, L.A. County, 1979-1982 Honda et a]., 1988 1371 216 216 Population Age Neighborhood 80/68 9. Local hospitals, American Health Foundation, New York City, 1965-1967 Wynder et al., 1971  217 200 Hospital No matching. Age of cases and controls not stated. 82/80 2.12 (1.2-3.62) No relation with amount (CPD), duration, or age started smoking. No difference in risk between current smokers and those who quit >25 years ago. In summary, "no causal relation." .78 (54-1.15) .81 (.55-1.20) .91 (.61-1.35) .92 (3-1.58) .76 (.51-1.13) No relation with amount (CPD) or duration of smoking. No relation with age started smoking. 1.9 (1.2-3.0) Vasectomy series: only men under 60 years of age. Retrospective case ascertainment (19% of cases had died). Positive association with duration of smoking. 55% of eligible cases were never interviewed. No relation with amount (CPD) of smoking. OR not adjusted for age. (Continued) Prostate Cancer and Smoking 251 TABLE 1. (Continued) Case-Control Studies: Prostate Cancer Prevalence and Smoking setting 10. Retirement community and tumor registry, L.A. County, 1972-1982 Reference Ross et al., 1987 (121 Ross et al., 1983  11. Local hospital, Tokyo, Japan, 1963-1978 Niijima and Koiso, 1980 1401 Cases" w 142 B 142 w 110 187 Controls' w No matching. cases of "almost identical" age compared to controls. 55/55? 139 139 Hospital Age. interviewer 117 186 (BPH) 110 Hospital Age, hospital, admission date Hospital 15. M.D.Anderson cancer Center, Houston, Texas, 1985-1987 16. Selected hospitals, Japan, 1976 Newell et al., 1989 1441 110 220 Hospital Age(1) Date of diagnosis M i s h i ~et al., 1985 1451 100 Population 100 18. Localhospitals, Minneapolis, St. Paul, 1976 Schumanet al., 1977 147l 40 w 47 B17 Hospital (BPH) 35 Population 43 Hospital Current smokers Former smokers' 60163 ? 269 Hospital B:l8 w Age(l)t race 176 W: 22 (%)a w: 110 Population 12. Roswell Park Kolonel Memorial and Institute, Buffalo, Winkelstein, 1977 141) New York, 1957-1964 13. h l h o s p i t a l s , Schwartzet Paris, France, al., 1%1  1954-onward Oishi et al., 14. Local hospitals, Kyoto, Japan, 1989 1431 1981-1984 Checkoway et al., 1987 (461 controls Age(% race Age(% race 200 OR (95% Cl)b cased 142 B 142 Hospital 17. Local hospital, Chapel W, NC, 1984-1985 Ever smokers among Matching variables (stratum s u e in Yeam) B 48/48 Ever smokers' Comments .9 1.1 No relation with amount (CPD) of smoking. .9 (31 discordant pairs) "No great difference" in smoking between cases and controls. 1.1 (Non ca Controls) 1.0 (ca Controls) No relation with smoking. 79/73 No relation with smoking habits. Current compared to noncurrent, former compared to nonformer smokers. Inverse relation with smoking. 1.36 (.76-2.45) .77 (44-1.35) .59 (A-1.03) 1.41 (.81-2.48) 1318 = 1.63 (matched pairs design) No matching. Age of cases and controls not stated. Age(3). race, admission date 83/87 83/64 (POP) 83/87 (Hasp) 1.71 (33-5.2) Black and white men combined. 2.79 (.%-9.29) (Population controls) .76 (.20-2.88) (Hospital controls) "W, white men; B, black men. bAdjusted for age at diagnosis unless otherwise indicated. Wsk categories vs. never cigarette smokers unless otherwise indicated. hand, it could be an overestimation resulting from inadequate control for smoking in the event that smoking and prostate cancer are related. Because high fat intake and smoking are highly correlated themselves , the latter association could spuri- ously inflate the association with prostate cancer and dietary fat. Whereas some reviews conclude that there is no association between smoking and prostate cancer [2-4,241, an analysis specifically focusing on this re- 252 Lumey lation reported a positive association , as have some case-control and cohort studies. There is a need, therefore, to undertake a systematic analysis of published reports on the relation between smoking and prostate cancer to clanfy this issue. The magnitude of the effect needs to be estimated, and a reasoned assessment as to whether an association should be regarded as causal needs to be made. In addition, several unresolved issues emerging from the present studies need to be identified and resolved. These concern conflicting findings emerging from retrospective (case-control) vs. prospective (cohort) studies, and from case-cohort studies with different (hospital vs. population) types of controls. The effects of publication bias, possibly favoring reports of positive as opposed to negative findings, and the effect of the choice of study populations (white vs. black men) also require quantification. Such a systematic analysis is presented hereunder. MATERIALS AND METHODS Study Materials From previous reviews [2-4,251 and from a Medline search from 1966 onward, we collected all publications which included data on the association between smoking and prostate cancer, regardless of primary study aim. These studies were arranged by study size (number of prostate cancer cases or deaths), separately for case-control and cohort studies. The latter include studies of incidence and of mortality. For case-control studies we listed the setting, the study reference, the number and type of cases and controls (hospital vs. population), the matching criteria for the controls (age, race, hospital), the proportion of ever smokers among cases and controls, and the estimated odds ratio (OR) for various categories of smokers (current, former, ever) vs. never smokers of cigarettes. In addition, we listed relevant comments, especially on the observed relations between duration and amount of smoking with prostate cancer (Table I). For cohort studies we listed the setting, the period when the cohort was assembled, cohort size, the number of incident cases or prostate cancer deaths, the follow-up period, the proportion of current smokers among the deaths or cases, the relative risk of prostate cancer mortality (or incidence) by level of smoking, and we compared former vs. never smokers. We also listed comments on the relation between duration and amount of smoking with prostate cancer (Tables I1 and 111). One relatively small case-control study (included in Table I as reference 43) was not further analyzed because it did not report on the risk of current, former, or ex-smokers relative to never smokers, but instead contrasted current with noncurrent, and former with nonformer smokers. Analytic Methods For case-control studies we plotted the ORs (age adjusted if provided, otherwise not adjusted as calculated from tabular data) for current, former, and ever smokers, by effective example size, in so-called funnel plots . This is a powerful graphical technique to assess the presence of publication bias (i.e., a relative absence of negative findings among the smaller published studies). Effective sample size (ESS) takes the different caseto-control matching ratio between studies into account and denotes the number of cases that would be required for equivalent study power if only one control per case were to be used. It is well known that multiple controls per case can significantly increase study power and hence some form of standardization of study size is required for a) a comparison of studies by size and for b) the optimal weighing of studies in statistical procedures. ESS was calculated inverting the sample size formula for multiple controls per case as given by Schlesselman [271. As an example, the ESS for the Yu et al.  case-control study among 989 white cases and 2,791 white controls would be 1,460, because a study with 1,460 cases and 1,460 controls has the same statistical power to detect a difference in risk factors between cases and controls. We then estimated the magnitude of two potential sources of divergent results among the published case-control studies, namely race (black vs. white men), and the choice of controls (hospital vs. population). Using published tabular data as available, the Mantel-Haenszel summary OR across studies within these categories, comparing current, former, and ever to never smokers, was then calculated. Tabular data were generally not available from smaller studies (ESS less than 200). For cohort studies, it was not possible to calculate a weighted average of the relative risk across studies for lack of data on follow-up time. Therefore, cohort studies will be discussed individually. RESULTS Case-Control Studies We identified 18 studies [28-471 from the literature, with a study size ranging from 40 to over 1,000 cases, as in the study camed out by Yu et al. (28) from the American Health Foundation (Table I). Duplicate reporting had occurred for several studies. Black men Prostate Cancer and Smokinn 253 TABLE II. Cohort Studies: Prostate Cancer Mortality and Smoking Current Cohort size prostate cancer deaths FollowUP (years) 293,916 4.607 2 %?E [@I 3,124 Rogot and Murra 1980 [&I 1,020 Kahn, 1966 582 Assembled Reference 1. US. 1954-1957 veterans, 1917-1940 Hsin et al 1991 748481 .' Setting sdeaths gt:F Relative risk (95% CI) by smoking level' 2 (96) 1 26 31 1.11 .97-1.27 (260) 26 ? 1.0 1.1 826 16 ? 1.22 1.46 568 40 18b .79 1.07 1.24 196 20 .R .80 .97 69 10 6Zb 319 3-5 63 3 1.15 1.23 1.05-1.27 1.09-1.38 (695) (374) 1.2 [511 52 2. Multiple 1972-1976 Risk Factor Intervention Cou al. 1 % ; 348J374 Dollet al., 1994 I541 34,440 4 Former smokers Comments 1.51 1.13 > 99% Whites 1.20-1.90 1.03-1.24 Level 1: 1-9 (78) (817) Level2 10-20 Level 3 21-39 1.6 Level 4 40+ CPD Smoking level as per status at study entry (1954-1957). No relation to age started smoking. Positive assodahon with number of years smoked cigarettes. Cigarette smokers who also took pi dcigadsmokeless , , ; go t excluded from McLaughlin report. Relative risk for current smokers/( e m of follow-up; 2.17/(2.5), 1.7/(8.5),1.31/(16), 1.18/(26). 93% white smokers Level 1 MRFlT initial screening Trial (MWTT) 3. British doctors 1951 Doll and Peto, 1976 I551 Doll and w, 1966 1561 Hammond, 1966 [57l L 4. American 1959-1964 Cancer z2 Level 1: 1-14 Level 2 15-24 Level 3 25t CPD Smoking levels as per status at follow-u No trend in smoking!evels. RR for current smokers: .99 (1994). 1.10 1 440.558 Level 1:Ever smokers 1.01 (131) 25 states 124 Hammond, 1964 [581 <3 71 52 134 5. Sweden, 1963 census sample 6. Hawaii, 196-1964 Japanese men 7. Minnesota, 1966 Lutheran Brotherhood Cohort Carstensen 25,129 et al., 1987 I591 Seversonet 7,999 al., 1969 [l;l 193 4-19 37 174 18-21 37 Hsin et al., 1990 7161 149 20 61 8. U.S. college alumni, entered School 1916-1950 9. ]a an sel)ect;.d health center Whittemore et al., 1985 47,271 107 16-50 Hirfama 1974 [iej 122,261 63 10 Weirand Dunn, 1970 [611 68,153 37 5-8 1%2 17,633 .97 (88) 1.11 (69) 1.1 (26) Age 40-69 Age 70-89 0.8 (31) 0.9 (15) Level 1: 1-7 Level 2: 8-15 Level 3 16+ CPD Level 1: Current smokers 1.7 .8-3.5 1.4 .4-4.4 (3) Level 1: 1-19 Level 2 20-29 Level 3:30+ CPD No dear dose response. 23% of subjects lost to follow-up. RR of ever smokers: 1.8 (1.1-2.9) No association Nested case control study .87 (6) 1.6 .8-3.5 (12) (11) [W 1965 .95 .61 Level 1: Current smokers diShiCtS 10. California, 1954-1957 union members .78 (37) Level 1: Ever smokers (aU forms of tobacco) "Number of deaths in brackets. bAmong entire cohort. were included in only 4 of these 18 studies [12,28,29, 461. There was a fair mix of hospital and population controls, and matching of cases with multiple controls (generally three or less) was common. In only two studies [34,46]were patients with benign prostate hypertrophy (BPH) used as controls. Matching variables typically included age (within 1-5 years), race, and date of diagnosis or a proxy thereof. The 254 Lumev TABLE 111. Cohort Studies: Prostate Cancer Incidence and Smoking Current smokers Cohort Setting Assembled 1969 1. Third U.S. National Cancer SWeY 2. Kaispr 1978-1985 Permanente, Northern CA 1976 3. Seventh Reference Relative risk (95% CI) by smoking level' Follow- among UD deaths (46) 1 2 3 .!a deaths (years) 7,518 257 Cross section ? .74 .74 Hiatt et al., 1% 1631 43,432 238 4.6 27 1 .o .6-1.6 (24) 1.9 1.2-3.1 (2.5) Millset al., 1989 [I51 14,000 180 6 ? .49 (3) Williams and Horn, 1977 size Pr'r.",:' Former smokers Comments Level 1: < 20 Level 2: 21-39 Level 3 40+ CPD [621 FLentists. 1.1 .8-1.5 (94) 23%blackmen Level 1: C = 20 CF'D Level 2: > 20 CPD 1.24 ,914.67 (79) Level 1: current smokers (1989 study) Level 1: 1-14 CA Level 2: 15+ CPD (1992 study, relates to ex-smokers only) Nonsienificant trend for amous and duration of smoking. Also, nonsi Scant relation (OR E 4 ) with duration (GE 10 years) of smoking. No association Mills and Beeson, 1992 [MI 4. us. college alumni, entered school 1916-1950 5. Retirement community and tumoi 1%2 Whittemore et al., 1984 47,271 243 (includes 107 deaths) 16-50 ? Ross et al., 5,106 138 3-7 41 161 1981-1985 .9 .7 Level 1: 1-10 Level 2: 11-20 Level 3 21+ CPD .7 1990 1-56] 5%County "Numberof cases in parentheses. proportion of ever smokers among the cases ranged from 48 to 95%. A funnel plot  of the ORs for ever smokers vs. never smokers by ESS shows that smaller studies generally report a positive association between smoking and prostate cancer. For larger studies, except the study carried out by van der Gulden et al. [MI, there is no such pattern (Fig. 1). This indicates that negative studies showing no association between prostate cancer and ever smoking are underrepresented in the published reports. Funnel plots for current and former vs. never smokers show a similar pattern. The weighted average across studies of the ORs for current (OR: .97), former (OR: .%), and ever (OR: 1.04) vs. never smokers stratified for type of control (hospital vs. population) and race suggests there is no association between smoking and prostate cancer (Table IV). However, studies with hospital controls tend to show a weak inverse, and studies with population controls a weak positive association for current and former compared to never smokers. The ORs are close to unity and statistically nonsignificant for both types of study. These patterns hold in both black and white men. Studies with hospital and population controls also differed with respect to the smoking habits of cases and controls. Compared to population controls, hospital controls were more likely to be current (40 vs. ! I]u7 - ! a 0 1 r .I ' 0 W loo0 1W Effective Sample She (Caws) Fig. 1. Smoking and prostate cancer. ORs for ever w. never smokers by ESS. Individual studies are numbered as in the reference list. Studies 12.28, and 29 have separate entries for black and white men. Study 47 has separate entries for hospital and population controls. 37%) or ever (80 vs. 71%) smokers, and less likely to be former (36% vs. 46%) smokers. An unexpected finding was that these patterns also hold for the cases in both groups of study. Smoking is more common in black compared to white men (Table V). Cohort Studies We identified 10 studies [16-18, 48-61] with a prostate cancer mortality follow-up from the litera- Prostate Cancer and Smoking 255 TABLE IV. Case-Control Studies, Summary ORs (95% CI) for Current, Former, and Ever Compared to Never Smokers by Type of Controls by Race* Current Former Ever OR CIS OR CIS OR CIS .94 1.04 .98 (.80-1.10) (37-1.25) (.87-1.11) .88 1.05 .94 (.74-1.04) 1.01 1.09 1.04 (.89-1.15) (.94-1.26) ( .98-1.15) .71 .76 .87 (.41-1.21) (.67-1.37) (.65-1.17) 1.36 1.25 1.28 (.74-2.49) (.88-1.79) (.94-1.73) .87 1.10 1.03 (.80-1.52) (. 78-1.35) .91 1.03 .97 .78-1.07 .87-1.21 .81-1.08 .91 1.10 .99 .77-1.07 .91-1.33 .87-1.12 1.00 1.09 1.04 .88-1.14 .95-1.24 .95-1.14 ~ Whites Hospital controls Population controls All Blacks Hospital controls Population controls All Combined Hospital controls Population controls All (.&4-1.31) ( .82-1.08) (.52-1.45) 95% CI and Summary ORs determinedby the Mantel-Haenszel method. Whites: Five studieswith hospital and six with population controls. Blacks: One study with hospital and one study with population controls. TABLE V. Case-Control Studies Percentage of Smokers (Current, Former, and Ever) for Cases and Controls (Hospital and Population) by Race* Smokers Controls Whites Hospital Population Blacks Hospital Population Combined Hospital Population Cases Whites Hospital Population Blacks Hospital Population Combined Hospital Population Current Former Ever 37 36 39 50 80 70 61 41 21 37 82 78 40 37 36 46 80 71 38 35 38 52 82 71 48 37 32 80 43 80 40 37 49 82 73 35 *Studiesweighed by ESS.Whites: six studies with population and six studies with hospital controls. Blacks: one study with population and one study with hospital controls. ture (Table 11)and 5 studies [15,62-661 with a prostate cancer incidence or prevalence follow-up (Table 111). The best known mortality studies include the U.S. Veterans study assembled between 1954 and 1967 [48-521, the British physicians study assembled in 1951 [54-561, and the American Cancer Society volunteer study assembled between 1959 and 1960 [57,58]. We also identified a report on prostate cancer deaths and smoking among men screened between 1972 and 1976 for the MRFIT cardiovascular disease risk factor intervention trial . The number of prostate cancer deaths ranges from 37 in a Californian Union Members study  to 4,607 in the U.S. Veterans study , and the number of new cases ranged from about 130 to 250. As best we know no significant number of black men was studied except in the Kaiser Permanente morbidity follow-up, which included 23% black men among 238 new cases . In this study results were not broken down by race, however. In several studies a repeated outcome assessment took place. Among the U.S. Veterans [48-521, prostate cancer mortality was assessed after 2 1/2, 8 1/2, 16, and 26 years of follow-up (when 66% of the original cohort was no longer alive), and among the British physicians [54-561 mortality was assessed at 10, 20, and 40 years of follow-up (when 60% of the original cohort was no longer alive). In both studies mortality follow-up was virtually complete. In nearly all cohort studies, excepting the British doctors, study subjects were not contacted at regular intervals, however: subjects’ vital status and cause of death (if applicable) were once ascertained through independent data sources such as insurance records. Therefore, prostate cancer mortality in all studies except the British physicians’ study was related to the subjects’ 256 Lumev smoking status at study entry. Generally, smoking status was compared for one to four levels of amount of cigarettes smoked per day (CPD) relative to never smokers, and for former smokers compared to never smokers. Seven of 10 cohort studies showed no association with amount of smoking (CPD) and prostate cancer mortality. Only the U.S. Veterans study , the h4RFIT screening cohort , and the Lutheran Brotherhood cohort  reported a positive association. As mentioned before, smoking habits at study entry in these three studies were related to cause of death after an extended follow-up period ranging from 16 to 26 years. The risk of prostate cancer mortality in the U.S.Veterans study is marginally elevated with a risk ratio (RR) of 1.18 (1.09-1.28) for current smokers and 1.13 (1.03-1.24) for former smokers. In the MRFIT study the reported RR was 1.3 (1.1-1.5) for the risk of current vs. nonsmokers. In the Lutheran Brotherhood cohort the RR was 1.9 (1.1-3.3) for former and 1.8 (1.1-2.5) for ever smokers. In the U.S.Veterans study there is a steady decrease in the relative risk of former vs. never smokers with increasing follow-up time. For current smokers the RR declined from 2.17 after 2 1/2 years of follow-up to 1.18 after 26 years of follow-up . Except for men smoking one pack a day or more in the Kaiser Permanente follow-up study , with a RR of 1.9 (1.2-3.1) compared to never smokers, no (or negative) associations between smoking and newly diagnosed prostate cancer cases were reported. DISCUSSION From our review of published case-control and cohort studies relating smoking habits to prostate cancer mortality and prevalence, a number of conclusions can be drawn. Because so few studies include black men, our conclusions refer to white men unless otherwise indicated. Published case-control studies show that the association between smoking (current, former, or ever) and prostate cancer is very weak, if at all present. The absence of any association holds in spite of a bias favoring the publication of positive studies. In addition, reports including elevated ORs are usually atypical in one or more respects. As an example, controls for the Dutch cancer registry study (OR = 2.12 for ever smokers)  were selected from patients with BPH. Since there is evidence that BPH is negatively associated with smoking , this choice of controls might well overestimate a smoking-prostate cancer association. The odds for ever smoking are also evaluated (OR = 1.71) in the second study with BPH controls .The positive Dutch study also had the largest proportion of ever smokers among cases (95%)and controls (90%)of all reported studies, and this tends to inflate the OR as nearly all cases and controls are smokers. For these and other reasons the authors of the Dutch study do not interpret the reported positive association as causal. The other study with a reported increased risk (OR = 1.9 for ever smokers) is atypical in that only men under age 60 were included since originally the study was designed to evaluate the relation between vasectomy and prostate cancer . This also explains why cases (of which 19% had died) were ascertained retrospectively, which could be a source of bias of unknown direction and magnitude. Contrary to other suggestions [48,63] we found that case-control studies with population controls are quite common. We documented 11 studies with hospital and 7 studies with population controls. In addition, we have no evidence that the positive results from case-control studies were exclusively due to the use of hospital controls. Whereas the prevalence of smoking is larger among controls in case-control studies with hospital controls, which has been observed previously [25,48,63,67], the apparent decrease in OR for prostate cancer for smokers combining all studies with population controls is only marginal (OR = .91 for current smokers and .91 for former smokers) because the proportion of smokers among the cases in these studies is also higher. Whereas variations in smoking rates among prostate cancer cases in different hospitals can be expected because of hospital-specific patient intake patterns, it is unclear why smoking rates among cases should be associated with the type of case-control study. The only explanation, other than random variation, could be that studies with hospital compared to studies with population controls are carried out in hospitals that have different patient characteristics. This requires further study. Whereas the use of population controls results in slightly increased OR estimates for current (OR = 1.03) and former (OR = 1.10) smokers, these results are still essentially indistinguishable from unity both from a clinical and a statistical perspective. For the evaluation of the association between smoking and prostate cancer, the two types of control both give a null result. Finally, none of the case-control studies showed a consistent relation between amount of smoking (pack years or CPD), duration of smoking, age started smoking, age quit smoking, or any other measure of exposure with prostate cancer. In summary, the absence of any relation between smoking and prostate cancer in nearly all case-control studies, and a plausible explanation for positive as- Prostate Cancer and Smoking sociations reported by the remainder, suggests that the relation, if at all present, is unlikely to be causal in white men. For black men the numbers studied are (too) small, but there may be a suggestion of an elevated OR among former smokers. This could further be explored. In line with the findings from case-control studies, most published cohort studies do not show a relation between smoking and prostate cancer incidence or mortality. There are three cohort studies that do report an association, and they need to be discussed in some detail as they happen to be among the reports with the largest number of cases. In all three studies (U.S.Veterans , MRFIT , and Lutheran Brotherhood ), information on smoking status was only collected at study entry. Mortality among the subjects was then ascertained after a follow-up period ranging from 16 to 26 years. Because smoking status at study entry is used as a proxy for smoking status at death, the study outcomes are likely to be biased for a number of reasons. The first reason is that classification at study entry does not take into account that, in the United States and elsewhere, there has been a dramatic decline in the proportion of smokers from 1950 to 1995. In the U.S.Veterans study, comparing mortality in 1966 and 1982, the proportion of men who reported smoking at study entry in 1954-1957 declined from 56% to 31%. The authors of this study estimate that more than 40% of the smokers may have quit after study entry [a]. In the British physicians study most doctors who smoked in 1951 had ceased to do so in 1991 and the prevalence of smoking between 1951 and 1991 declined from 62 to 18%.This could not be explained by aging of the cohort . (In the British study smoking histories were taken at 10-year intervals and no association was seen between smoking and prostate cancer.) This decline in smoking prevalence over time may also in part account for the apparent decrease in relative risk of current and former smokers in the U.S.Veterans study, from 2.17 (95% confidence interval [95% CI] not available) after 2 112 years of follow-up to 1.18 (1.09-1.38) after 26 years of follow-up (Fig. 2) and for the decline in relative risk over time. This could be expected if the excess risk of smoking were additive and the background risk increases as the cohort ages . The second reason why a comparison of smokers at study entry with subsequent mortality is likely to be biased is that there has been a dramatic change in the composition of the cigarettes smoked, from hightar, high-nicotine, nonfilter cigarettes smoked in the 1950s to low-tar, low-nicotine, filtered cigarettes smoked at present (Fig. 2). It may well be that puff size increased concomitantly, as it is inversely related 257 :I: - 30.0 28.0 - - 32.0 - 26.0 PL 16.0 - 14.0 - 18.0 12.0 10.0 t 2.3 2.1 P :H 1.5 - 1.3 - 1.1 - 0.9 - 0.7 - 0.5 - 1950 25 - l3 22.0 241) 20.0 2.1 lD55 1900 1WS 1970 1975 1900 1o#I a3 2OOO AlldahcbnnbOnUSmulo( Fig. 2. U.S. sales-weighted average tar and nicotine yields from all domestic brands on the U.S. market, 1954-1993. Abbreviations: RT, reconstituted tobacco; F, filter; ET, expanded tobacco. The numbers 85, 100, and I20 in the figure denote cigarette length in millimeters. (Figure reproduced, with permission. from Hoffmann D, Hoffmann I: Tobacco consumption and lung cancer. In Hanten HH (ed):“Advances in Basic and Clinical Research.” Boston: Kluwer Academic Publications, 1995, pp 1-42.) to the average nicotine content of a cigarette [69,70]. The net effect of these changes with respect to the cumulative tar exposure of smokers is hard to evaluate, but is likely to be associated with smoking level at study entry. If so, a major source of bias would be introduced. For the reasons given above, it is also clear that cohort studies relating smoking status in the remote past with prostate cancer mortality are not very relevant for today’s smokers, even if they were unbiased. An additional problem in cohort studies is the loss to follow-up of the study subjects. Whereas nearly all subjects were accounted for in the U.S.Veterans and the MRFlT follow-up studies, the Lutheran Brotherhood study lost 23% of its subjects to followup. The authors report “no signhcant differences” between subjects retained and subjects lost to follow-up but provide no further information . It would be of special interest to know the smoking patterns at study entry, by follow-up status and cause of death, to properly evaluate the results of this study. Additional information on mortality among those lost to follow-up could possibly be collected from other sources, for instance the National Death Index. This would indicate how loss to follow-up is related to smoking status at study entry and cause of death, which in turn permits an assessment of the likelihood of bias. Among cohort studies, current smokers are categorized in many different ways, using a variety of cutpoints for CPD. This leaves too much opportunity for ad hoc groupings after data inspection, and makes comparisons between studies impossible. It is therefore difficult to critically evaluate the reported doseresponse relations of amount of cigarettes smoked and prostate cancer mortality. For instance, in the MRFIT screening cohort , prostate cancer deaths are related only to two levels of smoking, smokers vs. nonsmokers. In the Lutheran Brotherhood study, based on much smaller numbers, an increased relative risk for ever smokers (1.8) and former smokers (1.9) is reported, but no dose response with CPD, categorized at 1-19,20-29, and 30 or more CPD. As a substantial proportion, maybe 40-50% of those smoking at study entry had quit smoking in the 1620-year period of follow-up  this would introduce, as an artifact, a dose-response relation between amount smoked and prostate cancer mortality, provided the heavy smokers at study entry are less likely to subsequently quit than the light smokers. Such differential quitting by exposure level is very plausible, as already noted by others . All things being equal, this might result in an underestimate of the effect of smoking per se, as has been stressed by Hsing et al. . CONCLUSIONS On the basis of data presently available, neither a clinically nor a statistically sigruhcant association between smoking and prostate cancer seems likely, but it cannot be ruled out entirely. In addition, reported positive associations between smoking and prostate cancer are unlikely to be causal. The positive associations reported by a limited number of cohort studies cannot be ignored because of their study size, but their results are hard to interpret because smoking status was only available at study entry, and during the follow-up period (ranging from 16 to 26 years) dramatic changes in smoking behavior and the composition of cigarettes took place. Because of the design used in these studies, the effects of these changes could not be evaluated appropriately. It should be relatively easy, however, to collect additional information on lifetime smoking patterns in several ongoing case-control and cohort studies. This is likely to provide a more definitive answer as to whether smoking and prostate cancer are causally related, and if so, how large the excess risk excess is likely to be. Because smoking is not likely to be re- lated to prostate cancer, the association between other purported risk factors and prostate cancer is not likely to be confounded by the incomplete measurement and control for smoking habits. Therefore, the collection of complete smoking information is less critical for studies of prostate cancer than in patients with other diseases. FUTURE NEEDS For case-control studies, a systematic evaluation of dose-response associations of smoking exposure and prostate cancer in future studies is recommended. These include, for smokers, age started smoking, number of years smoked, CPD, and brand of cigarettes smoked, and for nonsmokers also the number of years since quitting, if applicable. Information should also be collected on the amount and brand of cigarettes smoked and on the inhalation patterns in defined time periods, to allow for an estimate of lifetime tar exposure. The lifetime (or cumulative) tar exposure is calculated by summing over the different brands of cigarettes smoked in defined periods the tar content for each cigarette, the number of days the brand was smoked, and the number of cigarettes of each brand consumed per day . For cohort studies, a standard set of outcomes for reporting is needed. At a minimum these should include the relative risk of those smoking 1-9, 10-20, 21-30, 41 + CPD vs. never smokers, the risk of current (all CPDs aggregated) vs. never smokers, and the risk of former vs. never smokers. In addition, adequate information needs to be collected at study entry and at regular follow-up intervals to allow for an estimate of lifetime tar exposure. ORs and RRs should be provided with point estimates and 95% CIS accurate to the second decimal point, to allow inverse variance weighing of the point estimates. If, in future, case-control and cohort studies also focus on collecting adequate information on lifetime tar exposure it may be possible to obtain better risk estimates of the association between smoking and prostate cancer incidence and mortality. ACKNOWLEDGMENTS Dr.A. W. Hsing kindly provided additional information on the U.S.Veterans study and brought two of the reports on smoking and prostate cancer to my attention. The concept of effective sample size emerged from a fruitful discussion with Dr. E. Zang. Figure 2 on the changing composition of cigarettes over time was provided by Dr. D. Hoffmann. The tables were typed by Pat Lamb. Their help is gratefully acknowledged. Supported in part by Grant CA 17613 from the National Cancer Institute. Prostate Cancer and Smoking NOTE ADDED IN PROOF We identified an additional cohort study from Norway  which was added to the references. The study did not show an excess risk of smoking. REFERENCES 1. Wingo PA, Tong T, Bolden S: Cancer statistics, 1995. CA 45~8-30, 1995. 2. Nomura AMY,Kolonel LN: Prostate cancer: A current perspective. Epidemiol Rev 13:200-227, 1991. 3. Pienta KJ, Esper PS: Risk factors for prostate cancer. Ann Intern Med 118:793-803, 1993. 4. Key T Risk factors for prostate cancer. Cancer Surv 23~63-77,1995. 5. National Cancer Institute: “Cancer Statistics Review, 1973-1988.” Bethesda, MD: National Cancer Institute, NIH Publ. NO. 90-2789, 1990. 6. Kolonel L, Hankin JH, Nomura AMY: Multiethnic studies of diet, nutrition and cancer in Hawaii. In Hayashi Y, Nagao M, Sugimura T, et al., (eds): “Diet, Nutrition and Cancer; Proceedings of the 16th International Symposium of the Princess Takamatsu Research Fund 1985.” Tokyo: Japan Scientific Societies Press, 1986, pp 29-40. 7. Graham S, Haughey B, Marshall J, et al.: Diet in the epidemiology of carcinoma of the prostate gland. J Natl Cancer Inst 70687-692, 1983. 8. Ohno Y, Yoshida 0, Oishi K, et al.: Dietary beta-carotene and cancer of the prostate: A casetontrol study in Kyoto, Japan. Cancer Res 48:1331-1336, 1988. 9. Kaul L, Heshmat MY, Kovi J, et al: The role of diet in prostate cancer. Nutr Cancer 9:123-128, 1987. 10. Slattery ML, Schumacher MC, West DW, et al: Food consumption trends between adolescent and adult years and subsequent risk of prostate cancer. Am J Clin Nutr 52752-757, 1990. 11. Kolonel LN, Yoshizawa CN, Hankin JN: Diet and prostatic cancer: A casetontrol study in Hawaii. Am J Epidemioll27999-1012, 1988. 12. Ross RK, Shimizu H, Paganini-Hill A, Honda G, Henderson BE: Case-control studies of prostate cancer in blacks and whites in southern California. J Natl Cancer Inst 48:869-874, 1987. 13. Talamini R, LaVecchia C, Decarli A, et al.: Nutrition, social factors and prostatic cancer in a Northern Italian population. Br J Cancer 52:817-821, 1986. 14. West DW, Slattery ML, Robison LM, et al.: Adult dietary intake and prostate cancer risk in Utah: A casecontrol study with special emphasis on aggressive tumors. Cancer Caus Control 2:85-94, 1991. 15. Mills PK, Beeson WL,Phillips RL, et al.: Cohort study of diet, lifestyle, and prostate cancer in Adventist men. Cancer 64:598-604, 1989. 16. Hsing AW, McGlaughlin JK, Schuman LM, et al.: Diet, tobacco use, and fatal prostate cancer: Results from the Lutheran Brotherhood cohort study. Cancer Res 50: 6836-6840,1990. 17. Severson RK, Nomura AMY, Grove JS, et al.: A prospective study of demographics, diet and prostate cancer among men of Japanese ancestry in Hawaii. Cancer Res 49:1857-1860, 1989. 18. Hirayama T Epidemiology of prostate cancer with spe- 259 cia1 reference to the role of diet. Natl Cancer Inst Monogr 53:149-155, 1979. 19. Giovannucci E, Rimm EB, Colditz GA, Stampfer MJ, Ascherio A, Chute CC, Willett WC: A prospective study of dietary fat and risk of prostate cancer. J Natl Cancer Inst 85:1571-1579. 1993. 20. Whittemore AS, KolonelLN, Wu AH, et al.: Prostate cancer in relation to diet, physical activity, and body size in blacks, whites and Asians in the United States and Canada. J Natl Cancer Inst 86:652-661, 1995. 21. Margetts BM, Nelson M (eds): “Design Concepts in Nutritional Epidemiology.” New York: Oxford University Press, 1991. 22. Wynder EL, Stellman SD: The over-exposed control group. Am J Epidemiol 135:459-461, 1992. 23. Stellman SD: Chairman’s remarks. In Garfinkel L, Ochs 0, Mushinski M (eds): “Selection, Follow-up and Analysis in Prospective Studies: A Workshop.” Natl Cancer Inst Monogr 67145-147, 1985. 24. International Agency for Research on Cancer (IARC): IARC working group on the evaluation of carcinogenic risk of chemicals to humans. IARC Monogr 38:199-298, 1986. 25. Matzkin H, Soloway MS: Cigarette smoking: A review of possible associations with benign prostatic hyperplasia and prostate cancer. Prostate 22:277-290, 1993. 26. Light RJ, Pillemer DB: ”Summing Up: The Science of Reviewing Research.” Cambridge: Harvard University Press, 1984. 27. Schlesselman JJ: “Case Control Studies.” New York: Oxford University Press, 1982. 28. Yu H, Hams RE, Wynder EL: Case-control study of prostate cancer and socioeconomic factors. Prostate 13: 317-325, 1988. 29. Hayes RB, Pottern LM, Swanson GM, Liff JM, Schoenberg JB, Greenberg RS, Schwartz AG, Brown LM, Silverman DT, Hoover RN:Tobacco use and prostate cancer in blacks and whites in the United States. Cancer Caus Control 5221-226, 1994. 30. Siemiatycki J, Krewski D, Franco E, Kaiserman M: Associations between cigarette smoking and each of 21 types of cancer: A multi-site case-control study. Int J Epidemiol24:504-514, 1995. 31. Fincham SM, Hill GB, Hanson J, Wijayasinghe C: Epidemiology of prostatic cancer: A case-control study. Prostate 17189-206, 1990. 32. Slattery ML, West DW: Smoking, alcohol, coffee, tea, caffeine and theobromine: risk of prostate cancer in Utah (United States). Cancer Caus Control 4:559-563, 1993. 33. Elghany NA, Schumacher MC, Slattery ML, West DW, Lee JS: Occupation, cadmium exposure, and prostate cancer. Epidemiology 1:107-115, 1990. 34. van der Gulden JWJ, Verbeek ALM, Kok JJ: Smoking and drinking habits in relation to prostate cancer. Br J Urol 73:382-389, 1994. 35. Tavani A, Negri E, Franceschi S, Talamini R, La Vecchia C: Alcohol consumption and risk of prostate cancer. Nutr Cancer 2125-31, 1994. 36. Talamini R, Franceschi S, La Vecchia C, Guarneri S, Negri E: Smoking habits and prostate cancer: A casecontrol study in northern Italy. Prev Med 22:400-408, 1993. 37. Honda GD, Bernstein L, Ross RK, Greenland S, Gerkins V, Henderson BE: Vasectomy, cigarette smoking, 260 Lumey and age at first sexual intercourse as risk factors for prostate cancer in middle-aged men. Br J Cancer 5 7 326-331, 1988. 38. Wynder EL, Mabuchi K, Whitmore WF Jr: Epidemiology of cancer of the prostate. Cancer 28:344-360,1971. 39. Ross RK, Paganini-Hill A, Henderson BE: The etiology of prostate cancer: What does the epidemiology suggest? Prostate 4333-344, 1983. 40. Niijima T, Koiso K. Incidence of prostatic cancer in Japan and Asia. Scand J Urol Nephrol Suppl 55:17-21, 1980. 41. Kolonel L, Winkelstein W Jr: Cadmium and prostatic carcinoma. Lancet 2:566-567, 1977. 42. Schwartz D, Flamant R, Lellouch J, Denoix PF: Results of a French survey on the role of tobacco, particularly inhalation, in different cancer sites. J Natl Cancer Inst 261085-1108,1961. 43. Oishi K, Okada K, Yoshida 0, Yamabe H, Ohno Y, Hayes RB, Schroeder FH: Casecontrol study of prostatic cancer in Kyoto, Japan: Demographic and some lifestyle risk factors. Prostate 14:117-122, 1989. 44. Newell JE, Fueger JJ, Spitz MR, Babaian RJ: A casecontrol study of prostate cancer. Am J Epidemiol 130: 395-398, 1989. 45. Mishina T, Watanabe H, Araki H, Nakao M. Epidemiological study of prostatic cancer by matched-pair analysis. Prostate 6423-436, 1985. 46. Checkoway H, DiFerdinando G, H u h BS, Mickey DD: Medical, life-style, and occupational risk factors for prostate cancer. Prostate 10:79-88, 1987. 47. Schuman LM, Mandel J, Blackard C, Bauer H, Scarlett J, McHugh R Epidemiologic study of prostatic cancer: Preliminary report. Cancer Treat Rep 61:181-186, 1977. 48. Hsing AW, McLaughh JK, Hrubec Z, Blot WJ, Fraumeni JF Jr: Tobacco use and prostate cancer: 26-Year follow-up of US veterans. Am J Epidemiol133:437-441, 1991. 49. McLaughlin JK, Hrubec Z, Blot WJ,Fraumeni JF Jr: Smoking and cancer mortality among U.S.veterans: A 26-year follow-up. Int J Cancer 60:190-195, 1995. 50. Rogot E, Murray JL: Smoking and causes of death among U.S. veterans: 16 Years of observation. Public Health Rep 95:213-222, 1980. 51. Kahn HA: The Dorn study of smoking and mortality among U.S. veterans: Report on eight and one-half years of observation. Natl Cancer Inst Monogr 19:l125,1966. 52. Dorn HF: Tobacco consumption and mortality from cancer and other diseases. Public Health Rep 74:581593, 1959. 53. Coughlin SS, Neaton JD, Sengupta A: Cigarette smoking as a predictor of death from prostate cancer in 348,874 men screened for the Multiple Risk Factor Intervention Trial. Am J Epidemiol 143:1002-1006, 1996. 54. Doll R, Pet0 R, Wheatley K, Grey R, Sutherland I: Mortality in relation to smoking: 40 Years' observations on male British doctors. Br Med J 309:901-911, 1994. 55. Doll R, Pet0 R: Mortality in relation to smoking: 20 Years' observations on male British doctors. Br Med J 2:1525-1536, 1976. 56. Doll R, Hill AB: Mortality of British doctors in relation to smoking: Observations on coronary thrombosis. Natl Cancer Inst Monogr 19:205-268,1966. 57. Hammond EC: Smoking in relation to the death rates of one million men and women. Natl Cancer Inst Monogr 19~127-204, 1966. 58. Hammond E C Smoking in relation to mortality and morbidity. Findings in first thu-ty-four months of follow-up in a prospective study started in 1959. J Natl Cancer Inst 32:1161-1188, 1964. 59. Carstensen JM, Pershagen G, Eklund G: Mortality in relation to cigarette and pipe smoking: 16 Years' observation of 25,000 Swedish men. J Epidemiol Community Health 41:166-172, 1987. 60. Whittemore AS, Paffenbarger Jr Rs, Anderson K, Lee JE: Early precursors of site-specific cancers in college men and women. J Natl Cancer Inst 7443-51,1985. 61. Weir JM, Dunn Jr JF: Smoking and mortality: A prospective study. Cancer 25:105-112, 1970. 62. Williams RR, Horm JW: Association of cancer sites with tobacco and alcohol consumption and socioeconomic status of patients: Interview study from the third national cancer survey. J Natl Cancer Inst 58:525-547, . , . I lY//. 63. Matt RA, Armstrong MA,Klatsky AL, Sidney S: Alcohol consumption, smoking, and other risk factors and prostate cancer in a large health plan cohort in California (United States). Cancer Caus Control 5:66-72,1994. 64. Mills PK, Beeson WL: Tobacco use and prostate cancer: 26-Year follow-up of US veterans (Letter). Am J Epidemiol 135~326-327,1992, 65. Whittemore AS, Paffenbarger Jr RS, Anderson K, Lee JE: Early precursors of urogenital cancers in former college men. J Urol 132:1256-1261, 1984. 66. Ross RK, Bernstein L, Paganini-Hill A, Henderson BE: Effects of cigarette smoking on "hormone-related' diseases in a southern California retirement community. In Wald N, Baron J (eds): "Smoking and HormoneRelated Disorders." New York Oxford University Press, 1990, pp 32-54. 67. Olson SH, Kelsey JL, Pearson ThA, Levin B: Characteristics of a hypothetical group of hospital controls for a case-control study. Am J Epidemiol 139:302-310, 1994. 68. Mantel N: Tobacco use and prostate cancer: 26-Year follow-up of US veterans (Letter). Am J Epidemioll35: 327-328, 1992. 69. Muscat JE, Taioli E: Tobacco use and prostate cancer (Letter). Cancer Res 51367, 1991. 70. Qordevic MJ, Fan J, Ferguson S, Hoffmann D Self regulation of smoking intensity. Smoke yields of the lownicotine, low-'tar' cigarettes. Carcinogenesis 16:20152021, 1995. 71. Tverdal A, Thelle D, Stensvold I, Leten P, Bjartveit K Mortality in relation to smoking history: 13 years' follow-up of 68,OOO Norwegian men and women 35-49 years. J Clin Epidemiol46:475-487, 1993.