Obesity, alcohol consumption, smoking, and mortality

ELSEVIER

Obesity, Alcohol Consumption, Smoking, and Mortality PO-HUANG CHYOU, PHD, CECIL M. BURCHFIEL, DAN S. SHARP, MD, PHD, BEATRIZ L. RODRIGUEZ, AND ABRAHAM M. Y. NOMURA, MD

PHD,

KATSUHIKO

MD, PHD, J. DAVID

YANO, MD, CURB, MI),

PURPOSE: The goals of this study were to assessprospectively the impact of obesity, alcohol use, and smoking on total mortality and to test the etiologic hypothesis that subjects with two or more of these risk factors may experience an elevated risk of overall mortality. METHODS: Information on body mass index (BMI), alcohol intake, cigarette smoking, and other life-style factors was obtained from a cohort of 8006 Japanese-American men living in Hawaii. They were between 45 and 68 years of age at the initial examination (1965-1968). After 22 years of followup that included nearly 159,000 person-years of observation, 2667 deaths from all causes were identified. RESULTS: There was a significant quadratic (J-shaped) relation between BMI and overall mortality. A weaker J-shaped pattern in risk was also present for the intake of alcohol. A strong positive association was observed with pack-years of cigarette smoking. A synergistic interaction between BMI and alcohol was statistically significant (P = 0.0017). Specifically, men who had the lowest body mass (BMI -(: 2 1.2 1 kg/m*) and drank moderately to heavily ( 3 25 oz/mo) experienced a 63% excess risk (relative risk, 1.63; 95% confidence interval; 1.33 - 1.99) compared to a reference group composed of men who had intermediate body mass (BMI, 21.21- 26.30 kg/m’) and drank occasionally to lightly (0.01 - 24.99 oz/mo). The increase in risk due to the interactive effect of low BMI and high alcohol intake was stronger (and statistically significant) than when each of these risk factors was considered separately (excess risk, 28% and 2%, respectively). There was no significant interaction for BMI and cigarette smoking, for alcohol and cigarette smoking, or for the three factors combined. CONCLUSIONS: The most important finding of this study was that, in addition to confirming that cigarette smoking could shorten life, extreme (high or low) BMI values and high alcohol consumptaon are each potentially harmful to health, but even more so if moderate or heavy drinking is concomitant with low body mass, a possible indicator for low intake of nutrients. Ann Epidemiol 1997; 7:31 J-317. 0 1997 by Elsevier Science Inc. KEY WORDS: Body Mass Index, Alcohol, Cigarette Smoking, Mortality.

INTRODUCTION The potential deleterious health consequences of smoking, drinking, and obesity have been investigated in the past. Many studies have reported a positive linear dose-response relationship between cigarette smoking and mortality (l-4). Several investigators have suggested that heavy drinking increases the risk for mortality, while light-to-moderate drinking decreases the risk (5-7). However, the work of other researchers has not supported this view (8-13). A

From the Marshfield Medical Research Foundation, Marshfield Clinic, Marshfield. WI (P-H.0 the Honolulu Heart Proeram, Kuakini Medical Center, Honolul& HI (i.Y, B.L.R, 1.D.C); Divisionof Clinical Epidemiology, John A. Bums School of Medicine, University of Hawaii at Manoa (B.L.R, J.D.C); Honolulu Epidemiology Research Unit, Epidemiology and Biometry Program, Division of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, National Institute of Health, Bethesda, MD (C.M.B, D.S.S); and the Japan-Hawaii Cancer Study, Kuakini Medical Center. Honolulu, HI (A.M.Y.N). Address reprint requests to: Dr. J. David Curb, Honolulu Heart Program, Kuakini Medical Center, 347 N. Kuakini Street, Honolulu, HI 96817. Received June 26, 1996; accepted February 12, 1997. 0 1997 by Elsevier Science Inc. 655 Avenue of the Americas, New York, NY 10010

number of studies showed that the curve representing the risk of death in relation to obesity (most frequently expressed as body mass index (BMI), body weight, or relative body weight) is usually U-shaped or J-shaped (14-17). Few researchers have examined possible interactive effects of BMI, alcohol, and smoking upon mortality (4, 5, 18). Of these, none showed a statistically significant interaction for BMI and alcohol (5), for BMI and smoking ( 18), for alcohol and smoking (4), or for all three factors (5). However, studies on cancer of the upper aerodigestive tract found a much stronger direct associat,ion when the effects of both drinking and smoking were considered than when each was considered separately (19, 29). A cae-control study in Australia showed that the increased risk of rectal cancer in obese males was modified by beer intake (21). These findings suggest that alcohol, smaking, and obesity might be interrelated and act synergistically to increase the risk for disease morbidity and mortality. Because the separate relationships of obesity, alcohol, and smoking to mortality in middle age is still of substantial medical and public health interest, the present study was 1347s27Y7/97/$17.0@ :yj 1.’ .q: g’!{Y7)()[email protected]

312

Chyou et al. OBESITY, ALCOHOL

CONSUMPTION,

Selected Abbreviations

SMOKING, AND MORTALITY

and Acronyms

BMI = body mass index FEV, = forced expiratory volume in 1 second RR = relative risk

conducted to assessprospectively the association of each of these three specific personal characteristics on total mortality among middle-aged (45-68 years) Japanese-American men in Hawaii who participated in the Honolulu Heart Program. Many questions remain about the interrelationship between obesity, alcohol, and smoking; therefore, analyses are also presented that address the etiologic hypothesis that subjects having two or more of these risk factors may increase their overall mortality risk. MATERIALS

AND METHODS

The subjects for this study are American men of Japanese ancestry, born in the years 1900-1919 and residing on the Hawaiian island of Oahu. They were first identified by the Honolulu Heart Program in 1965 with use of the comprehensive 1942 Selective Service registration files (22). Of the 11,148 identified men, 8006 (71.8%) were interviewed and examined from 1965 to 1968, 180 (1.6%) died before they could be examined, and 2962 (26.6%) did not participate in the program. Analyses presented here utilized the following risk factor information collected at time of examination: age and date of birth, weight, standing height, alcohol intake, cigarette smoking history, serum cholesterol, forced expiratory volume in one second (FEV,), systolic blood pressure, and amount and type of physical activity. BMI was calculated as weight/height* (kg/m2). Alcohol intake was based on the usual monthly intake of beer, wine (including Japanese sake and fortified wines), and spirits (including whiskey, gin, brandy or other liquor) among current drinkers. The factors used to obtain estimates of alcohol (ethanol) content in all beverages consumed were those used in an earlier analysis: 3.7% for beer, 10% for wine, and 38% for spirits (23). Packyears were defined as the number of packs (1 pack = 20 cigarettes) smoked per day times the number of years of smoking. In addition, serum cholesterol values were determined by the Auto Analyzer N-24A method. Spirometry was performed to determine pulmonary function (FEV,) using modified standards of the American Thoracic Society that included back-extrapolation; correction for body temperature, pressure, and saturation with water vapor; and the use of a computer-linked digitizer (24). The physical activity index was calculated from answers to questions concerning work and recreational activities (25). The cohort has been carefully followed for mortality from all causes. Lethal events were ascertained through surveillance of hospital discharge records, newspaper obitu-

AUP Vol. 7. No. 4 Ma? lY97- 31 I-317

aries, and state death certificates. From the time of examination until January 1991, a total of 2667 deaths had been identified. Based on a 19-year follow-up survey of the study subjects since their examination, we found that only 1.3% of the men could not be located on Oahu. Consequently, the surveillance for mortality should be nearly complete. Mortality rates for levels of BMI (approximately quintiles), alcohol (nondrinkers and approximately tertiles of drinkers), and cigarette smoking (never-smokers and approximately tertiles of pack-years for smokers) were calculated by dividing the number of deaths in a specific level of the variable of interest by the total number of personyears contributed by individuals in that level, stratified by age at follow-up (45-54, 55-64, 65-74, and 3 75 years). The rates were then adjusted by the age constant, defined as the ratio of person-years contributed by individuals for each age group to those contributed by the entire analytical cohort. Next, the adjusted rates were summed across the age groups to obtain the age-adjusted rate for each level of BMI, alcohol, and cigarette smoking. Covariate-adjusted relative risks (RR) for mortality associated with BMI, alcohol use, and cigarette smoking were estimated by Cox proportional hazards regression models (26). The dose-response trend (both linear and quadratic) in risk for mortality with increasing level of BMI, alcohol use, and smoking was evaluated by a method utilizing the likelihood ratio test (I’ < 0.05 was considered as statistically significant). Similar analyses were performed using stratification by BMI, alcohol use, and smoking in order to assessthe possible combined effect of these three risk factors on mortality. In addition, tests of statistical significance were performed for interaction, with measuresof BMI, alcohol consumption, and cigarette smoking treated as continuous variables in various Cox regression models. We performed tests for the goodness-of-fit of Cox regression models and the necessary assumptions concerning proportional hazards ratios, and we found that they were met. RESULTS Table 1 shows the age-adjusted total mortality (rate per 1000 person-years) by level of BMI, alcohol intake, and cigarette smoking. The mortality rate was 18.06 for men who had the lowest BMI and 19.43 for men with the highest BMI, whereas rates for men with intermediate BMIs (from 21.21 to 26.30 kg/m*) ranged from 14.88 to 15.11. A similar pattern was also present for total mortality in association with the intake of alcohol. The rate was modestly high for nondrinkers, low for occasional (< 5 oz/mo) or light (524.99 oz/mo) drinkers, and highest for moderate-to-heavy (2 25 oz/mo) drinkers. A positive relationship was observed between cigarette smoking and mortality rate. The rate increased progressively with increasing intensity (packyears) of cigarette smoking.

TA3LE

1. Age-adjusted mortality rates by level of BMI, alcohol use, and smoking

Variable BMI (kg/m’) < 21.21 21.2S-2~.OO 2 3.01-24.611 24.61-26.X 2 16.11 Total Alcoht~l (oz/mo) Nondrinker -< 5 5 -24.99 325 Tot
No. of subjects at risk”

No. of deathsi

No. of PF

1593 1645 1582 1.607 8001

614 490 501 479 582 2666

30811 31465 32842 31723 32127 158.97C

2985 1693 1665 1645 7988

1059 492 486 620 2657

5839”; 345Zh 13808 32018 158,747

2242 1194 2609 1873 7918

569 329 896 8?3 2627

46128 241% 5156’ Ji581 157,427

1574

I’ The number of subjects at risk, the total number of deaths, and the person-years (PY) of follow-up varied
Calculations of risk factor-adjusted relative risk of death from all causes were based on a subject’s BMI, alcohol intake, and cigarette smoking (Table 2). Compared to men with average BMI (23.01-24&O kg/m’), the RR was significantly

TABLE 2. Covariate-adjusted relative risks (RR) for total mortality rates by level of BMI, alcohol use, and smoking Variable

RF

BMI (kg/m’) c 21.21 21.21.-23.00 2 3.01-24.60 24.61-26.10 2 26.31 P for quadratic rrend

1.13” 0.98 1.OO’ 1.01 1.29” 0.0098

1.28” 1.04 1.OP 0.99 1.22” < 0.0001

;a 2j P t;~ linear trend

1.OP 0.89” 0.86” 1.08 < 0.0001

l.O@ 0.94 0.8@ 1.02 0.0168

Smoking (pack-years) NC%W -: 25 15-49.09 3 53 I’ for tinear trend

1.O@ 1.17b’ 1.51” 1.137” < O.oool

1.00’ 1.18” 1.51” 1.78’ < 0.0001

Alcohol (o&no) Nl.mdrinker <.r -rJ i--24.99

RRh

li Adjusted for age, and other variables rncluded in the table. h Adjusted for age, FEVI, physical activity index, seturn cholesterol, systolic blood pressure, and other variabtes included in the table. i Referent group. J Significantly (P < 0.05) different from RR of 1.0.

due

to

(P < 0.05) greater for men who were in the highesr category of BMI, with adjustment for age, alcohol intake, and cigarette smoking. The risk was also higher for men with a BMI of < 21.21 kg/ m2. Furthermore, this curvilinear relationship between BMI and mortality was statistically significant. With respect to consumption of alcohol, chose who drank 5-24.99 ounces per month had a significantly reduced risk of mortality compared with nondrinkers after taking into account the effect of age, BMI, BM12, and c@rette smoking. On the orher hand, men who drank 2 25 ounces of alcohol per month experienced an increased risk compared with nondrinkers, but it was not statistically significant. AIthough a test for the overall positive association of alcohol intake with mortality was statistically significant, there did not appear to be a linear trend in risk. Furthermore, a statistical test for the nonlinear relationship between alcohol and mortality showed lack of significance. Table 2 (column 1) also shows that smokers had a greater risk of death than never-smokers with adjustment for age, BMI, BMI”, and alcohol use. In addition, a significanr dose-response trend in risk was present with increasing pack-years of cigarette use. The associations of BMI, alcoitxol, and cigarette smoking with mortality persisted even with further adjustment for other potential confounders including FEV,, physical activity index, serum cholesterol, and systolic blood pressure (Table 2, column 2). Furthermore, when subjects were stratified by age (< 55 years and .a 55 years), the association of BMI, alcohol, and cigarette smoking with mortality persisted for both age groups, except that. alcohot intake was no longer a significant risk factor for mortality among the older age group, with adjustment for rhe above-

314

Chyou et al. OBESITY, ALCOHOL

TABLE 3. Covariate-adjusted intervals) BMI (kg/d < 21.21

21.21-26.30

a 26.31

for total mortality

CONSUMPTION,

relative risks (95% confidence by level of BMI and alcohol used Alcohol (oz/mo)

Nondrinker 1.32 (1.10-1.58) [241/6071b 1.11 (0.97-1.27) [562/1732] 1.38 (1.17-1.64) [256/6461

< 25

3 25

1.22 (1.01-1.47) [197/606] 1.00’

1.63 (1.33-1.99) [174/3801 1.10 (0.94-1.29) [332/977] 1.26 (0.99-1.59) [I 14/2881

[572/20851 1.18 (0.99-1.42) [209/667]

AEP Vol. 7, No. 4 May 19Y7. 311-317

SMOKING, AND MORTALITY

Test for interaction: P = 0.0017

TABLE 4. Covariate-adjusted intervals)

for total mortality

BMI (kg/m’) < 21.21

21.21-26.30

a 26.31+

relative risks (95% confidence by level of BMI and smoking” Smoking (pack-years)

Never

< 50

3 50

1.20 (0.91-1.58) [86/3291b 1.oo’

1.75 (1.46-2.09) [322/866] 1.42 (1.22-1.65) 167612287) 1.78 (1.47-2.16) [227/650]

2.26 (1.84-2.78) [201/390] 1.82 (1.54-2.14) [452/1073] 1.99 (1.62-2.45) [180/4101

[319/1393] 1.28 (1.03-1.59) [164/5201

Test for interaction: P = 0.3591

aAdjusted for age, FEV,, physical activity index, serum cholesterol, systolic blood pressure, and pack-years of cigarette use. h Numbers in brackets are deaths/total subjects at risk. ’ Referent group.

aAdjusted for age, FEV,, physical activity index, serum cholesterol, systolic blood pressure, and alcohol. b Numbers in brackets are deaths/total subjects at risk. ’ Referent group.

mentioned covariates (data not shown). The age boundary of 55 years was selected on the basis of the mean age (mean, 54 years, range, 45-68 years) at initial examination. Although statistically significant, BMI, alcohol and cigarette smoking were not highly correlated with each other. For example, the Pearson’s correlation coefficient was only -0.03 (P = 0.002) between BMI and alcohol, -0.04 (P < 0.001) between BMI and pack-years of cigarette smoking, and 0.20 (P < 0.001) between alcohol and pack-years of cigarette smoking. Similarly, the correlation was -0.19 for FEVi and pack-years of cigarette smoking, 0.14 for BMI and serum cholesterol, and 0.25 for BMI and systolic blood pressure. Table 3 presents the estimated relative risks for mortality in association with the interaction of BMI and alcohol among the study subjects. We selected subjects who consumed an occasional/light amount of alcohol (< 25 oz/mo) and who had an intermediate BMI (21.21-26.30 kg/m2) as the referent group because men with either of these two characteristics had shown a Lower mortality. Compared to these men, the RRs were significantly greater than 1.0 in some, but not all, strata of BMI and alcohol. Of particular note was the observation that subjects who were moderateto-heavy drinkers and whose BMI was < 21.21 kg/m* had the greatest risk. Moreover, a test for the effect of interaction between BMI and alcohol on mortality was strongly sig nificant. Table 4 examines the risk of mortality according to the combined effects of BMI and cigarette smoking. Compared with never-smokers who had intermediate BMI values, the risk was significantly greater for men who smoked, regardless of BMI levels. The risk was also higher among never-smokers who had relatively lower or higher BMI level. Moreover, the greatest risk was observed for subjects who smoked 2

50 pack-years of cigarettes and whose BMI was < 21.21 kg/m’. However, a test for the effect of interaction between these two risk factors on mortality was not statistically significant. Estimates of covariate-adjusted RR for mortality in relation to alcohol consumption and cigarette smoking are presented in Table 5. Within each category of alcohol intake, increased cigarette smoking was associated with an increase in risk. However, the increase in risk due to the interactive effect of alcohol and smoking was not statistically sig nificant. The synergistic effect of BMI, alcohol, and cigarette smoking, considered together, on the risk of mortality was

TABLE 5. Covariate-adjustedrelative risks (95% confidence intervals) for total mortality by level of alcohol use, and smoking” Smoking (pack-years) Alcohol (oz/mo) Nondrinker

< 25

a 25

Never

< 50

a 50

1.04 (0.85-1.26) [308/10901h 1.00’

1.46 (1.21-1.75) [436/1248] 1.32 (1.10-1.58) [496/l 7441 1.66 (1.36-2.02) [287/802]

2.01 (1.66-2.45) [293/612] 1.68 (1.37-2.05) [266/646] 1.85 (1.51-2.26) [271/611]

[203/924, 0.90 (0.62-1.29) [5 712231

Test for interaction: P = 0.0902 a Adjusted for age, FIXI, physical activity index, serum cholesterol, systolic blood pressure, BMI, and BMl*. ’ Numbers in brackets are deaths/total subjects at risk. ‘ Referent group.

AED Vol. 7, No. 4 May 1997 3/l--31;

OBESITY, ALCOHOL CONSUMPTION,

TABLE 6. Covariate-adjustedrelative risks (95% confidence intervals) for total mortality by level of BMI, alcohol use and smoking” BMI (b/m’) Never smokers < 21.21

21.21-26.30

2 26.31

c[ 50 Pack-years < 21.21

21.21.-26.X

3 26.31

a50 Pack-years < 2L.21

21.21-26.30

2 26.31

Alcohol (oz/mo) Nondrinker

-=c25

= 25

1.24 (0.87-1.77) [53/1751” 0.91 (0.71-1.17) [166/656] 1.29 (0.96-1.73) [89/259]

0.96 (0.58-1.59) [24/121] 1.00’

0.94 (0.39-2.30) W301 0.75 (0.45-1.24) [25/128] 1.25 (0.71-2.21)

1.55 (1.17.-2.06) [ll Y/296] 1.43 (1.13-1.80) [227/718] 1.69 (1.26-2.26) [YO/2341

1.41 (1.06-1.86) [114/360] 1.29 (1.04-1.60) [293/1079] 1.46 (1.09-1.95) [89/305]

2.35 (1.76-3.14) [88/209] 1.34 (1.04-1.73) [153/4861 2.33 (1.61-3.37) [46/107]

2.18 (1.59-3.00) [65/129] 1.93 (1.5&2.47) [156/337] 2.30 (1.68-3.16) [72/1461

2.17 (1.54-3.07) [57/l 191 1.48 (1.14-1.90) [145/378] 1.95 (1.41-2.69) [64/149]

2.28 (1.65-3.15) [78/141] 1.91 (1.48-2.45) [150/3571 1.46 (1.00-2.14) 14311131

,128/607] 1.05 (0.73-1.50) [51/196]

P4/651

Test for interaction: P = 0.3698 d Adjusted for age, FEV,, phvsical activity index, serum cholesterol, and systolic blood pressure. h Numbers of deaths/total subjects at nsk Referent group

analyzed (Table 6). For example, among the heaviest drinkers with intermediate BMI, the RRs were 0.75, 1.34, and 1.9 1 for those who never smoked, smoked < 50 pack-years, and smoked 2 50 pack-years, respectfully, compared with men who were never-smokers at the time of examination, drank occasional/light amounts of alcohol, and who had an intermediate BMI. However, the interaction of these three risk factors was not statistically significant after taking into account the effect of age, FEV,, physical activity index, serum cholesterol, and systolic blood pressure.

DISCUSSION Results of the present study strengthen the conclusion from previous reports that obesity, alcohol intake, and cigarette smoking considered separately are risk factors for total mortalitv.

l..h~w 13 ai. SMOKING, AND MOKTALI~I-Y

315

Several researchers have reported a U or J-shaped relation between obesity and mortality (14,15), while others showed only that lean individuals had a higher mortality rate (18). None of these former studies adjusted for cigarette smoking and alcohol. We found that both lean (BMI, < 21.21 kg/m”) and obese (BMI, 26.31+ kg/ml) men had a significantly greater risk of death than those intermediate BMI. The observed association between BMI and mortality persisted after controlling for the independent effects of age, alcohol, cigarette smoking, FEV,, physical activity, serum cholesterol, and systolic blood pressure; this persistence suggests that the association was not due tc confounding by these variables. Some of these factors have heen previously shown to be significantly related to mortality among the study subjects (27, 28). We also found that men with the highest alcohol consumption had a slightly increased risk of death compared with nondrinkers, while those with occasional or light use had a reduced risk. These results are consisrent with findings of some of the previous studies (6, 7). In t-hose studies, an elevated risk was observed among heavy alcohol users, while a decreased risk was noted for light alcohol users. III contrast, other studies could not confirm this protective association (8-10). One of these studies suggested that a reduced risk among light drinkers observed in mosr crf the prospective studies might be due to misclassification 1\1asamong nondrinkers at baseline who became abstainer,: because of symptoms and disease (8). However, studies :&owed that light drinkers had lower mortality than nondrmkers after the effects of history of hypertension and diabetes have hem accounted for ( 11). Of interest were the results of our an.llyses showing a significant combined effect BMI and alcohol on mortality. Compared to occasional/light drinkers w~h intermediate BMI, men who drank 3 25 oz/mo and whc? h,ld the highest BMI experienced a 26% excess risk, hut I-his difference was not statistically significant. In contrast, 1he elevated risk was statistically significant for those who drank 3 25 oz/ mo and had the lowest BMI. The increase (63%) in risk due to the interactive effect of low BMI ;ind high alcohol intake was stronger than when they wc’rc considered separately (excess risk, 28% and 2%, respectively). Ws are not aware of any previous reports that directly demonstrated a significant interaction between BMI and alcohol on mortnlity. A mortality study in Denmark reportsd no significant interaction between BMI and alcohol ( 5 i. Because it is suspected that the effect crfprevalent disease, particularly cancer and smoking-related diseases,on individuals with low BMI could he an imporrant- t,isuc’,we repeated the analyses with exclusion of deaths occurring in the first 5 years of follow-up (data not shown). Thr: effect \>f interaction between AM1 and alcohol on nlh)rtality persisted (P = 0.0023). The biological mechanisms that- c:)ulJ ~~xpl:~nthe syncr-

316

Chyou et al. OBESITY, ALCOHOL

AEI’ Vol. 7, h’o. 4 CONSUMPTION,

SMOKING, AND MORTALITY

gistic effects of low BMI and relatively high consumption of alcohol on mortality are unclear. However, some studies suggested that, compared with nondrinkers, an increase in risk for cancer among heavy drinkers may be due in part to their poor nutritional intake (29). Other studies noted that an increase in risk for mortality in adult males with low BMI may be related to their low intakes and therefore low serum levels of vitamins (18). Thus, it is possible that heavy alcohol intake could enhance its detrimental effects on the health of lean middle-aged individuals whose diet may already be marginal, and hence these subjects may have been at increased risk of mortality as a result of nutritional deficiency and alcohol-related illnesses. Alternatively, it may be that thinness leads to subtle impairment of the body’s immune system (30) and an excessive use of alcohol could worsen the impairment, e.g., maldigestion, malabsorption, and tissue injury (31), and therefore predispose these subjects to early mortality. The mortality data in this study showed that smokers had a significantly greater risk for total mortality than their never-smoker counterparts. Our findings are in agreement with a number of previous studies (l-4). A cohort study that compared mortality by weight and smoking habits among 750,000 US men and women showed that male and female nonsmokers had the lowest risk in all weight categories, while the most obese male and female smokers of 2 20 cigarettes per day experienced the highest risk (RR, 2.21 and 2.73 for men and women, respectively) (14). Another prospective study showed that the effect of BMI on mortality was not modified by smoking (16). Neither of the two studies provided a precise test for interaction between BMI and smoking. Similarly, other investigations have shown that the impact of alcohol consumption on mortality persisted after smoking (6, 7, 11) or BMI (5) was taken into account. Our findings are in accord with those of the previous studies. We found that some of the RRs for mortality were greater in magnitude when stratifying by smoking and BMI or by smoking and alcohol than by smoking only, but tests for interaction between smoking and either BMI or alcohol were not significant. Some previous studies also tested for interaction between smoking and alcohol (32), or between smoking and BMI (18), but found no association with mortality. Few of the previous studies have examined the role of BMI, alcohol, and smoking, considered together, in relation to total mortality. Gronbak and associates reported no significant synergistic effect of these three exposures on mortality (5). We also found that the mortality risk was not strongly elevated by the simultaneous effect of atypical body mass at either extreme and heavy use of both alcohol and packyears of cigarette smoking. In general, characteristics such as BMI, alcohol intake, and cigarette smoking are subject to misclassification (e.g., ex-drinkers and former smokers who quit becauseof disease),

May 1YY7:111-317

vary over time (e.g., men classified as smokers at baseline include those who subsequently quit and those who smoked continuously), and have large inter- and intra-person variation (e.g., change in weight and hence BMI). It is likely that potential misclassification of these factors would underestimate the underlying associations with mortality. However, if one group (e.g., intermediate BMI and moderate drinkers) was more likely than another to change behavior in a positive direction (e.g., exercise more, quit smoking), this misclassification would bias the results towards a finding of lower risk in that particular group relative to others who did not make these beneficial changes. In summary, our data showed that cigarette smoking, alcohol intake, and BMI were each significantly associated with the risk of death from all causes. In addition, the present study indicates that the combination of high alcohol consumption and low BMI may particularly enhance the risk of mortality.

This work was partially supported by contract no. NOl-HC-05102 from the National Heart, Lung, and Blood Institute, and by Grant no. ROl CA 33644 from the National Cancer Institute, National Institute of Health, Bethesda, MD.

REFERENCES 1. Hammond EC, Horn D. Smoking and death rates-Report on fortyfour months of follow-up of 187,783 men. JAMA. 1958;166:1294-1308. 2. Doll R, Peto R. Mortality in relation to smoking: 20 years’ observations on male British doctors. Br Med J. 1976;2:1525-1536. 3. Best EWR, Josie GH, Walker CB. A Canadian studv of mortality in relation to smoking habits: A preliminary report. ‘Can J Public Health. 1961;52:99-106. 4. Rogot E, Murray JL. Smoking and causesof death among U.S. veterans: 16 years of observation. Public Health Rep. 1980;95:213-222. 5. Gronbak M, Deis A, Sorensen TIA, et al. Influence of sex. age. bodv mass index, and smoking on alcohol intake and mortality.’ BMJ. 1994;308:302-306. 6. Doll R, Peto R, Hall E, Wheatley K, Gray R. Mortality in relation to consumption of alcohol: 13 years’ observations on male British doctors. BMJ. 1994;309:911-918. 7. Farchi G, Fidanza F, Mariotti S, Menotti A. Alcohol and mortality in the Italian rural cohorts of the Seven Countries Study. Int J Epidemiol. 1992;21:74-82. 8. Shaper AC. Alcohol and mortality: A review of prospective studies. Br J Addiction. 1990;85:837-847. 9. Gordon T, Kannel WB. Drinking and mortality: The Framingham Study. Am J Epidemiol. 1984;120:97-107. 10. Berberian KM, van Duijn CM, Hoes AW, Valkenburg HA, Hofman A. Alcohol and mortality. Eur J Epidemiol. 1994;10:587-593. 11. Scherr PA, LaCroix AZ, Wallace RB, et al. Light to moderate alcohol consumption and mortality in the elderly. J Am Geriatr Sot. 1992;40:651-657. 12. Cullen K, Stenhouse NS, Weame KL. Alcohol and mortality in the Bussel Study. Int J Epidemiol. 1982;11:67-70. 13. Camacho TC, Kaplan GA, Cohen RD. Alcohol consumption and mortality in Alameda County. J Chronic Dis. 1987;40:229-236.

OBESITY, ALCOHOL CONSUMl’TTION, SMOKING. ih’l’

14. Lew EA, Garfinkel L. Variations m mortality by weight among 750,000 mm and women. J Chnm Dis. 1979;32:563-576. 15. Garfinkel L. i)verweight

and mortality. Cancer. 1986;58:1826-1829.

16. Wienpahl J. Ragland DR, Sidney S. Body mass Index and 15-year mortahty m a cohort of black men and women. J Clin Epidemiol. lY90;43:949-060.

24.

25.

1:. Wannamethee G, Shaper AG. Body weight and mortality in middle age British men: Impact of smoking. BMJ. 1989;299:1497-1502.

26.

18. Garn SM, Hawthorne VM, Pilkington JJ, Pesick SD. Fatness and mortality m the west of Scotland. Am J Clin Nutr. 1983;38:313-319.

27.

19. Vassallo A, Correa P, DeStefani E, et al. Esophageal cancer in Uruguay: A case-control study. JNCI. 1985;75:1005-1009.

28.

20. Kune GA, Kune S, Field B. et ~1. Oral and pharyngeal cancer, diet, smokmg, alcohol, sandserum viramin A and p carotene levels: A casec~mtrol study in men. Nutr Cancer. 1993;20:61-70.

29.

21. Kune GA, Kune S. Field B, Watson LF. Body weight and physical actwry as predictors of colcrrectal cancer risk. Nutr Cancer. 199@;13:9--17.

30.

22. Worth R. Kagan A. Ascertainment or men of Japanese ancestry in Hawaii through World War II Selective Service registration. J Chronic DIL 197@;23:38’1-397. 23. Pollxk

ES, Nomura AMY, Hedbnm LK, Stemmermann GN, Green

31. 32.

i. /ii.)u ~‘f .li t~k%-~-‘P,[iT?

317

SB. Prospective study of alcohol consumptuw 11~1 c.uxcr % Engl ] Med. 1984;110:617-621. Marcus EB, MacLean CJ, Curb JO, Johnson LK, ‘,‘ollmer WM, Huist AS. Reference values for FEV, in Japanese-Ame-l~-at~ men from 45 1~1 68 years of age. Am Rev Respir Dis. 1988;13H:l i’?i-l’~95. Abbott RD. Rodriguez BL, Burchfiel CM, Curb Jl). Physrcal actwxy molder middle-aged men and the reduced risk of xrokr: The flonoltdu Heart Program. Am J Epidemiol. 1993;139:88i--893. Cox DR. Regresslon models and life rabies (with .liscww(~n) J Royal Stat Sot Series [Bj. 1972;34:187-22C. Curb JD, Marcus EB, Reed DM, et al. Smoking. ~~~lrn~war\ functi~m. and mortahty. Ann Epidemiol. 1990;1:25--32. Stemmermann GN, Chyou P-H, Kagan A, N(r aura AM)‘, Yaw K. Serum cholesterol and mortality among Japanese-.~rnericat~ men: The Honolulu (Hawaii) Heart Progmm. Arch Intern Med. 199l;151:9hY--972. Ziegler RG, Morris LE. Blot WJ, Pottern LM, Hoovw K, Fraurnerri JF. Jr. Esophageal cancer among black men in ‘Yhhmgttm, I).(:. II. Role of nutrition. JNCI. 1981:67:1199--l 206 Chandra RK. Immunity and infection. In: J.M. Kmncy, K.N. Jqebhoy. G.L. Hill, O.E. C)wen, eds. Nutrttion and Mrtaholw~~ m I’.vient (:,w. Philadelphia: W. B. Saunders; 1988:598604. Roggin (;M, Iher FL, Kater RMH, et al. Malabscqwn m thi: chron~r alcoholic. Johns Hopkins Med J. 1969;125:32 t--a ii\. Klatsky AL, rZrmstrong MA, Friedman GD. Ak~~hol and mortality. Annals of Internal Medicine. 1992;117:646--651