- Jian-Min Yuan, research fellowa,
- Ronald K Ross, professora,
- Yu-Tang Gao, professorb,
- Brian E Henderson, professora,
- Mimi C Yu, professora
- aDepartment of Preventive Medicine, USC/Norris Comprehensive Cancer Center, University of Southern California School of Medicine, Los Angeles, California 90033, USA
- bDepartment of Epidemiology, Shanghai Cancer Institute, Shanghai 200032, People's Republic of China,
- Correspondence to: Dr Yuan
- Accepted 16 October 1996
Objective: To assess the risk of death associated with various patterns of alcohol intake.
Design: Prospective study of mortality in relation to alcohol consumption at recruitment, with active annual follow up.
Setting: Four small, geographically defined communities in Shanghai, China.
Subjects: 18 244 men aged 45-64 years enrolled in a prospective study of diet and cancer during January 1986 to September 1989.
Main outcome measure: All cause mortality.
Results: By 28 February 1995, 1198 deaths (including 498 from cancer, 269 from stroke, and 104 from ischaemic heart disease) had been identified. Compared with lifelong non-drinkers, those who consumed 1-14 drinks a week had a 19% reduction in overall mortality (relative risk 0.81; 95% confidence interval 0.70 to 0.94) after age, level of education, and cigarette smoking were adjusted for. This protective effect was not restricted to any specific type of alcoholic drink. Although light to moderate drinking (28 or fewer drinks per week) was associated with a 36% reduction in death from ischaemic heart disease (0.64; 0.41 to 0.998), it had no effect on death from stroke, which is the leading cause of death in this population. As expected, heavy drinking (29 or more drinks per week) was significantly associated with increased risks of death from cancer of the upper aerodigestive tract, hepatic cirrhosis, and stroke.
Conclusions: Regular consumption of small amounts of alcohol is associated with lower overall mortality including death from ischaemic heart disease in middle aged Chinese men. The type of alcoholic drink does not affect this association.
Moderate alcohol consumption is associated with a reduced risk of death from all causes and ischaemic heart disease in Western populations
No data are available for Chinese people, who drink little grape wine and have a low rate of ischaemic heart disease
In this study of middle aged Chinese men regular drinkers of small amounts of alcohol had a 19% lower death rate than non-drinkers
Death rates among moderate drinkers were lower for cancer and non-cancer causes, and the type of alcohol made no difference
Relative to non-drinkers, heavy drinkers had a 30% increased risk of death
In prospective cohort studies conducted in Western populations men and women who drink light to moderate amounts of alcohol have been found to have lower overall death rates than either non-drinkers or heavy drinkers.1 2 3 4 5 6 7 8 9 10 11 The observed protective effect of drinking on overall mortality is largely due to a reduced risk of fatal ischaemic heart disease, which accounts for roughly one third of all deaths in such populations. High consumption of wine has been suggested to be responsible for the low risk of ischaemic heart disease in France,12 and recent findings from a prospective study in Copenhagen support this hypothesis; drinking wine but not beer or spirits was found to be associated with reduced overall mortality in Danish people.13 On the other hand, several studies reported no differential effects of wine and other types of alcoholic beverages on risk of death.11 12 13 14 15 16 17 In a recent review of the relation between specific types of alcoholic drink and risk of coronary heart disease, Rimm et al concluded that there is no evidence that wine offers any more protection against ischaemic heart disease than beer or spirits.18
We examined prospectively the relation between alcohol consumption and mortality in a large cohort of middle aged Chinese men in Shanghai, China. In this cohort, ischaemic heart disease accounts for only about 9% of all deaths19 and grape wine is a minor source of alcohol among those who drink regularly. Therefore, our study population is a valuable one to evaluate further the relation between moderate alcohol intake and mortality.
Subjects and methods
Between January 1986 and September 1989 we invited all eligible male residents of four small, geographically defined communities from a wide area of the city of Shanghai to participate in a prospective, epidemiological study of diet and cancer. The men had to be aged 45-64 years and have no history of cancer. At recruitment we interviewed each subject using a structured questionnaire which included level of education, usual occupation, adult height and usual adult weight, history of tobacco and alcohol use, current diet, and medical history.
We asked each participant whether he had ever drunk alcoholic beverages at least once a week for six months or more. If the answer was yes, he was asked to provide the age at which he started to drink regularly and the usual amount of consumption of beer, wine, and spirits separately. If the subject was a former drinker the age at which he stopped drinking was recorded. One drink was defined as 360 g of beer (12.6 g of ethanol), 103 g of wine (12.3 g of ethanol), or 30 g of spirits (12.9 g of ethanol).20
Details of the follow up procedures for this cohort have been described.21 We routinely reviewed death certificates and cancer reports from the population-based Shanghai Cancer Registry, and all surviving members of the cohort were contacted each year.
For each man, years of follow up were counted from the date of recruitment to 28 February 1995 or the date of death or loss to follow up, whichever occurred first. Overall mortality was calculated according to various categories of alcohol consumption. Because of the small number of former drinkers in the study population, former and current drinkers could not be analysed separately. All rates were adjusted for age by using the person-year distribution of the entire cohort as the internal standard.
We used the Poisson regression method to examine the relation between alcohol consumption at recruitment and subsequent risk of death.22 Relative risks and their corresponding 95% confidence intervals were calculated for various categories of drinkers in reference to lifelong non-drinkers. Two statistical models were used to examine whether the observed relation between alcohol intake and mortality was specific to the type of drink consumed (beer, wine, or spirits). Model one examines the relation between a specific type of alcoholic drink and overall mortality while controlling for the consumption of the other two types (a covariate representing the total ethanol intake for the latter was added to the list of confounding variables). Model two tests for possible association between mortality and a specific type of alcoholic drink after adjusting for total ethanol intake (a covariate representing total ethanol intake was added to the list of confounding variables). Therefore, if the putative U shaped relation between alcohol intake and mortality is independent of type of drink consumed, model one will yield similar U shaped curves for beer, wine, and spirits while model two will show no additional association with beer, wine, or spirits after total alcohol intake has been taken into consideration. The other confounding variables adjusted for in the various regression models were age, cigarette smoking, and level of education.
We used the proportional hazards regression method to test for a curvilinear relation between alcohol intake and total mortality (linear and quadratic terms for number of drinks per week were included in the regression model).23 All quoted P values are two sided. Relative risks with two sided P values under 0.05 were considered to be significantly different from 1.0.
During the three year recruitment period 18 244 men enrolled in the study, and to date only 108 have been lost to follow up. At recruitment, 57% (10 471) of cohort members had never drunk alcoholic beverages regularly, 41% (7390) were current drinkers, and the remaining 2% (383) were former drinkers (table 1). Drinking was closely and positively associated with cigarette smoking and moderately related to age (positively) and level of education (negatively) (data not shown).
As of 28 February 1995, the cohort had been followed up for 122 300 person-years (an average of 6.7 years of follow up per man). During the follow up 1198 of the men had died (980/100 000 person-years). There were 498 deaths from cancer, 269 deaths from stroke, and 104 deaths from ischaemic heart disease. We attempted to review the medical records of all cancer cases and deaths. Sixty per cent (298) of the deaths from cancer were confirmed histopathologically, including 93% (27) of deaths from colorectal cancer, 88% (80) from stomach cancer, 81% (26) from upper aerodigestive tract cancer, 67% (98) from lung cancer, and 13% (13) from liver cancer; 97% of the remaining cancer cases showed raised α fetoprotein concentrations or a positive liver scan, or both, at diagnosis.
Of the men who died from stroke, 85% (229) had been admitted to hospital before dying; no evidence of medical care immediately before death was established for 9% (24) of these patients. The comparable figures, respectively, were 82% (85) and 13% (13) for ischaemic heart disease, 84% (59) and 11% (8) for chronic bronchitis and emphysema, and 94% (33) and 3% (1) for hepatic cirrhosis.
Total mortality by alcohol consumption
Moderate drinkers had a reduced risk of death compared with non-drinkers or heavy drinkers. After age, level of education, and cigarette smoking were adjusted for, the relative risk of death in those drinking 14 or fewer drinks a week compared with lifelong non-drinkers was 0.81 (95% confidence interval 0.70 to 0.94). On the other hand, men who consumed 43 or more drinks a week (that is, six or more drinks a day) had a 30% excess risk of death (1.30; 1.01 to 1.68) compared with lifelong non-drinkers. We tested for a curvilinear (U shaped) relation between alcohol intake and total mortality, and the result was significant (P for quadratic effect =0.001). Further adjustment for dietary factors had minimal effect on the relative risks. When men who had given up drinking or who had potentially life threatening illnesses (diabetes, emphysema, or hepatic cirrhosis) were excluded the relation between alcohol intake and overall mortality did not change (table 2).
Previously, we had found that there were fewer deaths in the first year of follow up than in subsequent years.18 In other words, terminally ill patients were less likely to participate in the cohort study than other eligible subjects. We repeated the analyses after excluding all deaths and person-years of follow up for the 12 months after recruitment. The relation between alcohol consumption and risk of death remained unchanged (table 2). We also examined the effect of length of follow up on the association (≤4 years v >4 years); no difference was detected (data not shown). Because excluding various groups of subjects did not affect our results all subsequent analyses were conducted on the entire cohort.
Table 3 presents the risk of mortality from all causes according to alcohol consumption and cigarette smoking. Within each smoking category, light to moderate drinkers (1-28 drinks a week) experienced a lower risk of death than lifelong non-drinkers or heavy drinkers (≥29 drinks a week). On the other hand, risk of death increased with increasing number of cigarettes smoked a day within each category of alcohol intake. The highest mortality risk was observed among heavy drinkers who also smoked cigarettes regularly.
Cause-specific mortality by alcohol consumption
Table 4 shows the relation between alcohol consumption and risk of death from all cancer and for the five commonest cancers. Cancer accounted for 42% of all deaths. After adjustment for age, level of education, and cigarette smoking, light to moderate drinkers had a non-significant 15% reduction in risk of death from any cancer relative to lifelong non-drinkers. As expected, heavy drinkers had a 3.7-fold increased risk of cancer of the upper aerodigestive tract, mainly oesophageal cancer (22). A non-significant 30-40% increase in risks of death from cancers of the stomach, colon, and rectum was observed in heavy drinkers.
Table 5 presents the association between alcohol intake and mortality from all non-cancer related causes and from the five commonest specific, non-cancer related causes. Stroke was the leading cause of non-cancer deaths in the study population, accounting for 22% of total deaths, while ischaemic heart disease accounted for only 9%. Compared with lifelong non-drinkers, light to moderate drinkers had a significant 20% reduction in mortality from all non-cancer related causes (0.80; 0.67 to 0.94), a 36% reduction in mortality from ischaemic heart disease (0.64; 0.41 to 0.998), and a 33% reduction in mortality from non-cancer related causes other than those specifically listed in table 5 (mainly other forms of heart disease and diabetes). Light to moderate drinking offered no protection from risk of death from stroke, and heavy drinking was associated with a significant 1.7-fold excess in risk of death from stroke (1.12 to 2.44). Adjustment for history of hypertension slightly diminished the effect of heavy drinking on risk of fatal stroke (1.53; 1.04 to 2.25). Heavy drinking was also positively associated with death from hepatic cirrhosis (2.99; 1.12 to 7.94).
We repeated all cause-specific analyses after excluding former drinkers, men with pre-existing serious illnesses, or deaths and person-year contribution from the cohort during the first 12 months after enrollment. The exclusions did not alter any of the relations described above. We also repeated all cause-specific analyses for the five commonest cancer sites after exclusion of cancer deaths without histopathological confirmation and for the five commonest non-cancer related causes after exclusion of deaths lacking evidence of medical care immediately before death. No material changes in results were observed.
Total mortality by type of alcoholic drink
Of the 7773 regular drinkers, 3500 (45%) drank beer (only 18 men reported consuming 29 or more drinks of beer a week), 4341 (56%) drank wine, and 3723 (48%) drank spirits. When we examined the relation between a specific type of alcoholic drink and overall mortality while simultaneously controlling for the total ethanol intake from the other two types as well as age, level of education, and smoking (model one, see methods for details), moderate drinkers, irrespective of the type of alcoholic drinks consumed, had a reduced risk of death relative to lifelong non-drinkers (table 6). Similarly, heavy drinkers of either wine or spirits showed a roughly 20% increase in mortality risk relative to non-drinkers. When we used model two (see methods) to test for possible type-specific effects on total mortality after accounting for total ethanol intake, the results were consistent with those from model one. We noted no additional effect from beer, wine, or spirits after total ethanol intake has been accounted for.
We found that Chinese men who consumed no more than 14 drinks a week had a significant 19% reduction in overall mortality relative to lifelong non-drinkers. The observed association was not explicable by confounding factors such as cigarette smoking, dietary intake, level of education, and age. In agreement with findings in Western populations,1 2 3 4 5 6 16 17 we observed a significant 36% reduction in mortality from ischaemic heart disease among light to moderate drinkers relative to lifelong non-drinkers. However, the protective effect of moderate drinking on total mortality in Shanghai men was not restricted to ischaemic heart disease, which accounts for only 9% of total deaths.
Bias is unlikely to explain the observed associations. The prospective study design precluded recall bias. A structured questionnaire was used to administer the interviews to ensure that all subjects were asked identical questions. We were also able to separate former drinkers from lifelong non-drinkers and thus eliminated the potential bias that former drinkers might have an increased risk of death and account for the raised mortality.24 Another proposed argument to explain the relatively high mortality among abstainers is that such individuals carry a greater “burden of ill health” due to lifelong health problems than light to moderate drinkers, regardless of their previous drinking status.25 We excluded all subjects with a history of potentially life threatening illnesses and no change in the results was observed. Our results also were unaffected by duration of follow up, supporting the notion that other unidentified bias is not likely to be responsible for the observed association.
There is evidence that the collected information is reasonably reliable. Diseases reported to be positively associated with heavy alcohol intake in Western populations (upper aerodigestive cancer, liver cirrhosis, and injury/accident) showed similar associations in our study. It is also important to note that loss to follow up was negligible (only 108 out of 18 244 subjects). Finally, it is unlikely that the observed associations between cause-specific mortality and alcohol intake were due to misclassification of cause of death. In 82% of cases cause of death was medically confirmed immediately before death, and exclusion of those without such confirmation did not affect the association between light to moderate alcohol intake and cause-specific mortality.
An ecological study suggested that high consumption of wine might be responsible for the lower risk of ischaemic heart disease in France.12 Recent findings from a prospective study in Copenhagen have shown that wine drinkers had a 50% reduction in mortality relative to non-drinkers.13 In the same study, beer drinking was not related to mortality and spirit consumption increased the risk.13 However, several studies have reported no differential effects of wine relative to other types of alcoholic beverages on mortality.11 14 15 16 17 Rimm et al recently concluded that much of the benefit is from alcohol rather than other components of each type of drink.18 We did not separate the grape wine from other wines drunk by Chinese, but data from a case-control study of nasopharyngeal carcinoma that we conducted in Shanghai in the late 1980s indicate that few Chinese men drink grape wine regularly. Among male controls in that study, only 6% of wine drinkers drank grape wine; the rest of them drank only rice wine (unpublished data). Our data did not show that intake of wine was more beneficial in reducing risk of death than consumption of beer or spirits.
In summary, light drinking was associated with a 19% reduction in overall mortality in middle aged men in Shanghai. This protective effect was not restricted to any specific type of alcoholic drink. There was also a 36% reduction in risk of death from ischaemic heart disease among light to moderate drinkers and significant increases in risks of cancer of the upper aerodigestive tract, stroke, and hepatic cirrhosis among heavy drinkers.
We thank X-L Wang, Y-L Zhang, and J-R Cheng of the Shanghai Cancer Institute for their help in data collection and management, the staff at the Shanghai Cancer Registry for their help in verifying cancer diagnosis, and Kazuko Arakawa of the University of Southern California for her help in data management and analysis.
Funding: Grants R01 CA43092 and R35 CA53890 from the National Cancer Institute, Bethesda, Maryland, USA.
Conflict of interest: None.