Objectives Using common alcohol consumption categories, to conduct a pooled analysis of six ongoing large-scale cohort studies in Japan in order to produce concrete estimates of the quantitative contribution of alcohol consumption to all-cause and major causes of mortality in the Japanese population.
Methods Of the 309 082 subjects, there were 35 801 deaths during 3 832 285 person-years of follow-up. Using a random-effect model, we conducted a meta-analysis of the HRs of each alcohol consumption category in each study, thereby obtaining pooled estimates for the risk of total and major causes of mortality due to alcohol consumption.
Results There was a J- or U-shaped association for the risk of total and major causes of mortality in men, and the risk of total and heart disease mortality in women. Compared with non-drinkers, there was a significantly lower risk for total mortality at an alcohol consumption level of <69 g/day, cancer mortality at <46 g/day, heart disease mortality at <69 g/day and cerebrovascular disease mortality at <46 g/day in men, and for total mortality at <23 g/day in women. In addition, mortality risk increased linearly with rising alcohol dose among drinkers. It was estimated that 5% of total mortality, 3% of cancer mortality, 2% of heart disease mortality and 9% of cerebrovascular disease mortality in men, but only 0–1% of these risks in women, could be prevented by reducing alcohol consumption to <46 g/day in men and <23 g/day in women.
Conclusion Maintaining alcohol consumption below 46 g/day in men and 23 g/day in women appears to minimise the risks of mortality in the Japanese population.
- Pooled analysis
- cohort study
- alcohol consumption
- alcohol & health
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
A number of studies have reported the health benefits of light to moderate alcohol consumption with respect to total mortality,1 2 and the risks of major causes of death such as cardiovascular disease,3 4 although these findings remain controversial.5 6 Heavy alcohol drinking has been found to be positively associated with cancer.7–9 In addition, the net balance of risks and benefits is likely to vary in different populations.5
In Japan, alcohol drinking is now recognised as an important and preventable public health problem. Alcohol consumption and the proportion of heavy drinkers had been increasing for decades until 1990, and have only recently reached a plateau.10 New public health policies should consider both qualitative and quantitative estimation of the effects of alcohol, not only on specific diseases, but also on total and major causes of mortality in the aggregate. However, because published studies use different alcohol consumption categories, meta-analysis for the purpose of quantitative assessment based on common alcohol consumption categories is not possible.11
In the present study, therefore, we conducted a pooled analysis of six ongoing large-scale cohort studies in Japan, using common alcohol consumption categories. The main purpose of this study was to estimate the quantitative contribution of alcohol drinking to total mortality and major causes of mortality (ie, mortality from cancer, heart disease and cerebrovascular disease) in the Japanese population, to permit future estimation of the burden of major diseases attributable to alcohol consumption in Japan.
In 2006, the Research Group for the Development and Evaluation of Cancer Prevention Strategies in Japan initiated a pooling project using original data from major cohort studies to evaluate the association between lifestyle and major forms of cancer in the Japanese population. This project was conducted in parallel with systematic reviews of the relevant literature.11 Topics for the pooled analysis were determined from discussions of the scientific evidence and public health implications.12 13 To ensure the quality and comparability of data, we established inclusion criteria for the present purpose. To be included, the study had to: (1) be a population-based cohort study conducted in Japan; (2) have started in the mid-1980s to mid-1990s; (3) have included more than 30 000 participants; (4) have obtained information on amount of alcohol consumed (g/day) using a validated questionnaire at baseline; and (5) have collected information on any-cause mortality during the follow-up period. Six ongoing studies met these criteria: (1) the Japan Public Health Center-based Prospective Study, Cohort I (JPHC-I),14 (2) the Japan Public Health Center-based Prospective Study, Cohort II (JPHC-II),14 (3) the Japan Collaborative Cohort Study (JACC),15 (4) the Miyagi Cohort Study (MIYAGI),16 (5) the Ohsaki National Health Insurance Cohort Study (OHSAKI)17 and (6) the Takayama Study (TAKAYAMA).18 Four of these studies have already published results on the association between alcohol intake and mortality in each cohort.7 8 19–21 For the present analysis of these cohorts, we used updated datasets with an extended follow-up period. When analysing individual results from each study, subjects with a previous history of any cancer, stroke or myocardial infarction, or with unknown alcohol drinking status were excluded. Table 1 presents selected characteristics of these studies. Each study obtained approval from a relevant institutional ethical review board, namely, that of the National Cancer Center (JPHC-I and JPHC-II), Aichi Medical University (JACC), Tohoku University Graduate School of Medicine (MIYAGI and OSAKI), and Gifu University Graduate School of Medicine (TAKAYAMA).
Subjects were followed from the baseline survey to the last date of follow-up for any cause of mortality in all included studies (table 1). Residence status in each study, including survival, was confirmed by using the residential registry. Information on the cause of death was obtained from death certificates provided by the Ministry of Health, Labour, and Welfare, in which cause of death was defined according to the International Classification of Disease, 10th version (ICD-10).22 Residence and death registration are required by law in Japan and the registries are believed to be complete.
Assessment of outcome
The outcome of the present study was all-cause mortality, including the three major causes of death in Japanese: cancer (ICD-10: C00–C97), heart disease (ICD-10: I20–I52), and cerebrovascular disease (ICD-10: I60–I69).
Assessment of exposure
Alcohol drinking status was assessed by self-administered questionnaires at baseline in each study. Although the style of the questions differed by study, each study calculated alcohol consumption in grams of ethanol per day for regular drinkers by collecting information on the types of beverage, frequency and amount of consumption in their questionnaires. Then, alcohol consumption was divided into categories by using identical cut-points across the studies: non-drinkers (never- and ex-drinker), occasional drinkers (<once/week), and regular drinkers (at least once/week: <23 g/day, 23–<46 g/day, 46–<69 g/day, 69–<92 g/day, ≥92 g/day for men; and <23 g/day, 23–<46 g/day, ≥ 46g/day for women). In Japan, ‘1 go’ is equivalent to approximately 180 ml of Japanese sake (rice wine), or 23 g of ethanol, and is the most common unit for measuring the amount of alcohol consumed. Correlation coefficients comparing alcohol consumption estimated from the questionnaire with dietary records (crude) were 0.77 in men and 0.55 in women from JPHC-I and JPHC-II,19 0.77 in men and 0.71 in women from MIYAGI,23 0.61 in men from OHSAKI,21 and 0.72 in men and 0.64 in women from TAKAYAMA.18 JACC, for which information on the validation of alcohol consumption was not available, used the same questions on alcohol consumption as the MIYAGI study. Ex-drinking and never-drinking non-drinkers were analysed separately in the JPHC-II, JACC, MIYAGI and OHSAKI studies; thus, additional analyses that subdivided non-drinkers into ex-drinkers and never-drinkers were conducted among these subjects.
Information on covariates included in multivariate analyses, such as smoking, body mass index, history of hypertension, history of diabetes, and leisure-time sports or physical exercise, was also obtained from the same questionnaire at the baseline survey in each study.
Person-years of follow-up were calculated starting from the date of the baseline survey in each study until the date of death or end of follow-up, whichever came first. We conducted separate analyses by sex. Each study estimated the HRs and their 2-sided 95% CIs for total cancer associated with each alcohol intake consumption category, by using a Cox proportional hazards model. The studies estimated age (years, continuous)-adjusted and area-adjusted (in JPHC-I, JPHC-II and JACC only) HRs. Additional multivariate adjustments were made for smoking (never smoker, past smoker, current smoker (men: 1–19 cigarettes/day, ≥20 cigarettes/day; women: current)), body mass index (<18.5, 18.5–<25, ≥25), history of hypertension (no, yes), history of diabetes (no, yes), and leisure-time sports or physical exercise (less than almost daily, almost daily), in addition to adjustment by age and area. We conducted further analysis that excluded deaths within 5 years of baseline and, in men, stratified analysis by smoking status (never smokers vs. current smokers). In addition, using four of these cohorts, namely, JPHC-II, JACC, MIYAGI and OHSAKI, we estimated risks in ex-drinkers and never-drinkers. An indicator term for missing data was created for each covariate. SAS V.9.1 and Stata V.11 were used for calculating the estimates.
A random-effects model was used to obtain a single pooled estimate of the HRs from the individual studies for each category.24 The study-specific HRs were weighted by the inverse of the sum of their variance and the estimated between-studies variance component.24 A study that had no cases for a category was not included in the pooled estimate for that category. The trend association was assessed in a similar manner: investigators from each study calculated the regression coefficient per 15 g increase in alcohol intake and its SE. These values from the individual studies were then combined using a random-effects model. We tested for and quantified the heterogeneity of the HRs for the highest category and the trend association of alcohol consumption association among studies by using the I2 statistic. Stata was used for the meta-analysis.
In addition, to express the impact of alcohol drinking on the risk of mortality, the population attributable fraction (PAF) (%) was estimated as pd(HR−1)/HR, where pd is the proportion of cases exposed to the risk factors.25
The present study included six ongoing large-scale population-based prospective cohorts comprising 309 082 subjects (144 012 men and 165 070 women) and 35 801 deaths (22 260 men and 13 541 women), including 13 274 deaths (37%) from cancer (8584 men and 4690 women), 4809 (13%) from heart disease (2831 men and 1978 women) and 4275 (12%) from cerebrovascular disease (2376 men and 1899 women) during 3 832 285 person-years of follow-up (average follow-up period: 12.4 years) (table 1).
Overall, 23% of men were non-drinkers; 44% habitually drank more than 23 g of alcohol per day and 3% more than 92 g. In women, 3% drank more than 23 g/day and 73% were non-drinkers.
In men (table 2), a U-shaped association was found. More specifically, there was a significantly lower all-cause mortality risk among subjects consuming <69 g/day, as compared with non-drinkers; there was no significant association in higher consumption categories. Similar patterns were observed when deaths within 5 years after baseline were excluded from the analysis and in stratified analyses by smoking status. In the analysis comparing ex-drinkers and never-drinkers, the HR of ex-drinkers was significantly higher than that of never-drinkers. Moreover, when compared with never-drinkers, an increase in risk was clearly observed in higher alcohol consumption categories.
In women (table 3), a similar U-shaped pattern was seen. The analysis comparing ex-drinkers and never-drinkers showed that, as compared with never-drinkers, the HR for ex-drinkers was significantly higher.
In men (table 2), a J-shaped pattern was observed, in which there was a significantly lower risk in subjects consuming <46 g/day, as compared with non-drinkers, and a significantly higher risk associated with the highest consumption category. Similar patterns were observed when deaths within 5 years after baseline were excluded from the analysis and in stratified analyses by smoking status. In the subgroup analysis of ex-drinkers and never-drinkers, the HR for ex-drinkers was significantly higher than that of never-drinkers, and the increased risk was clear in the highest alcohol consumption category when compared with that of never-drinkers. In women (table 3), no significant U-shaped pattern was found, except for the significantly increased risk in ex-drinkers.
Heart disease mortality
In men (table 2), heart disease mortality risk displayed a U-shaped pattern. There was a significantly lower risk in subjects consuming <69 g/day, as compared with non-drinkers. Similar patterns were observed when deaths within 5 years after baseline were excluded from the analysis, when analyses were stratified by smoking status, and when the analysis distinguished between ex-drinkers and never-drinkers, the former of which had a significantly higher risk than did the latter. In women (table 3), heart disease mortality followed a J-shaped pattern, with a significantly higher risk in those consuming ≥46 g/day of alcohol.
Cerebrovascular disease mortality
In men (table 2), cerebrovascular disease mortality followed a J-shaped pattern, with a lower risk in subjects consuming <46 g/day, as compared with non-drinkers, and a higher risk in those consuming ≥69 g/day. Similar patterns were observed when deaths within 5 years after baseline were excluded from the analysis, when analyses were stratified by smoking status, and when the analysis distinguished between ex-drinkers and never-drinkers, the former of which had a significantly higher risk than did the latter.
In women (table 3), there was no significant J-shaped association.
Population attributable fraction
Using estimated pooled HRs, we estimated the PAF attributable to alcohol consumption of ≥23 g/day and ≥46 g/day when subjects in these alcohol consumption categories decreased their consumption to <23 g/day and <46 g/day. In calculating these estimates, we assumed that non-drinkers (never- and ex-drinkers) and occasional drinkers (<once/week) did not change their amount of alcohol consumption. The results indicate that, in men, 4.8% of total mortality, 3.1% of cancer mortality, 1.9% of heart disease mortality and 8.6% of cerebrovascular disease mortality could be prevented if current drinkers reduced their alcohol consumption to <23 g/day, and that 4.6% of total mortality, 2.9% of cancer mortality, 1.8% of heart disease mortality and 8.8% of cerebrovascular disease mortality could be prevented if current drinkers reduced their alcohol consumption to <46 g/day. In women, the PAFs were much lower than in men: 0% of total mortality, 0% of cancer mortality, 0.6% of heart disease mortality and 0.7% of cerebrovascular disease mortality could be prevented if female current drinkers reduced their alcohol consumption to <23 g/day, and 0.1% of total mortality, 0% of cancer mortality, 0.6% of heart disease mortality and 0.8% of cerebrovascular disease mortality could be prevented if current drinkers reduced their alcohol consumption to <46 g/day.
In this pooled analysis, we found J- or U-shaped associations between the amount of alcohol consumed and the risks of total and major causes of mortality in men, and the risks of total and heart disease mortality in women. When ex-drinkers were excluded from the baseline category, the HRs for the low drinkers were reduced and those for the high drinkers were increased. We confirmed that, overall, alcohol consumption of <46 g/day in male and <23 g/day in female Japanese was associated with a lower mortality risk than those for non-drinkers and heavy drinkers. A similar pattern was observed when deaths within 5 years after baseline were excluded from the analysis and when analysis was stratified by smoking status. In addition, mortality risk appeared to increase linearly with rising alcohol dose among drinkers. Using these estimates, we calculated that 5% of total mortality, 3% of cancer mortality, 2% of heart disease mortality and 9% of cerebrovascular disease mortality in men, but only 0–1% of these risks in women, could be prevented by reducing alcohol consumption to <46 g/day in men and <23 g/day in women.
The present study has several strengths. It included most of the ongoing, large-scale, prospective cohorts in Japan. In addition, the birth generation of the study subjects in the cohorts overlapped. Therefore, pooling of these cohorts allows for stable summary quantitative estimates of the effect of alcohol consumption on premature death in middle-aged and elderly Japanese adults. At the same time, because this study was not based on a meta-analysis of published studies, the possibility of publication bias is small. In the studies included in this pooled analysis, alcohol consumption was measured before death, which precludes the possibility of selection and recall bias. In addition, the categories for alcohol intake and the covariates used were identical among studies, which removes a potential source of heterogeneity that can occur when conducting a meta-analysis of published studies.
However, there are several limitations that warrant consideration. As our analyses were conducted using only a baseline questionnaire, we were unable to consider changes over time in alcohol intake. Also, we were unable to assess alcohol consumption by type of beverage. Because questions on alcohol consumption differed among studies, heterogeneity of the distribution of alcohol categories between the studies could not be distinguished from misclassification caused by measurement error occurring in each study. In this pooled analysis, mortality was defined as an outcome; thus, it is difficult to distinguish whether alcohol consumption was associated with incidence, survival or both.
There is strong and consistent epidemiological evidence of a significant inverse association between low to moderate alcohol consumption and cardiovascular disease.26 Alcohol consumption is linked to changes in several vascular and biochemical factors that have potential cardioprotective benefits. Alcohol consumption is believed to reduce the risk of cardiovascular events primarily by increasing high-density lipoprotein (HDL) cholesterol levels, decreasing platelet aggregation via inhibition of prostaglandin synthesis and changes in fibrinogen, and lowering levels of tissue-plasminogen activator (t-PA) and PA inhibitor (PAI-1).27 28 However, the anticoagulant effect of alcohol, despite its beneficial effect on ischaemic stroke, may increase the risk of haemorrhagic stroke.29–31
Regarding cancer, the latest evaluation in 2007, by the International Agency for Research on Cancer, confirmed that alcoholic beverages are carcinogenic to humans (group 1) and concluded that the occurrence of certain sites of cancer, such as those of the oral cavity, pharynx, larynx, oesophagus, liver, colorectum and female breast, is causally related to alcohol consumption.32 33 The association between alcohol and cancer may be the result of high acetaldehyde exposure, which is considered carcinogenic.32 In addition, excessive alcohol intake impairs folate absorption and bioavailability by inhibiting expression of reduced folate carrier and decreasing hepatic uptake and renal conservation of circulating folate.34 Decreased folate status can contribute to aberrant DNA synthesis and methylation, which may increase the risk of cancer.35
Our results suggest that there is a specific level of alcohol consumption with the least risk, which was observed among men consuming <46 g/day and women consuming <23 g/day. Also, in drinkers, the increase in cancer risk appeared to be linear with respect to the amount of alcohol consumed. A previous meta-analysis estimated that 2–4 drinks/day (20–40 g/day) in men and 1–2 drinks/day (10–20 g/day) in women was inversely associated with total mortality1; these values roughly accord with the estimates for the present Japanese population. However, approximately half of all Japanese have an ALDH2-deficient phenotype. ALDH2 is a key enzyme in the conversion of acetaldehyde to acetate,36 and a deficiency in this enzyme results in greater acetaldehyde exposure.37 The high prevalence of ALDH2 deficiency indicates a need for caution in interpreting the results for non-drinkers, and for never- and ex-drinkers, because some of these subjects were unable to drink because of this deficiency, which would result in risk inflation in this category. ALDH2 deficiency potentially increases not only the risk of alcohol-related diseases, including cancer, but also the fraction of mortality attributable to high alcohol drinking in the population with a high prevalence of this deficiency.
With regard to sex differences in the protective effect of alcohol, previous studies examined possible mechanisms for this difference.1 2 Women have a higher blood alcohol concentration when men and women consume the same amount of alcohol.38 Also, women metabolise ethanol differently and have lower gastric alcohol dehydrogenase activity, resulting in higher blood ethanol levels.39 Women who consume moderate or high levels of alcohol have a higher risk than men of death from any cause, owing to their increased risk of cancer.9 We could not assess this association in detail, especially among female heavy drinkers, because the proportion of such women was too small, even in pooled analysis.
Finally, regarding public policy implications, maintaining light to moderate alcohol consumption, specifically <46 g/day in men and <23 g/day in women, may be a feasible public health recommendation for the Japanese population. However, it might not be appropriate to recommend that non-drinkers begin drinking, because of the observed increase in risk with rising alcohol dose among drinkers, the high prevalence of ALDH2 deficiency in Japanese subjects, and the fact that many abstainers have good reasons to refrain from drinking.40
In conclusion, our results suggest that maintaining alcohol consumption at <46 g/day (2 go) in men and <23 g/day (1 go) in women may minimise the risks of total mortality and mortality from major diseases in the Japanese population.
What is already known on this subject
A number of studies have suggested the health benefits of light to moderate alcohol consumption with respect to total and cardiovascular mortality, and the harmful effect of heavy alcohol drinking for certain sites of cancer.
However, because published studies use different alcohol consumption categories, meta-analysis for the purpose of quantitative assessment is not possible.
What this study adds
This study confirmed that maintaining alcohol consumption below 46 g/day in men and below 23 g/day in women minimises the risks of total mortality and mortality from major diseases in the Japanese population.
The authors gratefully acknowledge the assistance of Izumi Suenaga.
Research group members: Shoichiro Tsugane (principal investigator), Manami Inoue, Shizuka Sasazuki, Motoki Iwasaki, Tetsuya Otani (until 2006), Norie Sawada (since 2007), Taichi Shimazu, Taichi Yamaji (since 2007; National Cancer Center, Tokyo), Ichiro Tsuji (since 2004), Yoshitaka Tsubono (in 2003; Tohoku University, Sendai), Yoshikazu Nishino (Miyagi Cancer Research Institute, Natori, Miyagi), Kenji Wakai (Nagoya University, Nagoya), Keitaro Matsuo (since 2006; Aichi Cancer Center, Nagoya), Chisato Nagata (Gifu University, Gifu), Tetsuya Mizoue (National Center for Global Health and Medicine, Tokyo), Keitaro Tanaka (Saga University, Saga) and Akiko Tamakoshi (Aichi Medical University, Nagakute, Aichi).
Funding This study was supported by a grant for the Third Term Comprehensive Control Research for Cancer from the Ministry of Health, Labour and Welfare of Japan.
Competing interests None.
Ethics approval This study was conducted with the approval of the National Cancer Center, Aichi Medical University, Tohoku University Graduate School of Medicine and Gifu University Graduate School of Medicine.
Provenance and peer review Not commissioned; externally peer reviewed.