Many studies have demonstrated a U- or J-shaped relationship between alcohol intake and all-cause mortality. This association can be explained by a lower risk of cardiovascular disease (CVD) mortality in light to moderate drinkers.1^–4 The protective effect of light to moderate alcohol intake may be due to an increase in high-density lipoprotein (HDL) cholesterol and prevention of blood clotting and reduction in platelet aggregation.5 6 Red wine consumption may have an additional health benefit because of its polyphenolic compounds7 8 that interfere with the initiation, progression and rupture of atherosclerotic plaques9 and improve endothelial function.10 11 Although some epidemiological studies showed beneficial effects of wine consumption,12 13 the results of several other studies do not show an advantage of one type of alcoholic beverage over another.14 Studies have suggested that the beneficial effects of wine consumption compared with other beverages can mainly be attributed to differences in socioeconomic status, dietary and other lifestyle habits.15 16

In prospective cohort studies in which alcohol intake is only assessed at the baseline examination, consumption patterns are assumed to be relatively constant over the entire study period. However, it is unlikely that exposure measurements in the past accurately reflect long-term alcohol intake as consumption patterns usually change during life. To get correct estimates of the long-term effects of alcohol intake, repeated measures are needed. Moreover, the use of repeated measures, especially when using a cumulative average method, reduces within-subject variation over time and, thereby, misclassification in alcohol intake.17

In epidemiological studies, hazard ratios are commonly used to express the impact on mortality. As hazard ratios express effects for an exposed group relative to the effect of the unexposed group, they do not provide information regarding absolute health effects. Such insight can be obtained by the calculation of life expectancies and the number of life–years lost. Although concepts such as life expectancy are more informative and easier to understand, they are not reported frequently.

The objective of the present study is to assess the relationship between long-term intake of alcohol and the types of alcoholic beverages consumed and cardiovascular and all-cause mortality. To obtain more accurate effect estimates, we used seven repeated measures of the consumption of alcoholic beverages. In addition to hazard ratios, we also present our results in terms of differences in life expectancy.

METHODS

Study population

The Zutphen Study has been carried out since 1960 among middle-aged men in Zutphen, an old industrial town in the eastern part of The Netherlands with about 30 000 inhabitants. In 1960, a random sample was drawn of 1088 men born between 1900 and 1919 and living in Zutphen for at least 5 years. Of those men, 878 participated in the Zutphen Study (response rate 81%). Examinations were repeated in 1965, 1970, 1985, 1990, 1995 and 2000. In 1985, the group of 554 survivors was extended with a new random sample of men of the same age. Of the 1266 men who were invited in 1985, 939 men participated (response rate 74%). In every examination round, the participants who took part in both dietary and physical examinations were selected for the present study (1373 men).

Baseline data were collected in 1960 before the Helsinki Declaration was developed, and oral informed consent was obtained in view of follow-up data. In 1985 and 1990, the study was approved by the Medical Ethics Committee of the University of Leiden, The Netherlands, and in 1995 and 2000, by the Medical Ethics Committee of The Netherlands Organisation for Applied Scientific Research (TNO).

Assessment of alcohol and food consumption

The habitual alcohol consumption was assessed as part of the total diet, using the cross-check dietary history method,18 adapted to the Dutch situation.19 20 This method provides information about the participant’s usual food and alcohol consumption pattern, 6–12 months preceding the interview. From 1985 onwards, the information about the usual food and alcohol consumption pattern was limited to the month preceding the interview because consumption patterns from 1985 were much more complicated than those in the 1960s. The daily intake of alcohol, energy and other nutrients was calculated using food composition tables close to the year of measurement.21^–23 Alcohol intake was highly reproducible (r = 0.8),24 and ranking of the participants was considered valid, as shown by the strong correlation with HDL cholesterol.25

To assess the relationship between different types of alcoholic beverages and mortality, we calculated the alcohol intake from wine, beer or spirits respectively. Participants were grouped into three categories according to their intake of total alcohol and of alcohol from beer, wine or spirits: 0, >0–20 and >20 g/day. As studies have suggested that using non-drinkers as a reference group is unsuitable for studying the effects of alcohol on mortality because of the higher background risk among the former drinkers,26^–28 we created a time-dependent indicator variable for those men who stopped drinking during follow-up.

Assessment of potential confounders

Detailed information on type and amount of smoking was collected using standardised questionnaires. Cigarette, cigar or pipe smokers were divided into categories of never or long-term ex-smokers (ie, stopped smoking ⩾10 years ago), recent ex-smokers (ie, stopped smoking <10 years ago) and current smokers. Also, the number of years of cigarette smoking was calculated. During physical examinations, the men’s weight and height were measured and body mass index (BMI) was calculated. Information about the prevalence and history of clinical myocardial infarction (MI), stroke, diabetes mellitus (DM) and cancer was collected throughout the study. The men were classified into four levels of socioeconomic status (manual workers, non-manual workers, small business owners and professionals) according to occupation at baseline.

Case ascertainment

Participants were followed until death or censored on 30 June 2000. Three participants were lost to follow-up during the study and were censored after their last physical examination. Information about the causes of death was available from the official death certificate, the medical history collected from interviewing physicians or relatives of the dead person and other witnesses, and from abstracts of hospital and other medical records.29 The adjudication of the underlying causes of death was done by one clinical epidemiologist (Dr Alessandro Menotti, Association for Cardiac Research, Rome, Italy) and coded according to the International Classification of Diseases, 8th Revision (codes 410–414 for coronary heart disease (CHD), codes 430–438 for cerebrovascular diseases and codes 390–458 for CVD).30 Because the underlying cause of death in elderly people is often difficult to ascertain, we included primary and secondary causes of cardiovascular death.

Statistical analysis

Cox proportional hazard analyses with age as the time variable31 32 were performed using the PHREG procedure of SAS/STAT software (version 9.1; SAS Institute, Inc., Cary, NC, USA). Time at entry was age on 31 December in the year preceding the year in which the men participated in both dietary and physical examinations for the first time, that is the first measurement round; exit time was age at death, age when lost to follow-up or age when censored, whichever came first. We calculated the cumulative average alcohol intake to better represent long-term intake. With this method, exposure between 1960 and 1965 was calculated from the alcohol intake from the 1960 examination round; exposure between 1965 and 1970 was calculated from the average alcohol intake from the 1960 and 1965 examination rounds, and so on.17 For those men who were newly included in the study in 1985, information on alcohol intake was missing in the period 1960–1970. To take into account that average alcohol intake was higher in 1985 than in 1960–1970 and that taking cumulative averages excluding earlier intakes in those men who were new to the study in 1985 would overestimate their intake compared with men included in 1960, multiple imputation,33 five times, of alcohol intake and dietary covariates between 1960 and 1970 was carried out with an adapted version of predicted mean matching.34 The SAS code that was used for the multiple imputation can be downloaded from http://www.rivm.nl/sasmacros. Analyses on long-term alcohol intake were performed on five imputed datasets, and results were pooled using the MIANALYZE procedure of SAS/STAT software.

The covariates in the multivariable models included an indicator variable for former drinker, energy intake without alcohol (kcal/day), consumption of vegetables (g/day), fruit (g/day) and fish (g/day), intake of saturated and trans fatty acids (g/day), BMI (kg/m²), cigar or pipe smoking (never and long-term ex, recent-ex and current), cigarette smoking duration (divided by 10 years), the daily number of cigarettes smoked (divided by 10), prevalence of MI (yes or no), stroke (yes or no), DM (yes or no) and cancer (yes or no), and indicator variables for baseline socioeconomic status. In additional analyses, the multivariable models for, respectively, alcohol from beer, wine and spirits were additionally adjusted for total alcohol intake to investigate the independent effects of alcohol from one specified alcoholic beverage type. In the multivariable analyses, the cumulative average intake of all dietary covariates was calculated, and non-dietary covariates were updated at each measurement round. To test the proportional hazards assumption, a product term between alcohol intake and age was included in the model, and a p-value for interaction <0.10 was considered statistically significant.

We determined differences in life expectancy at age 50, that is average age at baseline, between men with different levels of cumulative average alcohol intake, by calculating the area under survival curves.35 To distinguish between the effects of alcohol intake and the effects of alcoholic beverage types, we compared the life expectancies for men who consumed alcohol from wine (>0 g/day) with those who consumed alcohol from beer and spirits and those who consumed no alcohol from the specified sources, using cumulative average intake at each measurement round. The men were included in the analysis during the period in which they met the requirements for the exposure categories concerned. Cox models, with age as the time variable and stratified by categories of amount (0, >0–20, >20 g/day) or source (no alcohol, alcohol from wine, alcohol from beer or spirits) of alcohol intake, were used to obtain the survival curves. The Cox models were adjusted for baseline covariates, that is dietary and smoking variables, BMI, prevalence of chronic diseases (ie, MI, stroke, DM and cancer) and socioeconomic status. As several participants used more than one type of alcoholic beverage in their usual diet and are, thereby, included in different exposure categories at the same time, the COVSANDWICH (AGGREGATE) statement was added to the PHREG procedure. Areas under the survival curves from the five imputed datasets were pooled,33 and 95% confidence intervals were obtained using the bootstrap method.36

RESULTS

Population characteristics

During 40 years of follow-up (mean survival age 77 years), 1130 of the 1373 men participating in the present study died (table 1). Of these deaths, 628 were cardiovascular disease (CVD) deaths, 348 were coronary heart disease (CHD) deaths and 139 were cerebrovascular deaths.

The percentage of alcohol users almost doubled from 45% in 1960 to 86% in 2000 (table 1). Among users, average alcohol consumption increased from 8 g/day in 1960 to 18 g/day among the survivors in 1985, after which it decreased to 14 g/day in 2000 (table 1). The percentage of wine users increased remarkably from 2% in 1960 to about 44% in 2000. This increase was observed at all levels of socioeconomic status. Among users, the average number of glasses consumed varied between a half and 1.5 glasses per day for wine, beer and spirits. With the exception of 1960, alcohol from spirits contributed the most to the total alcohol intake (fig 1). The correlation between alcohol from spirits and total alcohol intake varied between 0.67 and 0.86. For alcohol intake from wine and beer, the correlations were lower.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The contribution of alcohol from beer, wine and spirits to total alcohol intake within the Zutphen Study during 40 years of follow-up.

Alcohol consumption, alcoholic beverages and mortality

Long-term alcohol intake was significantly and inversely associated with mortality risk (table 2). Men with less than or equal to 20 g of long-term (i.e. cumulative average) alcohol intake had a 57% lower cerebrovascular mortality risk, a 30% lower CVD mortality risk and a 25% lower all-cause mortality risk compared with men with no alcohol intake. The associations for >20 g of long-term alcohol intake per day were weaker than those for ⩽20 g of long-term alcohol intake per day.

In the next step, the independent effects of long-term alcohol intake from wine, beer and spirits on mortality were estimated. After additional adjustment for total alcohol intake, ⩽20 g of long-term alcohol intake from wine per day—compared with no alcohol intake from wine—was inversely associated with CHD (HR 0.61, 95% CI 0.41 to 0.89), CVD (HR 0.68, 95% CI 0.53 to 0.86) and all-cause mortality risk (HR 0.73, 95% CI 0.62 to 0.87; table 3). In the models for CVD and all-cause mortality, ⩽20 g of long-term total alcohol intake remained inversely associated with mortality (HR CVD mortality 0.76, 95% CI 0.59 to 0.97; HR all-cause mortality 0.80, 95% CI 0.67 to 0.97). Additional analyses showed that the inverse association between wine consumption and mortality was present in all socioeconomic classes. Among manual workers, the HRs for long-term wine consumption were 0.63 (95% CI 0.31 to 1.30) for CHD, 0.78 (95% CI 0.49 to 1.23) for CVD and 0.78 (95% CI 0.54 to 1.12) for all-cause mortality. Among the professionals, the hazard ratios were comparable.

Long-term consumption of alcohol from beer and spirits was not independently related to mortality (table 3). However, given the strong correlation between alcohol from spirits and total alcohol intake, these findings should be interpreted with caution. Omitting the adjustment for the prevalence of chronic diseases changed our results slightly, but the overall conclusions remained the same.

Alcohol consumption, alcoholic beverages and life expectancy

Men with a long-term alcohol intake ⩽20 g/day (on average 6 g/day) had a 2.3 year (95% CI 0.5 to 4.2 years) longer life expectancy at age 50 compared with those who consumed no alcohol. For those men who consumed more than 20 g of alcohol per day (on average 29 g/day), life expectancy was 1.9 years (−1.1 to 4.9 years) longer compared with non-users.

Men who consumed alcohol from wine, on average 2 g/day, over a longer time period had a 2.5 years longer life expectancy at age 50 (−0.3 to 5.3 years) compared with those who consumed alcohol from beer or spirits (on average about 8 g/day; fig 2) and a 4.7 years (1.6 to 7.7 years) longer life expectancy compared with no alcohol users.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Survival curves for men with long-term consumption of alcohol from wine, beer or spirits, and no alcohol consumers within the Zutphen Study, adjusted for baseline energy intake without energy from alcohol, the number of cigarettes smoked, cigar or pipe smoking, intake of vegetables, fruit, fish, saturated and trans fatty acids, body mass index, prevalence of myocardial infarction, stroke, cancer and diabetes mellitus, and socioeconomic status.

DISCUSSION

Long-term light alcohol intake lowered cerebrovascular, total CVD and all-cause mortality risk, and was associated with a longer life expectancy compared with no alcohol intake. Independent of total alcohol intake, wine consumption was strongly and inversely associated with CHD, total CVD and all-cause mortality. For long-term wine consumers, consuming on average less than half a glass of wine per day, life expectancy at age 50 was about 5 years longer compared with no alcohol users.

The major strength of this study was the collection of detailed information on the consumption of different alcoholic beverages at each of seven examination rounds during 40 years of follow-up. This enabled us to study the effects of long-term, that is cumulative average, alcohol intake on mortality. The use of cumulative average intakes reduces within-subject variation over time and, thereby, measurement error. The detailed information on potential confounders such as smoking,37^–39 diet40 and socioeconomic status41 made it possible to study the independent effect of total alcohol intake and alcohol from different types of alcoholic beverages.

The present study also has some weaknesses. First, recent studies have observed that, among men, frequency of alcohol consumption was inversely associated with coronary heart disease risk, independent of the amount of alcohol consumed.42^–44 As information on alcohol consumption in the present study was collected as part of the usual diet, no data were available on drinking frequency. Second, average long-term alcohol intake in the present study was relatively low, and most participants used more than one type of alcoholic beverage in their usual diet. This may have led to less precise estimations of the effect of different types of alcoholic beverages on mortality. Third, for those men who were newly included in the study in 1985, information on alcohol intake was missing for the period 1960–1970. By multiple imputations of total alcohol intake, different types of alcoholic beverages and other dietary covariates in 1960–1970, we were able to counter an underestimation of cumulative average intake from 1985 onwards for those men who were newly included in the study. We repeated our analysis among the participants who were included in the study from 1960 (n = 875) and found similar associations between long-term total alcohol intake and different types of alcoholic beverages and mortality. Therefore, it is unlikely that the imputation of total alcohol intake, different types of alcoholic beverages and other dietary covariates among those men who were newly included in the study from 1985 biased our results.

Our results confirm the inverse association between moderate alcohol intake, CVD and all-cause mortality risk observed in other studies.38 39 45^–53 In contrast to these and other studies,38 39 45 53^–56 the highest exposure level was not associated with an increased mortality risk. However, average alcohol intake in the highest exposure level was relatively low (29 g/day) in the Zutphen Study, which may explain the lack of a positive association between mortality and a higher intake level. In their meta-analysis, Corrao et al estimated that the risk of CHD mortality is lowest at 16 g/day with a corresponding hazard ratio of 0.84.57 In the present study, the hazard ratios for CHD mortality were comparable but not statistically significant (table 2).

Smoking might confound the association between alcohol intake and mortality. Previous results from the Zutphen Study indicated that the number of cigarettes smoked as well as the duration of cigarette smoking are, independently of each other, associated with mortality.58 In our multivariable analyses, we adjusted for these aspects of smoking. So, it is not likely that the association between alcohol intake and mortality is explained by confounding due to smoking.

In the present study, wine consumption was associated with a lower mortality risk, independent of total alcohol intake. Some studies did not demonstrate a favourable effect of one specific type of alcoholic beverage over another,14 42 while others found an inverse and independent association between wine consumption and mortality.13 59^–61 In the present study, 70% of all wine consumed was red wine. This suggests that the cardioprotective effect of wine could be due to a protective effect of polyphenolic compounds in red wine, but other explanations cannot be ruled out.

Socioeconomic status might confound the association between wine consumption and mortality. However, in our multivariable analyses, we adjusted for socioeconomic status, based on occupation at baseline. At the start of the present study, the men were already in a later phase in their careers, and baseline socioeconomic status was considered to be a good indicator during the follow-up period. Furthermore, the increase in the percentage of wine users during follow-up was observed at all levels of socioeconomic status, and additional stratified analyses showed that the inverse association between wine consumption and mortality was present in all socioeconomic classes. These results suggest that the association between wine consumption and mortality cannot be explained by confounding due to socioeconomic status.

Long-term wine consumers had about 5 years longer life expectancy at age 50 compared with no alcohol users. Of these 5 years, about 2 years can be attributed to an effect of alcohol intake and is in accordance with the inverse association between long-term alcohol intake and all-cause mortality found in the present study. The remaining 3 years can be attributed to an effect of wine consumption. However, given the wide confidence interval, the effect of wine as such may be overestimated. To our knowledge, we are the first to study the effects of absolute alcohol intake and type of alcoholic beverage on life expectancy, and more studies are needed to verify our results.

In conclusion, long-term light alcohol intake is associated with a lower risk of cardiovascular and all-cause mortality risk and a longer life expectancy. The inverse associations between wine consumption and mortality remained after adjustment for total alcohol intake. Wine consumers had a 5 years longer life expectancy compared with no alcohol consumers; however, more studies are needed to draw conclusions on the strength of the association between wine consumption and mortality.