%20-%20https://sketch.com%20--%3e%3ctitle%3e3ECBBF50-FD08-412A-943D-2CFC5E025A16%3c/title%3e%3cdesc%3eCreated%20with%20sketchtool.%3c/desc%3e%3cg%20id='Stylesheet'%20stroke='none'%20stroke-width='1'%20fill='none'%20fill-rule='evenodd'%3e%3cg%20transform='translate(-926.000000,%20-668.000000)'%20id='BUTTONS'%3e%3cg%20transform='translate(100.000000,%20597.000000)'%3e%3cg%20id='BTN-/-RIGHTS'%20transform='translate(816.000000,%2065.000000)'%3e%3cg%20id='ICON-/-Rights'%20transform='translate(10.000000,%206.000000)'%3e%3cpath%20d='M13.5752841,0%20L1.34943182,0%20C0.59375,0%200,0.586666667%200,1.33333333%20L0,22.6666667%20C0,23.4133333%200.620738636,24%201.34943182,24%20L17.6505682,24%20C18.40625,24%2019,23.3866667%2019,22.6666667%20L19,5.36%20L13.5752841,0%20Z%20M13.5752841,1.89333333%20L17.0838068,5.36%20L13.5752841,5.36%20L13.5752841,1.89333333%20Z%20M1.34943182,22.6666667%20L1.34943182,1.33333333%20L12.1988636,1.33333333%20L12.1988636,5.36%20C12.1988636,6.10666667%2012.7926136,6.69333333%2013.5482955,6.69333333%20L17.6235795,6.69333333%20L17.6505682,22.6666667%20L1.34943182,22.6666667%20Z'%20id='Shape'%20fill='%23595959'%20fill-rule='nonzero'%3e%3c/path%3e%3ccircle%20id='Oval'%20stroke='%23595959'%20stroke-width='1.5'%20fill='%23595959'%20cx='9.5'%20cy='13.5355339'%20r='5.5'%3e%3c/circle%3e%3cpolyline%20id='Path-2'%20stroke='%23FFFFFF'%20stroke-width='1.5'%20transform='translate(9.500000,%2013.035534)%20rotate(45.000000)%20translate(-9.500000,%20-13.035534)%20'%20points='8%2016.0355339%2011%2016.0355339%2011%2010.0355339'%3e%3c/polyline%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
In 2010, the International Diabetes Federation estimated the global prevalence of diabetes mellitus at 6·6 % in adults (age range 20–79 years)(1). Diabetes is now one of the most common non-communicable diseases in the world and a major cause of premature illness and death in most countries(1). To prevent diabetic complications and premature mortality, patients are recommended to adopt a healthy lifestyle as part of their self-management strategy(Reference Buse, Ginsberg and Bakris2).
Within the general population, moderate alcohol consumption has been associated with a lower incidence of cardiovascular mortality(Reference Mukamal, Chen and Rao3) as well as overall mortality(Reference Di Castelnuovo, Costanzo and Bagnardi4) compared with non-consumers. Indeed, if people choose to drink alcohol, a moderate consumption – limited to one glass per d for women and two glasses per d for men – has been recommended for the general public by expert groups such as the World Cancer Research Fund(5). This upper limit is also recommended to individuals with diabetes mellitus(Reference Buse, Ginsberg and Bakris2). Moderate alcohol consumption has been suggested to reduce the risk of CHD morbidity and mortality in individuals with diabetes mellitus, relative to abstinence from alcohol(Reference Nakamura, Ueshima and Kadota6, Reference Koppes, Dekker and Hendriks7). However, a well-known difficulty in such comparisons is the use of non-consumers as the reference group. Indeed, it might include both lifelong abstainers but also former drinkers. This group has been shown to be very heterogeneous and, as a consequence, appears to be less healthy than moderate consumers(Reference Friesema, Zwietering and Veenstra8, Reference Rehm, Irving and Ye9). Thus, non-consumers may not be an ideal reference group in epidemiological investigations of alcohol and health outcomes.
As for individuals who have experienced CVD, individuals with diabetes may have modified their drinking habits after diagnosis(Reference Costanzo, Di Castelnuovo and Donati10). However, alcohol consumption in the past has hardly been investigated in relation to mortality in individuals with diabetes(Reference Koppes, Dekker and Hendriks7, Reference Howard, Arnsten and Gourevitch11). It has been reported in the general population that the relationship between alcohol consumption and mortality was attenuated when taking alcohol consumption in the past into account(Reference Friesema, Zwietering and Veenstra12, Reference Shaper, Wannamethee and Walker13). We investigated the association between alcohol consumption and mortality in individuals with diabetes using light alcohol consumption as a reference group and taking history of alcohol consumption into account.
Subjects and methods
Design and subjects
Within the European Prospective Investigation into Cancer and Nutrition (EPIC)(Reference Riboli, Hunt and Slimani14), a subcohort was defined of participants with a confirmed diagnosis of diabetes mellitus at baseline (1992–8), as has been described earlier(Reference Nothlings, Boeing and Maskarinec15, Reference Sluik, Boeing and Montonen16). Data were available from eleven out of twenty-three study centres in five out of ten European countries (Denmark, Germany, Italy, The Netherlands and Spain). Self-reports of diabetes at baseline were confirmed by a second source of information, i.e. contact to a medical specialist or practitioner, self-reported use of medication for diabetes treatment, repeated self-report of diagnosis during follow-up or by further inquiry of the participant, or record linkage to a diabetes registry. The study was conducted according to the guidelines laid down in the Declaration of Helsinki and was approved by ethical review boards of the single centres and the International Agency for Research on Cancer in Lyon, France. All subjects provided written informed consent.
Of 5457 initial self-reports, 4407 diabetes diagnoses were confirmed. A further 670 prevalent diabetes cases without self-reported diabetes at baseline were identified as a result of verification efforts in other projects within the EPIC. Thus, the overall cohort comprised 5077 individuals with confirmed diabetes at study entry. After exclusion of participants without follow-up data on vital status (n 23), participants in the highest or lowest 1 % of the ratio of energy intake:estimated energy requirement (n 133), deceased participants with missing date of death (n 1), and participants with missing information on the following: baseline questionnaire data (n 1), information on age at diabetes diagnosis (n 86), and information on present alcohol consumption (n 13) and lifetime alcohol consumption (n 35), the analytical sample included 4797 participants.
Exposure assessment
Information on alcohol consumption at recruitment was obtained from a FFQ applied during the baseline examination. The FFQ enquired about consumption of alcoholic beverages (cider, beer, wine and spirits) during the previous 12 months(Reference Riboli, Hunt and Slimani14). At recruitment, an 8 % sample (n 36 994) of EPIC study participants in each centre was randomly selected for a 24 h dietary recall interview that generated detailed information on all foods and beverages consumed during the day before the interview(Reference Slimani, Ferrari and Ocke17). This detailed information was used to estimate the average alcohol content (g/d) contained in ‘standard’ glasses in the EPIC countries, as they provide information on the type of glasses, glass volume and levels of filling. These average alcohol content values were estimated at the country level, separately for men and women, and by weighting for weekday v. weekend day, and seasonality. In addition, the 24 h dietary recall data were used to compute country-specific average consumption of alcohol per 100 g of beverage. This was done to transform average volumes (ml) of alcohol into average consumption (g), according to the particular alcohol subtypes consumed in the different countries.
Weekly consumption of wine, beer/cider, fortified wine and liquor (spirits) was assessed retrospectively at the ages of 20, 30, 40 and 50 years in a lifestyle factor questionnaire(Reference Bergmann, Schutze and Steffen18, Reference Schutze, Boeing and Pischon19). To estimate alcohol consumption at different ages and at recruitment, an algorithm was used that combined the information available on the length of time alcohol was consumed and the amount of alcohol consumed at these different ages for a given period. Rather than estimating the average lifetime consumption as a simple mean of alcohol intake at ages 20, 30, 40 years, etc., which would not capture information on different time lengths between ages where information is available, we calculated time lengths of alcohol consumption, and these were used in the analyses. The average alcohol content per glass was set at 12 g and cut-points were used of >0–6 g/d (0–0·5 glass, light consumption), >6–12 g/d (0·5–1 glass), >12–24 g/d (1–2 glasses), and in women>24 g/d (2 or more glasses) and in men >24–60 g/d (2–5 glasses) and >60 g/d (5 or more glasses). Prior alcohol consumption was defined as none, always moderate and sometimes heavy (consumption of 2·5 times the upper recommended limit at certain ages, i.e. >30 g/d in women and >60 g/d in men). In the EPIC, the amount of alcoholic beverages consumed in history by comparable age cohorts at certain ages was validated by comparison of self-reported data with the respective per capita measures from 1950 to 1995. Ethanol intake estimates were on average about 72 % compared with the per capita consumption data; only small differences in reproducibility between the various techniques existed and ranking individuals according to intake was acceptable(Reference Klipstein-Grobusch, Slimani and Krogh20).
Assessment of other covariates
Data on other sociodemographic and lifestyle factors, including smoking status, educational attainment, physical activity and medical history, were collected using a questionnaire or personal interview at baseline. Weight, height and waist circumference were measured at the baseline examination, and subsequently BMI as an index of general adiposity and the waist:height ratio as an index of abdominal adiposity were calculated(Reference Sluik, Boeing and Montonen16). Duration since diabetes diagnosis was calculated by subtracting the self-reported year of diagnosis or the exact date of diagnosis (only if the diagnosis was medically verified) from the age at baseline examination. Information on insulin therapy was either self-reported or obtained during medical verification.
Outcome ascertainment
Causes and dates of deaths were ascertained using record linkages with local, regional or central cancer registries, boards of health, or death indices (Denmark, Italy, The Netherlands, Spain). Germany identified deceased participants with follow-up mailings and subsequent inquiries to municipality registries, regional health departments, physicians and hospitals. Mortality data were coded according to the Tenth Revision of the International Classification of Diseases, Injuries and Causes of Death (ICD-10). For the cause-specific analyses, deaths from circulatory diseases (ICD-10 I00-I99), cancer (ICD-10 C00-D48) and all other known causes were grouped.
Statistical analyses
All statistical analyses were performed with SAS 9.2 (SAS Institute). Sex-specific categories of baseline alcohol consumption were defined as outlined previously. Hazard ratios (HR) and 95 % CI for total and cause-specific mortality were calculated separately for men and women using Cox proportional hazard regression models, where light alcohol consumption (>0–6 g/d) was used as the reference category. In addition, HR were calculated using non-consumers as a reference category, with and without inclusion of participants categorised as former consumers. Centre and age at recruitment in 1-year categories were entered as stratum variables. Age was used as the primary time variable, with entry time defined as the subject's age in years at recruitment and exit time defined as the subject's age in years at death or censoring (lost to follow-up or the end of the follow-up period, whichever occurred first).
HR were adjusted for diabetes duration (years), insulin use (yes or no), co-morbidities (cancer, heart disease or stroke), smoking status (never, former (quit ≤ 10, 11–20 or >20 years ago) or current (smoking duration ≤ 10, 11–20, 21–30, 31–40 or >41 years, < 15, 15–24 or >25 cigarettes smoked daily)), educational attainment (no, primary school, technical or professional school, secondary school, university or above), and employment status (yes, no or missing). A second model was additionally adjusted for three categories of alcohol consumption in the past: lifetime non-consumer (yes or no); always moderate (yes or no); sometimes heavy consumption in the past (yes or no). The influence of physical activity and dietary intake (fruit, vegetable, meat and fibre intake) on the model was investigated, but inclusion of these variables did not substantially change the effect estimates and were left out. A P value for non-linearity of the associations was calculated with a Wald χ2 test using the categories and adjustment models as outlined above using restricted cubic spline regression. A P value for trend was calculated using the median alcohol consumption within categories as the continuous variable. In addition, HR for an increment of 12 g/d (one glass) for consumers at baseline only was estimated. Finally, risk for specific causes of death was derived from competing risk models in which separate regression coefficients for different causes of death were compared using the Wald χ2 test and 95 % CI were derived from robust estimates of the covariance matrix(Reference Lunn and McNeil21, Reference Glynn and Rosner22).
Results
At baseline, median alcohol consumption was 16·0 g/d in men and 1·8 g/d in women. During a median follow-up of 9·2 years, 373 men and 160 women died. Of those 533 deaths, 127 were due to CVD, 109 due to cancer, 132 due to other known causes and in 165 cases, information on the cause of death was not yet verified or the cause was unknown.
Men who reported a high alcohol consumption had a higher waist:height ratio and the lowest level of education, whereas women with a high alcohol consumption were more likely to have a lower waist:height ratio, have acquired a higher level of education and be currently employed (Table 1). In terms of past alcohol consumption, the group of male light consumers appeared to be more homogeneous compared with non-consumers. A vast majority of light consumers reported to have always consumed alcohol moderately in the past. The same pattern was observed among the mortality cases.
(Mean values and standard deviations; numbers and percentages; medians and interquartile ranges (IQR))
* Denominator decreased due to missing values.
In men, alcohol consumption of 6 g/d or more was not associated with mortality compared with light consumption (Table 2). Adjustment for alcohol consumption in the past only marginally changed the estimates. Non-consumers at baseline seemed to be at a higher mortality risk than men consuming up to 6 g/d (HR 1·62, 95 % CI 1·05, 2·49). Further adjustment for alcohol consumption in the past attenuated this relationship. Compared with non-consumers at baseline, risk of death in light alcohol consumers was lower: multivariable HR was 0·62 (95 % CI 0·40, 0·95). Additional adjustment for past alcohol consumption slightly affected the associations. Further exclusion of former consumers from the comparison group, i.e. using lifetime non-consumers as the reference, lowered the risk estimates in men, but results were not significant anymore. HR were more imprecise since there were only a few case numbers among the lifetime non-consumers.
(Hazard ratios (HR) and 95 % confidence intervals)
* Age- and centre-stratified and adjusted for diabetes duration, insulin therapy, co-morbidities (heart disease, stroke, cancer), smoking status, duration and intensity, educational attainment, and employment.
† Model 1 additionally adjusted for alcohol consumption in the past.
Among women, likewise, no association between an alcohol consumption of 6 g/d or more and mortality was observed compared with light consumption and with or without adjustment for past alcohol consumption. Women who did not consume alcohol at baseline had a higher mortality risk, with a HR of 1·86 (95 % CI 1·12, 3·08). The HR changed only marginally when adjusted for past alcohol consumption. As in men, female light consumers had a lower mortality risk of 0·54 (95 % CI 0·33, 0·89) compared with non-consumers at baseline. Further exclusion of former consumers led to higher, non-significant but more imprecise risk estimates due to low case numbers.
Among current consumers only (men: n 2455 and women: n 1686), an increment of 12 g/d was not associated with mortality: HR was 1·02 (95 % CI 0·96, 1·14) in men (P for trend = 0·24) and 1·09 (95 % CI 0·93, 1·27) in women (P for trend = 0·15). From the restricted cubic spline regression, the associations among consumers and non-consumers also did not appear to be non-linear: P values for non-linearity were P = 0·85 in men and P = 0·27 in women.
Cause-specific analyses, although hampered by small numbers, showed no lower or higher risk for cardiovascular or cancer mortality (Table 3). Higher risks were observed for mortality due to other known causes, for men who were non-consumers at baseline and women whose consumption was above the recommended upper limit. Most frequent causes of death included in this category were diabetes mellitus, diseases of the digestive system (mainly the liver) and respiratory diseases. The competing risk model showed that for women who consumed more than 6 g/d, the risk of dying from other causes or CVD was significantly higher than cancer mortality risk.
(Hazard ratios (HR) and 95 % confidence intervals)
* Age- and centre-stratified and adjusted for diabetes duration, insulin therapy, co-morbidities (heart disease, stroke, cancer), smoking status, duration and intensity, educational attainment, and employment.
† Model 1 additionally adjusted for alcohol consumption in the past.
‡ Values were significantly different for risk estimate derived from competing risk model v. cancer mortality (P = 0·04).
§ Values were significantly different for risk estimate derived from competing risk model v. cancer mortality (P = 0·05).
∥ Values were significantly different for risk estimate derived from competing risk model v. cancer mortality (P = 0·02).
¶ Values were significantly different for risk estimate derived from competing risk model v. cancer mortality (P = 0·03).
Table 4 displays mortality risk in non-consumers at baseline in subcategories of lifetime non-consumers, always moderate consumers and sometimes heavy consumers. In men, higher risk of death was only observed in those who were categorised as individuals who sometimes drank heavily in the past (HR 3·65, 95 % CI 1·62, 8·24). Due to small numbers, only two categories could be built for women; the higher mortality risk was seen in those categorised as former drinkers (HR 2·56, 95 % CI 1·32, 4·93).
(Hazard ratios (HR) and 95 % confidence intervals)
* Age- and centre-stratified and adjusted for diabetes duration, insulin therapy, co-morbidities (heart disease, stroke, cancer), smoking status, duration and intensity, educational attainment, and employment.
Discussion
The present prospective multi-centric study of participants with diabetes mellitus showed no association between current alcohol consumption over 6 g/d and mortality risk compared with light consumption, also when past alcohol consumption was accounted for. Men and women who abstained from alcohol at baseline had a higher mortality risk compared with light consumers. The present study suggests that this higher risk was influenced by their past alcohol consumption rather than their current abstinence.
Several studies have investigated the association between alcohol and mortality in individuals with diabetes. A meta-analysis pooled six prospective studies and showed that mortality risk was decreased for a consumption of >0 and < 6 g/d compared with non-consumers (relative risk 0·64, 95 % CI 0·49, 0·82). Furthermore, risks of fatal and total CHD were decreased with an alcohol consumption of >0 and < 6, 6–18, and ≥ 18 g/d compared with non-consumers(Reference Koppes, Dekker and Hendriks7). Moreover, among 404 diabetic and glucose-intolerant Japanese men, non-daily consumers had a decreased mortality compared with never consumers. No association was observed for former or daily consumers(Reference Nakamura, Ueshima and Kadota6). Considering the different reference category, these results seem to be in line with the present findings.
Regardless of pre-existing diabetes, moderate alcohol consumption has been linked to a lower risk of atherosclerotic diseases and overall mortality compared with non-consumers, which could be explained by beneficial changes in lipid metabolism, endothelial function, inflammation, haemostatic balance, insulin sensitivity and blood pressure(Reference Djousse, Lee and Buring23). These favourable effects are particularly relevant in individuals with diabetes, in whom coronary risk factors are highly prevalent(Reference Howard, Arnsten and Gourevitch11, Reference van de Wiel24). In the general population, the relationship between alcohol and mortality has been consistently described as a J- or U-curve. This was confirmed in a meta-analysis of thirty-four prospective studies, where consumption of up to four drinks per d in men and up to two drinks per d in women was inversely associated with total mortality compared with non-consumers(Reference Di Castelnuovo, Costanzo and Bagnardi4).
We did not observe an association between a current alcohol consumption of more than 6 g/d and mortality risk compared with light consumers. However, the meta-analysis of Koppes et al. (Reference Koppes, Dekker and Hendriks7) has shown that consumption of 0–6 g/d may already give protection compared with abstinence. These findings demonstrate that a consumption above 6 g/d is not associated with mortality compared with a condition that is already considered beneficial. Thus, this does not mean that alcohol consumption is not associated with mortality in diabetic individuals, but that the difference between the protection associated with light consumption and a higher consumption is not significantly different.
These results show that the choice of reference category can have a substantial influence on the interpretation of an observed association between alcohol consumption and mortality. We chose light alcohol consumers as a reference group, because it has been shown that non-consumers (including former drinkers) as well as lifetime abstainers do not constitute a useful reference category. First, lifetime abstinence of alcohol is not normative in many high-income countries. As a consequence, this group is relatively small and members of this group differ from consumers in many other health determinants(Reference Rehm, Irving and Ye9). Second, the inclusion of former drinkers in the reference group when investigating alcohol consumption has been under discussion for over two decades(Reference Costanzo, Di Castelnuovo and Donati10, Reference Shaper, Wannamethee and Walker13). We observed that non-consumers had a higher mortality risk compared with light consumers at baseline when controlled for confounders including co-morbidities. In general, this can be explained by the following: (1) these individuals were lacking the favourable physiological effects of alcohol reported above or (2) these individuals quit due to health reasons. The second explanation supports the hypothesis that the group of non-consumers may include the so-called ‘sick-quitters’, which originates from findings of Shaper et al. (Reference Shaper, Wannamethee and Walker13). This hypothesis reads: a separation of abstainers into lifelong abstainers and former consumers leads to less pronounced or to a complete disappearance of beneficial effects observed in moderate consumers relative to non-consumers. Individuals who quit consuming alcohol due to health reasons are more vulnerable to mortality and, thus, may be responsible for some or most of the higher risk of abstainers(Reference Shaper, Wannamethee and Walker13, Reference Gmel, Gutjahr and Rehm25). However, Costanzo et al. (Reference Costanzo, Di Castelnuovo and Donati10) concluded in a review that the protective effect seen for moderate alcohol consumption does not appear to be substantially related to the fact that former consumers are included in the reference group. In their meta-analysis, Di Castelnuovo et al. (Reference Di Castelnuovo, Costanzo and Bagnardi4) showed that the protective effect of alcohol among the general population was lower when former consumers were excluded, but still significant. On the other hand, Friesema et al. (Reference Friesema, Zwietering and Veenstra12) found that lifetime alcohol consumption was not related to mortality or CVD incidence, whereas current consumption was associated with lower CVD and mortality risk compared with never drinkers. Since our population consisted of individuals with diabetes mellitus, their diagnosis or the existence of co-morbidities are plausible explanations for their decision to quit consuming alcohol. This would mean that their ill health rather than their alcohol consumption drove the association. If participants reduced their alcohol consumption after their diabetes diagnosis but before baseline assessment, this might have underestimated the association with mortality.
A limitation of the study was that the power was low. Therefore, it is possible that we failed to detect a true association between current alcohol consumption and mortality risk. In addition, information on current and past alcohol consumption was based on self-report, which might have introduced recall bias. Due to misreporting or the fact that the questions asked about alcohol consumption at defined ages only, misclassification into categories of lifelong abstainers, former and current consumers may have occurred. Moreover, no information was available about the onset and times and reasons for cessation of alcohol consumption. However, in a large cohort study, methods of assessing current and past alcohol consumption are restricted and a self-administered questionnaire has been judged to generate reliable and valid estimates(Reference Klipstein-Grobusch, Slimani and Krogh20, Reference Friesema, Veenstra and Zwietering26).
In conclusion, these results are in support of the current advice that individuals with diabetes mellitus can consume alcohol within the recommended upper limits. The increased mortality risk among non-consumers appeared to be affected by their past alcohol consumption rather than their current abstinence. In addition, the present study supports the hypothesis that former consumers, abstinent at point of entry into the study, do not constitute a useful reference category. The choice of reference category and the inclusion of former consumers can substantially affect the interpretation of risk estimates.
Acknowledgements
This study was supported by a European Foundation for the Study of Diabetes (EFSD)/Sanofi-Aventis grant. The sponsor did not have any influence on the contents of the manuscript. D. S., U. N. and H. B. designed the study; D. S. analysed and interpreted the data and drafted the manuscript. All authors critically revised the manuscript for important intellectual content. U. N. obtained funding for the study. None of the authors has any conflicts of interest to declare.
