Moderate alcohol drinking is not associated with risk of depression in older adults

The scarce research on the effects of moderate alcohol consumption on mental health among older adults suggests a protective effect against depression. We prospectively examined the association between patterns of moderate alcohol consumption, depression and psychological distress, using information from 5,299 community-dwelling older adults from the ELSA and Seniors-ENRICA cohorts. A Mediterranean drinking pattern (MDP) was defined as moderate alcohol intake (<40 g/day for men; <24 g/day for women) with a preference for wine and drinking only with meals. Depression was ascertained with the 10-item Geriatric Depression Scale (GDS-10), a self-report of clinically-diagnosed depression, or being on anti-depressant medication (Seniors-ENRICA); and with the 8-item Center for Epidemiologic Studies Depression Scale (CES-D) (ELSA). Psychological distress was assessed with the General Health Questionnaire-12 (GHQ-12). Compared to never drinkers, moderate drinkers showed comparable scores on the ENRICA-GDS-10 (PRR (95%CI): 1.03 (0.84–1.26)), the ENRICA-GHQ-12 (0.88 (0.73–1.06)), the ELSA-CES-D (0.92 (0.79–1.06)) and the ELSA-GHQ-12 (0.75 (0.55–1.01). The MDP was not associated with the GDS-10 or GHQ-12 scores, or with clinically-diagnosed depression; however drinkers with a preference for wine showed an increased number of psychological distress symptoms (1.31 (1.03–1.66)). In conclusion, we found no consistent protective association between moderate alcohol consumption and depression in older adults.

Given that the effects of alcohol on mental health might be context-specific, analyzing data from these two cohorts may facilitate determining whether any associations found differ between "Mediterranean" countries like Spain, where alcohol is traditionally used as a nutritional element and most drinking occurs during meals, and "Non-Mediterranean" countries like England, where alcohol is consumed in greater amounts and mainly used as a psychotropic-relaxant. In this work we also assessed the impact of the Mediterranean Drinking Pattern (MDP) on depression and psychological distress. The MDP, characterized by moderate alcohol intake, a preference for wine, and drinking only with meals, is the traditional pattern of alcohol consumption in Spain and other regions of the Mediterranean basin 20 . Previous studies among older adults have reported the MDP to be associated with a reduced risk of functional limitations 21 , frailty 22 and falls 23 , and with better scores on the Physical Component Summary of the12-item Short Form Health Survey 24 . However, its effects over mental health are unknown.

Methods
Study population and design. Seniors-ENRICA cohort. This cohort was established during 2008-2010 with individuals selected by multi-stage stratified random sampling from the non-institutionalized Spanish population aged ≥60 years (wave 1) 25 . For these analyses, we used information from 2,519 subjects participating in wave 2 (2012) and followed through wave 3 (2015) 26 . In these two waves, information regarding sociodemographic variables, health behaviors and morbidity was collected using a computer assisted telephone interview. There is evidence from Spanish population studies that telephone surveys yield similar distributions of health behaviors and morbidities as face-to-face interviews 27,28 . In particular, the Spanish telephone-administered geriatric depression scale has shown a high correlation with the validated personal administration of the scale 27 . An advantage of the telephone-administered geriatric depression scale is that it may increase participation, as depressed individuals are less likely to respond to fac-e-to-face questions on depression 27 . After telephone interviews, a diet history and a physical exam were performed by trained staff during at-home visits.
From the initial sample of 2,519 participants in 2012, 82 (3.3%) died and 616 (24.5%) were lost to follow-up; thus, in 2015 data on depression and psychological distress came from 1821 individuals. Participants remaining in the study were slightly younger, had higher educational level, fewer comorbidities and lower scores in the GDS and GHQ-12 scales. Never drinkers and ex-drinkers were more likely to be lost to follow-up. All participants gave informed written consent, and the Clinical Research Ethics Committee of the La Paz University Hospital in Madrid approved the study. All methods were performed in accordance with the relevant guidelines and regulations.
ELSA cohort. This cohort comprises people aged ≥50 years living in private households in England 29 . The present research analyzed data from Waves 0, 1 and 3. As in previous reports 30,31 , a combination of Wave 0 (drawn from households that had previously responded to the Health Survey from England in 1998, 1999 or 2001) and Wave 1 (which included those households from Wave 0 where at least one member was aged ≥50 years between March 2002 and March 2003), was used to define the baseline sample (n = 11,391). Information from the initial sample was collected using a personal face-to-face Computer Assisted Personal Interview, a nurse visit and a self-completion questionnaire. From the initial sample, we excluded those individuals aged <60 years (n = 4,166).
During follow-up, 241 participants died (3.3%), 1928 (26.7%) refused to participate or could not be contacted, and 488 (6.8%) became ineligible (mainly because they had moved outside of Britain or had been institutionalized). Thus, in 2006-2007, information on depression and physiologic distress was collected from 4,568 participants aged ≥60 years. Participants remaining in the study were more frequently women, younger, had higher educational level, lower tobacco consumption, fewer comorbidities and lower scores in the CES-D. The National Research Ethics Service (MREC/01/2/91) provided the ethical approval for ELSA.

Study variables. Alcohol consumption.
In the Seniors-ENRICA study, alcohol consumption was estimated based on a validated diet history, developed from the one used in the EPIC cohort study in Spain 32 . The diet history collected consumption data for 34 alcoholic beverages and used photographs depicting different glass sizes to help quantify serving sizes. Alcohol content, in grams, was then estimated using standard composition tables. According to the average alcohol intake, participants were classified as (1) never drinkers (this category also included occasional drinkers with average intake close to zero), (2) ex-drinkers (those reporting having stopped drinking for at least a year before the interview), (3) moderate drinkers and 4) heavy drinkers. We set the threshold between moderate and heavy drinking at ≥40 g/day for men and ≥24 g/day for women 33 . Among drinkers (categories 3 and 4 in the previous classification), a preference for a specific type of alcoholic beverage (wine or other) was acknowledged when more than 80% of alcohol intake was derived from such drink 22 . According to the consumption of alcohol with meals (lunch or dinner), drinkers (categories 3 and 4) were also classified into three groups: those who drank exclusively during mealtimes, those who drank exclusively outside meal times, and those who drank at any time. Finally, a Mediterranean Drinking Pattern (MDP) was defined as moderate average intake of alcohol (and no binge drinking), with wine preference and alcohol consumption only with meals 22 .
For the ELSA-cohort analyses, information on alcohol consumption and frequency was derived from Wave 0 but not from Wave 1 because this wave collected information only on the frequency of drinking. According to the average alcohol intake, and having defined 10 grams of pure alcohol as equivalent to an alcohol unit, participants were classified as never drinkers, ex-drinkers, moderate drinkers, and heavy drinkers. The threshold between moderate and heavy drinking was set at 28 units of alcohol/week in men and 21 units/week in women.
Depression. Depression was ascertained with three instruments: The ELSA study used the 8-item version of the CES-D) 35 . A higher score in the CES-D indicates a greater level of depression. (c) Self-reported depression: In the Seniors-ENRICA study this variable was defined as a positive answer to the question "Has a doctor ever told you that you have depression" or as being on antidepressant medication.
Psychological distress. The Seniors-ENRICA and the ELSA studies used the 12-item version of the General Health Questionnaire (GHQ-12), a screening tool for psychological distress. Individuals missing ≥2 items were excluded from the analyses. Higher scores in the GHQ-12 indicate higher psychological distress.
Other variables. Baseline self-reported data on age, sex, educational status, and tobacco consumption were collected in both cohorts. In the Seniors-ENRICA cohort, information on physical activity (metabolic equivalent-hour/week) and on the number of hours/day spent watching television was assessed using validated questionnaires 36,37 . In the ELSA study, participants were asked about the frequency (more than once a week, once a week, one to three times a month, or hardly ever/never) of participation in vigorous-, moderate-, and light-intensity activities. An overall measure of physical activity was generated by multiplying activity frequency by the metabolic equivalent value for each type of activity (6 for vigorous-, 3 for moderate-, and 1.5 for light-intensity activities, respectively). In the Seniors-ENRICA, adherence to the Mediterranean diet was captured using the Mediterranean Adherence Screener (MEDAS) index (excluding the alcohol component) 38 , and energy intake was calculated based on Spanish food composition tables. In both cohorts, baseline weight and height were measured with standard methods. BMI was calculated as (weight in kg)/(height in m) 2 . Normal weight was defined as a BMI <25, overweight as a BMI 25-29.9, and obesity as a BMI ≥30. Finally, in both cohorts, participants also reported whether they had previously been diagnosed with any of the following diseases: cardiovascular disease (ischemic heart disease, stroke, heart failure), diabetes, chronic lung disease (asthma, chronic bronchitis), and osteomuscular disease (osteoarthritis or arthritis); as well as the number of medications they were taking. In the Seniors-ENRICA, study staff compared self-reported information on medication use with medication packages stored at home.

Statistical analysis.
From the initial 1,821 participants in the Seniors-ENRICA cohort, we excluded 200 (11%) subjects with incomplete data on alcohol consumption or potential confounders. Additionally, for analyses on depression we excluded subjects lacking information on the GDS at baseline or follow-up (n = 411; 22.6%). For analyses on psychological distress we excluded individuals with missing data on the GHQ-12 at baseline or follow-up (n = 31; 1.7%). Similarly, from the initial 4,568 participants in the ELSA cohort, 440 (9.6%) subjects were excluded due to incomplete data on alcohol consumption or potential confounders. Also, we excluded individuals with missing information on the CES-D at baseline or follow-up (n = 157; 3.4%) for analyses on depression, and individuals with no data on the GHQ-12 at baseline or follow-up (n = 768; 16.8%) for analyses on psychological distress. Baseline socio-demographic, lifestyle, and clinical characteristics of individuals with and without missing information were similar, although the latter were on average younger, were more likely to be moderate drinkers, and showed a lower prevalence of diabetes and osteomuscular disease. Associations between alcohol consumption and depression or psychological distress were analyzed with generalized regression models because of the highly skewed distribution of the dependent variables (GDS-10, GHQ-12, and CES-D), which mirrored that of count variables (see Fig. 1). To account for over-dispersion and a higher number of zero values than would be expected under the Poisson distribution, we considered using either negative binomial or zero-inflated negative binomial models. Both approaches were compared using Vuong tests which suggested that zero-inflated negative binomial models provided a better fit of the data. Results were expressed as prevalence rate ratios (PRR) and their 95% confidence interval (CI). PRRs in this context represent the percent change in depression and psychological distress scales for one category of alcohol relative to the reference category, holding other variables constant. When the dependent variable was self-reported diagnosed depression, the analyses were performed using Poisson regression with a robust error variance, and results were expressed as Incidence rate ratios (IRR) and their 95% confidence interval.
We built three regression models with progressive levels of adjustment. Model 1 adjusted for age, sex, education, and the corresponding baseline values of the dependent variables. Model 2 further adjusted for tobacco consumption, physical activity, BMI, cardiovascular disease, diabetes, chronic lung disease, osteomuscular disease, and cancer in both cohorts. Additionally, the Seniors-ENRICA model 2 adjusted for time watching TV, the MEDAS score, total energy intake, and use of sleeping pills or other central nervous system-acting medications. Analyses of the influence of average alcohol consumption and the MDP were conducted in the total sample, whereas analyses of the influence of beverage preference and drinking with meals were performed only among drinkers. Models for specific drinking habits (i.e. beverage preference, drinking with meals or the MDP) were further adjusted for total alcohol consumption (grams/day).
We conducted separate sensitivity analyses. First, we excluded subjects with baseline cardiovascular disease or cancer as well as participants under anxiolytic, antipsychotic, or antiepileptic medication at baseline. Second, to facilitate comparisons between cohorts, we reran analyses with abstainers (never drinkers and ex-drinkers) as the reference category, and reduced the threshold between moderate and heavy intake to 15 g/day for both genders. Third, we split moderate consumption into two categories: below or above the median average alcohol intake. Fourth, we evaluated the risk of depression as defined by a score above 3 in the CESD or GDS using logistic regression. Finally, we modeled the association between alcohol consumption patterns and changes (defined as difference between follow-up and baseline values) in the GDS-10, GHQ-12, and CES-D, using the VCE (bootstrap) command in Stata to estimate confidence intervals based on 100 resamples (see Supplementary Tables  and 2). Our results are presented for both sexes combined because sex-alcohol drinking patterns interactions terms included in regression models failed to reach statistical significance and findings were similar in each sex.
Statistical analyses were performed using Stata ® , version 13.1.
Data availability. Data from the Seniors-ENRICA can be accessed upon request to the authors. Data from the ELSA study are available upon request to the UK Data Service

Results
The prevalence of ex-drinkers, moderate drinkers and heavy drinkers at baseline was 7%, 62% and 5% in Seniors-ENRICA, and 5%, 80% and 10% in ELSA, respectively. and Compared to never-drinkers, ex-drinkers were more often male, had higher education, and were more likely to report unhealthy characteristics (obesity, cardiovascular disease, cancer, and use of sleeping pills and other central nervous system-acting medications) (please see Table 1). Both moderate and heavy drinkers were younger, more frequently male, more educated, and, despite lower levels of physical activity, healthier (lower prevalence of obesity, chronic disease, and medication use, and more favorable baseline scores on the GDS-10, GHQ-12 and CESD) ( Table 1). According to the GDS (GDS >3) and CESD (CESD >3) scores, at baseline 7.7% (Seniors-ENRICA) and 14.3% (ELSA) of participants had a diagnosis of depression. Similarly, a physician diagnosis of depression was reported at baseline by 7.5% of participants in the Seniors-ENRICA.
In the Seniors-ENRICA study, drinkers with a preference for wine, who drank only with meals, or who presented a MDP pattern were more likely to be female, slightly older, more physically active, and less frequently obese than their counterparts (Table 2). Those same three types of drinker had a lower prevalence of smoking, higher MEDAS scores, a lower prevalence of respiratory disease, and a higher prevalence of cardiovascular disease, diabetes and osteomuscular disease than their counterparts. The most favorable scores in both the GDS-10 and the GHQ-12 were observed among drinkers with a preference for alcoholic beverages other than wine, those who drank with and outside of meals, and those with no MDP ( Table 2).
The distributions of the GDS-10, CES-D, and GHQ-12 scores at baseline and at follow-up can be observed in Fig. 1. As aforementioned, the most common score in all scales was zero and distributions were all highly skewed to the right. After a 2.8 (ENRICA) and a 7.4-year-mean follow-up (ELSA), 4% and 10% of participants developed incident depression (GDS or CESD scores >3), respectively. Table 3 displays the results for the prospective association between alcohol consumption patterns and follow-up GDS-10 and GHQ-12 scores in the Seniors-ENRICA study. Compared to never-drinkers, moderate alcohol consumers showed similar scores in the GDS-10 (PRR: In the Seniors-ENRICA, we found no evidence of an association between moderate alcohol intake (IRR: 1.14; 95%CI: 0.57-2.29) or the MDP (IRR: 1.84; 95%CI: 0.80-4.25) and incidence of self-reported depression (defined as a physician-diagnosis of depression or as being on antidepressant medication) (Supplementary Table 3). Results for heavy drinking, beverage preference, or drinking with meals and self-reported depression could not be interpreted due to a low number of incident self-reported depression cases (Supplementary Table 3).
Results for the prospective association between average alcohol consumption and follow-up CES-D and GHQ-12 scores in the ELSA cohort can be found in Table 4. Compared to never drinkers, both moderate and heavy drinkers showed similar percent changes in depression and psychologic wellbeing scales. In this sense, the PRR (95%CI) for CESD-D was 0.92 (0.79-1.06) in moderate drinkers, and 1.00 (0.84-1.20) in heavy drinkers; the corresponding results for the GHQ-12 were 0.75 (0.55-1.01) and 0.84 (0.58-1.21), respectively.
Sensitivity analyses yielded similar results to those presented in the main tables (data not shown).

Discussion
Overall, we found no consistent protective association between moderate alcohol consumption and depression in older adults. Moreover, in the Seniors-ENRICA cohort, adherence to certain alcohol consumption patterns such as drinking only with meals or the MDP failed to show an association with either of the studied mental health outcomes. However, results showed some non-significant tendency to a lower risk of depression (Spanish sample) and psychological distress (both samples) associated with heavy drinking, which should be further investigated in studies with a larger number of heavy drinkers. Previous research on the link between moderate alcohol intake and mental health in adults has yielded inconsistent results. In the National Psychiatric Morbidity Survey in England, which followed 2,413 adults for 18 months, moderate drinking (defined as drinking <21 or <14 units/week in men or women, respectively, or having an AUDIT score <8 points), showed no association with incident anxiety or depression (measured with the Clinical Interview Schedule-Revised instrument) when compared to abstaining 17 . By contrast, Norway's Nord-Trøndelag Health Survey of 38,930 individuals with a mean age of 48 years, ascertained that low-to-moderate drinkers had a lower risk of case-level anxiety and depression (based on Hospital Anxiety and Depression Rating Scale) than non-drinkers (self-reported abstainers and non-abstainers currently consuming no alcohol). Unfortunately, the study's partial cross-sectional design hindered drawing causal inference 39 . Also, a Spanish cohort study of 13,619 university graduates (mean age: 38 years) followed for 10 years showed that female -but not male-moderate drinkers (5-15 g/day) were less likely to self-report suffering from depression than abstainers 15 . However, a study of midlife alcohol consumption of the Whitehall II cohort of British civil servants, concluded that there was no association between drinking patterns, including a moderate (<40 g/day) one, and risk of depression. The 28-year follow-up study consisted of 7,478 men and women between the ages of 35 and 55 years and depression was assessed by the GHQ-30 40 . More recently, results based on 3,201 individuals aged 18-65 years, who were primary care attenders in the WHO collaborative study of Psychological Problems in General Health Care, suggested a possible protective effect of moderate (defined as AUDIT <8 points) drinking in the risk of developing a new onset of depression and generalized anxiety disorder (GAD) 14 .
To our knowledge only one study has focused on older adults. Gea et al. followed 5,505 high-risk men and women (55 to 80 y) from the PREDIMED Trial. Unlike the WHO collaborative study, participants in PREDIMED were initially free of depression or a history of depression, and did not have any history of alcohol-related problems. After 7 years of follow-up, moderate drinkers (5-15 g/day) of both genders were less likely to self-report a diagnosis of depression than abstainers 16 . Specifically, wine consumption in the range of 2-7 drinks/week decreased the risk of incident depression in this population 16 . Results in our study using abstainers (never drinkers and ex-drinkers) as the reference category or changing the threshold between moderate and heavy intake to 15 g/day showed no association between moderate alcohol intake and depression. A partial explanation for the discrepant results between the PREDIMED study and ours might be that participants in the PREDIMED trial were primary-care patients participating in an intensive group-based health education intervention; thus, we speculate that the intervention-related social support very likely geared towards a greater adherence to a Mediterranean diet (including moderate drinking) 41 , received by PREDIMED participants may have contributed to a lower depression risk 42 . Moreover, greater adherence to a healthy diet also reduces the incidence of cardiovascular disease, diabetes and other chronic diseases which, in turn, are strong risk factors or triggers for depression.
In some cases, differences between clinic-based studies 14,16 -which found a protective association between moderate alcohol intake and mental health-and studies in community samples 40 which did not find such association (i.e. ours), could be partially explained by the self-medication hypothesis, whereby alcohol intake is used to ameliorate mental and other health problems; this phenomenon could operate more strongly in clinical than community samples due to their worse health status. Our work has several strengths compared to published research. First, unlike previous studies [14][15][16] , we distinguished between never drinkers and ex-drinkers. Since ex-drinkers might have quit drinking due to ill health, including depression, studies incorporating ex-drinkers in the reference category may overestimate the protective effect of moderate alcohol drinking on depression. Second, we adjusted our analyses for a larger number of lifestyle and morbidity variables than previous reports 14,15,17 . Third, as far as we are aware of, this is the first study to provide detailed analyses on the association between different patterns of moderate alcohol consumption, including the MDP, and mental health outcomes 16 . Fourth, we used a validated diet history to estimate alcohol consumption in the Seniors-ENRICA cohort. In fact, the Pearson correlation coefficient between alcohol consumption data from the diet history and seven 24-h recalls over the course of one year was a satisfactory 0.65 32 . Finally, we obtained consistent results using two cohorts characterized by different patterns of alcohol consumption and different follow-up periods. This is important because it suggests that the association, or lack thereof, between moderate alcohol intake and depression does not differ by cultural and social context, i.e., it may be similar in "wet and "dry" countries. Additionally, the ELSA 7.4-year mean follow-up suggests that the failure to detect an association is not due to failing to follow individuals "long enough to see an effect".
The main limitation of the study was that alcohol consumption was self-reported with the corresponding risk of recall error and social desirability bias. It is known that questions on alcohol habits are commonly subject to under-reporting, particularly among heavy drinkers 43 . In this sense, although numbers were similar to those reported in previous population-based studies 44,45 , the percentage of heavy drinkers in our cohorts was rather low (5% in Spain and 10% in the UK). In any case, there is evidence that telephone interviews are less affected by social desirability bias than personal interviews. Second, we lacked repeated measurements of alcohol intake  Table 3. Prospective association between alcohol consumption patterns and Geriatric Depression Scale (GDS) and the General Health Questionnaire-12 (GHQ-12) scores for older adults followed-up for 2.8 years; Seniors-ENRICA cohort study. PRR: Prevalence rate ratio; CI: Confidence interval. a Moderate alcohol consumption with preference for wine and drinking only with meals. Model 1: Adjusted for the age, sex, educational level (≤primary, secondary, university) and baseline: Geriatric Depression Scale or General Health Questionnaire −12 scores. Model 2: Additionally adjusted for: tobacco smoking (never/former/current), physical activity (MET-h/week), time watching TV (h/week), MEDAS index (excluding alcohol), total energy intake (Kcal/ day), Body Mass Index (≤25, 25-29.9, ≥30 kg/m 2 ), cardiovascular disease, respiratory disease, osteo-muscular disease, diabetes, cancer, sleeping pills and central nervous system-acting medications. Models for specific drinking habits (i.e. beverage preference, drinking with meals or the MDP) were further adjusted for average alcohol consumption (grams/day). during follow-up to update the information on exposure to alcohol. Third, although we used validated depression and psychological distress scales, these were not confirmed by a clinical diagnosis. Thus, we cannot rule out the existence of diagnostic bias, which effect on these results remains unknown. However, in the Seniors-ENRICA, were we had information on previous physician-diagnoses of depression or being under antidepressant treatment (which was confirmed against drug packages at home), we observed a similar prevalence of depression either using this information or the GDS scale. Fourth, our results may not be generalizable to institutionalized older adults. Finally, in both cohorts we observed an association between heavy drinking and decreased number of psychological distress symptoms approaching statistical significance. However, the small number of heavy drinkers in this cohort makes these results unreliable. Table 4. Prospective association between average alcohol consumption and Center for Epidemiologic Studies Depression Scale (CES-D) and the General Health Questionnaire-12 (GHQ-12) scores for older adults followed-up for 7.4 years; English Longitudinal Study of Ageing (ELSA). PRR: Prevalence rate ratios; CI: Confidence interval. Model 1: Adjusted for the age, sex, educational level (≤primary, secondary, university) and baseline CESD scores or GHQ-12 scores. Model 2: Adjusted additionally for tobacco smoking (never/ former/current), physical activity, BMI (≤25, 25-29.9, ≥30 kg/m 2 ), cardiovascular disease, respiratory disease, osteomuscular disease, diabetes, and cancer.