Alcohol consumption predicts incidence of depressive episodes across 10 years among older adults in 19 countries

Abstract

Alcohol consumption is increasing in many countries, and excessive alcohol consumption is particularly increasing among older adults. Excessive alcohol consumption causes morbidity and mortality, especially among older adults, including an increased risk of depressive episodes. We review the mechanisms through which alcohol consumption may affect depression, and argue that the effects of alcohol consumption on depressive episodes among older adults are understudied. We harmonized data among older adults (≥50 years) on alcohol consumption, depressive episodes, and an array of risk factors across 10 years and 19 countries (N=57,276). Alcohol consumption was categorized as current or long-term abstainer, occasional, moderate and heavy drinking at an average of 2.3 follow-up time points. Depressive episodes were measured through the CES-D or EURO-D. Multi-level Cox proportional frailty models in which the random effect has a multiplicative relationship to hazard were estimated with controls for co-occurring medical conditions, health behaviors, and demographics. Long-term alcohol abstainers had a higher hazard of depressive episodes (HR=1.14, 95% C.I. 1.08–1.21), as did those reporting occasional (HR=1.16, 95% C.I. 1.10–1.21) and heavy drinking (HR=1.22, 95% C.I. 1.13–1.30), compared with moderate drinking. Hazard ratios were attenuated in frailty models; heavy drinking, however, remained robustly associated in a random-effects model with a frailty component (HR=1.16, 95% C.I. 1.11–1.21). Interactions were observed by gender and smoking status: long-term abstainers, women’s, and smokers’ (HR for interaction, 1.04, 95% C.I. 1.00–1.07) hazards of depressive episodes increased more than what would be expected based on their multiplicative effects, when compared to moderate drinking, non-smoking men. Excessive alcohol consumption among older adults is a concern not only for physical, but also for mental health. Physician efforts to screen older adults for excessive alcohol use is critical for mental health to remain strong in aging populations.

1. Introduction

Alcohol consumption varies considerably across countries, and available evidence indicates that use is generally increasing across the world. Through 2017, the WHO has been collecting information on per-capita consumption information, both recorded consumption from sales and taxation records or through production reports, as well as estimates of unrecorded consumption (e.g., through homemade beverages) (Lachenmeier & Rehm, 2009), throughout member states (Manthey et al., 2019). While changes over time are heterogeneous, on average alcohol abstinence has decreased by an average of 6.5%, and per-capital consumption increased by approximately 9%, across the world through 2017 (Manthey et al., 2019).

Growing evidence also indicates that the global increase in alcohol consumption is disproportionately among older adults (Keyes, Jager, Mal-Sarkar, et al., 2019). Historical increases of alcohol consumption in older age are due in part to adults living longer and healthier, given that predominant reasons for declining drinking across the adult lifespan are the development of chronic diseases and contraindications with medication (Naimi, Stockwell, Zhao, et al., 2017). Data from the United States indicate that the prevalence of binge drinking among those age 50–64 increased from 15.5% in 2005–2006 to 19.1% in 2013–2014, and among those over 65 increased from 8.1% to 9.0% (Han, Moore, Sherman, Keyes, & Palamar, 2017). Alcohol disorders among adults >65 are also rapidly increasing (one could cite the numbers here?) (Grant, Chou, Saha, et al., 2017). In cross-country harmonized survey data, current abstainers among adults >50 decreased in 17 of the 22 countries (Calvo et al., 2019). Confirmatory evidence indicates increases in alcohol consumption among older adults in China and many European countries as well (Breslow, Castle, Chen, & Graubard, 2017; Han et al., 2017; Kim, Lee, Lee, et al., 2012; World Health Organization, 2014). As populations are getting older, and drinking more, sustained research that aims to address potential health consequences of alcohol consumption in older age is critical.

Increases in alcohol consumption, globally, and among older adults in particular are concerning, given that heavy alcohol use is a major contributor to global morbidity and mortality (Rehm et al., 2009), especially among those in older age. Across all ages, heavy alcohol use is estimated to contribute to 3.8% of deaths and 4.6% of disability-adjusted life years, an estimate that is likely growing as alcohol consumption becomes more common. Yet alcohol can cause arguably greater harm in older ages. Alcohol is metabolized more slowly in older age compared with younger ages (Cederbaum, 2012; Meier & Seitz, 2008), and percentage of body weight that is water decreases (Cederbaum, 2012; Wang, Nicholson, Jones, Fitzhugh, & Westerfield, 1992), leading to higher blood alcohol levels at lower levels of consumption. Medication usage that is contraindicated with alcohol consumption substantially increases at older ages, and is associated with morbidity and mortality (Lehmann & Fingerhood, 2018). Falls and other injuries related to alcohol consumption become more prevalent (Lehmann & Fingerhood, 2018). In the United States, the majority of deaths from chronic (~78%) and a substantial proportion of deaths from acute (~40%) alcohol consumption occur after age 55 (CDC, 2015).

While alcohol consumption and its association with health and mortality have been extensively studied (Stockwell et al., 2016), available data indicate heavy alcohol consumption is a cause of increased morbidity and mortality, and that the association between alcohol consumption and health varies by factors such as gender and co-occurring health behaviors such as smoking (Kerr, Greenfield, Bond, Ye, & Rehm, 2011; Keyes & Miech, 2013; Rehm & Sempos, 1995); the extent to which these moderating factors would also affect the association between alcohol consumption and depressive episodes is critical, given that both gender (i.e., being female) and smoking are aggravating the mortality/morbidity risk (Fluharty, Taylor, Grabski, & Munafò, 2017; Piccinelli & Wilkinson, 2000).

Among the potential health correlates of heavy alcohol use, depressive disorder, comprised by multiple periods of depressive episodes, are of particular concern (Ramsey, Engler, & Stein, 2005). Depressive episodes are serious clinical disorders characterized by relatively unrelenting periods of sadness and anhedonia as well as somatic symptoms such as sleep and appetite changes (Diagnostic and Statistical Manual of Mental Disorders, 2013). Depressive symptoms that may not rise to the level of depressive disorder diagnoses also may be disabling (Preisig, Merikangas, & Angst, 2001; Wells, Burnam, Rogers, Hays, & Camp, 1992). The relationship between alcohol use and depressive episodes has been extensively studied in young populations (Cairns, Yap, Pilkington, & Jorm, 2014; Holahan, Moos, Holahan, Cronkite, & Randall, 2003; Pedrelli, Shapero, Archibald, & Dale, 2016), and among adults evidence indicates a bidirectional and reciprocal relationship between excessive alcohol use and depressive episodes (Boschloo, Vogelzangs, van den Brink, et al., 2012; Graham, Massak, Demers, & Rehm, 2007; Hasin & Grant, 2002; Rodgers et al., 2000); however, there is limited investigation of alcohol use as a predictor of depressive episodes in older adults specifically. There are also depressive symptoms that might not reach the level of disorder but are a serious burden in older age as they face increased losses of close relationships, loneliness, and illness.

The prevalence of depressive episodes varies across countries (Bromet, Andrade, Hwang, et al., 2011; Kessler & Bromet, 2013). Data from the World Health Organization World Mental Health Survey Initiative, for example, found a past-year prevalence vary between approximately 10–20% across high-income countries, with Germany and Italy exhibiting the lowest prevalence while the United States and France exhibited the highest prevalence. Prevalence in low-income countries was generally lower than high-income countries, ranging from 6.5% in China to 18.4% in Brazil (Kessler & Bromet, 2013). While the reasons underlying cross-cultural differences in depressive episodes have not been extensively investigated, the higher prevalence of depressive episodes in high-income countries is notable, and some have hypothesized that depressive episodes are to some degree an “illness of affluence,” or that within-country inequality or relative comparison may to some degree underlie higher prevalence (Bromet et al., 2011), although such hypotheses remain speculative. These studies have not, however, explicitly examined depressive episodes among older adults. Such data are important, because depression-related morbidity is high (Friedrich, 2017), as is mortality directly related to depressive symptoms and other mental health problems such as suicide (Chesney, Goodwin, & Fazel, 2014; Cuijpers & Smit, 2002; Gilman et al., 2017). In the United States, among all suicide deaths, approximately 17% are estimated to occur among individuals over 65, with men overrepresented in death by suicide compared with women (Stone, Simon, Fowler, et al., 2018).

Epidemiological studies of the relationship between alcohol use and depressive episodes have focused on whether their correlation is produced by (i) shared common causes, (ii) alcohol use leading to depression, (iii) depression leading to alcohol use, or (iv) a combination of multiple pathways. Most of this work has not specifically focused on older adults. With regard to alcohol use and depression sharing common causes, there is evidence that there are common genetic as well environmental causes of both heavy alcohol use and depression, though available data focus on adolescents and young adults over older adults. For example, stressful life events such as relationship instability and relationship loss, job loss, financial instability, and traumatic events, predict increases in both heavy alcohol use and depressive symptoms (Keyes, Hatzenbuehler, Grant, & Hasin, 2012; Keyes, Hatzenbuehler, & Hasin, 2011). However, rigorous epidemiological studies including twin designs that control for genetic sources of confounding (Kendler, Heath, Neale, Kessler, & Eaves, 1993; Kuo, Gardner, Kendler, & Prescott, 2006; Lyons, Schultz, Neale, et al., 2006), and other observational studies have used a range of additional confounder control methods such as covariate adjustment with varying measures assessed as fixed and time-varying effects (Brook, Brook, Zhang, Cohen, & Whiteman, 2002; Crum, Green, Storr, et al., 2008; de Graaf, Bijl, Spijker, Beekman, & Vollebergh, 2003; Falk, Yi, & Hilton, 2008; Fergusson, Boden, & Horwood, 2009; Grant & Harford, 1995; Sihvola et al., 2008), indicate that a direct relationship remains after controlling for sources of confounding.

With regard to the temporal ordering of heavy drinking and depressive symptoms, available evidence indicates a bidirectional and mutually reinforcing relationship. The strongest relationship, however, was found for heavy drinking predicting the onset of depressive episodes (Boden & Fergusson, 2011). For example, in a population-based adult cohort in Denmark with 26 years of follow-up, depressive episodes were more commonly documented subsequent to onset of alcohol use disorders compared with depressive symptoms before the onset of alcohol use disorders. This suggests that while the relationship between depressive symptoms and heavy drinking may be bidirectional and reinforcing, the risk of depressive episodes after the onset of heavy drinking is higher than vice versa (Flensborg-Madsen et al., 2009). Additional studies in other population-based cohorts with extensive follow-up have documented similar longitudinal patterns of associations, supporting the proposition that heavy alcohol use is a cause of new depressive episodes (Fergusson et al., 2009).

The mechanisms through which heavy alcohol use prospectively predicts onset of depressive symptoms include both socially and biologically mediated pathways. Consequences of chronic heavy drinking that may elicit mood changes, aggression (Heinz, Beck, Meyer-Lindenberg, Sterzer, & Heinz, 2011), attentional and productivity deficits include relationship, job, and financial instability (Keyes et al., 2012), which in turn may trigger negative mood and depressive symptoms. Further, biological underpinnings of the relationship between heavy drinking and depressive symptoms have been evaluated in both human and animal models, and reviewed previously in the literature (Boden & Fergusson, 2011). Chronic and heavy alcohol consumption can cause depressive episodes by affecting neural regulation involved with depressive episodes such as dopamine and GABA receptors (Banerjee, 2014; Davies, 2003; Koob, Roberts, Schulteis, et al., 1998; Schuckit, 1994). Further, enzymes involved in folate metabolism may be affected by chronic heavy alcohol use, which may in turn lead to depressive symptoms (McEachin, Keller, Saunders, & McInnis, 2008), and sleep patterns that may be disrupted because of drinking may also underlie the development of depressed mood (Sjoholm et al., 2010). Taken together, the available evidence indicates that while heavy drinking and depressive episodes share common causes, there is a residual relationship that is not explained by common causes, with heavy alcohol use predicting the onset and persistence of new depressive episodes. The mechanisms, highlighted above, through which this effect is hypothesized to be mediated include social factors, such as social role loss and dysfunction, as well as an array of biologically mediated characteristics involved with neural regulation, folate metabolism, and sleep.

Given that drinking is increasing among older adults of recent cohorts, a population among whom stressful life events are already high (Brilman & Ormel, 2001; Chou & Chi, 2000; Luppa, Sikorski, Luck, et al., 2012), continued investigation regarding the level and extent of dynamic and time-varying alcohol consumption that may influence depressive episode risk is critical for public health of older populations. Life-course models of alcohol consumption demonstrate that initiation of drinking behavior begins in adolescence and early adulthood and patterns of drinking are often established during the transition to middle adulthood (Merline, Jager, & Schulenberg, 2008; O’Neill, Britton, Brunner, & Bell, 2017; Schulenberg & Maslowsky, 2009). But drinking behavior in older age is still dynamic (Perreira & Sloan, 2001), with events such as remarriage, divorce, widowhood, and retirement predicting increases, even after controlling for problem-drinking histories. Further, intervention programs that focus on older age are effective (Schonfeld et al., 2015).

In addition to characterizing the overall relationship between levels of alcohol consumption and risk for depressive episodes, understanding how this risk differs across countries (Rehm et al., 2009) provides a cross-cultural lens that is important for understanding potential points of intervention at a policy level. Both alcohol consumption and risk of depressive episodes vary substantially across the world. The level of alcohol consumption and heavy drinking in a particular population is also determined by cultural norms around alcohol consumption (countries in which there is a cultural of heavier drinking, have, unsurprisingly, more heavy drinkers) (Skog, 1985), taxes and trade policy (Chaloupka & Tauras, 2011; Wagenaar, Salois, & Komro, 2009; Xu & Chaloupka, 2011), as well as other country-level factors such as laws around alcohol outlets and marketing, as well as country economy and employment (Anderson, Chisholm, & Fuhr, 2009). Depressive episodes also vary by country-specific factors such as education and access to resources for housing and financial stability (Seedat, Scott, Angermeyer, et al., 2009), employment opportunities and country wealth (Bromet et al., 2011; Lorant et al., 2003), as well as exposure to traumatic events such as war and terrorism (Salguero, Fernández-Berrocal, Iruarrizaga, Cano-Vindel, & Galea, 2011). Despite the importance of these population-level factors, however, characterizing variation in these associations across countries is limited by the availability of harmonized data alcohol use and other measures among older adults.

Harmonization of survey data from across different countries to elucidate associations between alcohol use and health outcomes represents an advance in the existing literature. Existing cross-country comparisons are primarily limited to sales and administrative data on alcohol consumption as a proxy for patterns of consumption (Poznyak et al., 2013; Rehm & Poznyak, 2015). These sources can be augmented with survey sources as available (Slade, Chiu, Glantz, et al., 2016), which allow for a depth of understanding regarding cross-country differences. Harmonizing data can be methodologically complex, however, as questions may cover different time frames (e.g., past week, past year), and may have different questions with regard to patterns of alcohol use resulting in measurement invariance. Assessing such variation in the harmonization process, providing methodological overview of harmonization assumptions, and validity assessments comparing harmonized measures to existing alcohol consumption assessments provide a framework to utilizing varied data sources together in order to advance the literature.

To elucidate the risk of depressive episodes based on alcohol consumption among older adults, we harmonized data on alcohol consumption, depressive episodes, as well as an array of risk factors across 19 countries and across an average of 4.23 years of observation in order to systematically assess the relationship between alcohol consumption and depressive episodes among older adults. We assess (i) variation in depressive episodes by country, (ii) include interactions by gender and smoking, and test (iii) the strength of association between alcohol consumption and depressive episodes after control for time-fixed and time-varying confounders. Further, we provide a descriptive examination of the variation in the strength of the relationship between alcohol consumption and depressive episode risk across countries.

2. Methods

2.1. Data and sample

Longitudinal data from 19 countries collected between 2004 and 2014 were drawn from five ongoing cohort studies of individuals aged 50 and over. We only included data from individuals who did not report depressive episodes at baseline, in order to estimate onset of new episodes of depressive episodes during the study period. Out of 85,523 individuals in the initial sample, 22,563 were dropped because of depressive episodes (see below for definition of depressive episode used in this exclusion criterion) at baseline, and 5684 because of missing data, resulting in an analytic sample of 57,276 individuals.

Table A1 lists countries, years and span of data collection, and the sample size (number of observations and number of unique individuals) among those who provided information on alcohol consumption at least one time. The surveys included were: Survey of Health, Ageing and Retirement in Europe (SHARE; Austria, Belgium, Czech Republic, Denmark, Estonia, France, Germany, Israel, Italy, Netherlands, Poland, Slovenia, Spain, Sweden, Switzerland) (Borsch-Supan, Brandt, Hunkler, et al., 2013), China Health and Retirement Longitudinal Study (CHARLS; China) (Zhao, Hu, Smith, Strauss, & Yang, 2014), English Longitudinal Study of Aging (ELSA; England) (Steptoe, Breeze, Banks, & Nazroo, 2013), Korean Longitudinal Study of Aging (KLOSA; Korea) (Jang, 2015), and Health and Retirement Study (HRS; United States) (Sonnega et al., 2014).

2.2. Alcohol consumption measurements

We harmonized drinking status across surveys, as discussed in other publications (Calvo et al., 2019). To summarize, although alcohol questions and responses varied across questionnaires, we systematically developed and tested a harmonization procedure resulting in a drinking status variable that is comparable across countries and time. We classified individuals as moderate, occasional, and heavy drinkers using harmonized measures of frequency, quantity, drinks per day, and binge drinking. While measures were in some cases heterogeneous across surveys, all surveys asked current drinking status, and the quantity and frequency of current consumption patterns. Quantities were harmonized across surveys by converting drinks to standard sizes and ethanol contents, and converting frequency of drinking into days per week (Kelly & Mozayani, 2012; NIAAA, n.d.). Most surveys (CHARLS, KLOSA, MHAS, SHARE, and HRS in wave 4 an onward) also asked if the respondents have ever consumed alcohol, which allows us to separate current from lifetime abstainers. In surveys that did not ask this question (ELSA and EPS), we separated current from long-term abstainers based on the number of subsequent surveys in which they reported “no consumption.” We used information on drinks per day to classify moderate, occasional, and heavy drinking status. Heavy drinking was based on quantity and frequency of consumption, and was supplemented by a question on binge drinking asked in HRS, MHAS, and SHARE. Our definition of heavy drinking combines regular heavy drinking and single episodes of binge drinking following well-validated clinical cut-points (NIAAA, 2005): moderate drinking was defined as drinking 1 or more days per week, with drinks per day ≤3 for men or ≤2 for women, and no binge drinking (>4 for men or >3 for women) in a single day. Heavy drinking was defined as >3 drinks per day on drinking days for men or >4 in a single day, and women have >2 drinks per day on drinking days or any instance of >3 in a single day. Validation analyses indicate that the ranking of countries from high to low prevalence of heavy drinking corresponds to external data sources (Calvo et al., 2019). Table 1 provides a descriptive overview of the proportion of long-term and current abstainers, occasional, moderate, and heavy drinkers, at baseline and the last wave of follow-up. Overall, moderate drinking was the most prevalent drinking category at baseline and last-wave of follow-up, with 38.5% (95% C.I. 38.3–38.8) reporting moderate drinking at baseline, and 38.2% (95% C.I. 37.8–38.6) reporting moderate drinking at last-wave follow-up.

Table 1.

Descriptive statistics of a sample of adults age 50 and over in 19 countries N=57,276.

	Baseline		Last-wave follow-up
	Mean	95% CI	Mean	95% CI
Depressive episode %	0.0	0.00–0.00	23.9	23.6–24.3
Long-time abstainer %	18.3	18.14–18.52	17.2	16.9–17.5
Current abstainer %	14.2	14.03–14.38	20.2	19.9–20.5
Occasional drinker %	22.9	22.67–23.09	17.6	17.2–17.9
Moderate drinker %	38.5	38.29–38.77	38.2	37.8–38.6
Heavy drinker %	6.1	5.95–6.18	6.8	6.7–7.1
Smoker %	19.2	18.99–19.37	15.7	15.4–16.0
Underweight %	1.4	1.35–1.47	1.7	1.6–1.8
Normal weight %	37.4	37.17–37.64	38.2	37.7–38.5
Overweight %	38.9	38.61–39.09	38.7	38.3–39.1
Obese %	22.3	22.13–22.54	21.5	21.2–21.9
ADLs any %	6.8	6.64–6.89	9.0	8.8–9.2
Very bad self-reported health %	5.8	5.72–5.95	8.3	8.1–8.5
Bad self-reported health %	21.3	21.09–21.49	25.2	24.9–25.6
Regular self-reported health %	37.1	36.85–37.32	36.9	36.5–37.3
Good self-reported health %	25.6	25.42–25.86	21.8	21.5–22.2
Very good self-reported health %	10.2	10.00–10.30	7.9	7.5–7.9
Heart %	17.1	16.91–17.28	19.1	18.7–19.4
Diabetes %	15.2	15.03–15.38	16.6	16.3–16.9
Lung %	6.8	6.69–6.93	8.1	7.9–8.3
Arthritis %	38.2	37.94–38.42	38.8	38.4–39.2
High-blood pressure %	47.6	47.35–47.84	50.5	50.1–50.9
Stroke %	4.8	4.70–4.91	6.0	5.8–6.2
Cancer %	10.1	9.97–10.27	10.9	10.6–11.1
Female	52.6	52.38–52.87	52.2	51.8–52.6
Age in years	63.9	63.89–63.98	68.3	68.2–68.4
≥50, <60 years group %	28.9	28.85–29.1	14.9	12.6–16.3
≥60, <70 years group %	39.9	39.65–40.23	27.8	25.4–29.3
≥70, <80 years group %	20.4	70.05–70.81	39.4	36.5–41.7
≥80 years group %	10.6	10.46–10.76	18.9	14.7–20.2
Primary education uncompleted %	9.1	8.99–9.27	8.4	8.2–8.6
High school education uncompleted %	23.2	23.03–23.45	26.9	26.6–27.3
High school education completed %	39.2	38.91–39.39	37.0	36.6–37.4
College education uncompleted %	5.3	5.22–5.44	5.6	5.4–5.7
College education completed or more %	23.2	22.94–23.36	22.1	21.8–22.4
Married %	74.4	74.18–74.62	73.1	72.7–73.4
Single %	3.7	3.61–3.8	3.9	3.7–4.0
Divorced %	8.2	8.09–8.36	8.1	7.9–8.4
Widowed %	13.7	13.5–13.84	14.9	14.6–15.2
Worker %	40.3	40.09–40.56	28.5	27.8–28.5
Unemployed %	3.0	1.93–2.07	3.9	3.6–4.2
Retired %	55.2	54.91–55.4	63.8	62.4–65.2
Disabled %	1.6	1.52–1.64	3.8	3.6–4.0

Notes: Unweighted statistics for a longitudinal sample of adults measured from 2004 to 2014. Heart related diseases or heart attack, diabetes or high-blood sugar, lung disease except asthma, arthritis or rheumatism, high-blood pressure or hypertension, stroke or transient ischemic attack, and cancer or malignant tumor except skin cancer.

2.3. Depressive episode measurements

We included self-report measures of depressive episodes and not clinical diagnoses of major depression. Depressive episodes were measured as a dichotomy indicating previously established clinical thresholds (Andresen, Malmgren, Carter, & Patrick, 1994; Börsch-Supan, Brugiavini, Jürges, et al., 2008; Castro-Costa, Dewey, Stewart, et al., 2007; Radloff & Teri, 1986; Steffick, 2000) for the Center for Epidemiologic Studies Depression (CES-D) and the European Union Depression (EURO-D) scales. Specifically, for each individual we dichotomized the items of the depression scales, next calculated the sum score of depressive symptoms, and finally used a cut-off of ≥3 depressive symptoms for the 8-item CES-D scale (HRS), and ≥4 for both the 10-indicator CES-D scale (CHARLS and KLOSA) and the 12-indicator EURO-D scale (SHARE). Whenever a participant’s sum score of self-reported depressive symptoms falls above these cut-offs, a depressive episode is recorded.

2.4. Covariates

We included a wide range of potential confounding variables that were successfully harmonized across cohorts. Covariates included in the present analysis are detailed in Table 1, and included respondent-reported time-invariant demographics such as age at baseline, gender, and education, as well as time-varying demographics such as marital status, and employment status.

We also included time-varying smoking status (any current smoking versus none) and health status as reported by the respondent including body mass index (BMI) categories (obese BMI ≥ 30, overweight 25 ≤ BMI<30, normal weight 18.5 ≤ BMI<25, and underweight BMI<18.5), limitations on activities of daily living (ADLs, measured as a count of any difficult for bath, dress, or eat), self-reported health, and a set of dichotomies for self-reported chronic conditions indicating whether the respondent was ever diagnosed by a doctor with: heart related diseases including heart attack, diabetes or high-blood sugar, lung disease except asthma, arthritis or rheumatism, high-blood pressure or hypertension, stroke or transient ischemic attack, and cancer or malignant tumor except skin cancer. Gender and smoking status were tested as effect measure modifiers given our previous research suggesting interaction with alcohol use in predicting health outcomes (Keyes, Calvo, Ornstein, et al., 2019).

2.5. Statistical analysis

All data management and analyses were done in Stata 15 MP. Baseline year varied across countries ranging from 1998 (United States) through 2011 (China, Estonia, and Slovenia). On average, countries began follow-up 3.47 years after baseline. All results show year of baseline assessment on the X-axis. Prior to analysis, we weighted each country sample to 10,000 individuals according to country-specific age and sex distributions, in order to allow each country to contribute equally to the estimation process (otherwise, results would be driven primarily by countries with larger sample sizes, such as the United States). We first estimated Kaplan Meier curves for the survival time to new depressive episodes across the follow-up time of the respondents, combining data on all countries, as well as the overall Kaplan Meier survival curves stratified by baseline drinking status. We also examined Kaplan Meier curves stratified by each country. Next, we estimated semiparametric Cox proportional hazards models for the association, first unadjusted, then adjusted for all aforementioned covariates. We tested proportional hazards assumptions for all covariates in the Cox proportional hazards models. The results are presented in Table A2. Covariates that were significantly non-proportional over time were transformed by multiplying by the natural logarithm of time. We conducted a series of Cox proportional hazards models, iteratively including time-invariant and time-varying predictors, allowing for random intercepts and slopes for each country using time-invariant frailty models as well as time-varying-predictors frailty models, and present results for each of these models to determine robustness to variation in modeling assumptions and techniques, as well as country-specific and country-pooled effects. Finally, we tested for multiplicative interactions between time-varying alcohol consumption categories as gender as well as time-varying smoking status by including cross-product terms in the Cox proportional hazards models.

3. Results

Fig. 1 presents the overall Kaplan Meier curve for risk of depressive episode incidence. These curves are generated from the start of follow-up through end of the study, death, or occurrence of a depressive episode, and are thus averaged across age. Age is subsequently controlled in the subsequent Cox regression models. Overall, the median survival time (i.e., time to depressive episode among those with no depressive episode at baseline) was approximately 3000 days (8.2 years), and approximately 23.9% of respondents developed a new depressive episode by the end of the follow-up period. Fig. 1B presents Kaplan Meier curves stratified by baseline drinking status, which differed across category (Log-rank test for equality of depressive episode functions: Chi²=1623.23; P-value<0.001). Long-term abstainers had the highest risk of depressive episodes during the study period through approximately 3000 days, after which depressive episode risk converged across long-term abstainers, current abstainers, and heavy drinkers. Moderate and occasional drinkers had the lowest risk of depressive episode throughout the study period.

Fig. 1.

Kaplan Meier curves for depressive episodes in a longitudinal sample of adults age 50 and over in 20 countries N=57,276. (A) Overall depression survival rate. (B) Depression survival rates by baseline alcohol consumption category.

Table 2 includes the results of semiparametric Cox models estimating hazard rates of new episodes of depression among those with no depressive episode at baseline, adjusted for covariates (unadjusted estimates are provided in Table A3). Model 1 allows the intercept of the hazard model to be country-specific and includes time-invariant predictors. Model results indicated that long-term alcohol abstainers had a higher hazard of depressive episode across the study period (HR=1.14, 95% C.I. 1.08–1.21), as did those reporting occasional (HR=1.16, 95% C.I. 1.10–1.21) and heavy drinking (HR=1.22, 95% C.I. 1.13–1.30), compared with moderate drinking. Models 2 through 4 included the following additional statistical components: a country random component using a time-invariant predictors frailty model (Model 2); country-specific time-varying predictors (Model 3); and both country random components and time-varying-predictors frailty model (Model 4). The magnitude of the identified associations tended to decrease toward the null, with long-term and current abstaining as well as occasional drinking increasing the hazard of depressive episodes by 1.01, 1.01, and 1.02, respectively. Heavy drinking remained robustly associated with increases in depressive episode risk (HR=1.16, 95% C.I. 1.11–1.21). By country, depressive episode rates across the study period comparing the non-parametric Kaplan Meier curves to the semiparametric predicted baseline hazard rate from the Cox proportional model based on both a stratified baseline hazard rate and a frailty model are presented in Fig. A1, demonstrating that results are largely robust to alternative approaches; however, we generally observe that the unadjusted Kaplan curves overestimate depressive episodes, and that the frailty model smooths out sharp drops in the survival curves (e.g., Germany) that occur when non-parametric models are used. Further, there is heterogeneity in the variation across modeling choices by country, notably for Estonia and the United States (Fig. A1).

Table 2.

Results of semiparametric Cox models estimating depressive episode rates N=57,276.

	Model 1		Model 2		Model 3		Model 4
	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Alcohol consumption (Ref: Moderate drinker)
Long-term abstainer	1.14***	1.08–1.21	1.16***	1.10–1.23	1.01*	1.00–1.02	1.01	1.00–1.01
Current abstainer	1.03	0.98–1.09	1.03	0.98–1.09	1.00	0.99–1.00	0.99	0.99–1.00
Occasional drinker	1.16***	1.10–1.21	1.15***	1.10–1.21	1.01***	1.01–1.02	1.01***	1.00–1.02
Heavy drinker	1.22***	1.13–1.30	1.22***	1.14–1.31	1.02***	1.01–1.03	1.02***	1.01–1.0:
Smoker	1.18***	1.13–1.22	1.18***	1.13–1.23	1.16***	1.11–1.21	1.16***	1.11–1.21
Weight categories (Ref: Normal weight)
Underweight	1.11	0.99–1.24	1.14*	1.02–1.28	1.10	0.98–1.23	1.12*	1.00–1.26
Overweight	0.96	0.93–1.00	0.96	0.93–1.00	0.99***	0.99–1.00	0.99***	0.98–0.99
Obese	0.93**	0.88–0.97	0.93**	0.88–0.97	0.99***	0.98–0.99	0.99***	0.98–0.99
ADLs any	1.37***	1.30–1.45	1.40***	1.32–1.48	1.39***	1.31–1.46	1.42***	1.34–1.49
Self-reported health (Ref: Very good)
Very bad	4.03***	3.63–4.47	4.10***	3.70–4.55	4.17***	3.76–4.63	4.30***	3.87–4.77
Bad	2.78***	2.53–3.06	2.82***	2.56–3.09	2.86***	2.61–3.15	2.92***	2.66–3.21
Regular	1.80***	1.64–1.97	1.79***	1.64–1.96	1.83***	1.67–2.00	1.83***	1.67–2.01
Good	1.27***	1.16–1.41	1.27***	1.15–1.40	1.28***	1.16–1.41	1.28***	1.16–1.41
Chronic conditions
Heart	1.10***	1.05–1.15	1.09***	1.05–1.14	1.01***	1.01–1.02	1.01***	1.01–1.02
Diabetes	1.05*	1.00–1.10	1.06*	1.01–1.11	1.01 *	1.00–1.01	1.01*	1.00–1.01
Lung	1.14***	1.08–1.21	1.13***	1.07–1.20	1.15***	1.08–1.21	1.14***	1.07–1.20
Arthritis	1.21 ***	1.16–1.25	1.20***	1.15–1.24	1.22***	1.17–1.26	1.20***	1.16–1.25
High-blood pressure	0.99	0.96–1.03	1.00	0.96–1.03	1.00	0.99–1.00	1.00	0.99–1.00
Stroke	1.19***	1.11–1.27	1.18***	1.10–1.26	1.20***	1.12–1.28	1.20***	1.12–1.28
Cancer	1.06*	1.00–1.13	1.05	0.99–1.12	1.06*	1.00–1.13	1.05	0.99–1.12
Female	1.53***	1.47–1.59	1.53***	1.47–1.59	1.54***	1.48–1.60	1.55***	1.49–1.61
Age group at baseline (Ref: 50–60 years)
>60, <70 years	0.92***	0.88–0.96	0.91***	0.87–0.95	0.98***	0.98–0.99	0.98***	0.98–0.99
>70, <80 years	1.05	1.00–1.11	1.04	0.99–1.10	1.00	1.00–1.01	1.00	0.99–1.01
>80 years	1.25 ***	1.16–1.35	1.23***	1.14–1.33	1.03***	1.02–1.04	1.02***	1.01–1.03
Educational status (Ref: College completed)
Primary uncompleted	1.26***	1.16–1.36	1.28***	1.18–1.39	1.28***	1.18–1.39	1.32***	1.22–1.43
High school uncompleted	1.11***	1.05–1.16	1.11***	1.05–1.17	1.12***	1.06–1.18	1.13***	1.08–1.19
High school completed	1.04	0.99–1.10	1.04	0.99–1.09	1.05	1.00–1.10	1.05	1.00–1.10
College uncompleted	1.01	0.92–1.10	1.05	0.96–1.15	1.01	0.93–1.11	1.06	0.97–1.15
Marital status (Ref: Married)
Divorced	1.09**	1.02–1.17	1.10**	1.03–1.18	1.01*	1.00–1.02	1.01*	1.00–1.02
Single	0.94	0.86–1.03	0.94	0.86–1.03	0.94	0.86–1.03	0.94	0.86–1.03
Widowed	1.00	0.95–1.05	1.01	0.97–1.07	1.00	0.99–1.01	1.00	0.99–1.01
Labor force status (Ref: Worker)
Retired	0.90***	0.86–0.94	0.90***	0.86–0.94	0.98***	0.98–0.99	0.98***	0.98–0.99
Unemployed	1.33***	1.20–1.47	1.29***	1.16–1.43	1.29***	1.16–1.43	1.25***	1.12–1.38
Disabled	1.11*	1.01–1.22	1.12*	1.02–1.23	1.01	1.00–1.02	1.01	1.00–1.02

Notes: Weighted results with robust CI:

^***P<0.001,

^**P<0.01,

^*P<0.05.

Model 1: baseline hazard was calculated stratified by country using time-invariant predictors. Model 2: baseline hazard was calculated for all countries and multiplied by a country random component using time-invariant predictors frailty model. Model 3: baseline hazard was calculated stratified by country using time-varying predictors. Model 4: baseline hazard was calculated for all countries and multiplied by a country random component using time-varying-predictors frailty model.

Other covariates were associated with increased and decreased hazard of depressive episodes in expected directions. In Model 4, increased hazard of depressive episodes was observed for smokers, those underweight, those reporting any restrictions on activities of daily living, all self-reported health categories compared to those who report very good health, and across many chronic diseases. Across demographics, depressive episode hazard is increased among female respondents, those with low levels of education, divorced respondents, and those unemployed. Depressive episode hazard decreased among those overweight and obese compared to normal weight, those aged 60–70 years compared with 50–60, and those who are retired from the labor force compared to those still working.

Table 3 overviews the results of interaction tests across gender and smoking status. Interactions between smoking status and alcohol consumption categories as well as gender and alcohol consumption categories were observed, after adjusting for all model covariates included in previous tables (coefficients for model covariates are not presented in Table 3, but are found in Table A4). When examining multiplicative interaction, observed interactions indicate when the occurrence of the outcome, in this case depressive episodes, is equal, more, or less than what would be expected given their expected multiplicative effects. That is, given a hypothetical risk factor that is associated with two times the hazard of the outcome, and a second risk factor that is associated with four times the hazard of the outcome, a hazard ratio of 8 for those exposed to both would be indicative of no interaction (because we would expect a multiplicative effect in the absence of the interaction of 8, or 2 × 4=8). Thus, hazard ratios for the joint effect of the two factors more than 8 indicates more than would be expected, and hazard ratios less than 8 indicate indicates less than would be expected. In Model 6 in which the baseline hazard was calculated for all countries and multiplied by a country random component using a time-varying-predictors frailty model, individuals who both smoked and were long-term abstainers (HR for interaction, 0.96, 95% C.I. 0.95–0.98), or current abstainers drinkers (HR for interaction, 0.96, 95% C.I. 0.96–0.99), had less of an increase in the hazard of depressive episodes compared to what would be expected based on their multiplicative effects, when compared to non-smoking moderate drinkers. Further, individuals who were both long-term abstainers and women (HR for interaction, 0.95, 95% C.I. 0.92–0.97) had less of an increase in the hazard of depressive episodes compared to what would be expected based on their multiplicative effects, when compared to moderately drinking men. Finally, we observed a three-way interaction, whereby individuals who were both long-term abstainers, women, and smokers (HR for interaction, 1.04, 95% C.I. 1.00–1.07) had more of an increase in the hazard of depressive episodes compared to what would be expected based on their multiplicative effects, when compared to moderately drinking, non-smoking men.

Table 3.

Results of semiparametric Cox models estimating depressive episode rates with moderators N=57,276.

	Model 5		Model 6
	HR	95% CI	HR	95% CI
Alcohol consumption (Ref: Moderate drinker)
Long-term abstainer	1.04***	1.03–1.06	1.04***	1.02–1.05
Current abstainer	1.01*	1.00–1.03	1.01	1.00–1.02
Occasional drinker	1.01	1.00–1.02	1.01	1.00–1.02
Heavy drinker	1.01	1.00–1.03	1.01	0.99–1.02
Smoker	1.19***	1.12–1.27	1.18***	1.11–1.26
Female	1.59***	1.49–1.68	1.61***	1.52–1.71
Two-way interaction with smoking
Long-term abstainer*smoker	0.97***	0.95–0.98	0.96***	0.95–0.98
Current abstainer*smoker	0.98**	0.96–0.99	0.98**	0.96–0.99
Occasional drinker*smoker	1.00	0.99–1.01	1.00	0.99–1.01
Heavy drinker*smoker	1.03*	1.00–1.05	1.03*	1.00–1.05
Two-way interaction with gender
Long-term abstainer*female	0.93***	0.90–0.96	0.95***	0.92–0.97
Current abstainer*female	0.99	0.97–1.01	1.00	0.98–1.02
Occasional drinker*female	1.01	0.99–1.02	1.01	0.99–1.03
Heavy drinker*female	1.00	0.98–1.02	1.00	0.98–1.02
Three-way interaction with gender and smoker
Long-term abstainer*female*smoker	1.05**	1.02–1.09	1.04**	1.00–1.07
Current abstainer*female*smoker	1.00	0.97–1.03	0.99	0.97–1.02
Occasional drinker*female*smoker	0.99	0.98–1.01	0.99	0.97–1.01
Heavy drinker*female*smoker	1.00	0.96–1.03	1.00	0.96–1.03
Covariates	Yes		Yes

Notes: Weighted results with robust CI:

^***P<0.001,

^**P<0.01,

^*P<0.05.

The coefficients for covariates are presented in Table A4. Omitted variables: moderate drinker, 50–60 years, normal weight, very good health, college or more, married, worker. Model 5: baseline hazard was calculated stratified by country using time-varying predictors. Model 6: baseline hazard was calculated for all countries and multiplied by a country random component using time-varying-predictors frailty model.

Fig. 2 provides hazard ratios from the interaction models to examine the hazard of depressive episodes among subgroups by gender, smoking, and drinking status with a reference group of male, non-smoking moderate drinkers. Those with the highest hazard of depressive episodes were women during periods of both heavy drinking and smoking. Smoking males had a higher hazard of depressive episodes regardless of drinking status compared with moderately drinking and non-smoking males.

Fig. 2.

Predicted hazard ratios of depression rates by alcohol consumption, smoking, and gender. Notes: Hazard ratios from Model 6, Table 3. Omitted variables consist of individuals 50–60 years old, married, working, with college completed, very good health, no chronic diseases, and regular weight. [***] P<0.001, [**] P<0.01, [*] P<0.05.

Fig. 3 displays the country-specific risk of depressive episodes across the study period for each country. For simplicity of presentation, we contrast the Kaplan Meier curve for time to depressive episode in the two groups with the lowest and highest overall risk based on alcohol consumption status, gender and smoking, based on the interaction results: males who were moderate drinkers and non-smokers in a given time period, and female who were long-term abstainers and smokers in a given time period. Countries have heterogeneous survival rates, with some countries experiencing overall low rates of depressive episodes, such as England, Estonia, and Israel, whereas others have high rates of depressive episodes, such as Austria, Spain, Slovenia, Switzerland and the Czech Republic. Further, we observed differences in the strength of the association between alcohol consumption and depressive episode risk across counties, suggesting heterogeneity of effects by country.

Fig. 3.

Predicted depressive episode rates by country comparing moderate drinking, male non-smokers (lowest risk group, solid gray line) to long-term abstaining, female smokers (highest risk group, dashed black line). Notes: Predicted baseline depressive episode rates based on Table 3, model 6 for individuals 50–60 years old, married, working, with college completed, very good health, no chronic diseases, and regular weight.

4. Discussion

We documented that among adults 50 years and older across 19 countries, alcohol consumption is differentially associated with the incidence of depressive episodes across drinking categories. Rates of depressive episodes were highest among high and low alcohol consumers, that is, long-term alcohol abstainers, and heavy drinkers. Those reporting moderate drinking patterns had the lowest risk of a depressive episode. These findings hold when controlling for health status, demographics, and other health behaviors (time-varying and constant), but are heterogeneous across two established risk groups for depressive episodes: smokers and women. Indeed, women showed a higher incidence of depressive episodes throughout the study period, and we found that moderate and occasional drinking is less inversely associated among women than among men. Similarly, smoking (for both gender) increased the risk for a depressive episode, with synergistic effects among those who also engage in heavy drinking. Given that depressive episodes are associated with significant suffering, as well as increased rates of mortality, these findings suggest that clinicians should screen for alcohol use among older patients, and provide timely and accurate health information about the mental and physical health risks of certain alcohol consumption patterns.

The increased risk of new depressive episodes during periods of heavy drinking among older adults is of particular concern. The co-occurrence of substance use with deterioration of mental health is well-documented across the life course, with excessive alcohol use hypothesized to have a bidirectional and reinforcing relationship with depressive symptoms (Sullivan, Fiellin, & O’Connor, 2005). This is particularly concerning among older adults, among whom depression and suicide risk is already increased compared to other adults, and among whom medications and co-occurring chronic health conditions may exacerbate the adverse health effects of alcohol. While drinking declines in older age, heavy drinking remains a prevalent health concern. Screening tools for older adults are reliable and well-validated (Schonfeld et al., 2015), including Screening, Brief Intervention, and Referral to Treatment (SBIRT) and should be wide spread among clinicians interacting with older adults regardless of other morbidities, but perhaps especially among older adults that are relatively active enough to engage in heavy drinking with regularity.

In contrast, our data also indicate that moderate drinkers have lower risks for depressive episodes compared to both long-term abstainers and heavy drinkers. This finding could be attributable to long discussed stress-dampening effect of alcohol consumption (Sayette, 1999). However, the possibility of methodological artifact should be considered. The generally “J” shaped curve between alcohol consumption and depressive episodes mirrors the common epidemiological findings for alcohol consumption and mortality (Connor, 2006; Rehm, Roerecke, & Room, 2016; Rimm, Klatsky, Grobbee, & Stampfer, 1996), in particular cardiovascular risk that has been extensively evaluated in the literature (Leong, Smyth, Teo, et al., 2014; Liang, Mente, Yusuf, et al., 2012). Evidence is accumulating refuting the notion that the mechanism underlying the “J” shaped curve is causal from moderate consumption to longer longevity and lower risk of cardiometabolic outcomes, and more likely to be the product of residual confounding (Cho et al., 2016; Holmes, Dale, Zuccolo, et al., 2014; Naimi, Brown, Brewer, et al., 2005; Naimi et al., 2017; Stockwell et al., 2016; Taylor, Lu, Carslake, et al., 2015; Zhao, Stockwell, Roemer, Naimi, & Chikritzhs, 2017). Yet, Keyes, Calvo, et al. (2019) using the Health and Retirement Study, found that the inverse association for moderate drinking held when evaluating alcohol consumption as a time-varying exposure and that substantial uncontrolled confounding would be necessary to eliminate the associations to the null, but that does not preclude such levels of confounding to be possible, underscoring the difficulties of examining alcohol consumption in cohort data. For risk of depressive episodes in particular, moderate alcohol consumption is associated with a lower level of risk factors for depression in older age such as divorce and loss of social support (Perreira & Sloan, 2001) multi-morbidity (Han, Moore, Sherman, & Palamar, 2018), and other health behaviors (Naimi et al., 2005, 2017), and as such, continued focus on evaluating the mechanisms underlying observed associations is critical. In sum, the state of the literature is such that patients should be regularly queried about alcohol consumption, and at this point, there is no safe level of consumption that is recommended as promoting mental health. Indeed, the notion that alcohol consumption does not have a safe or recommended minimum level is increasingly being supported by a growing evidence base, including large observational studies (Wood, Kaptoge, Butterworth, et al., 2018) and quasi-experimental studies (Cho et al., 2016; Holmes et al., 2014; Millwood, Walters, Mei, et al., 2019; Taylor et al., 2015) suggesting that even low levels of alcohol consumption increase risk for adverse health outcomes.

Variation in the association between alcohol consumption and depressive episodes by gender and smoking status also underscore that universal statements about the safety of alcohol use in older age need to be considered within the context of other risk factors. As is known in the literature, we also observed that women in particular are at increased risk for depressive episodes regardless of drinking status. Alcohol use among women, particularly those in older age, is associated with a greater burden of health problems than for men at a similar drinking level (Hanna, Chou, & Grant, 1997; Holman, English, Milne, & Winter, 1996; Urbano-Marquez et al., 1995), as well as additional health concerns such as breast cancer (Bagnardi, Rota, Botteri, et al., 2013). Thus, in these data, women are conferred no decrease in risk for depressive episodes across any category of alcohol consumption, and other literature indicate that across every level of alcohol consumption, women experience increases in health risks (Stampfer, Colditz, Willett, Speizer, & Hennekens, 1988), including depressive episodes. Other factors particularly influencing women over 50, such as the menopausal transition (Freeman, 2010), are associated with increased risk of new episodes of depression across alcohol consumption categories (Milic, Glisic, Voortman, et al., 2018). Even though women consume less alcohol than men at a population level (Keyes, Jager, et al., 2019), a focus on ensuring women, who drink excessively, have access to resources to reduce their risk for depressive episodes, continues to be a health need.

Overall, we find that alcohol consumption is associated with the onset of depressive episodes differentially across drinking categories and interacting with smoking and gender, but these associations are heterogeneous across countries. Cross-country variation points to the importance of further exploring contextual factors shaping the consequences of alcohol consumption. Contexts in which consuming alcohol may be more or less harmful for different individuals may depend on drinking cultures and regulation, access to health and housing, financial and economic crises, and exposure to war, terrorism, or mass violence. Nevertheless, it is also worth considering that the main effects of alcohol on depressive episodes were observable in every country, suggesting some evidence of biological processes through which heavy drinking has a deleterious effect on mental health. Further, these results underscore the need for cross-country analyses in harmonized data sources in order to observe the extent to which associations are general versus specific across place. In sum, our future work will focus on estimating the characteristics of countries that moderate the risk of depressive episodes, as well as the relation between alcohol consumption and depressive episodes.

Limitations of the present study should be carefully considered. As outlined above, these are observational data and alcohol consumption is self-reported. Individuals in the present study are aged 50 and older, which is a broad age range within which there may be heterogeneous effects with respect to the association between alcohol consumption and depressive symptoms. Our analyses controlled for age, thus the results represent the average relationship across all age groups. Further analyses to examine age-specific trends would elucidate particular ages within which alcohol-related harms may emerge. This is particularly salient when considering the effects of selective survival in studies of older adults and alcohol consumption. In particular, heavy drinkers, and those with severe alcohol use disorders, have higher mortality, thus cohorts of older adults are likely to have selection effects that increase with age. As such, age-stratified examinations of older adult drinking should be done with caution, underscoring our approach of age-adjusted associations, presenting the average across all age groups. Further, the potential for residual confounding remains a concern in all studies of alcohol consumption and health. In these data, alcohol use, especially moderate drinking, is associated with better health status and lower levels of risk factors for depression, especially in older age (Naimi et al., 2005). These results may be the product of residual confounding, given that moderate drinking is associated with a wide range of other health-promoting behaviors, better overall health, as well as higher socio-economic status. While we controlled for such factors in our analyses, complete confounder control is difficult in observational studies, thus results for moderate drinking should be interpreted with caution. Finally, while we combined lifetime and long-term abstainers from alcohol due to necessities of the harmonization process, we did not have sufficient information on reasons for quitting drinking, including whether the individual had a history of alcohol abuse or dependence that prompted abstention, in order to separate groups in more detail. Further, we harmonized data on depression across seven different longitudinal cohort studies with some variation in how depression was measured, including the CES-D and EURO-D scales. We attempted to mitigate the potential for misclassification and measurement invariance to influence results by our multi-component estimation method that allows countries to have their own random components, but future efforts to harmonize measures at the outset of data collection would aid in cross-cultural comparisons.

In summary, excessive alcohol use causes substantially greater harm in older ages than in younger ages (CDC, 2015; Chen & Yoon, 2016; Lehmann & Fingerhood, 2018). Available data indicate that alcohol consumption is increasing among older adults in a wide variety of countries (Breslow et al., 2017; Han et al., 2017; Kim et al., 2012; World Health Organization, 2014), as their populations are living longer lives. As we document in the present study, excessive alcohol consumption among older adults is a concern not only for physical well-being, but also mental health, including depressive episodes. Efforts to include regular, informed, and empirically-based query and information to older adults in routine as well as specialty medical care is increasingly important to ensure that mental health and physical well-being remain vital and strong in aging populations.

Appendix

Table A1.

Number of observations and individuals by country and year.

Countries	Year	Number of observations	Number of individuals
Austria	2007, 2011, 2013	6583	3167
Belgium	2007, 2011, 2013	6953	3026
China	2011, 2013	2940	1485
Czech Republic	2007, 2011, 2013	6787	3124
Denmark	2007, 2011, 2013	4714	1883
England	2008, 2012	4526	2289
Estonia	2010, 2011, 2013	5848	2956
France	2006, 2011, 2013	6567	2993
Germany	2007, 2011, 2013	2563	1113
Israel	2010, 2013	1806	915
Italy	2007, 2011, 2013	4981	2084
Korea	2006, 2008, 2010, 2012	14,731	5229
Netherlands	2007, 2011, 2013	4890	2007
Poland	2007, 2012	1299	669
Slovenia	2011, 2013	2576	1295
Spain	2007, 2011, 2013	4178	1890
Sweden	2007, 2011, 2013	4337	1791
Switzerland	2007, 2011, 2013	5383	2408
United States	2004, 2006, 2008, 2010, 2012, 2014	65,795	16,952
Total		157,457	57,276

Table A2.

Test of proportional hazards assumption.

Variables	P-value
Drinking categories	0.000***
Smoking	0.397
ADLs ever	0.113
Weight category
Underweight	0.950
Overweight	0.000***
Obese	0.000***
Self-reported health	0.225
Chronic conditions
Heart	0.034*
Cancer	0.682
High-blood pressure	0.001***
Diabetes	0.000***
Lung	0.393
Stroke	0.477
Arthritis	0.098
Female	0.051
Age groups	0.001***
Educational status	0.422
Marital status
Divorced	0.001***
Single	0.029
Widowed	0.000***
Working status
Retired	0.000***
Unemployed	0.170
Disabled	0.000***

Notes: P-value<0.05 indicates a time-varying covariate. These were included in Table 2, models 3 and 4, interacted with time logged.

Table A3.

Unadjusted hazard ratios by alcohol consumption on semiparametric Cox models to estimate depressive episode rates.

	Model 1		Model 2
Variables	OR	95% CI	OR	95% CI
Drinking categories (Ref: Moderate drinker)
Long-term abstainer	1.75	1.66–1.85	1.78	1.69–1.87
Current abstainer	1.50	1.42–1.58	1.51	1.43–1.59
Occasional drinker	1.41	1.35–1.48	1.42	1.36–1.48
Heavy drinker	1.19	1.11–1.27	1.20	1.12–1.29

Notes: Weighted results with robust CI. Omitted variables: moderate drinker, 50–60 years, normal weight, very good health, college or more, married, worker. Model 1: baseline hazard was calculated stratified by country using time-varying predictors. Model 2: baseline hazard was calculated for all countries and multiplied by a country random component using time-varying-predictors frailty model.

Table A4.

Results of semiparametric Cox models estimating depressive episode rates with moderators N=57,276.

	Model 5		Model 6
Variables	OR	95% CI	OR	95% CI
Weight categories (Ref: Normal weight)
Underweight	1.09	0.97–1.22	1.12	1.00–1.25
Overweight	0.99***	0.99–1.00	0.99***	0.99–1.00
Obese	0.99***	0.98–0.99	0.99***	0.98–0.99
ADLs any	1.38***	1.31–1.46	1.41 ***	1.34–1.49
Self-reported health (Ref: Very good health)
Very bad	4.21***	3.79–4.67	4.33***	3.90–4.80
Bad	2.89***	2.63–3.17	2.94***	2.68–3.23
Regular	1.84***	1.68–2.01	1.84***	1.68–2.02
Good	1.29***	1.17–1.42	1.28***	1.16–1.41
Chronic conditions
Heart	1.01***	1.01–1.02	1.01***	1.00–1.02
Diabetes	1.01*	1.00–1.01	1.01*	1.00–1.01
Lung	1.14***	1.08–1.21	1.13***	1.07–1.20
Arthritis	1.22***	1.17–1.26	1.20***	1.16–1.25
High-blood pressure	1.00	0.99–1.00	1.00	0.99–1.00
Stroke	1.19***	1.11–1.27	1.19***	1.11–1.27
Cancer	1.06	1.00–1.12	1.05	0.99–1.11
Age group at baseline (Ref: 50–60 years)
≥ 60, <70 years	0.98***	0.98–0.99	0.98***	0.98–0.99
≥ 70, <80 years	1.00	1.00–1.01	1.00	0.99–1.01
≥ 80 years	1.03***	1.01–1.04	1.02***	1.01–1.03
Educational status (Ref: College completed)
Primary uncompleted	1.29***	1.18–1.39	1.33***	1.22–1.44
High school uncompleted	1.13***	1.07–1.19	1.14***	1.08–1.20
High school completed	1.05	1.00–1.11	1.05	1.00–1.11
College uncompleted	1.01	0.93–1.11	1.06	0.97–1.16
Marital status (Ref: Married)
Divorced	1.01*	1.00–1.02	1.01*	1.00–1.02
Single	0.94	0.86–1.03	0.94	0.86–1.03
Widowed	1.00	0.99–1.01	1.00	0.99–1.01
Labor force status (Ref: Worker)
Retired	0.98***	0.98–0.99	0.98***	0.98–0.99
Unemployed	1.29***	1.16–1.43	1.24***	1.12–1.38
Disabled	1.01	1.00–1.02	1.01	1.00–1.02

Notes: Weighted results with robust CI:

^***P<0.001,

^**P<0.01,

^*P<0.05.

Main effects of alcohol consumption, smoking, and gender, as well as interaction terms between these variables are presented in Table 3. Model 5: baseline hazard was calculated stratified by country using time-varying predictors. Model 6: baseline hazard was calculated for all countries and multiplied by a country random component using time-varying-predictors frailty model.

Fig. A1.

Kaplan Meier and Predicted depressive episode rates by country. Notes: Predicted baseline depression rates based on Table 2, models 5 and 6, for men, 50–60 years old, married, working, with college completed, very good health, no chronic diseases, and regular weight. BHR stands for Baseline Hazard rate.