Article Text
PDF
Abstract
Background Some of the previously reported health benefits of low-to-moderate alcohol consumption may derive from health status influencing alcohol consumption rather than the opposite. We examined whether health status changes influence changes in alcohol consumption, cessation included.
Methods Data came from 571 current drinkers aged ≥60 years participating in the Seniors-ENRICA cohort in Spain. Participants were recruited in 2008–2010 and followed-up for 8.2 years, with four waves of data collection. We assessed health status using a 52-item deficit accumulation (DA) index with four domains: functional, self-rated health and vitality, mental health, and morbidity and health services use. To minimise reverse causation, we examined how changes in health status over a 3-year period (wave 0–wave 1) influenced changes in alcohol consumption over the subsequent 5 years (waves 1–3) using linear/logistic regression, as appropriate.
Results Compared with participants in the lowest tertile of DA change (mean absolute 4.3% health improvement), those in the highest tertile (7.8% worsening) showed a reduction in alcohol intake (β: –4.32 g/day; 95% CI –7.00 to –1.62; p trend=0.002) and were more likely to quit alcohol (OR: 2.80; 95% CI 1.54 to 5.08; p trend=0.001). The main contributors to decreasing drinking were increased functional impairment and poorer self-rated health, whereas worsening self-rated health, onset of diabetes or stroke and increased prevalence of hospitalisation influenced cessation.
Conclusions Health deterioration is related to a subsequent reduction and cessation of alcohol consumption contributing to the growing evidence challenging the protective health effect previously attributed to low-to-moderate alcohol consumption.
- alcohol
- health status
- cohort study
- older adults
- reverse causation
Statistics from Altmetric.com
Introduction
There is a dynamic debate about the relationship between alcohol consumption and health. Although alcohol use is one of the leading causes of burden of disease,1 2 observational studies have reported a beneficial effect of low-to-moderate alcohol consumption on all-cause mortality, especially cardiovascular mortality.3–5 However, potential selection biases and residual confounding are important methodological issues undermining confidence in such results.6–8 Furthermore, the direction of causation between alcohol consumption and health is unclear. In general, studies have reported worse lifestyle and health characteristics in lifetime abstainers and ex-drinkers than in regular drinkers,9–12 an association that becomes stronger as health deteriorates with age. Also, many studies have attributed reduction or cessation of alcohol consumption to poor health,13–17 suggesting that part of the beneficial effect of low-to-moderate alcohol observed in cohort studies may be due to health status influencing alcohol consumption rather than the opposite. This phenomenon might be accentuated in older, compared with younger adults, due to faster and greater health deterioration.
Our main aim was to examine whether changes in health status influence changes in alcohol consumption, including cessation, in older adults. We analysed data from a cohort of community-dwelling older drinkers in Spain, a Mediterranean country where alcohol consumption patterns, lifestyle and health characteristics differ from those found in countries where most previous research has been conducted.13–17 We selected only current drinkers in order to minimise potential selection biases, residual confounding and reverse causation.18 Moreover, to further mitigate potential reverse causation, we evaluated the influence of changes in health status over a 3-year period (wave 0–wave 1) on changes in alcohol consumption over the following 5 years (waves 1–3). Lastly, as a measure of health status, we used a multidomain health deficit accumulation (DA) index. This is important because, as far as we are aware of, ours is the first study investigating which specific dimensions of health have greater influence on changes in drinking patterns.
Methods
Study design and population
Data come from the Seniors-ENRICA cohort, whose participants were selected between June 2008 and October 2010 by stratified cluster sampling of the community-dwelling adult population of Spain. At baseline (wave 0), a computer-assisted telephone interview was used to obtain data on sociodemographic factors, lifestyle and morbidity. Additionally, two home visits were conducted to collect blood and urine samples, perform a physical examination and obtain a diet history.19 20 Participants aged 60 years and above were followed for a median of 8.2 years (range: 6.8–9.1) during which we performed three additional follow-ups: wave 1 in 2012, wave 2 in 2015 and wave 3 in 2017. Because no alcohol consumption data were collected in wave 1, all analyses for this work are based on waves 0 and 1 for health-related variables and waves 1 and 3 for alcohol consumption variables.
All study participants provided written informed consent.
Study variables
Health status
Based on the procedure used by Rockwood et al in creating a frailty index,21 22 we calculated a DA index for each wave using a total of 52 health deficits. The index comprised four dimensions or domains: (1) physical and cognitive functional impairments (22 items); (2) self-reported health and vitality (7 items); (3) mental health (6 items); and (4) morbidities and use of health services (7 items). Most deficits in the DA index were assessed dichotomously (1 point if present and 0 otherwise), except for cognitive functioning, self-rated health, vitality, mental health, body mass index (BMI) and use of outpatient healthcare. These deficits were instead scored according to severity (0 points for no deficit, 0.25–0.75 points for mild-to-moderate deficits and 1 point for severe deficit). The overall and domain-specific DA scores were calculated as the total sum of points assigned to each health deficit divided by the number of deficits considered and further multiplied by 100 to obtain a range from 0% to 100%. The complete list of health deficits, and associated scores, are found in online supplementary table 1, and a detailed description of the construction of the DA index domains is provided in the methodological appendix.
Supplementary file 1
Change in overall and domain-specific DA indices from wave 0 to wave 1 (median follow-up: 3.2 years; range: 1.8–4.6) was calculated as the score for each index at wave 1 minus the corresponding score at wave 0. Thus, a negative absolute change indicates health improvement and a positive change denotes health deterioration. Then, we calculated sex-specific tertiles independently for each index, resulting in different tertile ranges for each of them. The lowest tertile included those participants with health improvement, the intermediate tertile included those with little or no change and the highest tertile included those with health deterioration.
Alcohol consumption
In waves including the alcohol data module, usual consumption of alcoholic beverages in the preceding year was estimated with a validated diet history developed from the one used in the EPIC cohort study in Spain.19 23 This diet history collected information on 34 alcoholic beverages and used photographs to better quantify portion sizes. Alcohol content of each beverage was estimated using standard composition tables. At each wave, study participants were classified as never drinkers (average alcohol intake of 0 g/day), ex-drinkers (average alcohol intake of 0 g/day who answered ‘I used to drink alcohol, but I quit’ when asked which statement best described their alcohol consumption) or current drinkers (average alcohol intake >0 g/day).
For current drinkers at wave 1, change in alcohol intake from wave 1 to wave 3 (median follow-up: 5.0 years; range: 4.4–5.4) was calculated as the average intake at wave 3 minus the average intake at wave 1. Also, current drinkers at wave 1 who quit drinking from wave 1 to wave 3 were identified for the analyses examining cessation of alcohol consumption.
Potential confounders
For these analyses, we have used information on potential confounders at wave 1, such as sociodemographic and lifestyle characteristics including sex, age, educational level, tobacco smoking, leisure-time physical activity expressed as metabolic equivalents of task-hour/week, usual time spent watching TV (as a measure of sedentarism) and diet quality assessed with the Mediterranean Diet Adherence Screener.24 Also, we measured weight and height in standardised conditions25 to calculate the BMI as the weight (in kilograms) divided by the squared height (in metres).
Statistical analysis
From the initial sample of 1296 current drinkers at wave 1, 75 (5.8%) died and 544 (42.0%) were lost to follow-up at wave 3. From the remaining 677 participants (52.2%), we excluded 103 without information on alcohol consumption at wave 3, and an additional three with missing data on potential confounders. Thus, the final analytical sample comprised 571 individuals.
The association of change in overall and domain-specific DA indices from wave 0 to wave 1 with change in average alcohol intake from wave 1 to wave 3 was summarised with β coefficients and their 95% CIs, obtained from linear regression. Two models were tested: model 1 adjusted for sex, age and educational level, and model 2 further adjusted for tobacco smoking, leisure-time physical activity, time watching TV, diet quality and BMI. Models with additional adjustment for change in alcohol intake from wave 0 to wave 1 were also fitted.
To examine the association between DA change (from wave 0 to wave 1) with the risk of quitting alcohol consumption from wave 1 to wave 3, we fit logistic regression models to obtain ORs and their 95% CIs. Models were adjusted for the same variables as mentioned above.
In order to identify specific health deficits with greater influence on alcohol consumption changes, analyses were replicated for each item of the DA index. We also assessed whether study results varied by sex by testing interaction terms defined as the product of change in DA indices (tertiles) and sex. None of interactions reached statistical significance. Thus, results for men and women are presented combined. Lastly, we replicated the analyses imputing missing values of alcohol intake at wave 3 and potential confounders for participants who we followed-up until wave 3 using multiple imputation by chained equations.
Statistical significance was set at two-sided p value <0.05. Analyses were performed with Stata V.13.1.
Results
Health status of study participants declined from wave 0 to wave 1, with a mean change in overall DA index of 1.5% (95% CI 1.1 to 2.0). However, there was a substantial variation across tertiles of DA change, with a mean value of –4.3% (health improvement) for those in the lowest, 1.2% in the intermediate and 7.8% in the highest tertile (health deterioration). Participants in the highest tertile of change in the DA index were older, had a lower education, were more frequently never smokers, performed less physical activity and had a higher BMI than those in the lowest and intermediate tertiles. Regarding alcohol consumption, participants reduced their average alcohol intake from wave 1 to wave 3 (mean change: –2.29 g/day; 95% CI –3.47 to –1.10). This reduction was slighter among those in the lowest tertile of DA change and progressively grew more substantive as health deteriorated (table 1).
Characteristics of study participants at wave 1 by tertiles of change in the overall DA index during 3.2 years (wave 0 to wave 1)
The association between change in overall and domain-specific DA indices from wave 0 to wave one and change in average alcohol intake (g/day) from wave 1 to wave 3 is shown in table 2.
Association of change in overall and domain-specific DA indices during 3.2 years (wave 0–wave 1) with change in alcohol intake (g/day) in the subsequent 5 years (wave 1–wave 3)
Compared with participants in the lowest tertile of change in the overall DA index, those in the highest tertile reduced their alcohol intake (β: –4.32 g/day; 95% CI –7.00 to –1.62; p value for trend=0.002). The results were consistent when the model was further adjusted for change in alcohol intake from wave 0 to wave 1 (β: –4.15 g/day; 95% CI –6.77 to –1.54; p value for trend=0.002). Also, analyses using imputed values confirmed the results obtained in the main analysis (β: –3.77 g/day; 95% CI –6.46 to –1.07 for the highest vs the lowest tertile of change in the overall DA index; p value for trend=0.006). Increased functional impairment was associated with a decrease in alcohol intake (β: –3.25 g/day; 95% CI –5.88 to –0.63, for the highest vs lowest tertile; p value for trend=0.022). We also found some tendency to reduce intake in those reporting increased self-rated health and vitality problems, as well as in individuals presenting an increase in morbidity and use of health services, though it failed to reach statistical significance.
A total of 124 (21.7%) participants quit drinking from wave 1 to wave 3. Compared with those in the lowest tertile of change in the overall DA index, those in the intermediate and the highest tertiles were more likely to quit drinking (OR: 1.99; 95% CI 1.10 to 3.59 and OR: 2.80; 95% CI 1.54 to 5.08, respectively; p value for trend=0.001) (table 3). The results were not materially modified after additional adjustment for change in alcohol intake from wave 0 to wave 1 (OR: 1.99; 95% CI 1.09 to 3.59 for the intermediate vs the lowest tertile of change in the overall DA index, and OR: 2.75; 95% CI 1.50 to 5.02 for the highest vs the lowest tertile; p value for trend=0.001). Analyses using imputed values led to results similar to those obtained in the main analysis (OR: 2.43; 95% CI 1.34 to 4.40 for the highest vs the lowest tertile of change in the overall DA index; p value for trend=0.004). A worsening in the functional, self-rated health and vitality, and morbidity and health services use domains suggested a tendency to increase the odds of alcohol cessation, though none reached statistical significance (table 3).
Association of change in overall and domain-specific DA indexes during 3.2 years (wave 0–wave 1) with quitting alcohol consumption in the subsequent 5 years (wave 1–wave 3)
When we explored the association of individual items with changes in alcohol consumption, we found that self-rated health was the item that contributed the most to the association between the DA index and decrease in alcohol consumption (β: –7.10 g/day; 95% CI –13.30 to –0.90). Another item that suggested a relationship with alcohol reduction was a decrease in walking speed (β: –3.09 g/day; 95% CI –6.35 to 0.18).
Again, self-rated health was the single component to influence the outcome quitting alcohol the most (OR: 4.48; 95% CI 1.17 to 17.08) followed by onset of diabetes (OR: 2.71; 95% CI 1.19 to 6.14) and stroke (OR: 13.04; 95% CI 1.22 to 139.87), and increased prevalence of hospitalisation (OR: 2.04; 95% CI 1.05 to 3.95).
Discussion
In this study of older drinkers in Spain, health deterioration was associated with both a decrease in alcohol intake and even cessation of alcohol consumption. A worsening in the functional, self-rated health and vitality and morbidity and health services use domains of the DA suggested an influence on both reduction and cessation of alcohol intake. This hints at a synergistic contribution to the overall association between DA deterioration and drinking patterns. The main contributor to drinking reduction or cessation was self-rated health, whereas onset of diabetes or stroke and an increased prevalence of hospitalisation also influenced the odds of cessation.
A number of previous studies have found an association between poor health and reduction or cessation of alcohol consumption in middle-aged and older adults.13–17 26 However, these studies evaluated the relationship between baseline health status, or changes in health from baseline to follow-up, and changes in alcohol consumption during the same follow-up period. This approach does not entirely account for reverse causation, because changes in health and in drinking behaviour may overlap in time, and it cannot be determined with certainty whether deteriorating health leads to reduced alcohol consumption or the opposite. Our study examined the association between changes in health status in a 3-year period and changes in alcohol consumption over the subsequent 5 years. By avoiding any time overlap, we ensured that changes in health status preceded changes in alcohol intake.
Increased functional impairments were related to a decrease in alcohol intake, with a reduction in walking speed as the major contributor of this domain to this change. However, our analyses failed to reveal a statistically significant association between functional impairments and cessation of alcohol consumption. Previous research has reported a lower risk of functional limitations associated with alcohol intake versus abstention in older adults,27–29 whereas ex-drinkers were at a higher risk of impairment.29 Also, alcohol use has been associated with lower risk of frailty,30–32 and it is worth noting that slow gait is one of the main diagnostic criteria of frailty.33 However, despite their longitudinal design, previous studies27–29 did not account for changes in alcohol intake during follow-up, so reverse causation might have played a role in such associations. In fact, our results suggest that functional deterioration might have an impact on drinking habits, exerting a greater influence on the amount of alcohol consumed than on cessation.
Worsened self-rated health, the health status measure, together with onset of chronic diseases, most commonly used in previous research, had a significant effect reducing alcohol intake. This specific result supports those based on the English Longitudinal Study of Ageing cohort, which revealed an association between worsened perceived health and decline in drinking frequency.16 Other authors, however, have reported a lower prevalence of worsened self-rated health in participants who quit drinking17 or no association at all.14 Despite the prominent effect of the self-rated health item, we failed to detect a statistically significant association between the self-rated health and vitality domain and change in alcohol intake. This might be explained by the negative association between some of this domain’s items (eg, interference of health with social activities) and decrease in drinking.
Onset of diabetes or stroke and increased prevalence of hospitalisation were related to alcohol cessation. Molander et al 14 reported an association between hospitalisations in the previous year and a lower consumption of alcohol, and Liang and Chikritzhs15 found that a diabetes diagnosis increased the likelihood of lowering intake or quitting drinking. It is known that alcohol intake may lead to hypoglycaemia in treated patients with diabetes (particularly those on insulin treatment), so this may contribute to explain the association between diabetes onset and reduced drinking. Although previous studies observed decreased drinking levels in association with onset of other chronic conditions or major medical events,13–15 we did not.
Lastly, changes in mental health did not appear to influence variations in drinking patterns. The relationship between alcohol and mental health is complex: on one hand, depression has been identified as a risk factor for heavy drinking34; on the other hand, low-to-moderate drinking has been associated with a lower risk of mental health problems, including depression and anxiety.35–37 Whereas the former relationship may be the result of using alcohol for self-medicating, thus increasing alcohol intake as mental health deteriorates, the latter might reflect an intake reduction to avoid interactions with medications prescribed to treat health problems, including depression.
Strengths and limitations
The main strengths of this study are its longitudinal design with over 8 years of median follow-up and the design used that clearly establishes the temporality of the association and minimises reverse causation. Furthermore, including only current drinkers reduces residual confounding as well as potential selection biases, usually difficult to control in observational studies since life-time abstainers generally present worse lifestyle and health characteristics than regular drinkers, and ex-drinkers may have stopped drinking due to poor health. Regarding our measures, alcohol consumption data come from a validated diet history, with a Pearson correlation coefficient between the diet history and seven 24-hour recalls for alcohol consumption over 1 year of 0.65.23 Lastly, rather than the most commonly used measure of health, self-rated health, we used the DA index, also widely used, which has shown to predict adverse outcomes such as death, institutionalisation or hospitalisation in older adults.21
Among the limitations of the study is its small sample size, which may have limited the power to detect existing associations for certain individual deficits previously described in the literature, such as cancer or hypertension.13–15 17 Moreover, compared with those who were excluded from the analyses because of losses to follow-up or missing data in important variables, those who were included had a younger age, higher educational level, better diet, lower sedentary time and a lower DA index at baseline, as well as a lower increase in the DA index from wave 0 to wave 1; thus, it is likely that our results underestimate the actual study associations. Also, as in any observational study, we cannot entirely rule out residual confounding, despite the measures taken to reduce it.
Conclusions
Health deterioration is linked to a subsequent decrease and cessation of alcohol consumption in older drinkers. This finding adds to recent scepticism regarding the protective health effect previously attributed to low-to-moderate alcohol consumption. Therefore, alcohol consumption should not be recommended for health improvement. This is particularly salient in older populations given the high prevalence of chronic conditions aggravated by alcohol38 and the frequent use of alcohol-interacting medications.39
Our findings confirm self-rated health, one of the most commonly used measures of health status in previous research, as the single most influential predictor of drinking behaviour. Morbidity predicts alcohol cessation, and functional limitations (particularly slower walking speed) might influence the amount of alcohol consumed. Thus, they both should also be considered when evaluating the association between alcohol consumption and health in older adults.
What is already known on this subject
Observational studies have reported a beneficial effect of low-to-moderate alcohol consumption on all-cause mortality, especially cardiovascular mortality.
However, the direction of causation between alcohol consumption and health is unclear because part of the potentially beneficial effect of alcohol may be due to health status influencing alcohol consumption rather than the opposite.
What this study adds
This study shows that health deterioration predicts a decrease and cessation of alcohol consumption in older drinkers. Specifically, self-rated health is the single most influential predictor of drinking behaviour.
Morbidity predicts alcohol cessation, and functional limitations (particularly slower walking speed) also influence the amount of alcohol consumed. These findings add to recent scepticism regarding the protective health effect previously attributed to low-to-moderate alcohol consumption.
References
Footnotes
Contributors FR-A and E-LG conceived the study; RO and EG-E performed the statistical analyses; RO and FR-A drafted the manuscript; all authors reviewed the manuscript for important intellectual content; RO and FR-A had primary responsibility for final content. All authors read and approved the final manuscript.
Funding This work was mainly supported by grant no. 02/2014 from the Plan Nacional sobre Drogas (Ministry of Health of Spain). Additional funding was obtained from FIS grants 12/1166, 16/609 and 16/1512 (Instituto de Salud Carlos III, State Secretary of R+D+I and FEDER/FSE), CIBERESP, and the Salamander Project (JPI-A Healthy Diet for a Healthy Life, State Secretary of R+D+I PCIN-2016-145).
Disclaimer The funding agencies had no role in study design, data collection and analysis, interpretation of results, manuscript preparation or in the decision to submit this manuscript for publication.
Competing interests None declared.
Patient consent Obtained.
Ethics approval The Clinical Research Ethics Committee of ‘La Paz’ University Hospital in Madrid approved the study.
Provenance and peer review Not commissioned; externally peer reviewed.