Moderate Alcohol Consumption and Risk of Heart Failure Among Older Persons

Context Heavy consumption of alcohol can lead to heart failure, but the relationship between moderate alcohol consumption and risk of heart failure is largely unknown.

Objective To determine whether moderate alcohol consumption predicts heart failure risk among older persons, independent of the association of moderate alcohol consumption with lower risk of myocardial infarction (MI).

Design Prospective cohort study conducted from 1982 through 1996, with a maximum follow-up of 14 years.

Setting and Participants Population-based sample of 2235 noninstitutionalized elderly persons (mean age, 73.7 years; 41.2% male; 21.3% nonwhite) residing in New Haven, Conn, who were free of heart failure at baseline. Persons who reported alcohol consumption of more than 70 oz in the month prior to baseline were excluded.

Main Outcome Measure Time to first fatal or nonfatal heart failure event, according to the amount of alcohol consumed in the month prior to baseline.

Results Increasing alcohol consumption in the moderate range was associated with decreasing heart failure rates. For persons consuming no alcohol (50.0%), 1 to 20 oz (40.2%), and 21 to 70 oz (9.8%) in the month prior to baseline, crude heart failure rates per 1000 years of follow-up were 16.1, 12.2, and 9.2, respectively. After adjustment for age, sex, race, education, angina, history of MI and diabetes, MI during follow-up, hypertension, pulse pressure, body mass index, and current smoking, the relative risks of heart failure for those consuming no alcohol, 1 to 20 oz, and 21 to 70 oz in the month prior to baseline were 1.00 (referent), 0.79 (95% confidence interval [CI], 0.60-1.02), and 0.53 (95% CI, 0.32-0.88) (P for trend = .02).

Conclusions Increasing levels of moderate alcohol consumption are associated with a decreasing risk of heart failure among older persons. This association is independent of a number of confounding factors and does not appear to be entirely mediated by a reduction in MI risk.

Heavy alcohol consumption can have toxic effects on the heart that result in heart failure.^1,2 The relationship between moderate alcohol consumption and risk of heart failure, however, is largely unknown. One study of patients hospitalized for myocardial infarction (MI) reported that compared with patients consuming little or no alcohol, those reporting moderate consumption had a nonsignificantly lower odds of developing heart failure during hospitalization.³ Investigators have also shown that among persons with existing left ventricular systolic dysfunction, light to moderate drinking is a favorable prognostic factor.⁴ However, prospective community-based studies of moderate alcohol consumption and heart failure risk among persons free of heart failure at baseline have not been conducted. One would naturally expect such studies to show that moderate alcohol consumption is associated with a lower risk of heart failure, simply because moderate drinking is associated with a reduced risk of MI^5,6 and MI is a major risk factor for heart failure. Yet, moderate alcohol consumption, perhaps by lowering blood pressure^7,8 or by promoting neurohormonal changes that prevent the clinical onset of heart failure,^9,10 might reduce heart failure risk via pathways that do not necessarily involve MI.

Demonstration of an association between moderate alcohol consumption and lower heart failure risk that is independent of the association of alcohol consumption with lower MI risk would be important, because it would indicate that moderate alcohol consumption protects against adverse cardiovascular processes beyond those related to ischemic heart disease. Furthermore, the establishment of such an association in older persons might be of particular importance from a public health standpoint, because light to moderate drinking is engaged in by a notable proportion of older persons (around 20%),¹¹ and heart failure is one of the leading causes of hospitalization of the elderly.¹² In the current study, we used a community-based sample of elderly persons to prospectively examine whether moderate alcohol consumption, compared with no alcohol consumption, was associated with a lower risk of heart failure, independent of any reduction in MI risk.

The current study is based on data from the New Haven, Conn, site of the Established Populations for the Epidemiologic Study of the Elderly program (EPESE), funded by the National Institute on Aging.¹³ The New Haven EPESE is a prospective cohort study of community-based persons aged 65 years and older. The cohort was assembled in 1982 by obtaining a stratified probability sample of persons from 3 New Haven housing strata: public housing for the elderly, private housing for the elderly, and general housing in the community. Sampling from these strata yielded 2812 persons who agreed to participate in the study (the response rate was 82%). At baseline, trained interviewers administered a structured, 75-page questionnaire to all participants to obtain information on a variety of factors. We excluded 311 persons who were considered potential heart failure cases at baseline, defined as use of both loop diuretics and cardiac glycosides or self-reported symptoms of orthopnea based on affirmative answers to the following 2 questions: "Do you get shortness of breath when you are lying down flat?" and "Does this shortness of breath improve when you sit up, or do you use extra pillows at night to prevent it?" We additionally excluded 219 persons with missing data on alcohol consumption or other study variables used in multivariable analyses. Because the number of persons reporting heavy alcohol consumption was small (n = 47), which made it difficult to analyze heavy consumers as a separate group, and because we were primarily interested in the effect of moderate alcohol consumption on heart failure risk relative to little or no consumption, we further excluded the 47 heavy consumers (see below for description of alcohol consumption levels). After all exclusions, the sample size for the study was 2235. Participants were followed up from 1982 to 1996 (maximum follow-up time was 14 years).

At baseline, interviewers asked participants how many bottles or cans of beer, glasses of wine, and drinks of liquor they had consumed in the preceding month. From this information, we calculated the total intake of pure alcohol (in ounces) for the month to be: 0.48 × the number of bottles or cans of beer + 0.60 × the number of glasses of wine + 0.675 × the number of drinks of liquor. This formula was based on a national survey of alcohol consumption taken around the time of the baseline interview¹⁴ and has been used in a prior study of alcohol consumption and coronary heart disease.¹⁵ Participants were categorized into the following 3 groups based on the total ounces consumed in the prior month: 0, 1 to 20 oz (30-600 mL) (up to approximately 1 or 1.5 drinks per day), and 21 to 70 oz (630-2100 mL) (about 1.5 to 4 drinks per day). The middle category was chosen because studies have shown that consumption of less than 1 or 1.5 drinks per day can have protective effects against adverse cardiovascular outcomes.^16,17 The 21- to 70-oz category was chosen because some evidence suggests that consumption of more than 2 drinks per day may also have protective effects against adverse cardiovascular outcomes.⁶ Those persons consuming more than 70 oz in the month before baseline were considered heavy consumers and were excluded from analyses.

Our analyses controlled for the effects of other study measures thought to be associated with the risk of heart failure including age, sex, race (white vs nonwhite), education, angina, self-reported history of physician-diagnosed MI or diabetes, and hypertension based on measured levels of blood pressure and antihypertensive medication use (no hypertension: systolic blood pressure [SBP] <140 mm Hg, diastolic blood pressure [DBP] <90 mm Hg, and no antihypertensive medications; stage I hypertension: SBP ≥140 to 160 mm Hg or DBP ≥90 to 100 mm Hg in persons not taking antihypertensive medications; stage II hypertension: SBP ≥160 mm Hg or DBP ≥100 mm Hg or current use of antihypertensive medications). We also considered baseline pulse pressure in our analyses because this factor shows a strong, graded association with heart failure risk in the New Haven EPESE.¹⁸ Baseline smoking status (current vs past or never) and body mass index (BMI; 4 categories: ≤23 kg/m², 24-27 kg/m², ≥28 kg/m², and missing) were also controlled for. In addition to controlling for baseline history of MI, we also controlled for the development of MI during follow-up. Ascertainment of MI during follow-up in the New Haven EPESE has been described previously.¹⁹

The principal outcome in this study was the occurrence of a heart failure event (nonfatal or fatal). Although some persons in the New Haven EPESE experienced multiple heart failure events during follow-up, we only considered the first event in our analyses. Nonfatal events were identified through surveillance of 2 New Haven area hospitals where over 90% of the hospital admissions for the New Haven EPESE cohort occurred. Surveillance of these 2 hospitals was supplemented by information from the Health Care Financing Administration. Participants whose primary or first 3 secondary hospital discharge diagnoses included International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes 428, 402.01, 402.11, 402.91, 404.01, 404.13, 404.91, or 404.93 were identified and their records were reviewed after follow-up had been completed. To be considered a nonfatal heart failure case, hospitalization records needed to indicate the following: (1) symptoms of heart failure, such as dyspnea or orthopnea; (2) radiographic (eg, cardiomegaly on x-ray) or physical (eg, pulmonary rales more than basilar, S₃ gallop) evidence of heart failure; (3) heart failure as 1 of the 3 major precipitating factors for hospitalization; and (4) occurrence of heart failure symptoms prior to or within 24 hours of hospitalization. Heart failure has been defined in a similar manner by other studies.^20,21 Identification of fatal heart failure events was made by examining death certificates, which were coded by a single nosologist. Death certificates in which the underlying cause of death was listed as ICD-9-CM code 428, 402.01, 402.11, 402.91, 404.01, 404.13, 404.91, or 404.93 were considered to be fatal heart failure events. Ascertainment of vital status was nearly complete (>99%).

The first step in the analysis was to examine bivariate associations between the 3 alcohol consumption groups and the other study measures described above. These bivariate associations were assessed by χ² tests when the other study measures were categorical, and by analysis of variance when the other study measures were continuous (age, years of education, and pulse pressure). We then constructed Kaplan-Meier curves and used the log-rank test to compare each alcohol group's unadjusted cumulative probability of surviving free of heart failure during follow-up. Next, we analyzed the association between alcohol consumption and risk of heart failure in multivariable statistical models that adjusted for the effects of the other study measures. Multivariable adjustment was accomplished by Cox proportional hazards regression,²² with time to first heart failure event as the outcome of interest. In the multivariable Cox models, the group reporting no consumption in the prior month was treated as the referent level and the other 2 alcohol groups were entered as dummy variables. To determine if a linear trend existed between levels of alcohol consumption and risk of heart failure, we ran multivariable Cox models with an ordinal alcohol consumption variable that assigned participants the median ounces of alcohol consumed by their respective alcohol groups. The development of MI during follow-up was treated as a time-dependent variable in the Cox models.

We also conducted several secondary analyses. First, the group reporting no alcohol consumption at baseline may have included persons who quit drinking as a result of cardiovascular symptoms or illness. It has been argued that inclusion of such "sick quitters" could lead to a spurious association between alcohol consumption and reduced risk of subsequent cardiovascular events.²³ Investigators have attempted to rule out the possibility of this type of spurious association by looking at the relationship between alcohol and cardiovascular events after (1) excluding events within the early part of the follow-up period,^15,17,24 (2) excluding persons with a history of coronary heart disease at baseline (rather than simply controlling for it at baseline, which could result in residual confounding due to variations in severity of the pre-existing coronary heart disease),²⁴ or (3) excluding persons who reported that their level of alcohol consumption at baseline was noticeably lower than it had been in the past.¹⁵ To see if our results were influenced by sick quitters, we examined the relationship between alcohol consumption and heart failure after making the 3 types of exclusions noted above. Second, we analyzed whether our results were affected by controlling for time-dependent covariates that measured diabetes status, presence of hypertension, and pulse pressure during follow-up (3, 6, and 12 years after baseline). Third, recent evidence suggests that the protective effect of alcohol against coronary heart disease may depend on the type of alcoholic beverage consumed, with a higher level of protection being seen among wine drinkers.¹⁶ To see if the risk of heart failure depended on the type of alcohol consumed, we ran separate analyses for wine, beer, and liquor.

The New Haven EPESE was a complex survey that involved sampling weights, stratification, and clustering. These aspects of the survey can have effects on point estimates and SEs, and these effects have been accounted for in previous New Haven EPESE studies by using SUDAAN (Research Triangle Institute, Research Triangle Park, NC) for data analyses.¹⁹ However, we found that for the alcohol variables, results in SAS (SAS Institute, Cary, NC) and SUDAAN were very similar. Because the results were similar, and because SUDAAN did not allow us to incorporate time-dependent covariates in Cox proportional hazards regression models, we used SAS for all statistical analyses.

The average age of the study population was 73.7 years. Approximately 41.2% of participants were male, and 21.3% were nonwhite. Approximately 50.0% of persons reported no alcohol consumption in the month before baseline, while 40.2% reported consumption of 1 to 20 oz and 9.8% reported consumption of 21 to 70 oz. Table 1 shows bivariate associations between alcohol consumption and the other study variables. Those consuming higher levels of alcohol were significantly younger, more likely to be male, less likely to be nonwhite, and had higher educational levels. They were also less likely to have a history of diabetes, less likely to have stage II hypertension at baseline, and more likely to be current smokers. Those who consumed higher levels of alcohol also appeared to be less likely to have a low BMI or missing data on BMI. Those who consumed 1 to 20 oz in the month before baseline were significantly less likely to have a history of MI compared with those with lower or higher levels of consumption.

During follow-up, a total of 281 persons experienced a first heart failure event (28 of these events were fatal). Table 2 shows crude heart failure event rates, according to alcohol consumption level. Those reporting higher levels of alcohol consumption had lower rates of heart failure. For those consuming no alcohol, 1 to 20 oz of alcohol, and 21 to 70 oz of alcohol, the event rates per 1000 person-years of follow-up were 16.1, 12.2, and 9.2, respectively. Although not included in our analyses, the group reporting more than 70 oz of alcohol consumption had a comparatively low heart failure rate (7.0 events per 1000 person-years of follow-up). Unadjusted survival curves showing each group's cumulative probability of surviving free of heart failure are presented in Figure 1. A log-rank test indicated that the difference between these curves was significant (P = .01).

To examine the association between alcohol consumption and heart failure after adjusting for other factors, we ran multivariable Cox proportional hazards models (Table 3). After adjustment for age and sex (model 1), the relative risks (RRs) of heart failure were 1.00 (referent), 0.71 (95% confidence interval [CI], 0.56-0.92), and 0.47 (95% CI, 0.29-0.76) (P for trend = .01) for those consuming no alcohol in the past month, 1 to 20 oz of alcohol, and 21 to 70 oz of alcohol, respectively. On additional adjustment for race, education, angina, history of MI and diabetes, MI during follow-up as a time-dependent covariate, hypertension, pulse pressure, BMI, and current smoking status (model 2), the RRs for increasing levels of alcohol consumption were 1.00 (referent), 0.79 (95% CI, 0.60-1.02), and 0.53 (95% CI, 0.32-0.88) (P for trend = .02).

We then conducted several sets of secondary analyses. First, we attempted to rule out possible bias due to the inclusion of sick quitters in the group who reported no consumption of alcohol at baseline. We therefore ran fully adjusted Cox models that excluded the first 2 years of follow-up, excluded persons with a history of MI at baseline, or excluded persons who reported having curtailed their alcohol consumption prior to baseline. After making these exclusions, we found that increasing levels of alcohol consumption remained associated with a decreasing risk of heart failure. For example, when excluding the first 2 years of follow-up, the risk of heart failure was 1.00 (referent), 0.82 (95% CI, 0.62-1.09), and 0.54 (95% CI, 0.32-0.92) (P for trend = .03) for those consuming no alcohol, 1 to 20 oz, and 21 to 70 oz of alcohol, respectively. After excluding those with a history of MI at baseline, results were 1.00 (referent), 0.76 (95% CI, 0.57-1.00), and 0.47 (95% CI, 0.26-0.85) (P for trend = .02). And after excluding those who reported heavier alcohol consumption at some time in the past, the results for the 3 levels of alcohol consumption were 1.00 (referent), 0.76 (95% CI, 0.55-1.05), and 0.31 (95% CI, 0.15-0.67) (P for trend = .002).

Second, we attempted to determine if incident diabetes during follow-up, or hypertension and pulse pressure during follow-up, could explain our results. To make this determination, we ran a fully adjusted Cox model with time-dependent covariates for diabetes, hypertension, and pulse pressure. Results of this model were similar to our main analyses; RRs were 1.00 (referent), 0.79 (95% CI, 0.61-1.03), and 0.55 (95% CI, 0.33-0.90) (P for trend = .03) for nondrinkers, those consuming 1 to 20 oz of alcohol, and those consuming 21 to 70 oz of alcohol, respectively.

Third, we analyzed whether the association between alcohol consumption and risk of heart failure differed according to the type of alcoholic beverage consumed. Because of the small number of persons drinking more than 20 oz of any 1 type of beverage in the month before baseline, consumption levels for each type of beverage were simply categorized as "none" vs "any." The results of the beverage-specific analyses are presented in Table 4. Consumers of each type of beverage showed similar reductions in heart failure risk compared with nonconsumers after adjustment for other factors. None of the specific beverage types showed a significant association with heart failure after controlling for total alcohol consumption.

Fourth, we sought to determine whether there were significant interactions between alcohol consumption and other major study variables. Given the relatively small number of events in the group consuming 21 to 70 oz of alcohol in the month before baseline, the power to detect interactions in this group was somewhat limited. Thus, we created a dichotomous "none" vs "any" alcohol consumption variable and assessed interactions between this dichotomous variable and other selected study variables (sex, race, diabetes history, and hypertension levels). In fully adjusted models, however, we failed to find that any of these interactions were significant.

Finally, we also examined the relationship between alcohol consumption and total mortality (Table 5). For those consuming 0, 1 to 20, and 21 to 70 oz of alcohol, the RRs of total mortality were 1.00 (referent), 0.81 (95% CI, 0.72-0.91), and 0.75 (95% CI, 0.62-0.90) (P for trend = .01) after adjustment for all of the other study variables. Furthermore, we noticed that controlling for heart failure during follow-up helped explain a notable amount of the relationship between moderate alcohol intake and reduced total mortality risk (data not shown). Moderate alcohol consumption was also associated with a reduced risk of hospitalization for any cause, but the reduction was not significant and was not nearly as substantial as the reduction in risk that was seen for heart failure (data not shown).

The main finding of this study was that compared with no alcohol consumption, increasing levels of alcohol consumption in the low to moderate range were associated with a lower risk of heart failure among community-based older persons. This association was observed after controlling for a number of factors including age, sex, race, education, angina, history of MI and diabetes, MI during follow-up, hypertension, pulse pressure, BMI, and smoking. A prior study had shown that moderate alcohol consumption was associated with a lower odds of heart failure among persons hospitalized for MI, although the association was not significant.³ The current study demonstrates for the first time that moderate alcohol consumption is significantly associated with a lower risk of heart failure in a community-based population. Our study also showed that different types of alcoholic beverages (beer, wine, liquor) were associated with similar reductions in heart failure risk, suggesting that it is pure alcohol, and not the type of beverage, that is associated with lower heart failure risk. Overall, moderate drinking did not appear to be a harmful factor in this study but was, instead, arguably beneficial in that it was associated with a reduced risk of total mortality.

What could explain our finding that moderate alcohol consumption is associated with a lower risk of heart failure? First, our results may have been due to confounding from basic demographic factors such as age, sex, or socioeconomic status. As noted above, however, our finding was observed after adjustment for age and sex. It was also observed after adjustment for race and education levels, both of which can be regarded as rough indicators of socioeconomic status. Thus, confounding from demographic factors does not appear to be a likely explanation for our results.

Second, the group reporting no consumption in the month before baseline may have included persons who quit drinking as a result of symptoms associated with cardiovascular disease. This could lead to a spurious association between alcohol consumption and lower risk of cardiovascular events, including heart failure. Our main analyses minimized the possibility of this spurious association by excluding persons with heart failure at baseline, based on symptoms and medication use, and by controlling for history of MI at baseline. Furthermore, our finding persisted in secondary analyses that excluded persons with a history of MI at baseline, excluded heart failure events within the first 2 years of follow-up, and excluded those who reported having curtailed their consumption of alcohol. These secondary analyses helped to further rule out the possibility that our finding reflected a scenario in which illness caused people to quit drinking at baseline.

Third, alcohol consumption may have had physiological effects that reduced the risk of MI, thereby resulting in lower heart failure risk. Such a pathway seems reasonable since, on one hand, alcohol consumption may reduce the risk of MI by increasing high-density lipoprotein cholesterol levels,²⁵ decreasing platelet aggregation,²⁶ and increasing fibrinolytic activity²⁷ and, on the other hand, MI is a common antecedent of heart failure. Nevertheless, our study found that alcohol consumption was associated with lower heart failure risk, even after controlling for baseline history of MI and MI during follow-up. This suggests that a reduction in MI risk may not substantially explain the association we observed. However, during follow-up, we only controlled for MI events that resulted in hospitalizations. We did not control for other ischemic events that did not result in hospitalizations, but which could have progressed toward heart failure, such as silent/unrecognized MI and chronic or unstable angina. Thus, it is possible that the association between moderate alcohol consumption and lower heart failure risk is mediated by a reduction in the risk of these other ischemic events.

Fourth, diabetes and hypertension are strong risk factors for heart failure. Recent evidence suggests that modest alcohol consumption may reduce the risk of diabetes.²⁸ There is also evidence that modest alcohol consumption is associated with lower blood pressure levels in rats^29-31 and humans.^7,8,31,32 Even among people with heart failure, modest alcohol consumption lowers blood pressure in the short term.³³ Yet, we controlled for hypertension, pulse pressure, and diabetes history at baseline and during follow-up and still found an inverse relationship between moderate alcohol consumption and heart failure. As such, alcohol-induced effects on blood pressure parameters and diabetes may not substantially account for our results.

Fifth, sporadic reports suggest that alcohol may have effects on several neurohormones that are thought to influence the progression of heart failure. Norepinephrine and arginine vasopressin are neurohormones that may foster the progression of heart failure.^34,35 Investigators have noted that low-dose alcohol consumption may blunt the effects of norepinephrine in rats³⁶ and may decrease plasma levels of arginine vasopressin in humans⁹ over the short term. It has also been reported that modest alcohol consumption may lead to short-term increases in plasma levels of atrial natriuretic peptide,^9,10 a neurohormone that might act against the progression of heart failure.³⁷ The neurohormonal effects of moderate alcohol consumption that are noted above would be consistent with a protective effect of moderate alcohol consumption against heart failure. However, the effects of modest alcohol consumption on neurohormonal factors are not well-established, and any attempt to use such factors to explain the association of modest alcohol consumption with reduced heart failure risk is speculative.

Sixth, our study findings may have been affected by misclassification of alcohol consumption and/or heart failure events. Given the prospective design of our study, however, any misclassification of alcohol consumption would likely have been nondifferential. Similarly, it seems likely that misclassification of heart failure deaths would have been nondifferential, as these deaths were classified without knowledge of a person's self-reported alcohol intake. Misclassification of heart failure hospitalizations was also—in all likelihood—nondifferential, because there is no obvious reason why misclassification of these hospitalizations would depend on moderate drinking status. Thus, although there may have been some misclassification of alcohol consumption and heart failure in this study, this misclassification was probably nondifferential in nature. Nondifferential misclassification biases associations toward the null and could not, therefore, explain away the protective association we found between moderate drinking and lower heart failure risk. Instead, such misclassification suggests that we may have underestimated the protective effect of moderate drinking against heart failure.

Although our results suggest that moderate alcohol consumption may reduce the risk of heart failure, it is important to point out that our investigation was not a true experiment but was, instead, an observational study. As such, our results may have been due to some confounding factor that we failed to control for. Moderate drinking may simply be correlated with a healthy diet or other healthy lifestyle practices. We had no information on diet and cannot rule out that it or other unmeasured factors confounded our results. Therefore, our results should be interpreted with caution.

In conclusion, this study provides evidence that increasing levels of moderate alcohol consumption are associated with a decreasing risk of heart failure among community-based older persons. This association is independent of a number of potential confounding factors and does not appear to be mediated by a reduction in the risk of MI. The community-based nature of our study population suggests that the results of this study may have wide applicability. Heavy consumption of alcohol can lead to negative cardiovascular outcomes such as higher blood pressure,⁸ cardiomyopathy,² and sudden cardiac death.³⁸ Thus, individuals should continue to be cautioned against drinking excessive amounts of alcohol. Nevertheless, this study adds to the growing evidence that moderate consumption of alcohol may be beneficial to the cardiovascular system.