Heavy drinking occasions in relation to ischaemic heart disease mortality— An 11–22 year follow-up of the 1984 and 1995 US National Alcohol Surveys

income, marital status, geographical region, depression score, survey period or other drug use were small. Conclusions Among former and current drinkers, heavy drinking occasions should be taken into account when examining the complex association of alcohol consumption on IHD mortality risk.

Background The relationship between alcohol consumption and ischaemic heart disease (IHD) risk is complex and several issues remain unresolved because many studies used rather crude exposure measures often based on one or two questions. The objective of this study was to investigate the association between heavy drinking occasions and IHD mortality while controlling for average daily alcohol intake and separating former drinkers from lifetime abstainers.

Methods
Cox regression analyses were used with IHD mortality as the outcome in a sample of 9934 participants of the US National Alcohol Surveys conducted in 1984 and 1995.

Results
To the end of 2006, 326 deaths from IHD were recorded in the 11-to 22-year follow-up period. Any past heavy drinking occasions in former drinkers [hazard ratio (HR) ¼ 2.06; 95% confidence interval (95% CI): 1.10-3.85] compared with former drinkers without such drinking occasions, and any heavy drinking occasion in current drinkers at baseline (HR ¼ 2.05; 95% CI: 1.03-3.98) compared with current drinkers with average daily intake of one to two drinks, were associated with higher IHD mortality in men and any heavy drinking occasions among drinkers of up to 1 drink average consumption in women with similar effect size. Confounding effects from age, race, education, employment, income, marital status, geographical region, depression score, survey period or other drug use were small.

Introduction
The relationship between alcohol consumption and ischaemic heart disease remains controversial. Since evidence from long-term randomized trials is unavailable, one has to rely mostly on observational studies to investigate this relationship, of which many have shown some form of a beneficial association on heart disease. [1][2][3] Generally, this relationship is described as a curvilinear (J-shape) with a relatively wide range of average daily alcohol intake associated with a potential cardioprotective effect without taking into account differential risks for former or heavy episodic drinkers. 1 For different reasons-such as study design, exposure and covariate assessment and statistical analysis-a potential cardioprotective association has been questioned many times. [4][5][6][7] Furthermore, substantial heterogeneity across studies (e.g. for region, sex and mortality vs morbidity) seen when examining average daily alcohol intake suggests that important confounders or effect modifiers have not been well accounted for in prior studies. 1 One possible explanation for differential risk (beyond sex and outcome) seems to be episodic heavy drinking occasions. 8 In most of the larger cohorts, however, patterns of episodic heavy drinking were either not assessed or were rare; 9 furthermore, assessment of alcohol intake typically relied on few items, often only current usual quantity and frequency. In addition, most studies have not separated lifetime abstainers from former drinkers, whose IHD mortality risk has been shown to be significantly higher in comparison to lifetime abstainers. 10 We wanted to test the hypothesis that there would be an increased risk for IHD mortality from heavy alcohol consumption in former drinkers and current episodic heavy drinkers compared with respective drinking groups without such drinking behaviour after stratifying by sex in a follow-up study of a nationally representative US sample designed for assessment of alcohol intake patterns.

Methods Subjects
Two cross-sectional waves (1984 and 1995) of the US National Alcohol Survey (NAS) were used as baseline data. 11  Exposure assessment All questions on alcohol exposure were implemented identically in both surveys.

Drinking status
Participants who answered 'I have never had wine/ beer/whiskey or liquor' to beverage-specific questions on how often they consumed alcohol were classified as lifetime abstainers. Those who reported consumption of 'less than once a year' for all types of beverages were classified as former drinkers. Former drinking behaviour was further distinguished into those who never consumed more than five drinks at one time and those who exceeded this limit based on a question about the largest number of drinks ever consumed. Analysing this distinction between past drinking behaviour was not possible in women because of the small number of former drinkers with past heavy drinking occasions.

Volume of alcohol consumption
Usual frequency of drinking days (no reference period given) of alcohol intake by beverage (wine/beer/ liquor) was assessed with nine answer categories (from highest: three or more times a day, twice a day, once a day, nearly every day, 3 or 4 times per week, once or twice a week, two or three times a month, about once a month, less than once a month but at least once a year, less than once a year, never). In addition, answers to questions on the proportion of drinking as many as five to six drinks per occasion, three to four drinks per occasion and one to two drinks per occasion consumed recently by beverage type (answer options among drinkers of such beverages: nearly every time, more than half the time, less than half the time, once in a while) were used by a summation algorithm to derive total average daily alcohol intake. 13,14 For the main analyses, the average daily number of drinks was converted into grams per day and categorized into the following categories of total average daily alcohol volume, with cut-points reflecting occasional drinkers, those having one to two drinks, three to four drinks and five or more drinks (assuming 14 g of pure alcohol content per US standard drink 15 ): 0 , 0.1-2.49, 2.5-28, 428-56 and 456 g/day. In women, the highest category was taken as 414 g/day, as the distribution of cases did not allow for a further distinction of categories.

Current heavy drinking occasions
We used the following indicators for current heavy drinking occasions in current drinkers at baseline: any positive answer to usual consumption of 5 or more drinks on one occasion by type of beverage (which also were used for the calculation of total average alcohol volume), or (because heavy drinking occasions involving more than 1 beverage could be missed) positive answers to specific questions on more than 12 and 8-11 drinks per drinking day in the past 12 months (any alcoholic beverage). 13,14 Covariates The list of potential confounders was restricted to the following because others, such as dietary intake other than alcohol, physical activity or BMI, were not assessed in the original surveys: race (White, Black, Hispanic and Asian), current smoking status (at least once every 52 months often vs not at all), marital status (yes vs no), education (less than high school, completed high school, some college and completed college), employment status (full time, retired and others), income (<US$10 000, 10 000 to <20 000, 20 000 to <40 000 and 540 000), born outside of the USA among Hispanics, geographical region (Northeast, Midwest, Pacific, South, and Mountain), survey (1984 vs 1995), depressive symptoms (CES-D 516) 16,17 and illegal drug use (at least once every 52 months often vs not at that level).

Outcome assessment
Out of a total of 1554 deaths from all-causes, 326 deaths (3.3% of the total analysis sample, 20.9% of all recorded deaths in the analysis sample) due to IHD (ICD-9: 410-414; ICD-10: I20-I25) were recorded until the end of 2006.

Statistical analyses
Time in number of days from the date of the interview to either IHD death or 31 December 2006 was used in Cox proportional hazard regression analyses. All analyses were conducted separately by sex and adjusted for age (years). Proportional hazards assumptions were tested using Schoenfeld residuals. Age was the only variable violating this assumption and therefore an interaction term between age and natural log-time was included in all models. First, the association between total average daily alcohol consumption at baseline and IHD mortality with current non-drinkers as the reference group was estimated. Then, current non-drinkers were divided into lifetime abstainers and former drinkers, with lifetime abstention as the reference group. Successively, indicator variables for former drinker with any heavy drinking occasion in the past (among men only) and current drinkers of one or two drinks average daily consumption with any episodic heavy drinking occasion in the year before the baseline interview were added to this model. Sensitivity analyses were conducted excluding deaths in the first 2 years of follow-up to avoid reverse causality problems when already diseased subjects had changed their consumption prior to baseline, and IHD deaths beyond 75 years of age at the time of death because of potentially compromised accuracy of the cause of death ascertainment. 18 We further tested the heavy drinking contrasts in subgroups defined by age, smoking status, other drug use, survey and depressive symptoms where this was possible, based on the number of cases. Statistical analyses were conducted in SPSS version 17.0 (SPSS Inc., Chicago, IL, USA).

Results
Mean [standard deviation (SD)] follow-up time was 15.1 (6.1) years and mean age at the time of IHD death was 71.1 (13.2) years in men, and 16.0 (6.1) years of follow-up, 75.1 (14. 3) years of age in women. Among men, 29.6% were non-drinkers at baseline, of whom slightly more than half were former drinkers; 65.1% (967) of current drinkers with 2.5-28 g/day average daily intake were classified as drinkers with episodic heavy drinking occasions (Table 1). Among women, 37.3% (1034) of all non-drinkers at baseline were former drinkers and <10% (532) of the women in the sample reported average daily alcohol consumption beyond 14 g/day ( Table 2). Although the distributions of several potential confounders were quite different across drinking groups, the effect estimates from regression analyses were not influenced by these differences. Table 3 shows the relationship between average per day alcohol consumption and IHD mortality. Regarding men, with non-drinkers as the reference group, a somewhat U-shaped curve emerged (Model 1). Having lifetime abstainer as the reference group (Model 2, fully adjusted), the hazard ratio (HR) for former drinking was 1.26 [95% confidence interval (95% CI): 0.78-2.02] and the curve shifted upwards, bringing point estimates for average daily alcohol intake categories closer to 1 for average intake up to 56 g/day resulting in a J-shaped curve. In women, former drinking showed a small decreased risk of IHD mortality. CIs were generally wide and included 1 for all estimates in Table 3.
Among men, adding contrasts depicting any former heavy drinking occasions among former drinkers and episodic heavy drinking occasions in current drinkers of 2.5-28 g/day average daily intake showed substantial risk increases with similar effect sizes [HR ¼ 2.06 (95% CI: 1.10-3.85) and 2.02 (1.03-3.98), respectively] (Table 4). Indeed, of all covariates included in the regression models, these two heavy drinking contrasts were most strongly associated with IHD mortality risk (other than age), whether indicated by P-value or effect size. With lifetime abstainer as FOLLOW-UP OF THE 1984 AND 1995 US NATIONAL ALCOHOL SURVEYS   any current heavy drinking in women with 2.5-14 g/ day average intake showed similar effect size compared with men. Additional adjustment for covariates did not substantially change any of the alcohol estimates in the analysis. Using the fully adjusted models, sensitivity analyses did not show substantial changes. Separate subgroup analyses revealed that the associations for former or current heavy drinking were very similar in those who did not use other drugs. However, the association of former heavy drinking was restricted to older men (465 years), and the association of current heavy drinking episodes to those between 40 and 65 years. In each age group, the respective associations were similar when the sample was simultaneously restricted to non-smokers and men with CES-D <16. The association of current heavy drinking episodes was also similar across surveys (1984 and 1995). Former heavy drinking occasions showed stronger associations in the 1984 sample, and no association in the 1995 sample. Such subgroup analyses were not possible in women because of the small number of cases in each cell.

Discussion
This study, using more detailed measurement of exposure than previous studies, showed the importance of distinguishing between drinking patterns in both former drinkers and current drinkers with low average intake (on average two drinks per day or less, usually thought to be beneficial for heart disease) in men. An almost identical association was found in women for current drinkers up to one drink per day. However, owing to low power due to the small number of IHD deaths in drinking categories, CIs were wide, particularly among women. Nevertheless, the association of former drinking and current heavy drinking was consistent across sexes and in several sensitivity analyses and subgroup analyses. Although confounding was small in all models, particular age seems to be a strong influence on the association of alcohol consumption on IHD mortality risk in general, and particularly for heavy drinking occasions. Given these findings, current average daily alcohol consumption alone does not seem to adequately capture the alcohol-IHD risk relationship, and particularly for men the deleterious role of any heavy drinking in former drinkers as well as in current drinkers of one to two drinks on average was apparent.

Limitations
Although assessment of alcohol intake in surveys seems to be reasonably valid, 14,19 it generally underestimates total per capita alcohol consumption in a population compared with sales statistics despite a correlation between the two over time. 20,21  All models were adjusted for alcohol categories as displayed in Model 2 in Table 3; fully adjusted models further include smoking status, race, education, employment status, marital status, income, survey, region, depression symptoms (CES-D 516), born outside the USA (among Hispanics only) and other drug use.
a Within average daily alcohol consumption of 2.5-28 g/day in men and 2.5-14 g/day in women.
NA: not applicable.
FOLLOW-UP OF THE 1984 AND 1995 US NATIONAL ALCOHOL SURVEYS However, we used several items to assess heavy alcohol consumption in the sample and multiple beverage-specific items to derive total daily average alcohol intake. Both these factors generally result in higher estimated total alcohol consumption in population surveys. 14,22-29 Importantly, we used multiple items for identifying drinkers with heavy drinking occasions at baseline and thus have confidence that we have separated those with potentially deleterious patterns from drinkers with no such drinking behaviour. However, as in many other studies, we relied on baseline assessment of alcohol exposure and, although past drinking behaviour was taken into account in a crude measure (five or more drinks ever in former drinker), alcohol consumption over the lifecourse might have changed during follow-up or prior to baseline more than we were able to capture with our measurements. More detailed assessment of frequency of heavy drinking occasions over the life course certainly would help to shed light into the alcohol-heart relationship, given that both moderate drinking and heavy drinking have been shown to vary over time. 30 Furthermore, even lifetime abstainers can be misclassified with a one time only measurement. 31,32 Although accuracy of cause of death ascertainment is diminished at higher age at time of death, excluding deaths occurring beyond 75 years of age revealed similar effects to the main analysis. Confounding was small in all models and indicators for heavy drinking episodes among both former and current mid-volume (one to two drinks per day) drinkers showed substantial effect sizes; however, our list of potential confounders was not complete and we cannot exclude the possibility that these unmeasured confounders, such as medical history, dietary factors other than alcohol or physical activity, explain the results of this study. Although many CIs included 1, small sample size might have precluded us from detecting significant effects in some drinking groups, rather than the effect estimates not being important. This is supported by the sensitivity and subgroup analyses. Lastly, using observational data, we can only examine associations and so avoid statements on causality of effects.

Implications
A recent meta-analysis showed a significantly elevated pooled relative risk (RR) for IHD death in former drinkers compared with long-term abstainers in both sexes without taking into account specific former drinking behaviour. 10 Few studies have addressed former drinking behaviour and these were mostly based on former average consumption. [33][34][35][36] The present study indicates that it is important to not only separate former drinkers from lifetime abstainers, but also distinguish between former drinkers who do and do not have past heavy drinking behaviour.
A recent meta-analysis showed an RR of 1.45 (95% CI: 1.24-1.73) for drinkers with usual heavy drinking occasions compared with current drinkers without usual episodic heavy drinking, using relatively crude measures of heavy drinking identified by usual quantity-frequency measurements without the detail in exposure measurement used in the present study. A recent study comparing the risk of myocardial infarction and coronary deaths from binge drinking in relation to regular drinkers in France and Northern Ireland showed similar effect sizes with an RR of 1.81, 95% CI: 1.05-3.11 for binge drinking in comparison with regular, non-binge drinkers in Northern Ireland, similar to our findings. 37 Our results not only confirm this increased risk, but also suggest that the difference in risk might be strongest in low consumption of up to and including two drinks per day on average. This suggests that heavy drinking occasions should be avoided altogether and when engaged in, may not confer any cardioprotective effect even at low levels of average consumption (HR compared with lifetime abstainer ¼ 1.17, 95% CI: 0.65-2.08). High and rising prevalence of binge drinking in many countries give reason for public health concern. 38,39 Possible adverse effects of heavy drinking occasions on heart disease include increased blood pressure, fibrinolytic factors and ventricular arrhythmia after cessation of drinking, 40 in contrast to a cardioprotective association of regular moderate alcohol consumption through an increase in high-density lipoproteins, inhibition of platelet activation and fibrinolytic factors. 41 A comprehensive review concluded that low-density lipoproteins are increased by heavy drinking episodes, and no increase in high-density lipoprotein levels, 42 although the evidence is inconsistent. 43 Nevertheless, we cannot comment on the influence of episodic heavy drinking occasions as we defined them beyond 28 g/day average daily alcohol intake in men and 14 g/day in women simply because there were not enough participants who exhibited average daily alcohol consumption of three to four drinks per day without any heavy drinking occasions. A potential cardioprotective association seemed to be evident up to 72 g/day average alcohol intake in the most recent meta-analysis, 1 suggesting a potentially beneficial association in drinkers beyond one to two drinks average intake. It is possible that the beneficial effect on IHD is strongest at higher intake than two drinks per day on average even when heavy drinking episodes were present. However, this remains speculative because there are little data with which to investigate episodic heavy drinking in this range of average intake. Furthermore, for any particular individual, the alcohol-IHD relationship cannot be seen in isolation from other disease outcomes because even at low levels of alcohol intake the effect on many other disease outcomes is detrimental, 44,45 and three to four drinks on average is not a safe consumption level from a clinical or public health perspective as it has been shown to be detrimentally associated with many other disease outcomes. 46

Conclusions
This study confirms earlier research and gives further weight to arguments that the relationship between alcohol consumption and IHD mortality is quite complex and even nonlinear modelling of only current average volume is insufficient to reflect this complexity. Moreover, for investigating IHD mortality risk associations, it confirmed the importance of separating lifetime abstainers, former drinkers and the latter's past drinking behaviour, in addition to heavy drinking occasions among low to moderate drinkers. It seems that heavy drinking occasions are an important component in determining IHD mortality risk in past or present alcohol consumers.