Alcohol Drinking and Total Mortality Risk

Introduction 

Total mortality has been studied as one global measure of the effects of alcohol drinking. In 1926, Pearl (1) reported that heavy drinkers had the highest risk of death but that light-moderate drinkers were at lower risk than alcohol abstainers; this was probably the first report of the J-shaped alcohol-mortality relation. Pearl had no biological explanation for the possible benefit of lighter drinking and concluded that it was “not harmful.” In recent decades, a number of studies and several meta-analyses 2, 3, 4, 5 have confirmed this J-shaped alcohol-mortality relation and shown that it is a composite of disparate relations of alcohol drinking to various conditions. Heavy drinking—herein operationally defined as usual daily intake of 3 or more standard-sized drinks—is associated with excess mortality from a number of cardiovascular and noncardiovascular causes. The lower mortality risk associated with lighter drinking is largely attributable to lower risk of coronary heart disease (CHD) and other atherothrombotic vascular conditions. Substantial evidence for several plausible biological mechanisms for protection by alcohol against CHD has contributed to widespread acceptance of a causal hypothesis. These include effects via high-density-lipoprotein (HDL) cholesterol, antithrombotic actions, better endothelial function, and other benefits 6, 7, 8, 9, 10.

In the absence of randomized controlled trial data, unresolved confounding might be involved in the apparent protection by light-moderate alcohol intake. Such confounding would represent an alternative explanation for lower risk of light drinking or higher risk of abstainers, but has not been demonstrated despite many attempts. One hypothesis has been that a spurious apparent relation might be the result of inclusion of persons who become abstainers because of illness (“sick quitters”) among the abstainer referent group. Originally propounded in the 1980s by Sharper, Wannamethee, and Walker (11), the “sick quitter” argument was recently reopened by a meta-analysis (12) that suggested that this potential problem may also involve inclusion of a second high-risk group among abstainers, one composed of persons reporting drinking less than monthly. It was suggested that these “occasional” drinkers might include high-risk persons who reduced their intake because of illness (12).

Prior Kaiser-Permanente studies 

Our first report about alcohol and total mortality risk (13) utilized a nested case-control design with four matched groups of persons with different alcohol drinking categories. Persons in each group were individually matched to the 2015 persons who reported 6 or more drinks per day. The J-curve results were clear, as was the fact that lower risk of light drinkers was due to fewer CHD deaths. However, the data source was 1964–1968 health examinations, which failed to separate ex-drinkers from lifelong abstainers and lumped all light-moderate drinkers into the “2 or fewer per day” category.

To correct these potential sources of bias for future work, from 1978–1985, we collected more detailed alcohol intake information from 127,212 persons who voluntarily took health examinations at Northern California Kaiser Permanente facilities and who provided self-classified ethnicity (14). The alcohol queries created clear separation of lifelong abstainers from former drinkers and subdivided light-moderate drinkers into three categories. One of these was a large category of persons who reported less than one drink per month. The drinking category distributions of these subjects were as follows: lifelong abstainers = 15,272; ex-drinkers = 4,124; “occasional” (<1 drink per month) = 27 ,007; more than1 drink per month but less than 1 drink per day = 4,303; 1–2 drinks per day = 23,110; 3–5 drinks per day = 8,419; and more than 6 drinks per day = 1,977.

Two published reports of total mortality are based on these data 15, 16. Both used Cox proportional hazards models with 7 covariates and included separate ex-drinker and occasional drinker categories. With 4501 deaths through 1988 (15) and 16,431 deaths through 1998 (16), both analyses confirmed the J-curve demonstrated increased risk of death among ex-drinkers and showed neither increased nor decreased risk among occasional drinkers.

Updated mortality analysis 

Results and Specific Comments 

Covariate effects and sex differences 

Comparison of age-sex adjusted and fully adjusted models (Fig. 1, left panel) shows an increased risk for ex-drinkers in both, no increased risk for occasional drinkers in either, and a J-curve for light-moderate and heavier drinking. The fully adjusted model exhibits less increase in risk for ex-drinkers and heavy drinkers and more apparent benefit at light-moderate drinking. Stepwise models (not shown) indicated that the major factor involved was control for smoking.

View Large Image Download to PowerPoint

Figure 1 Comparison of total mortality, age-sex adjusted, and fully adjusted model (left panel); fully adjusted data stratified by sex (right panel).

Data stratified by sex (Fig. 1, right panel) show more apparent benefit of light drinking in women. Explanations are not evident, but could include differences in biological effects of alcohol or in drinking pattern.

Ethnic differences 

Data for whites, African Americans, and Asian Americans (Fig. 2) indicate that the apparent benefit at lighter drinking is greatest in white persons, but a statistically significant lower risk at light drinking is also present in African Americans. For Asian Americans, who have a much larger proportion of abstainers and a smaller proportion of deaths attributed to CHD, there is borderline reduced mortality risk at light drinking and a greater increased risk among ex-drinkers and heavy drinkers.

View Large Image Download to PowerPoint

Figure 2 Fully adjusted models for blacks and whites (left panel) and Asians (right panel).

Interval between baseline data and death 

All relations appear to become attenuated with passage of time (Fig. 3, left panel). This probably is due to a general reduction of alcohol intake in this population (18), resulting in less harm from heavy drinking and less benefit from light-moderate drinking.

View Large Image Download to PowerPoint

Figure 3 Fully adjusted model for interval between baseline data and death (left panel); fully adjusted model for cardiovascular (CV) and noncardiovascular (non-CV) deaths (right panel).

Cardiovascular and noncardiovascular deaths 

The previously observed marked disparity 15, 16 is again evident (Fig. 3, right panel). The cardiovascular curve is substantially due to CHD deaths. For cardiovascular mortality, even past drinkers and heavy drinkers do not have much increase in risk. This relative specificity argues against a confounded explanation for lower CHD risk among light-moderate drinkers.

Differences according to baseline age 

With few cardiovascular deaths, persons younger than 40 years old at baseline show no benefit at light intake and have significantly increased risks starting at a reported intake of 1 to 2 drinks per day (Fig. 4, left panel). In middle life (baseline age between 40 and 60 years) the J-curve appears. Among older persons (Fig. 4, right panel), the benefits are more evident and the risks of heavy drinking are not great, but they are less favorable than those of light-moderate intake.

View Large Image Download to PowerPoint

Figure 4 Fully adjusted model for differences according to baseline age: Deaths before age 60 (left panel); deaths at or after age 60 years (right panel).

Differences according to likelihood of under-reporting 

Under-reporting of alcohol intake is widely believed to be present (19), but it is virtually impossible to measure and control for. The presumed effect is to systematically reclassify some true “heavy” drinkers as apparent “light-moderate” drinkers. Several likely effects on outcomes would be a lowered apparent threshold for harm, creation of an apparent continuous relation in place of a true threshold relation, and lessening of the apparent benefit of light-moderate drinking.

We tried to identify a group of persons among moderate drinkers that was more likely to be under-reporters, inferred from computer-stored information about lifetime responses to queries about alcohol and about alcohol-related diagnoses. This subset included persons who, on another occasion, indicated intake of three or more drinks per day or who ever had a diagnosis of an alcohol-related condition; a detailed description has been published (19). Twenty-seven percent of persons reporting 1 to 2 drinks per day fell into this more likely under-reporter group. These persons were about twice as likely to have high aspartate aminotransferase levels (19), a fact indirectly pointing to probable heavy drinking. The likely under-reporting group accounted for much of the relation of moderate drinking to increased prevalence of systemic hypertension (19).

Fig. 5 shows mortality among non–heavy drinkers separately for those in the suspected under-reporter group and those not in that group. Reduced mortality risk is concentrated in those unlikely to be under-reporters. The suspected under-reporters show little if any apparent mortality benefit.

View Large Image Download to PowerPoint

Figure 5 Fully adjusted model for mortality among non-heavy drinkers, separately for those in the suspected under-reporter group and those not in that group.

Conclusions 

This new analysis reconfirms that the relation of alcohol drinking to total mortality is J-shaped, with increased risk for persons reporting more than 3 drinks per day and reduced risk for lighter drinkers. The reduced risk of lighter drinkers is due almost entirely to lower risk of death from cardiovascular disease. Former drinkers are at increased risk, especially for noncardiac death. Infrequent (occasional) drinkers have risk similar to that of lifelong abstainers. There are sex and ethnic differences in the magnitude of the alcohol-associated risk, although the general shape of the relation is similar.

Age differences are substantial, with greater increased alcohol-associated risks among younger persons. Apparent benefit from light-moderate drinking is not seen before middle life. Among confounders, control for associated smoking is of special importance. The apparent magnitude of benefit at lighter drinking is probably reduced by systematic under-reporting.