01 45-6008/00/2401-0072$03 00/0 ALCOHOL ISM CLINICAL AND EXPERIMFNI AI RESEARCII

Vol. 24, No. 1 January 2000

Alcohol Consumption, Alcohol Dependence, and All-Cause Mortality

Deborah A. Dawson

Background: This study examined the effects of alcohol consumption and DSM-IV alcohol dependence on the risk of mortality.

Methods: Data from the 1988 National Health Interview Survey Alcohol Supplement were matched to the National Death Index for the years 1988 to 1995 (baseline n = 37,682 U.S. adults age 2 2 5 linked to 3,586 deaths). All mortality analyses were based on proportional hazards models that adjusted for age, sex, raceiethnicity, marital status, education, income, labor force status, body mass index, smoking status, and poor health indicators at baseline.

Results: When dependence was not considered and all past-year abstainers were used as the reference group, both light and moderate drinkers exhibited a reduced risk of mortality, with hazards ratios of 0.76 (0.68-0.84) and 0.84 (0.74-0.96). Heavy drinkers had about the same risk of dying as did past-year abstainers, and very heavy drinkers had an increased risk that was not significant (OR = 1.17, CI = 0.93-1.47). When lifetime abstainers were used as the reference category, the protective effect of moderate drinking fell short of significance, and there were nearly significant increased risks among former drinkers and vety heavy drinkers. When dependence was considered, light and moderate drinkers without depen- dencc had a reduced mortality risk regardless of reference group, and there was no significant effect among heavy or very heavy drinkers without dependence. Among dependent drinkers, there was no protective effect of light or moderate drinking, and very heavy drinkers had a significantly increased risk (OR = 1.56 relative to past-year abstainers and 1.65 relative to lifetime abstainers).

Conclusions: Because alcohol dependence nullifies the protective effect of light and moderate drinking, it is important to understand its role as an independent risk factor for mortality. Differences between dependent and nondependent drinkers who drank comparable amounts suggest that this risk may result from longer and heavier drinking histories before baseline, more severe health problems at baseline, more heavy episodic drinking, and, possibly, differences in beverage preference.

Key Words: Mortality, Consumption, Dependence, Alcoholism.

H E ASSOCIATION BETWEEN alcohol consumption T and all-cause mortality has been studied extensively. Although the findings of past research have been inconsis- tent, the bulk of studies have found a J-shaped curve (see reviews in Anderson 1994; Poikolainen, 1995.) That is, relative to abstainers, light to moderate drinkers have dem- onstrated a reduced risk of mortality whereas heavy drink- ers have demonstrated an increased risk of mortality. Re- cent literature supporting a J-shaped curve between alcohol consumption and all-cause mortality includes studies con- ducted in a number of countries. Yuan et al. (1997), who followed 18,244 Chinese men age 45 to 64 years for 6 years, found a significant reduction in mortality at a consumption level of 1 to 14 drinks per week, a nonsignificant reduction

From the Division of Biometry and Epidemiology, National Institute on Alcohol Abuse & Alcoholism, National Institutes of Health, Bethesda, Maryland.

Received for publication March 31, 1999; accepted October 20, 1999. Reprint requests: Deborah A. Dawson, Ph. D., NIAAAIDBE, W i k o Bldg.,

Suite 514, 6000 Executive Blvd. MSC 7003, Bethesda, MD 20892-7003; Fax: 301-443-8614; E-mail: ddawson@willco.niaaa.nih.gov

Copyright 0 2000 by the Research SocieQ on Alcoholism.

72

at 15 to 28 drinks per week, and an increase in risk at more than 28 drinks per week (all relative to lifetime abstainers). In the Copenhagen city heart study, which followed 13,285 Danish men and women between the ages of 30 and 79 for 10 to 12 years, GronbEk et al. (1994) found that individuals consuming 1 to 6 drinks per week had the lowest mortality levels, with significantly increased risks among baseline abstainers and those consuming more than 42 drinks per week. A sample of 12,321 British male physicians who were followed for 13 years indicated that the lowest mortality rates occurred among those consuming 8 to 14 standard drink units per week and that mortality among those con- suming 1 to 14 units per week was significantly lower than mortality among baseline abstainers (Doll et al., 1994). A French sample of 34,014 middle-aged men followed for an average of 9.3 years similarly reported the lowest mortality rates among those consuming 22 to 54 g of ethanol (roughly 1-2.5 drinks) per day (Renaud et al., 1998).

Other recent studies have reported nonsignificant J-shaped mortality curves. In a registry-based sample of 9,057 Swedish adult twins followed from 1975 to 1992 (Andreasson and Brandt, 1997), the lowest mortality rates

Alcohol Clin Exp Re,, Vol 24, No 1, 2000. pp 72-81

ALCOHOL CONSUMPTION, DEPENDENCE, AND ALL-CAUSE MORTALITY 73

were among those consuming 1 to 250 g (1-11 drinks) per month; however, the increased risk of death among abstain- ers (OR = 1.2) was not significant. Two other recent small prospective studies have reported similar findings of a non- significant J-shaped curve (Cullen et al., 1993; Keil et al., 1997), and a Dutch study that examined frequency but not quantity of drinking reported a nonsignificant reduction in risk among occasional, frequent, and daily drinkers com- pared to nondrinkers (Berberian et al., 1994). Studies such as these neither support nor contradict the J-shaped curve at the individual study level; however, they contribute to the significant J-shaped curve that has resulted when data are combined across surveys in meta-analyses (Duffy, 1995: Holman et al., 1996). When Holman et al. reanalyzed 16 studies of mostly European populations, they found that relative to abstainers, men drinking

74 DAWSON

Linkage with death records was performed by the National Center for Health Statistics, based on identifiers collected at the time of the NHIS interview, such as the respondents social security number (SSN), date of birth, and so forth. These identifiers were combined in various ways to produce 12 alternative criteria for selecting a possible match with a death record, for example, (1) SSN plus first name both match, (2) SSN plus last name both match, (3) month and year of birth and first and last names all match, and so forth. All cases that met even 1 of the 12 criteria for a match were assigned to one of five classes, ordered by the certainty of the match. In the first class, where there were matches on SSN, first, middle, and last names, sex, state of birth, and month and year of birth (i.e., on all matching data), every match was considered to be a true match, that is, an actual death. In the fifth class, where the SSN was known but did not match, all matches were considered to be false matches. In the intervening three classes, a score was devised to represent the probability of a true match. This score was a weighted function of the 12 matching criteria, with positive weights used for criteria that were met and negative weights for criteria that were not met. A cutoff level was then selected within each class to maximize the proportion of records correctly classified and min- imize the number of records incorrectly classified. Using two large na- tional population surveys where longitudinal follow-up provided an indc- pendent ascertainment of vital status (i.e., of whether the respondent was dead or alive at the end of the follow-up period), this matching procedure yielded accurate matches for 94% of female respondents and 97% of male respondents (National Center for Health Statistics, 1997).

When the NHIS sample was matched with death records for 1988 to 1995, approximately one third of all potential matches fell below the required cutoff and were treated as nondeaths. Cases with insufficient linkage data were removed from the sample, leaving a total of 42,910 cases with linked alcohol, smoking, and mortality data. These cases were re- weighted to match the original 1988 NHIS poststratification estimates of the adult population by age, sex, and race (Massey et al., 1989).

This analysis was restricted to the 37,682 cases where the respondents were 25 years of age or older at baseline. Younger respondents, many of whom were below the legal drinking age when originally interviewed, were excluded from the analysis to minimize the possibility that baseline drink- ing habits would differ substantially from those over the follow-up period and to facilitate comparisons with other surveys of adult samples. The total number of deaths among the analytical sample was 3,586, or 9.5% of all cases. This corresponds to a weighted mortality estimate of 8.3%, with the differential reflecting the oversampling of Blacks, whose mortality rates were higher than those of non-Blacks.

Measures

The 1988 NHIS collected detailed information on alcohol intake from individuals who were classified as past-year drinkers. These were individ- uals who in screening reported having had at least 12 alcohol drinks in the year preceding interview. Other categories of drinkers defined by the screening questions were lifetime abstainers (drank less than 12 drinks in their lives), lifetime infrequent drinkers (drank 12 or more drinks in their lives but never drank 12 or more drinks in any one year), and former drinkers (drank 12 or more drinks in at least 1 year, but not the year preceding interview). In this analysis, lifetime abstainers were assumed to have consumed no drinks in the year preceding interview. Lifetime infre- quent drinkers and former drinkers were asked how many drinks they had consumed in the preceding year (range = 0-11). This number was con- verted to an average daily ethanol intake by assuming a standard drink size of 0.54 oz of ethanol (the mean drink size reported by this sample) and dividing by 365.

Past-year drinkers were asked how often they usually drank (no time reference period specified, recorded in exact number of days per week, month, or year) and how many drinks they usually consumed on those days. They also were asked whether they had consumed any alcohol in the 2 weeks preceding the interview and the date of their last drink. For a 2 week reference period (either the 2 weeks preceding interview or the most recent 2 weeks in which a drink was consumed), they were asked number

of drinking days, usual quantity of drinks per drinking day, and usual drink size in separate series of questions for beer, wine, and liquor. Based on these data, usual volume of intake was ascertained in one of three ways:

(1) For individuals who provided detailed 2 week data and whose lasr drink occurred during the 2 wceks immediately preceding the interview (80.8% of all past-year drinkers), annual volume of ethanol intake was calculated as the product of annualized frequency (26 times the 2 week frequency) times quantity of drinks per drinking day times drink size in ounces times an ethanol conversion factor of 0.045 for beer, 0.121 for wine, and 0.409 for liquor (DISCUS, 1985; Kling, 1989; Modern Brewery Age, 19Y2; Turner, 1990; Williams et al., 1993). These volumes were summed across beverage type and divided by 365 to yield average daily ethanol intake. Because interviewing was conducted continuously throughout the year, the estimation of annual intake from a 2 week reference period should not entail any seasonal bias.

(2) For the 1.5% of past-year drinkers who had missing data for the 2 week reference period but did provide usual frequency and quantity of drinks (no time reference period specified), average daily ethanol intake was estimated as the product of this annualized frequency and quantity (assuming 0.54 oz of ethanol per drink), divided by 365.

(3) For the 17.7% of past-year drinkers who provided detailed 2 week data but whose most recent drink was not consumed during the 2 weeks immediately preceding the interview, multiplying the 2 week consumption levels times 26 would have overestimated of annual frequency of drinking. Moreover, the date of last drink was missing for more than two thirds of these individuals. After comparing the results of various ways to estimate annual intake volume for these drinkers, I decided to multiply their 2 week consumption times 13 to obtain annual frequency, which was multiplied, as in (I), by number of drinks, drink size, and the appropriate ethanol conversion factor and then summed across beverages. In essence, these individuals were assumed to drink half as frequently as those whose last drink had been consumed in the 2 weeks preceding the interview. This resulted in a volume of at least 13 drinks in the past year, which ensured that they met the threshold of >12 drinks required to be defined as a past-year drinker. When this approach was compared to the alternative of constructing an estimate based on usual frequency and quantity of drink- ing, as in method 2, the cases were evenly divided in terms of which approach yielded the highest volume estimate. The median value of the ratio of the estimates yielded by these two approaches was 1.01.

Past-year drinkers also were asked the past-year frequencies of con- suming 2 5 and 2 9 drinks on a single day (with the difference between the two representing the frequency of drinking 5-8 drinks). If one of these two frequencies was missing, it was imputed on the basis of the other non- missing frequency. If both were missing, they were imputed on the basis of average daily ethanol intake. Consumption on these heavy drinking days was incorporated into an adjusted estimate of average daily ethanol intake by replacing the unadjusted average daily intake with an intake of 3.24 oz (6 standard drinks) on days of consuming 5 to 8 drinks and an intake of 6.48 oz (12 standard drinks) on days of consuming 2 9 drinks and then recalculating the daily average. (If this reduced average daily intake, the original value was retained.) This adjusted average daily ethanol intake formed the basis for the categories of past-year consumption that were used in this analysis:

Both current and former drinkers were asked frequency of drinking and usual quantity of drinks consumed during their period of heaviest drink- ing. Again assuming a standard drink size of 0.54 oz ethanol, these variables were multiplied together to estimate average daily ethanol intake during the period of heaviest consumption. This intake was categorized by using the same categories and definitions as listed previously for past-year consumption, to describe the drinking categories of former drinkers.

Alcohol dependence was classified in accordance with the DSM-IV criteria (American Psychiatric Association, 1994) and was measured using a set of 17 symptom item indicators taken from a larger list of 41 alcohol-related problems. (Although the list of indicators was designed before the publication of the DSM-IV criteria, it contained items that were sufficient to generate DSM-IV diagnoses.) To be classified with alcohol dependence, an individual had to meet three or more of the seven

ALCOHOL CONSUMPTION, DEPENDENCE, AND ALL-CAUSE MORTALITY 75

Table 1. Number and Characteristics of Adults Age 25 and Older, By Drinking Status at Baseline % With alcohol % Dying

in follow- Number of % Distribution Average daily ethanol dependence in cases at of cases at intake during year year preceding U P baseline baseline preceding baseline baseline interval

Total Past-year abstainers

Lifetime abstainers Lifetime infrequent drinkers Former drinkersa

Light/moderateb Heavyhery heavyb

Past-year drinkers Lightd Moderated Heavy* Very heavyd

37,682 19,102 7,148 4,579 7,375 4,999 1,335

18,580 9,557 6,422 1,402 1,086

100.0 (0.0) 49.5 (0.4) 17.8 (0.3) 11.8 (0.2) 19.9 (0.3) 15.5 (0.2) 4.4 (0.1)

50.5 (0.4) 26.0 (0.3) 17.6 (0.3)

3.0 (0.1) 3.9 (0.1)

0.32 (0.1)

76 DAWSON

Table 2. Adjusteda Mortality Hazard Rate Ratios (HRRs) for Adults Age 25 and Older by Drinking Status at Baseline

Model 1 Model 2

HRR 95% CI HRR 95% CI Past year abstainers 1 .OO (Reference) -

Lifetime abstainers - 1 .OO (Reference)

Former drinkers Lifetime infrequent drinkers - 0.93 (0.82 -1.06)

Lightlmoderate - 1.09 (0.97 -1.24) Heavyhery heavyb - 1 .I 8 (0.99 -1.42)

Past-year drinkers Light 0.76 (0.68 -0.84) 0.78 (0.69 -0.89) ModerateC 0.84 (0.74 -0.96) 0.88 (0.76 -1.01) HeavyC 0.94 (0.73 -1.19) 0.98 (0.76 -1.26) Very heavyC 1 .I7 (0.93 -1.47) 1.23 (0.96 -1.56)

a Adjusted for age, sex, racekthnicity, marital status, education, income, labor force participation, body mass index, and smoking status.

Based on intake during period of heaviest drinking. Based on intake during year preceding baseline interview.

Table 3. Adjusted Mortality Hazard Rate Ratios (HRRs) for Adults Age 25 and Older By Drinking Status and DSM-IV Alcohol Dependence at Baseline

Model 1 Model 2

HRR 95% CI HRR 95% CI Past year abstainers 1 .OO (Reference) -

Lifetime abstainers - 1 .OO (Reference)

Former drinkers Lifetime infrequent drinkers - 0.93 (0.82 -1.06)

Light/moderateb - 1.09 (0.97 -1.24) Heavyhery heavyb - 1 .I 8 (0.99 -1.42)

Past-year drinkers Not dependent

LightC 0.75 (0.67 -0.84) 0.78 (0.69 -0.88) Moderate 0.82 (0.72 -0.94) 0.86 (0.79 -0.93) Heavy 0.88 (0.69 -1.12) 0.92 (0.69 -1.23) Very heavy 1.02 (0.80 -1.30) 1.07 (0.98 -1 .I 7)

LightC 1.15 (0.87 -7.52) 1.20 (0.89 -1.61) Moderate I .26 (0.95 -1.67) 1.32 (1.01 -1.72) HeavyC 1.35 (0.75 -1.92) 1.42 (0.96 -2.10) Very heavyC 1.56 (1.17 -2.08) 1.65 (1.38 -1.98)

Dependent

negative for dependence during that time period. Among past-year drinkers, the proportions classified with depen- dence rose from 2.4% of light drinkers to 42.4% of very heavy drinkers.

The proportions of individuals dying between 1988 and 1995 were highest for past-year abstainers (11.3%). Among these, former drinkers were the most likely to have died (12.2%), especially those who had been heavy or very heavy drinkers (13.5%), whereas lifetime infrequent drinkers were the least likely to have died (9.4%). Past-year very heavy drinkers also were among those with the highest mortality levels, with 10.0% having died. There were few significant differences in the proportions dying among these groups. However, all exhibited higher levels of mor- tality than did light to heavy past-year drinkers. Among these, the proportions dying varied from 4.8% of the light drinkers to 6.3% of the heavy drinkers. Thus, the crude mortality data revealed a U-shaped pattern with respect to volume of ethanol intake at baseline.

Table 2 presents mortality hazard rate ratios adjusted for demographic and socioeconomic characteristics, body mass index, smoking history, and poor health at baseline. The first model used all past-year abstainers (i.e., lifetime ab- stainers, lifetime infrequent drinkers and former drinkers) as the reference category against which the mortality risks of past-year drinkers were measured. The results of this model indicated a protective effect of both light drinking (HRR = 0.76, CI = 0.68-0.84) and moderate drinking (HRR = 0.84, CI = 0.74-0.96). Heavy drinking did not significantly affect mortality (HRR = 0.94, CI = 0.73- 1.19). The effect of very heavy drinking suggested a slight increased risk, but it was not statistically significant (HRR = 1.17, CI = 0.93-1.47). Thus, even after adjust- ment for a variety of potential confounders, the adjusted mortality figures continued to exhibit a U-shaped pattern, with a suggestion of a J-shaped increase in risk at the heaviest intake levels.

As indicated in model 2, switching to lifetime abstainers as the reference group yielded few significant changes in

a Adjusted for age, sex, race/ethnicity, marital status, education. income, labor force participation, body mass index, and smoking status.

Based on intake during period of heaviest drinking. Based on intake during year preceding baseline interview.

these findings. The most important change was that the protective effect of moderate drinking fell just short of significance (HRR = 0.88, CI = 0.76-1.01) when com- pared to lifetime abstention. In addition, former drinkers demonstrated an increased risk that fell short of signifi- cance but showed a clear dose response, varying from 1.09 for former light or moderate drinkers to 1.18 for former heavy or very heavy drinkers. When all former drinkers were combined (data not shown), they had a significantly increased risk of mortality relative to lifetime abstainers (HRR = 1.21, CI = 1.09-1.35). (This hazard rate ratio exceeds that for either of the former drinking groups shown in Table 2 because of the exceptionally high mortality rates among former drinkers whose level of intake during their period of heaviest drinking was unknown.)

To examine how alcohol dependence would affect the mortality risks associated with different consumption levels (Table 3), a dichotomous measure of past-year dependence was added to the proportional hazards models. In both models, dependence exerted a significant positive effect on the risk of mortality (HRR = 1.53 in model 1 and 1.54 in model 2, data not shown). Rather than create separate categorical measures for dependent and nondependent drinkers at each volume level, I constructed models to test for interactions between dependence and volume of intake. Because tests revealed no significant interactions between the categorical consumption measures and dependence, I calculated hazard rate ratios for dependent individuals by exponentiating the sum of the parameters for their respec- tive consumption levels and dependence; those for nonde- pendent individuals were calculated by exponentiating the parameters for their consumption levels alone. This ap- proach resulted in fewer model parameters and smaller confidence intervals than would have been obtained with separate categorical measures of volume for dependent and

ALCOHOL CONSUMPTION, DEPENDENCE, AND ALL-CAUSE MORTALITY 77

Table 4. Characteristics of Past-Year Drinkers Age 25 and Older, by Drinking Status and DSM-IV Alcohol Dependence at Baseline: Standardized for Age and Sex

Light drinkers Moderate drinkers Heavy drinkers Very heavy drinkers

Dependent Nondependent Dependent Nondependent Dependent Nondependent Dependent Nondependent ~~

Past year consumption Average daily intake Frequency of drinking Quantity per drinking day Days drank 2 5 drinks Days drank 2 9 drinks % Intake in beer % Intake in wine % Intake in liquor Period of heaviest drinkinq Average daily intake Frequency of drinking Quantity per drinking day Age at first drink Past-year health indicators Bed days % Hospitalized % In poor health % With major limitations % With preexisting conditions

0.13 (

78 DAWSON

drinking was greater only among moderate and heavy drinkers, their quantity of drinks per drinking day was consistently higher across volume categories, as were their frequencies of drinking five or more and nine or more drinks.

Considering the period of heaviest drinking, dependent drinkers reported volumes of consumption that were 1.6 to 2.6 times greater than those of their nondependent coun- terparts. Again, this difference was driven more by differ- entials in the quantity consumed per drinking day than in the overall frequency of drinking. Dependent drinkers also started drinking at a younger age, meaning that they had experienced a slightly longer exposure to ethanol at any given age.

In terms of health, dependent drinkers reported more days spent in bed due to poor health and a higher preva- lence of hospitalization in the year preceding the NHIS interview. The measure of baseline poor health that was used as a control in this analysis was based on three com- ponents: reporting health as fair or poor (as opposed to good, very good, or excellent), reporting major limitation of activity, and/or reporting any of six preexisting medical conditions. Among heavy and very heavy drinkers, those with dependence were more likely than those without de- pendence to report being in poor health or to report pre- existing conditions. Among light to heavy drinkers, those with dependence were more likely to report major limita- tion of activities. Thus, within the control for poor health that was used in this analysis, dependent drinkers seem more likely than nondependent drinkers to have experi- enced multiple and more serious indicators.

DISCUSSION

This analysis of prospective data from a sample of U.S. adults 25 years of age and older at baseline indicated that alcohol consumption and alcohol dependence exerted sig- nificant and independent effects on the risk of all-cause mortality over a 7.5 year follow-up period. Considered separately, the effect of consumption was a U-shaped curve, with a near-significant increase in mortality among the heaviest drinkers, and the effect of dependence was positive and independent of volume of consumption. To- gether, these findings revealed that among nondependent drinkers, light and moderate drinking was protective and heavier drinking did not significantly affect the risk of dying; among dependent drinkers, the effect of alcohol consumption was never protective and often increased the risk of death. These findings indicate the importance of considering dependence in addition to consumption when interpreting the implications of drinking for all-cause mor- tality. Although the possibility of greater underreporting of consumption among dependent drinkers cannot be ruled out, this studys comparison of dependent and nondepen- dent drinkers suggested several alternative reasons for the positive mortality risk associated with alcohol dependence.

First, the comparison of dependent and nondependent drinkers suggested that the frequency of heavy episodic drinking may be positively associated with all-cause mor- tality. This would be expected for mortality from external causes such as injuries, but this study, in which external causes contributed only 5% of all deaths during the obser- vation period, suggests that heavy drinking occasions may increase the risk of other types of death as well. This finding is supported by the recent meta-analyses conducted by Leino et al. (1998) and Fillmore et al. (1998), which found higher all-cause mortality hazard rate ratios associated with heavy quantity of intake per drinking day than with high frequency of drinking. Similarly, a recent analysis by Kau- hanen et al. (1997) found that men whose usual intake was six or more beers per drinking day had a higher risk of all-cause mortality than those whose usual intake was three or fewer beers, even after adjusting for the higher overall volume of intake of the former. Heavy episodic drinking is not limited to dependent drinkers, and this aspect of drink- ing pattern should be investigated in conjunction with vol- ume of intake for all drinkers.

Second, the comparison of dependent and nondepen- dent drinkers suggested that drinking at baseline may not fully capture the harm or benefits conferred by earlier patterns of drinking. When available, drinking history (to at least distinguish former drinkers from lifetime ab- stainers) and measures of lifetime or heaviest intake and lifetime dependence might explain mortality risks that seem contradictory in light of baseline consumption pat- terns. In this study, former drinkers exhibited a mortality risk that increased with their level of heaviest intake and that fell just short of being significantly higher than that for lifetime abstainers. In this sense, their risk curve paralleled that for dependent drinkers, suggesting that prior dependence may have been a factor in many former drinkers decisions to stop drinking. (In a prospective study of U.S. medical patients between the ages of 55 and 65, Mertens et al., 1996, found that fully two thirds of baseline abstainers had former alcohol problems.) Studying both baseline and prior consumption levels (i.e., drinking trajectories) and the incidence and remis- sion of dependence among all drinkers, not just former drinkers, might provide additional insights into how drinking and mortality are linked.

In a related point, it has been argued (e.g., Shaper, 1990) that the inclusion of former drinkers in the reference cat- egory of baseline abstainers might bias findings because of the possibility of drinking cessation having been linked to health problems associated with the risk of dying. Using the drinking history obtained in the NHIS, I was able to deter- mine that the distinction between baseline and lifetime abstinence as the reference group against which consump- tion levels were assessed was of limited importance in this U.S. sample of adults 25 years of age and older. It had only a small effect on the magnitudes of the hazard rate ratios for different consumption levels but did make a difference

ALCOHOL CONSUMPTION, DEPENDENCE, AND ALL-CAUSE MORTALITY 19

in whether some of the ratios with borderlinep values could be considered significant.

It is possible that with a larger sample or a longer follow-up period, the choice of reference group might make the difference between a U-shaped and J-shaped curve for all-cause mortality; that is, it might affect whether the increased risk among the heaviest drinking group(s) is sig- nificant or not. Given the near-significant increased risk of mortality among former drinkers compared to lifetime ab- stainers, the choice of reference group also might be more important in populations where former drinkers make up a larger proportion of baseline abstainers, that is, where lifetime abstinence is less common. This is by no means certain, though, because the probability of lifetime abstain- ers being selected with respect to poor health, social isola- tion, or other mortality risk factors is likely to be greater when abstinence is less common.

A third point suggested by the comparison of dependent and nondependent drinkers has to do with baseline health indicators. Although this studys control for baseline poor health was intended to rule out this factor as a confounder of the effect of dependence, there were many differences in severity not captured by the baseline measure used. It is likely that even more differences would have been docu- mented if the NHIS had collected information on comorbid psychiatric and drug use disorders that are known to be disproportionately prevalent among individuals with alco- hol dependence (Grant and Harford, 1995; Kessler et al., 1994; Merikangas and Gelernter, 1990; Regier et al., 1990) or on illness-related stressors and associated coping re- sponses (Mertens et al., 1996). Although this study found little change in the effects of consumption and dependence when baseline health was excluded as a control, the impact of including versus excluding it might have been greater had a more sensitive health measure or measures been used.

In any event, the decision about whether to adjust mor- tality hazards estimates for baseline health should not be made on empirical grounds but rests on the extent to which baseline health problems are considered to have resulted from prior alcohol consumption. The best possible solution would be to measure both the direct effects of consumption at baseline and the indirect effects of prior consumption through their effect on baseline health. Thus baseline con- ditions not attributable to prior consumption would be retained as controls, and those resulting from prior con- sumption would be treated as intervening variables., Even this optimal approach fails to capture changes in drinking that may occur between the baseline measure and the end of the observation period, illustrating the limits of estimat- ing the mortality risks associated with alcohol consumption from data taken at a single time point.

Finally, the comparison of dependent and nondependent drinkers revealed that at light and moderate consumption levels, dependent drinkers consumed significantly less of their overall ethanol intake in the form of wine. In view of

studies that have found a protective effect of light or mod- erate wine consumption (a finding that is by no means consistent across studies; see White, 1996), this suggests that beverage preference might be another factor associ- ated with dependence being a positive risk factor for mor- tality.

The ranges of ethanol intake at which this study found protective and risky effects with respect to all-cause mortality were in line with those reported in other stud- ies. For example, both the meta-analysis by Holman et al. (1996) and the large-scale American Cancer Society pro- spective study (Thun et al., 1997) found that the increase in mortality risk did not begin until an average of four or more drinks per day were consumed, the same level at which this study found the first significant and near- significant increased risks. However, in both the studies by Thun et al. and Holman et al., this finding was re- stricted to specific subpopulations, individuals age 30 to 59 with low cardiovascular risk in the former case and men in the latter case.

Holman et al. found a lower threshold for increased risk among women (with a significantly increased risk at intake levels of more than drinks per day) that was not supported in this study. Although I did not a priori stratify the sample by age and sex, I tested for and failed to detect any significant interaction between sex and the volume of intake consumption levels. (The only near- significant interaction with sex suggested a less strongly protective effect of light drinking among male compared to female drinkers, and no impact on the risk at the heavy drinking level of approximately two to four drinks per day that distinguished men and women in the study by Holman et al.) Additional research is needed to clar- ify whether the risky volume thresholds do indeed differ for men and women. Gender differentials in the selec- tivity of drinking and in sensitivity to reporting baseline health conditions need to be examined. It is also impor- tant to adjust volume measures for atypical heavy drink- ing, as was done in this study, and to test for the signif- icance of any differences in threshold rather than assuming that a stratified analysis is justified.

These findings may not be generalizable to popula- tions other than that used for analysis, which consisted of U.S. adults 25 years of age and over. All-cause mortality curves reflect the curves for the most common causes of death, which vary substantially by age and among cul- tures with different diets and social practices. For the study sample, coronary heart disease was the primary cause of death, and its J-shaped association with alcohol consumption has been widely documented. Populations where mortality is predominantly associated with infec- tious diseases or external causes such as accidents and suicide-and these populations include U.S. adolescents and young adults as well as the populations of many developing countries-cannot be assumed to experience

80 DAWSON

alcohol-related mortality risks comparable to those de- scribed in this study.

REFERENCES American Psychiatric Association (1994) Diagnostic and Statistical Manual

of Mental Disorders. 4th ed. American Psychiatric Association, Wash- ington, DC.

Andcrson P (1994) Are there net health benefits from moderate drinking? All-cause mortality. Contemp Drug Probl 21:143-161.

Andreasson S, Brandt L (1997) Mortality and morbidity related to alcohol.

Berberian KM, van Dujin CM, Hoes AW, Valkenburg HA, Hofman A (1994) Alcohol and mortality: Results from the EPOZ follow-up study. Eur J Epidemiol 10:587-593.

Boffetta P, Garfinkel L (1990) Alcohol drinking and mortality among men enrolled in an American Cancer Society prospective study. Epidemiol-

Bunn JY, Booth BM, Cook CA, Blow FC, Fortney JC (1994) Relationship between mortality and intensity of inpatient alcoholism treatment. Am J Public Health 84:211-214.

Callahan CM, Tierney WM (1995) Health services use and mortality among older primary care patients with alcoholism. J Am Geriatr Soc

Camargo CA, Hennekens CH, Gaziano JM, Glynn RJ, Manson JE, Stampfer MJ (1997) Prospectivc study of moderate alcohol consump- tion and mortality in US male physicians. Arch Intern Med 157:79-85.

Chyou P-H, Burchfiel CM, Yano K, Sharp DS, Rodriguez BL, Curb JD, Nomura AMY (1997) Obesity, alcohol consumption, smoking and mor- tality. Ann Epidemiol 7:311-317.

Cullen KJ, Knuiman MW, Ward NJ (1993) Alcohol and mortality in Brusselton, Western Australia. Am J Epidemiol 137:242-248.

Davis MA, Ncuhaus JM, Moritz DJ, Lein D, Barclay JD, Murphy SP (1994) Health behaviors and survival among middle-aged and older men and women in the NHANES I epidemiologic follow-up study. Prev Med 23:369-376.

Deleyiannis FWB, Thomas DB, Vaughan TJ, Davis S (1996) Alcoholism: Independent predictor of survival in patients with head and neck cancer. J Natl Cancer Inst 88542-549.

DISCUS (Distilled Spirits Council of the United States) (1985) Annual Statistical Review 19841198.5. Distilled Spirits Industry, Washington, DC.

Doll R, Peto R, Hall E, Wheatley K, Gray R (1994) Mortality in relation to consumption of alcohol: 13 years observations on male British doctors. Br Med J 309:911-918.

Duffy JC (1995) Alcohol consumption and all-cause mortality. Int J Epidemiol 24:lOO-105.

Fillmore KM, Golding JM, Graves KL, Kniep S, Leino EV, Romelsjo A, Shoemaker C, Ager CR, Allebeck P, Ferrer H (1998) Alcohol consump- tion and mortality: 111. Studies of female populations. Addiction 93: 219 -229.

Finney JW, Moos RH (1991) Long-term course of alcoholism: I. Mortal- ity, relapse and remission rates and comparisons with community con- trols. J Stud Alcohol 53:44-54.

Gordon T, Doyle J (1987) Drinking and mortality. Am J Epidemiol 125:263-270.

Gordon T, Kannel WB (1983) Drinking habits and cardiovascular disease: The Framingham Study. Am Heart J 105:667-673.

Grant BF, Harford TC (1995) Comorbidity between DSM-IV alcohol use disorders and major depression: Results of a national survey. Drug Alcohol Depend 39:197-206.

Gronhaek M, Deis A, Sorensen TIA, Becker U, Borch-Johnsen K, Miiller C, Schnohr P, Jensen G (1994) Influence of sex, age, body mass index, and smoking on alcohol intake and mortality. Br Med J 308:302-306.

Hampl K, Hajin J (1992) Mortality of alcohol abusers, in 36th International Congress on Alcohol and Drug Dependence, vol 2; pp 1146-1 152, Inter- national Congress on Alcohol and Drug Dependence, Glasgow, Scot- land.

Alcohol Alcohol 32:173-178.

OW 1~347-348.

43: 1378-1 383.

Holman CD, English DR, Milne E, Winter MG (1996) Meta-analysis of alcohol and all-cause mortality: A validation of NHMRC recommen- dations. Med J Aust 164:141-145.

Kagan A, Yano K, Rhodcs G, McGee DL (1981) Alcohol and cardiovas- cular disease: The Hawaiian experience. Circulation 64(Suppl 111):27- 31.

Kauhanen J, Kaplan GA, Goldberg DE, Salonen JT (1997) Beer binging and mortality: Results from the Kuopio ischemic heart disease risk factor study, a prospective population based study. Br Med J 3152346- 851.

Keil U, Chambless LE, Doring A, Filipiak B, Sticber J (1997) The relation of alcohol intake to coronary heart disease and all-cause mortality in a beer-drinking population. Epidemiology 8:150-156.

Kessler RC, McGonagle KA, Zhao S, Nelson CB, Hughes M, Eshlenian S, Wittchen H-U, Kendler K (1994) Lifetime and 12-month prevalence of DSM-111-R psychiatric disorders in the United States: Results from the National Comorbidity Survey. Arch Gen Psychiatry 51:s-19.

Klatsky AL, Armstrong MA, Friedman GD (1 992) Alcohol and mortality. Ann Intern Med 117:646-654.

Kling W (1989) Measurement of ethanol consumed in distilled spirits.

Langle G, Mann K, Mundle G, Schied HW (1993) Ten years after: The post-treatment course of alcoholism. Eur Psychiatry 8:95-100.

Leino EV, Romelsjo A, Shoemaker C, Ager CR, Allebeck P, Fcrrer H, Fillmore KM, Golding JM, Graves KL, Kniep S (1998) Alcohol con- sumption and mortality: 11. Studies of male populations. Addiction 93205-21 8.

Lewis CE, Smith E, Kercher C, Spitznagel E (1995) Assessing gender interactions in the prediction of mortality in alcoholic men and women: A 20-year follow-up study. Alcohol Clin Exp Res 19:1162-1172.

Massey JT, Moore TF, Parsons V, Tadros W (1989) Design and estima- tion from the National Health Interview Survey, 1985-1994. National Center for Health Statistics, Vital Health Stat 2(110):1-33.

Merikangas KR, Gelernter CS (1990) Comorbidity for alcoholism and depression. Psychiatr Clin North Am 13:613-632.

Mertens JR, Moos RH, Brennan PL (1996) Alcohol consumption, life context, and coping predict mortality among late-middle aged drinkers and former drinkers. Alcohol Clin Exp Res 20:313-319.

Modern Brewery Age (1992) Beer analysis. Modern Brewery Age 43(37):

National Center for Health Statistics (1 997) National Health Interview Survey Multiple Cause of Death Public Use Data File. National Center for Health Statistics, Centers for Disease Control, Hyattsville, MD.

Poikolainen K (1995) Alcohol and mortality: A review. J Clin Epidemiol 48:455-465.

Regier DA, Farmer ME, Rae DS, Locke BZ, Keith SJ, Judd LL, Goodwin FK (1 990) Comorbidity of mental disorders with alcohol and other drug abuse. Results from the Epidemiologic Catchment Area (ECA) Study.

Rehm J, Sempos CT (1995) Alcohol consumption and all-cause mortality. Addiction 90:471-480.

Renaud SC, GuCguen R, Schenker J, dHoutaud A (1998) Alcohol and mortality in middle-aged men from Eastern France. Epidemiology

Romelsjo A, Karlsson G, Henningsohn L, Jakobsson SW (1993) Preva- lence of alcohol-related mortality in both sexes: Variation between indicators, Stockholm 1987. Am J Public Health 832338-844.

Rossow I, Amundsen A (1997) Alcohol ahuse and mortality: A 40-year prospective study of Norwegian conscripts. Soc Sci Med 44:261-267.

Serdula MY, Koong S-L, Williamson DF, Anda RF, Madans JH, Klein- man JC, Byers T (1995) Alcohol intake and subsequent mortality: Findings from the NHANES I follow-up study. J Stud Alcohol 56:233- 239.

Shah BV, Barnwell BG, Bieler GS (1997) SUDAAN Users Manual. Re- lease 7:5., Research Triangle Institute, Research Triangle Park, NC.

J Stud Alcohol 501456-460.

52-53.

J A M 26412511-2518.

9:184-188.

ALCOHOL CONSUMPTION, DEPENDENCE, AND ALL-CAUSE MORTALITY 81

Shaper AG (1990) Alcohol and mortality: A review of prospective studies. Br J Addict 85:837-847.

Spinatsch M (1992) Prediction of long-term outcome of male alcoholics after inpatient treatment: The case of a clinical population in German- speaking Switzerland. Int J Addict 27:1087-1103.

Thun MJ, Peto R, Lopez A, Monaco JH, Henley SJ, Heath CW, Doll R (1997) Alcohol consumption and mortality among middle-aged and elderly U.S. adults. N Engl J Med 337:1705-1714.

Turner C (1990) How much alcohol is in a standard drink? An analysis of 125 studies. Br J Addict 85:1171-1175.

Wells JE, Walker ND (1990) Mortality in a follow-up study of 616

alcoholics admitted to an inpaticnt alcoholism clinic 1972-1976. N Z Med J 103:l-3.

Yuan J-M, Ross RK, Gao Y-T, Henderson BE, Yu MC (1997) Follow up study of moderate alcohol intake and mortality among middle-age men in Shanghai, China. Br Med J 314:18-23.

White IR (1996) The cardioprotective effects of moderate alcohol con- sumption. Br Med J 312:1179-1180.

Williams GD, Clem D, Dufour M (1993) Apparent Per Capita Alcohol Consumption: National, State, and Regional Trends, 1977-1991. Alcohol Epidemiologic Data System (AEDS) Report No. 27. National Institute on Alcohol Abuse and Alcoholism, Rockville, MD.