Content area
To determine the influence of six determining variables on the shape of the risk curve between alcohol and all-cause mortality.
Based on a systematic search with clear inclusion criteria, all articles on alcohol and all-cause mortality until 2000 were included.
Precision-weighted pooling of relative risks (RRs); precision-weighted hierarchical analysis.
For pooling: RRs for different categories of average volume of drinking, lifetime abstainers and ex-drinkers. For hierarchical analysis: on first level: consumption in grams of pure alcohol per day; on second level: length of follow-up time in months; per capita consumption; average age, proportion of abstainers, average volume of drinking, and variability of average volume of drinking at baseline.
RR of former and current drinkers for all-cause mortality compared to abstainers.
The main hypotheses could be confirmed for males: Ex-drinkers had a higher mortality risk than lifetime abstainers; the higher and the more diverse the average volume of alcohol consumption, the wider the dip of the curve; the older the persons at baseline, the more pronounced the protective effect; and the longer the follow-up time, the less pronounced the protective effect. Except for average volume of drinking effects for females went in the same direction but with one exception did not reach significance.
There are systematic influences on the shape of the risk curve between alcohol and all-cause mortality. The overall beneficial effect of light to moderate drinking remained under all scenarios, indicating a high validity of the overall shape despite the heterogeneity between studies.
*u*n*s*t*u*r*t*u*r*e**t*e*x*t*
MORTALITY How stable is the risk curve between alcohol and all-cause mortality and what factors influence the shape? A precision-weighted hierarchical meta-analysis
Gerhard Gmel1, Elisabeth Gutjahr2 & Ju" rgen Rehm3,4 1Swiss Institute for the Prevention of Alcohol and Drug Problems, Lausanne; 2Addiction Research Institute, Zurich, Switzerland; 3Centre for Addiction and Mental Health, Toronto; 4Department of Public Health Sciences, University of Toronto, Toronto, Canada
Accepted in revised form 6 November 2002
Abstract. Objective: To determine the influence of six determining variables on the shape of the risk curve between alcohol and all-cause mortality. Methods: Data: Based on a systematic search with clear inclusion criteria, all articles on alcohol and allcause mortality until 2000 were included. Statistical methods: Precision-weighted pooling of relative risks (RRs); precision-weighted hierarchical analysis. Variables: For pooling: RRs for different categories of average volume of drinking, lifetime abstainers and ex-drinkers. For hierarchical analysis: on first level: consumption in grams of pure alcohol per day; on second level: length of follow-up time in months; per capita consumption; average age, proportion of abstainers, average volume of drinking, and variability of average volume of drinking at baseline. Outcomes measures: RR of former and current drinkers for allcause mortality compared to abstainers. Results: The main hypotheses could be confirmed for males: Exdrinkers had a higher mortality risk than lifetime abstainers; the higher and the more diverse the average volume of alcohol consumption, the wider the dip of the curve; the older the persons at baseline, the more pronounced the protective effect; and the longer the follow-up time, the less pronounced the protective effect. Except for average volume of drinking effects for females went in the same direction but with one exception did not reach significance. Conclusions: There are systematic influences on the shape of the risk curve between alcohol and all-cause mortality. The overall beneficial effect of light to moderate drinking remained under all scenarios, indicating a high validity of the overall shape despite the heterogeneity between studies.
Key words: Alcohol, All-cause mortality, Average consumption, Ex-drinker, Gender, Heavy drinking occasions, Hierarchical meta-analysis, Pooled analysis
Background and Objectives There is substantial evidence that alcohol can be detrimental to health (for overviews see Refs. [1-3]). However, alcohol also has health benefits [4], most notably the beneficial effect of moderate drinking on coronary heart disease (CHD, eg. [5]). All-cause mortality can be considered as the best single summary indicator for severe health consequences of alcohol consumption [4], where both detrimental and beneficial consequences can be seen.
Both kinds of relationship between drinking and allcause mortality are by no means new. Pearl [6] in his seminal work in the 1920s already found both a beneficial effect of moderate drinking and a detrimental effect of heavy drinking on all-cause mortality. Summarizing the evidence, previous meta-analyses have revealed a J-shaped curve: abstainers had a higher mortality risk than moderate drinkers, with heavier drinkers displaying the highest risk for mortality [7-10].
This is not surprising, as the overwhelming majority of individual level studies have found this shape
of a curve as well [10], and that most exceptions from this rule can be plausibly explained.
In particular, the following potential confounders have been discussed but so far not yet systematically analysed.
Sick-quitter hypothesis: A separation of abstainers into lifetime abstainers and ex-drinkers leads to less pronounced or to a complete disappearance of beneficial effects [11-13].
The reasoning behind this effect stems from the fact, that some drinkers quit drinking because of health reasons. These drinkers are more vulnerable for mortality and thus may be responsible for some or most of the higher risk of abstainers compared to moderate drinkers.
Length of follow-up hypothesis: Ceteris paribus, the longer the follow-up time, the less pronounced the effects.
European Journal of Epidemiology 18: 631-642, 2003.\Delta
2003 Kluwer Academic Publishers. Printed in the Netherlands.
For theoretical reasons, relative risks (RRs) will converge to 1 for older ages closer to death. This effect has been empirically found in many epidemiological studies. As cohort studies in epidemiology tend to start with 40-60-year-olds, this effect is relevant and translates into the above-specified hypothesis. Moreover, most cohort studies have one measurement of exposure and then outcome measurement several years later. Exposure is assumed to be stable over time [14]. The longer the follow-up, the higher the chance of change in exposure, increasing unsystematic measurement error, and decreasing the effect sizes on the relationship between exposure and outcome.
Cultural variation hypothesis: Cultures or groups with higher regular consumption have a nadir of the curve associated with more reduction of risk and at larger consumption levels (see also Refs. [15, 16]).
This hypothesis is based on the fact, that these cultures typically have patterns of drinking regularly and with meals without drinking in binges and to intoxication [17-19]. These patterns have been associated with relatively lesser harm on acute outcomes and benefits on CHD [20-22].
Age hypothesis: The older the cohort at baseline, the more pronounced the protective effect of alcohol.
In established market economies, with increasing age, the contribution of cardiovascular disease to mortality in general and the contribution of CHD in particular is more pronounced [23]. As the effect on CHD is mostly responsible for the protective effect (e.g. [15]), it is reasonable to conclude that the protective effect will be more pronounced with increasing age.
Based on the data set of Rehm et al. [10], these hypotheses were empirically tested. This procedure followed the advice of Greenland [24] or Petitti [25] to first establish the overall effect in meta-analysis, and then to examine sources of systematic variation to explain heterogeneity across studies. The analyses were restricted to the age groups 45 and above at baseline or largely exceeding 45-years during the follow-up time, as only in these age groups is there sufficient ischemic disease to expect a curvilinear relationship for all-cause mortality.
Methods The procedures for selecting studies, extrapolating information on average volume of drinking, deriving the creation of precision weights and calculating the risk curve were described in detail in Rehm et al. [10]. In short, a computer-assisted search in major data banks has been complemented by requests for relevant papers at alcohol epidemiology conferences, and with letters to key persons in the field. Finally, we
included all studies in the prior meta-analyses of the field [7-9].
This section will describe the operationalization of the potential factors of heterogeneity and the specific procedures used to calculate their influence on the shape of the risk curve.
Table 1 gives an overview of studies included in the analysis and some statistical information on the variables included in the equation.
RR estimates were logarithmized to permit estimation of models that force the regression equation through the origin, as the log of an RR of 1 is 0 (for no consumption or abstinence). The J-shape is often estimated by means of simple quadratic functions. In the present study this would estimate a symmetric function going through the origin. Hence, the minimum risk (nadir) would be estimated in the middle between abstention and the consumption at which the RR is again equal to that of abstainers. Most studies on the association between alcohol consumption and all-cause mortality, however, find the minimum risk already at very low levels of consumption, whereas the risk of consumption exceeds that of abstainers at fairly high consumption levels, thus contradicting a symmetric functional relationship around the nadir (see Table 2 and Figure 1). As a result, the level of alcohol consumption associated with the estimated nadir would be too high. Furthermore, as argued by Goetghebeur and Pocock [76] fitting a simple quadratic function for J-shaped associations in epidemiology is often not a valid test for the upturn.
Therefore, a mixture of two functions estimated the J-shape in the present study. The first function used the logs of alcohol consumption (in grams per day +1) to estimate the left part, and the second used alcohol consumption linearly to estimate the upturn after the nadir. One gram was added, as the logarithm of 0 (abstention) is undefined, but the logarithm of 1 is 0 and thus the regression equation without an intercept estimates a curve going through the origin. Furthermore, the logarithmic transform compresses values larger than 1, thus the higher the alcohol consumption the lower the additional impact of a unit change from the logarithmic part of the regression equation, whereas the additional impact of a unit change remains constant for the linear part. As the consequence, with a negative coefficient for the logarithmic part that is absolutely greater than the positive coefficient of the linear part, at low values of alcohol consumption the downturn of the risk curve would be modelled through the higher impact of the logarithmic part, whereas the upturn would be modelled through the higher impact of the linear part. Hence, the following functional relationship was used:
LogdhRRThic 1/4 b1c logdhxic th 1Th th b2cxic th eic; dh1Th
where i is the index of consumption level; c the index of studies; x the alcohol consumption in grams per day.
632
Table 1. Included
studies and their
characteristics
Author and year
of publication
Sample
Sex Follow-up
time
in months
Per capita
consumptionin
litres
of
pure ethanol
Averagevolume
in
grams a
day
SD of
averagevolume
Proportionof abstainers
Meanage
at
baseline
Year of
baselinesurvey
Reference:lifetimeabstention
Andreasson and Brandt,
1997
[26]
Female 228
6.0
8.0 6.6 26%
33.0 1973
No
Andreasson and Brandt,
1997
[26]
Male 228
6.0 13.7 10.1 21%
33.0 1973
No
Blackwelder et al.,
1980
[27]
Male 96 6.9
17.9 12.5 48%
55.5 1967
No
Bofetta and Garfinkel,
1990
[28]
Male 144
5.9 25.3 20.9 59%
50.5 1959
No
Brenner et al.,
1997
[29]
Male 84 10.8
54.2 38.2 7%
42.8 1987
No
Camacho et al.,
1987
[30]
Female 180
7.1
9.2 7.2 33%
55.0 1965
No
Camacho et al.,
1987
[30]
Male 180
7.1 13.2 11.4 18%
53.6 1965
No
Camargo et al.,
1997
[31]
Male 128
7.5 10.2 7.8 26%
53.2 1983
No
Carmelli et al.,
1995
[32]
Male 288
7.5 26.8 17.4 18%
46.1 1968
No
Colditz et al.,
1985
[33]
Mixed 57 8.0
12.7 12.5 39%
73.1 1976
No
Cullen et al,
1993
[34]
Female 276
8.9
9.8 11.6 47%
56.9 1966
Yes
Cullen et al,
1993
[34]
Male 276
8.9 25.0 20.0 21%
57.5 1966
Yes
De Labry
et al.,
1992
[35]
Male 144
7.9 25.5 16.8 15%
49.5 1973
No
Deev et al.,
1998
[36]
US-sample
Female 156
7.9
7.2 3.7 29%
52.7 1974
No
Deev et al.,
1998
[36]
Russian sample
Female 156
5.4
4.1 2.7 39%
53.2 1976
No
Deev et al.,
1998
[36]
US-sample
Male 156
7.9 17.0 11.4 13%
48.9 1974
No
Deev et al.,
1998
[36]
Russian-sample
Male 156
5.4 15.2 11.2 7%
48.9 1976
No
Doll et al.,
1994
[37]
Male 156
7.4 23.8 18.4 12%
62.0 1978
No
Dyer et al.,
1977
[38]
Male 120
5.6 21.8 20.3 8%
47.8 1958
Yes
Farchi et al.,
1992
[39]
Male 180
13.2
86.2 48.3 0%
54.7 1965
Yes
Fillmore et al.,
1998
[40]
US 67
Female 84 6.8
6.7 4.0 25%
46.8 1967
Yes
Fillmore et al.,
1998
[40]
Frisco 64
Female 240
7.2
7.9 4.6 23%
46.8 1964
Yes
Fillmore et al.,
1998
[40]
US 81
Female 120
7.7
7.3 4.3 23%
46.2 1981
Yes
Friedman and Kimball,
1986 [41]
Non-smokers Female 288
6.2
8.7 10.4 52%
44.5 1950
No
Friedman and Kimball,
1986 [41]
Smokers Female 288
6.2 12.5 12.9 25%
44.5 1950
No
Friedman and Kimball,
1986 [41]
Non-smokers Male 288
6.2 19.9 24.5 25%
44.5 1950
No
Friedman and Kimball,
1986 [41]
Light smokers
Male 288
6.2 23.1 24.3 17%
44.5 1950
No
Friedman and Kimball,
1986 [41]
Heavy smokers
Male 288
6.2 31.6 30.0 13%
44.5 1950
No
633
Table 1. (Continued
)
Author and year
of publication
Sample
Sex Follow-up
time
in months
Per capita
consumptionin
litres
of
pure ethanol
Averagevolume
in
grams a
day
SD of
averagevolume
Proportionof abstainers
Meanage
at
baseline
Year of
baselinesurvey
Reference:lifetimeabstention
Fuchs et al.,
1995
[42]
Female 144
7.7 10.2 11.2 32%
46.6 1980
Yes
Garfinkel et al.,
1988
[43]
Female 144
6.0 17.0 14.8 86%
50.9 1960
No
Goldberg et al,.
1994
[44]
Male 252
7.5 14.7 15.5 36%
57.5 1967
No
Goldberg et al,.
1994
[44]
Male 180
7.9 14.4 15.1 45%
70.0 1973
No
Gordon and Doyle,
1987 [45]
18 years
follow-up
Male 216
6.0 27.5 28.4 33%
46.5 1954
No
Gordon and Doyle,
1987 [45]
10 years
follow-up
Male 120
6.0 40.5 33.6 25%
46.5 1954
No
Gro/nbaek et al.,
1994
[46]
Mixed 132
9.6 23.1 26.9 23%
51.7 1977
No
Gro/nbaek et al.,
1998
[47]
Female 138
10.7
7.4 10.2 50%
60.5 1983
No
Gro/nbaek et al.,
1998
[47]
Male 138
10.7
13.9 33.0 35%
60.5 1983
No
Hart et al.,
1999
[48]
Male 252
7.1 23.5 18.2 32%
48.2 1972
No
Hoffmeister et al.,
1999
[49]
Female 72 10.6
23.4 20.2 48%
54.1 1986
No
Hoffmeister et al.,
1999
[49]
Male 72 10.6
27.8 22.5 20%
52.6 1986
No
Keil et al.,
1997
[50]
Female 96 10.6
18.8 12.0 44%
54.3 1984
No
Keil et al.,
1997
[50]
Male 96 10.6
45.3 30.8 13%
54.4 1984
No
Kittner et al.,
1983
[51]
Male 144
7.2 42.2 34.0 76%
55.0 1967
No
Kivela
"et al., 1989
[52]
Female 120
6.3
7.3 4.9 34%
64.5 1974
No
Klatsky et al.,
1981
[53]
Female 120
6.8 55.6 37.1 28%
43.3 1966
No
Klatsky et al.,
1981
[53]
Male 120
6.8 52.7 35.5 24%
45.4 1966
No
Klatsky et al.,
1992
[54]
Female 72 7.9
7.0 12.9 21%
40.9 1982
Yes
Klatsky et al.,
1992
[54]
Male 72 7.9
15.9 19.0 11%
42.1 1982
Yes
Kono et al.,
1986
[55]
Male 216
4.9 33.5 26.0 31%
47.5 1965
Yes
Lazarus et al.,
1991
[56]
Female 120
8.0
54.6 1974
Yes
Lazarus et al.,
1991
[56]
Male 120
8.0
64.5 1974
Yes
Leino et al.,
1998
[57]
US 67
Male 84 6.8
11.9 6.9 26%
45.2 1967
Yes
Leino et al.,
1998
[57]
US 69
Male 48 6.9
11.9 6.9 20%
42.0 1969
Yes
Leino et al.,
1998
[57]
Frisco 64
Male 240
7.2 13.7 7.2 13%
45.3 1964
Yes
Leino et al.,
1998
[57]
Frisco 67
Male 204
7.5 13.7 6.9 10%
38.9 1967
Yes
Marmot et al.
1981
[58]
Male 120
6.2 20.7 17.7 34%
52.0 1968
No
Maskarinec et al.
1998
[59]
Female 192
7.6 17.7 18.6 87%
53.5 1978
No
Maskarinec et al.
1998
[59]
Male 192
7.6 26.4 29.3 66%
51.2 1978
No
Mertens et al.,
1996
[60]
Mixed 48 7.0
23.6 18.2 19%
61.4 1989
No
Miller et al.,
1990
[61]
Male 108
4.1 25.2 27.8 72%
50.0 1977
Yes
Paunio et al.,
1994
[62]
Male 56.4 7.4
24.3 20.5 11%
58.3 1986
No
Rehm, 1996 [63]
Only used to estimate
the mortality
risk of lifetime
abstention
vs. ex-drinker
634
Rehm and Sempos,
1995 [64]
Female 180
7.9
4.6 5.8 46%
68.0 1972
Yes
Rehm and Sempos,
1995 [64]
Male 180
7.9 10.3 15.7 28%
68.0 1972
Yes
Rehm et al.,
1993
[65]
Female 156
11.0
10.7 8.9 0%
45.9 1976
No
Rehm et al.,
1993
[65]
Male 156
11.0
29.5 28.0 0%
43.2 1976
No
Renaud et al.,
1998
[66]
Male 144
13.0
61.3 40.4 11%
48.9 1981
No
Renaud et al.,
1999
[67]
Only used to estimate
the mortality
risk of lifetime
abstention
vs. ex-drinker
Scherr et al.
1992
[68]
East Boston
Mixed 60 7.9
17.2 11.5 33%
73.0 1982
No
Scherr et al.
1992
[68]
Iowa
Mixed 60 7.9
16.1 10.6 65%
73.6 1982
No
Scherr et al.
1992
[68]
New Haven
Mixed 60
7.9 17.0 11.4 38%
73.6 1982
No
Simons et al.,
1996
[69]
Female 48 6.5
8.7 6.5 48%
61.3 1988
No
Simons et al.,
1996
[69]
Male 48 6.5
17.2 15.6 22%
61.3 1988
No
Suhonen et al.,
1987
[70]
Male 60 6.3
7.7 4.1 21%
51.3 1975
No
Tsubono et al.,
1993
[71]
Mixed 48 6.5
29.4 21.0 67%
61.3 1988
Yes
Tsugane et al.,
1999
[72]
Male 72 6.6
42.3 25.4 23%
49.4 1990
No
Wannamethee and Shaper,
1997 [73]
Male 117.6
7.5 16.1 19.6 10%
55.1 1984
Yes
Wannamethee and Shaper,
1999 [74]
Only used to estimate
the mortality
risk of lifetime
abstention
vs. ex-drinker
Yuan et al.,
1997
[75]
Male 84 2.8
30.3 28.1 55%
29.7 1988
Yes
635
Hierarchical linear models The impact of study characteristics on the J-shape association was estimated by means of hierarchical linear models. Equation (1) describes the so-called first level relationship. Analysis was performed by using precision-based weights for each data point at the first
level. Hierarchical linear models do not assume b1c and b2c in Equation (1) to be fixed, but varying across studies (slopes as outcomes see Ref. [77])
b1c 1/4 c10 th c11wkc; dh2Th where wk are second level variables (study characteristics). The same applies to b2c. Significant associations at the second level explain partly the variation of findings across studies, and, hence, the heterogeneity between studies. Study characteristics used as second level variables were follow-up time, per capita consumption, average volume of alcohol consumption for drinkers, standard deviation (SD) of average volume of alcohol consumption for drinkers, proportion of abstainers, and mean age at baseline. Per capita consumption data came either from the World Drinking Trends [78] or the Global Status Report on Alcohol [79] and was averaged over the follow-up period. The other five variables were directly derived from the publication. Data of these variables were either directly provided in the paper or recalculated by using sample size information and interval centers of consumption categories.
One of the questions in meta-analytic hierarchical linear models is, whether the slopes should be estimated randomly varying or non-randomly varying. Randomly varying slopes can be estimated by inclusion of an error term in Equation (2). Firstly, random coefficient models assume multivariate normality of errors, which may be particularly doubtful in the present context. In addition, the available studies hardly present a random sample of potential studies. Secondly, random effect models are more heavily affected by biases such as publication bias. Thirdly, random effect models usually do not change point estimates (see Ref. [24] for these and further arguments). Without an error distribution slopes across studies may appear more variable than they really are, due to chance variation [77]. In the present study, however, this is required as it favours the detection of cross-study heterogeneity. Hence, given the small sample size a greater variability of slopes helps against type II errors, i.e. to overlook potential factors to explain study heterogeneity.
Results Table 2 gives an overview of the results for women and men (see also Ref. [10]). Clearly, the J-shape appears for both sexes, however with different levels of consumption associated with minimal risk and with females experiencing higher risks at relative lower consumption rates beyond moderate drinking.
Both risk curves confirm prior results (e.g. [7]). In addition, we found a strong effect for ex-drinkers, especially for females. Thus, female ex-drinkers had a 44% elevated risk, and males a 21% elevated risk. This result strongly supports the sick-quitter hypothesis.
Table 2. Relative mortality risk and 95% CIs for different categories of average volume - females and males mean age 45 and over at baseline
RR Lower 95% CI 95% CI Upper Females
Ex-drinker 1.44 1.28 1.61 Abstainers 1.00 Reference category >0-10 g 0.87 0.84 0.89 >10-30 g 1.01 0.99 1.04 >30-50 g 1.40 1.34 1.47 >50 g 1.43 1.34 1.53
Males
Ex-drinker 1.21 1.10 1.32 Abstainers 1.00 Reference category >0-10 g 0.85 0.83 0.87 >10-20 g 0.80 0.78 0.82 >20-30 g 0.91 0.89 0.94 >30-40 g 0.96 0.93 1.00 >40-70 g 1.04 1.01 1.07 >70-110 g 1.27 1.23 1.31 >110 g 1.46 1.33 1.60
Figure 1. Impact of follow-up time on risk curve, males. Black circles indicate the results of the meta-analysis (for explanations see text).
636
An inspection of the confidence intervals (CIs) reveals that the interval is smallest for the beneficial effect of the lowest drinking category (i.e., drinking between 0 and 10 g of pure alcohol on average), both for males (interval: 0.04) and females (interval: 0.05). Whereas the estimates in this category are based on most and most consistent information, the largest drinking category shows more substantial variation (interval for males: 0.27 and for females: 0.19) reflecting the more insecure database. The seeming inconsistency in the higher consumption categories for females may reflect just this methodological problem.
With respect to the other hypotheses and influencing factors, Figure 1 gives an impression of the influence of length of follow-up time on the shape of the curve for males. The middle line represents the relationship at the median of all the follow-up times included: 135 months. The other lines give the relationship at the 25th and 75th percentiles of follow-up time, representing 75 and 189 months. As hypothesized, the longer the follow-up time, the less pronounced the J-shape of the curve. Length of followup time does not seem to influence the location of the nadir with respect to average volume consumed or the point where the curve crosses again the line,
where RR equals to 1, i.e. when the risk corresponds to the risk of abstainers.
The parameters of the nadir and of the crossing of the equivalence line, where the risk corresponds to the risk of abstainers, has been used to describe the curve in Table 3. All parameters are given in comparison to the nadir of the median for each variable. Concretely, for the first line in Table 3: for the 25th percentile with a length of stay of 75 months, in comparison to the median stay of 135, the average volume of alcohol consumption for the nadir stays practically the same (i.e. shifted 0.06 g/day to the right). The RR is 0.05 lower for the 25th percentile compared to the risk for the median (RR estimates of 0.797 vs. 0.845), which means that the beneficial effect is more pronounced, i.e. farther away from 1. The curve crosses the line of same risk as abstainers at the point, which is one g/day higher than the point for the median (at 52 g/day compared to 51 g/day). The sign of the values are derived from subtracting the corresponding values of the median from the values for the 25th percentile. The two significances denote the fact, that the overall follow-up time has a significant impact on the overall curvature. Thus, with the three values in Table 3, the curvature can be clearly described in its main
Table 3. Shape of the risk relationship between alcohol and all-cause mortality as influenced by selected factors - males
Shift of g/day for nadir
Shift of RR for nadir
Shift of equivalent risk to abstainer
Effect and significance of c11*
Effect and significance of c21*
Follow-up time in 25th perc.: 75 months 0.06 \Delta 0.05 1 0.00134 \Delta 0.000044; months (median: 135) 75th perc.: 189 months \Delta 0.11 0.05 0 t = \Delta 2.9;
df = 194; p = 0.004
t = \Delta 2.8; df = 194; p = 0.006
Per capita consumption 25th perc.: 6.21 l \Delta 0.69 0.00 0 0.007023 \Delta 0.000759 in liters of pure ethanol (median: 7.28)
75th perc.: 7.93 l 0.49 0.00 2 t = 0.952;
df = 194; p = 0.341
t = \Delta 4.17; df = 194; p < 0.001
Average volume in 25th perc.: 14.83 g/day \Delta 0.58 0.00 \Delta 3 0.00117 \Delta 0.000106 grams a day (median: 23.05)
75th perc.: 29.02 g/day 0.49 0.00 2 t = 1.031;
df = 194; p = 0.303
t = \Delta 3.913; df = 104; p < 0.001
SD of average volume 25th perc.: 11.71 \Delta 1.38 0.01 \Delta 7 0.000283 \Delta 0.000211 (median: 19.32) 75th perc.: 27.93 2.16 \Delta 0.01 11 t = 0.136;
df = 194; p = 0.892
t = \Delta 3.891; df = 194; p < 0.001
Proportion of abstainers 25th perc.: 13% 1.31 0.00 7 \Delta 0.074882 0.008065 (median: 22%) 75th perc.: 35% \Delta 1.31 0.00 \Delta 7 t = \Delta 1.014;
df = 194; p = 0.311
t = 3.759; df = 194; p < 0.001
Mean age at baseline in 25th perc.: 45.87 years \Delta 1.29 0.02 \Delta 6 \Delta 0.003385 \Delta 0.000026 years (median: 50.87) 75th perc.: 57.49 years 1.69 \Delta 0.02 10 t = \Delta 1.04;
df = 194; p = 0.299
t = \Delta 0.247; df = 194; p = 0.805
*The significances here denote the effect of variables on both parameters of the model: the log parameter (b1c), which controls the upturn to the left and the linear parameter (b2c), which mainly controls the upturn to the right (see Methods section).
637 characteristics: the average volume of drinking associated with minimal risk, the level of risk in comparison to abstention, and the level of drinking, where the risk is again the same as for abstainers. For females (see Table 4 below), the risk curves look similar but the differences are less pronounced and the overall differences do not become significant.
Different indicators were used to operationalize drinking patterns. Overall, for males: the higher the average volume and the higher the variability, the wider the dip (e.g. area before the curve crosses the line of equivalent risk with abstainers).
Consequently, the nadir of minimum risk can be found at higher average volumes of drinking. However, the nadir was not estimated deeper, i.e. the level of maximal protection did not change. In the same logic were the changes due to proportion of abstainers, although this variable did not achieve significance. Finally, mean age at baseline changed both the location of the nadir and the width of the curve. The older, the more pronounce the protective effect. Overall for males, the average volume of consumption associated with minimum risk varied between 9 and 15 g/day of pure alcohol and the RRs associated
between 0.80 and 0.89. Thus, under all scenarios there was a protective effect of light to moderate drinking. The variation for the crossing of unity risk was wider ranging between 35 and 66 g/day of pure alcohol.
The curves for females in general went in the same direction but did not become significant except for variability of drinking, and, marginally, for average volume, indicating, contrary to men, that the optimum level of drinking decreases with increasing volume (see Table 4). Overall, for females, the average volume of consumption associated with minimum risk varied between 3 and 13 g/day of pure alcohol and the RRs associated between 0.86 and 0.97; with the curve crossing the unity line between 11 and 59 g/day. Overall, the dip was much less pronounced than for males and risk increased at lower level of consumption. Variability also was larger, but even under these circumstances all scenarios showed a protective effect.
Discussion The overall J-shape could be corroborated in this analysis for both gender. This shape can be seen as
Table 4. Shape of the risk relationship between alcohol and all-cause mortality as influenced by selected factors - females
Shift of g/day for nadir
Shift of RR for nadir
Shift of equivalent risk to abstainer
Effect and significance of c11*
Effect and significance of c21*
Follow-up time in 25th perc.: 66 months 1.03 )0.03 5 0.00065 )0.000017 months (median: 132) 75th perc.: 180 months )0.91 0.02 )4 t = 0.613;
df = 86; p = 0.540
t = )0.35; df = 86; p = 0.725
Per capita consumption 25th perc.: 6.54 l )0.45 0.01 )2 )0.005376 )0.00022 months (median: 132) 75th perc.: 7.98 l 0.11 0.00 0 t = )0.245;
df = 86; p = 0.807
t = )0.253; df = 86; p = 0.807
Average volume in 25th perc.: 7.39 g/day 0.16 )0.01 0 0.006671 )0.000319 grams a day (median: 10.18)
75th perc.: 17.45 g/day )0.53 0.02 )3 t = 1.058;
df = 86; p = 0.291
t = )1.762; df = 86; p = 0.078
SD of average volume 25th perc.: 6.17 )0.50 0.00 )2 0.003103 )0.000503 (median: 11.16) 75th perc.: 13.86 0.36 0.00 1 t = 0.407;
df = 86; p = 0.684
t = )2.321; df = 86; p = 0.020
Proportion of 25th perc.: 25% 0.16 0.00 1 0.052256 )0.000698 abstainers (median: 34%) 75th perc.: 48% )0.25 0.01 )1 t = 0.446;
df = 86; p = 0.655
t = )0.15; df = 86; p = 0.881
Mean age at baseline 25th perc.: 46.38 years )3.35 0.04 )14 )0.006017 )0.000336 in years (median: 53.50) 75th perc.: 61.36 6.78 )0.07 34 t = )0.899;
df = 86; p = 0.369
t = )1.029; df = 86; p = 0.304
*The significances here denote the effect of variables on both parameters of the model: the log parameter (b1c), which controls the upturn to the left and the linear parameter (b2c), which mainly controls the upturn to the right (see Methods section).
638 resulting mainly from the beneficial effects of alcohol consumption on CHD and, to a smaller degree, ischemic stroke, and the detrimental effects on more than 50 disease conditions (for overviews [80, 81]).
The main hypotheses related to influencing factors could be confirmed for males: Ex-drinkers had a higher mortality risk than lifetime abstainers; the higher the average volume, the wider the dip of the curve; the older the persons at baseline, the more pronounced the protective effect; and the longer the follow-up time, the less pronounced the protective effect.
Overall, however, the factors used in the present study did not explain much of the heterogeneity between studies. There is still quite a lot of variability unexplained. Thus, further research is necessary to look into the heterogeneity of study results.
This heterogeneity may be mainly related to two points: the assessment of alcohol varying widely between studies, and the influence of patterns introducing variability at the same volume of drinking. Both variables have implications for future studies in alcohol epidemiology: assessment in cohort studies must standardize and reflect the developments in assessment made in survey research (e.g. special issue of the Journal of Substance Abuse on measuring alcohol consumption, 2000), both for volume and patterns [82]. There are several factors influencing the measurement of alcohol consumption [83]. For example, different modes of administration of interviews (telephone, face-to-face, mailed) resulted in different consumption values within the same sample. Also, non-response bias may vary from study to study, and the measurement instruments (e.g. graduated frequency vs. quantity frequency approaches) and corresponding drink sizes may greatly vary between studies [84]. If standardized measures are available in the future, future meta-analysis may be able to include both average volume and patterns, and heterogeneity will likely decrease. Another source of heterogeneity stems from the differences in adjusting for potential effect modifiers. In the present study we included studies, which at least adjusted for age. However, other factors such as smoking, dietary factors, or physical activity may be important for the association between alcohol consumption and mortality. These factors could not be dealt within the present study as either many studies did not use any other control variable but age, or too few studies used the same control variables.
Effects for females went in the same directions but were not significant. Two reasons may be responsible for this result. First, there are fewer female studies and thus the power to detect effects is smaller. Secondly, there are indications that patterns of drinking play a less important role for females than for males (e.g. [18, 85]).
Finally, what do the results mean from a public health point of view? All scenarios revealed a protective effect at low doses of alcohol. However, the heterogeneity of results beyond low doses clearly indicate, that other factors influence the relationship between average volume of consumption and mortality. Part of these influences may be heavy drinking occasions which could not adequately be operationalized in this study. Both epidemiological and physiological research indicate that such occasions are linked to detrimental vascular outcomes including CHD [20, 86], thus reducing the beneficial effects. In fact, if all of the alcohol is consumed in binges, the beneficial effect of average light to moderate drinking may even disappear [85]. In other words, in an extreme population, where all the drinking takes place in the form of heavy drinking occasions, the curve may no longer be J-shaped. However, this extreme case is quite unlikely in the usual population of cohorts, where middle class professionals are the rule. This may explain part of the homogeneity with respect to lower average drinking categories.
Acknowledgements The Swiss National Science Foundation has supported this work with a grant to the first and third author (# 3200-061721.00/1). Some of this work was undertaken during the preparation of a report, Evidence Regarding the Level of Alcohol Consumption Considered to be Low Risk for Men and Women, for the Australian Commonwealth Department of Health and Aged Care, October 1999 (see Ref. [1]). We would like to acknowledge the helpful comments of Robin Room on an earlier version of this text as well as the help of the Swiss Institute for the Prevention of Alcohol and Drug Problems, especially of Elisabeth Grisel, in technical assistance.
References
1. Single E, Ashley MJ, Bondy SJ, Rankin JG, Rehm J,
with the assistance of Dobbins M. Evidence regarding the level of alcohol consumption considered to be lowrisk for men and women. Final Report. Canberra: Report prepared for Australian Commonwealth Department of Health and Aged Care, 1999, October 1999. 2. US Department of Health and Human Services. Ninth
Special Report to the US Congress on Alcohol and Health from the Secretary of Health and Human Services. Rockville, MD: US Department of Health and Human Services, Public Heath Service, National Institute of Health (NIH), National Institute on Alcohol Abuse and Alcoholism (NIAAA), 1997. 3. US Department of Health and Human Services. Tenth
Special Report to the US Congress on Alcohol and Health from the Secretary of Health and Human Services. Rockville, MD: US Department of Health and Human Services, Public Health Service, National Institutes of Health (NIH), National Institute on Alcohol Abuse and Alcoholism (NIAAA), 2000.
639 4. Ashley MJ, Rehm J, Bondy SJ. Measuring health risks
and benefits of alcohol. In: US Department of Health and Human Services (ed), 10th Special Report to the US Congress on Alcohol and Health. Rockville, MD: US Department of Health and Human Services, National Institute on Alcohol Abuse and Alcoholism, 2000. 5. Chadwick DJ, Goode JA (eds), Alcohol and Cardiovascular Diseases: Novartis Foundation Symposium 216. New York: John Wiley & Sons, 1998. 6. Pearl R. Alcohol and Longevity. New York, NY:
Knopf, 1926. 7. English DR, Holman CDJ, Milne E, et al. The Quantification of Drug Caused Morbidity and Mortality in Australia, 1992. Canberra: Commonwealth Department of Human Services and Health, 1995. 8. Holman CDA, English DR, Milne E, Winter MG.
Meta-analysis of alcohol and all-cause mortality: A validation of NHMRC recommendations. Med J Aust 1996; 164: 141-145. 9. White IR. The level of alcohol consumption at which
all-cause mortality is least. J Clin Epidemiol 1999; 52: 967-975. 10. Rehm J, Gutjahr E, Gmel G. Alcohol and all-cause
mortality: A pooled analysis. Contemp Drug Probl 2001; 28: 337-361. 11. Shaper AG. Alcohol and mortality: A review of prospective studies. Br J Addict 1990; 85: 837-847. 12. Shaper AG. A response to commentaries: The effects of
self-selection. Br J Addict 1990; 85: 859-861. 13. Shaper AG, Wannamethee SG, Walker M. Alcohol
and mortality in British men: Explaining the U-shaped curve. Lancet 1988; 8623: 1267-1273. 14. Sempos CT, Flegal KM, Johnson CL, Loria CM,
Woteki CE, Briefel RR. Issues in the long-term evaluation of diet in longitudinal studies. J Nutr 1993; 123: 406-412. 15. Rehm J, Bondy SJ. Alcohol and all-cause mortality: An
overview. Novartis Found Symp 1998; 216: 223-232; discussion 232-236. 16. Rehm J. Alcohol consumption and mortality. What do
we know and where should we go? Addiction 2000; 95: 989-995. 17. Room R, Ma" kela" K. Typologies of the cultural position of drinking. J Stud Alcohol 2000; 61(3): 475-483. 18. Rehm J, Monteiro MG, Room R, et al. Steps towards
constructing a global comparative risk analysis for alcohol consumption: Determining indicators and empirical weights for patterns of drinking, deciding about theoretical minimum, and dealing with different consequences. Eur Addict Res 2001; 7(3): 138-147. 19. Rehm J, Gmel G. Applying principles of comparative
risk analysis to substance abuse-related burden. Eur Addict Res 2001; 7(3): 95-97. 20. Puddey IB, Rakic V, Dimmitt SB, Beilin LJ. Influence
of pattern of drinking on cardiovascular disease and cardiovascular risk factors - A review. Addiction 1999; 94: 649-663. 21. Rossow I, Pernanen K, Rehm J. Accidents, suicide
and violence. In: Klingemann H, Gmel G (eds), Mapping the Social Consequences of Alcohol Consumption. Dordrecht: Kluwer Academic Publishers, 2001; 93-112. 22. Walsh GW, Rehm J. Daily drinking and harm. Contemp Drug Probl 1996; 23: 465-478.
23. Murray CJL, Lopez A. The Global Burden of Disease:
A Comprehensive Assessment of Mortality and Disability from Diseases, Injuries and Risk Factors in 1990 and Projected to 2020. Cambridge, MA: Harvard School of Public Health, 1996. 24. Greenland S. Meta-analysis. In: Rothman KJ, Greenland S (eds), Modern Epidemiology. Philadelphia, PA: Lippincott-Raven Publishers, 1998; 643-673. 25. Petitti DB. Meta-Analysis, Decision Analysis, and
Cost-Effectiveness Analyses. Methods for Quantitative Synthesis in Medicine. 2nd ed. New York/Oxford: Oxford University Press, 2000. 26. Andre' asson S, Brandt L. Mortality and morbidity related to alcohol. Alcohol Alcohol 1997; 32(2): 173-178. 27. Blackwelder WC, Yano K, Rhoads GG, Kagan A,
Gordon T, Palesch Y. Alcohol and mortality: The Honolulu Heart Study. Am J Med 1980; 68(2): 164- 169. 28. Boffetta P, Garfinkel L. Alcohol drinking and mortality among men enrolled in an American cancer society prospective study. Epidemiology 1990; 1(5): 342- 348. 29. Brenner H, Arndt V, Rothenbacher D, Schuberth S,
Fraisse E, Fliedner TM. The association between alcohol consumption and all-cause mortality in a cohort of male employees in the German construction industry. Int J Epidemiol 1997; 26(1): 85-91. 30. Camacho TC, Kaplan GA, Cohen RD. Alcohol consumption and mortality in Alameda County. J Chronic Dis 1987; 40(3): 229-236. 31. Camargo CA, Hennekens CH, Gaziano JM, Glynn RJ,
Manson JE, Stampfer MJ. Prospective study of moderate alcohol consumption and mortality in US male physicians. Arch Int Med 1997; 157: 79-85. 32. Carmelli D, Swan GE, Page WF, Christian JC. World
war II - veteran male twins who are discordant for alcohol consumption: 24-year mortality. Am J Public Health 1995; 85(1): 99-101. 33. Colditz GA, Branch LG, Lipnick RJ, et al. Moderate
alcohol and decreased cardiovascular mortality in an elderly cohort. Am Heart J 1985; 109(4): 886-889. 34. Cullen KJ, Knuiman MW, Ward NJ. Alcohol and
mortality in Busselton, Western Australia. Am J Epidemiol 1993; 137(2): 242-248. 35. De Labry LO, Glynn RJ, Levenson MR, Hermos JA,
LoCastro JS, Vokonas PS. Alcohol consumption and mortality in an American male population: Recovering the U-shaped curve-findings from the Normative Aging Study. J Stud Alcohol 1992; 53(1): 25-32. 36. Deev A, Shestov D, Abernathy J, Kapustina A, Muhina N, Irving SH. Association of alcohol consumption to mortality in middle-aged US and Russian men and women. Ann Epidemiol 1998; 8(3): 147-153. 37. Doll R, Peto R, Hall E, Wheathley K, Gray R. Mortality in relation to consumption of alcohol: 13 years' observations on male British doctors. Br Med J 1994; 309: 311-918. 38. Dyer AR, Stamler J, Paul O, et al. Alcohol consumption, cardiovascular risk factors, and mortality in two Chicago epidemiologic studies. Circulation 1977; 56(6): 1067-1074. 39. Farchi G, Fidanza F, Mariotti S, Menotti A. Alcohol
and mortality in the Italian rural cohorts of the Seven Countries Study. Int J Epidemiol 1992; 21(1): 74-81.
640 40. Fillmore KM, Golding JM, Graves KL, et al. Alcohol
consumption and mortality. III. Studies of female populations. Addiction 1998; 93(2): 219-229. 41. Friedman LA, Kimball AW. Coronary heart disease
mortality and alcohol consumption in Framingham. Am J Epidemiol 1986; 124(3): 481-489. 42. Fuchs C, Stampfer MJ, Colditz GA, et al. Alcohol
consumption and mortality among women. N Engl J Med 1995; 332(19): 1245-1250. 43. Garfinkel L, Boffetta P, Stellman SD. Alcohol and breast
cancer: A cohort study. Prev Med 1988; 17: 686-693. 44. Goldberg RJ, Burchfiel CM, Reed DM, Wergowske G.
A prospective study of the health effects of alcohol consumption in middle-aged and elderly men. The Honolulu Heart Program. Circulation 1994; 89(2): 651-659. 45. Gordon T, Doyle JT. Drinking and mortality. The
Albany Study. Am J Epidemiol 1987; 125(2): 263-270. 46. Gro/nbaek M, Deis A, So/rensen TIA, et al. Influence of
sex, age, body mass index, and smoking on alcohol intake and mortality. Br Med J 1994; 308: 302-306. 47. Gro/nbaek M, Deis A, Becker U, et al. Alcohol and
mortality: Is there a U-shaped relation in elderly people? Age Ageing 1998; 27: 739-744. 48. Hart CL, Smith GD, Hole DJ, Hawthorne VM. Alcohol consumption and mortality from all causes, coronary heart disease, and stroke: Results from a prospective cohort study of Scottish men with 21 years of follow-up. Br Med J 1999; 318: 1725-1729. 49. Hoffmeister H, Schelp EP, Bo" hning D, Dietz E, Kirschner W. Alkoholkonsum in Deutschland und seine gesundheitlichen Aspekte. Berlin/Heidelberg: Springer Verlag, 1999. 50. Keil U, Chambless LE, Do" ring A, Filipiak B, Stieber J.
The relation of alcohol intake to coronary heart disease and all-cause mortality in a beer-drinking population. Epidemiology 1997; 8(2): 150-156. 51. Kittner SJ, Garcia-Palmieri MR, Costas R, Jr, CruzVidal M, Abbott RD, Havlik RJ. Alcohol and coronary heart disease in Puerto Rico. Am J Epidemiol 1983; 117(5): 538-550. 52. Kivela" S-L, Nissinen A, Ketola A, Punsar S, Puska P,
Karvonen M. Alcohol consumption and mortality in aging or aged Finnish men. J Clin Epidemiol 1989; 42(1): 61-68. 53. Klatsky AL, Friedman GD, Siegelaub AB. Alcohol
and mortality. A ten-year Kaiser-Permanente Experience. Ann Int Med 1981; 95(2): 139-145. 54. Klatsky AL, Armstrong MA, Friedman GD. Alcohol
and mortality. Ann Int Med 1992; 117(8): 646-654. 55. Kono S, Ikeda M, Tokudome S, Nishizumi M, Kuratsune M. Alcohol and mortality: A cohort study of male Japanese physicians. Int J Epidemiol 1986; 15(4): 527-532. 56. Lazarus NB, Kaplan GA, Cohen RD, Leu D-J.
Change in alcohol consumption and risk of death from all causes and from ischaemic heart disease. Br Med J 1991; 303: 553-556. 57. Leino EV, Romelsjo" A, Shoemaker C, et al. Alcohol
consumption and mortality. II. Studies of male populations. Addiction 1998; 93(2): 205-218. 58. Marmot MG, Shipley MJ, Rose G, Briony JT. Alcohol
and mortality: A U-shaped curve. Lancet 1981; 8220: 580-583.
59. Maskarinec G, Meng L, Kolonel LN. Alcohol intake,
body weight, and mortality in a multiethnic prospective cohort. Epidemiology 1998; 9(6): 654-661. 60. Mertens JR, Moos RH, Brennan PL. Alcohol consumption, life context, and coping predict mortality among late-middle-aged drinkers and former drinkers. Alcohol Clin Exp Res 1996; 20(2): 313-319. 61. Miller GJ, Beckles GLA, Maude GH, Carson DC.
Alcohol consumption: protection against coronary heart disease and risks to health. Int J Epidemiol 1990; 19(4): 923-930. 62. Paunio M, Heinonen OP, Virtamo J, et al. HDL cholesterol and mortality in Finnish men with special reference to alcohol intake. Circulation 1994; 90(6): 2909-2918. 63. Rehm J. Alcohol and all-cause mortality (Letter to the
Editor). Int J Epidemiol 1996; 25(1): 215-216. 64. Rehm J, Sempos CT. Alcohol consumption and allcause mortality. Addiction 1995; 90: 471-480. 65. Rehm J, Fichter MM, Elton M. Effects on mortality of
alcohol consumption, smoking, physical activity, and close personal relationships. Addiction 1993; 88: 101- 112. 66. Renaud SC, Gue' guen R, Schenker J, d'Houtaud A.
Alcohol and mortality in middle-aged men from Eastern France. Epidemiology 1998; 9(2): 184-188. 67. Renaud SC, Gue' guen R, Siest G, Salamon R. Wine,
beer, and mortality in middle-aged men from eastern France. Arch Int Med 1999; 159(16): 1865-1870. 68. Scherr PA, LaCroix AZ, Wallace RB, et al. Light to
moderate alcohol consumption and mortality in the elderly. J Am Geriatr Soc 1992; 40(7): 651-657. 69. Simons LA, Friedlander Y, McCallum J, Simons J.
Alcohol intake and survival in the elderly: a 77 month follow-up of the Dubbo study. Aust N Z J Med 1996; 26: 662-670. 70. Suhonen O, Aromaa A, Reunanen A, Knekt P. Alcohol
consumption and sudden coronary death in middleaged Finnish men. Acta Med Scand 1987; 221: 335-341. 71. Tsubono Y, Fukao A, Hisamichi S. Health practices
and mortality in a rural Japanese population. Tohoku J Exp Med 1993; 171: 339-348. 72. Tsugane S, Fahey MT, Sasaki S, Baba S. Alcohol
consumption and all-cause and cancer mortality among middle-aged Japanese men: Seven-year follow-up of the JPHC Study Cohort I. Am J Epidemiol 1999; 150(11): 1201-1207. 73. Wannamethee SG, Shaper AG. Lifelong teetotalers, exdrinkers and drinkers: Mortality and the incidence of major coronary heart disease events in middle-aged British men. Int J Epidemiol 1997; 26(3): 523-531. 74. Wannamethee SG, Shaper AG. Type of alcoholic drink
and risk of major coronary heart disease events and allcause mortality. Am J Public Health 1999; 89(5): 685- 690. 75. Yuan J-M, Ross RK, Gao Y-T, Henderson BE,
Yu MC. Follow up study of moderate alcohol intake and mortality among middle aged men in Shanghai, China. Br Med J 1997; 314: 18-23. 76. Goetghebeur EJT, Pocock SJ. Detection and estimation of J-shaped risk-response relationships. J Roy Stat Soc 1995; 158(Part 1): 107-121. 77. Bryk AS, Raudenbush SW. Hierarchical Linear Models: Applications and Data Analysis Methods. Newbury Park: Sage Publications, Inc., 1992.
641 78. Productschap voor Gedistilleerde Dranken. World
Drink Trends. Henley-on-Thames: NTC Publications Ltd., 1999. 79. WHO. Global Status Report on Alcohol. Geneva:
World Health Organization (WHO), Substance Abuse Department, 1999. 80. Gutjahr E, Gmel G, Rehm J. Relation between average
alcohol consumption and disease: An overview. Eur Addict Res 2001; 7(3): 117-127. 81. Rehm J, Room R, Monteiro M, et al. Alcohol as a risk
factor for burden of disease. In: WHO (ed), Comparative Quantification of health risks: Global and regional burden of disease due to selected major risk factors. Geneva: WHO Regional Office, in press. 82. Rehm J. Re: `Alcohol intake assessment: the sober
facts'. Am J Epidemiol 2000; 151(4): 436-438. 83. WHO. International Guide for Monitoring Alcohol
Consumption and Related Harm. Geneva: World Health Organization (WHO), Department of Mental Health and Substance Dependence, 2000.
84. Knibbe RA, Bloomfield K. Alcohol consumption estimates in survey in Europe: Comparability and sensitivity for gender differences. J Subst Abuse 2001; 22(1): 23-38. 85. Rehm J, Greenfield TK, Rogers JD. Average volume of
alcohol consumption, patterns of drinking and allcause mortality. Results from the US National Alcohol Survey. Am J Epidemiol 2001; 153(1): 64-71. 86. McKee M, Britton A. The positive relationship between alcohol and heart disease in eastern Europe: Potential physiological mechanisms. J Roy Soc Med 1998; 91: 402-407.
Address for correspondence: Dr Gerhard Gmel, Swiss Institute for the Prevention of Alcohol and Drug Problems, Avenue de Ruchonnet 14, CH 1003 Lausanne, Switzerland Phone: +41-21-3212959; Fax: +41-21-3212940 E-mail: [email protected]
642
Copyright (c) 2003 Kluwer Academic Publishers