R E V I E W
Drug and Alcohol Review (July 2010), 29, 437–445 DOI: 10.1111/j.1465-3362.2009.00153.x
REVIEW
Alcohol as a risk factor for liver cirrhosis: A systematic review and meta-analysis dar_153
437..445
JÜRGEN REHM1,2,3,4, BENJAMIN TAYLOR1,2, SATYA MOHAPATRA1, HYACINTH IRVING1, DOLLY BALIUNAS1,2, JAYADEEP PATRA1 & MICHAEL ROERECKE1,2 1
Centre for Addiction and Mental Health,Toronto, Canada, 2Dalla Lana School of Public Health, University of Toronto, Toronto, Canada, 3TU Dresden, Dresden, Germany, and 4Department of Psychiatry, University of Toronto,Toronto, Canada
Abstract Introduction and Aims. Alcohol is an established risk factor for liver cirrhosis. It remains unclear, however, whether this relationship follows a continuous dose–response pattern or has a threshold.Also, the influences of sex and end-point (i.e. mortality vs. morbidity) on the association are not known. To address these questions and to provide a quantitative assessment of the association between alcohol intake and risk of liver cirrhosis, we conducted a systematic review and meta-analysis of cohort and case–control studies.Design and Methods. Studies were identified by a literature search of Ovid MEDLINE,EMBASE,Web of Science, CINAHL, PsychINFO, ETOH and Google Scholar from January 1980 to January 2008 and by searching the references of retrieved articles. Studies were included if quantifiable information on risk and related confidence intervals with respect to at least three different levels of average alcohol intake were reported. Both categorical and continuous meta-analytic techniques were used to model the dose–response relationship.Results. Seventeen studies met the inclusion criteria.We found some indications for threshold effects.Alcohol consumption had a significantly larger impact on mortality of liver cirrhosis compared with morbidity. Also, the same amount of average consumption was related to a higher risk of liver cirrhosis in women than in men. Discussion and Conclusions. Overall, end-point was an important source of heterogeneity among study results.This result has important implications not only for studies in which the burden of disease attributable to alcohol consumption is estimated, but also for prevention. [Rehm J,Taylor B, Mohapatra S, Irving H, Baliunas D, Patra J, Roerecke M. Alcohol as a risk factor for liver cirrhosis: A systematic review and meta-analysis. Drug Alcohol Rev 2010;29;437–445] Key words: alcohol, liver cirrhosis, mortality, morbidity, meta-analysis. Introduction The causal impact of alcohol on liver disease and cirrhosis in particular has long been known. As early as in 1785, Benjamin Rush, in his famous Inquiry into the Effects of Ardent Spirits upon the Human Body and Mind, listed liver conditions as the second ‘usual’ disease consequence of habitual drinking of spirits and compared the effects of alcohol on the liver with the preying on Prometheus’ liver by the vulture [1]. Liver cirrhosis is the most important single fatal chronic disease condition caused by alcohol consumption globally, with approximately 15% of all alcohol-attributable deaths in
2004 were as a result of liver cirrhosis [2]. Thus, it comes as no surprise that liver cirrhosis has always been included in standard alcohol-attributable disease lists [3,4]. The empirical literature is likewise unequivocal— several meta-analyses have assessed the association between alcohol consumption and liver cirrhosis [5–8] and all reached a general consensus that there is convincing evidence that heavy alcohol consumption is associated with increased risk of liver cirrhosis. However, several questions remain. First is the exact nature of the relationship, that is, whether there is a continuous dose–response relationship or a threshold is
Jürgen Rehm PhD, Professor (CAMH—Centre for Addiction and Mental Health), Benjamin Taylor MSc, Research Analyst, Satya Mohapatra PhD, Research Analyst, Hyacinth Irving MA, Research Method Specialist, Dolly Baliunas MSc, PhD Student, Jayadeep Patra PhD, Scientist, Michael Roerecke MSc, PhD Student. Correspondence to Prof Dr Jürgen Rehm, Centre for Addiction and Mental Health, 33 Russell Street Room 2035, Toronto, ON, Canada M5S2S1. Tel: +416 5358501 ext. 6378; Fax: +416 2604156; E-mail:
[email protected] Received 22 May 2009; accepted for publication 22 August 2009. © 2010 Australasian Professional Society on Alcohol and other Drugs
438
J. Rehm et al.
not clear. Results from the classic study of Lelbach and the meta-analyses by Corrao et al. [5–7,9] suggest a continuous curve of increasing risk of liver cirrhosis with increasing volume of alcohol consumption without any evident threshold. Corrao et al. found a significant increase in the risk of liver cirrhosis at 25 g of pure alcohol per day compared with abstainers and overall a monotonic, almost exponential increase of relative risks (RR) for the range of consumption considered [6]. On the other hand, a Danish group [10] concluded, that there was a threshold seen at a consumption level of greater than five drinks (60 g of pure alcohol) on average per day. People drinking that heavily had a 27-fold increased mortality from alcoholic cirrhosis in men and a 35-fold increased mortality from alcoholic cirrhosis in women compared with the Danish general population. However, there was no further dose– response relationship and no additional risk for drinking a much higher volume than 60 g day-1. The second unanswered question concerns the influence of sex on the association between alcohol consumption and risk of liver cirrhosis. With the exception of one meta-analysis by Corrao et al. [5], previous meta-analyses have presented their findings mainly undifferentiated by sex. English et al. [8], for example, found more than a ninefold increase in the risk of liver cirrhosis for more drinking more than 40 g of pure alcohol in both sexes compared with abstainers, indicating no gender difference in RR for heavy consumption categories. Nevertheless, they did state different risks for men and women drinking between 20 and 40 g of pure alcohol per day. The third question is whether the risk of liver cirrhosis varies by the end-point under consideration, that is, morbidity versus mortality. In the meta-analyses by Corrao et al. [5–7] there was a high degree of heterogeneity among the studies, possibly indicating that the end-point was an important source of variation among study results. However, to our knowledge, no systematic analysis has been conducted to investigate this question. However, there are biological reasons why liver cirrhosis mortality should be more strongly affected by drinking alcohol (compared with morbidity). Alcohol drinking, especially heavy consumption, has been shown to worsen existing liver disease considerably and to have detrimental effects on the immune system, thus negatively affecting the course of existing liver disease and increasing the chance of death [11–15]. In the present study, we aimed to address these three unanswered questions and to quantify the risk of liver cirrhosis associated with increasing alcohol consumption based on an updated systematic review and metaanalysis of observational studies. The search for the last meta-analysis published in 2004 [6] was conducted © 2010 Australasian Professional Society on Alcohol and other Drugs
more than 10 years ago, and we included seven studies not previously included in the former study. In particular, we wanted to test for a threshold effect by conducting a meta-analysis on all study results with less than 12 g and between 12 and 24 g of pure alcohol per day and quantify the alcohol–liver cirrhosis relationship by sex and by end-point. The results will not only be used in the Comparative Risk Assessment for alcohol within the Global Burden of Disease 2005 Study [16], but also will have implications for public health measures, such as guidelines for low risk drinking. Methods Case definition Liver cirrhosis is a chronic disease of the liver characterised by the replacement of normal tissue with fibrous tissue and the loss of functional liver cells. Cirrhosis results from permanent damage or scarring of the liver that leads to a blockage of blood flow through the liver and prevents normal metabolic and regulatory processes. Cirrhosis can also lead to an inability of the liver to perform its biochemical functions (http://www.liver. ca/Liver_Disease/Adult_Liver_Diseases/Cirrhosis_of_ the_Liver.aspx; accessed 11 November 2008). Only studies with clinical defined assessment of morbidity and death certificates for mortality were included (i.e. no, self-reports). Depending on the different periods in which the data were collected, several different revisions of the International Classification of Diseases (ICD) were used to define liver cirrhosis. The codes used for potential inclusion were: 581 in ICD7 and 571 in ICD8 and ICD9, and K70, K74 in ICD10. We found no studies based on ICD10. Literature search We performed a literature search for studies in humans of the association between alcohol consumption and liver cirrhosis in multiple electronic bibliographic databases from January 1980 to January 2008, including: Ovid MEDLINE, EMBASE, Web of Science, CINAHL, PsychINFO, ETOH and Google Scholar. The search was conducted using any combination of the key words: alcohol, alcohol consumption, alcohol intake, heavy drinking, liver diseases and liver cirrhosis. In addition, we manually reviewed the content pages of the major epidemiological journals and the reference lists of relevant and review articles. No language restrictions were imposed. Inclusion and exclusion criteria Studies were included in the meta-analysis if: (i) they had a case–control or cohort design, that is, a stronger
Alcohol and liver cirrhosis
level of control than with a cross-sectional study; (ii) hazard ratios, RR or odds ratios and their 95% confidence intervals (CIs) (or information allowing us to compute them) were reported; (iii) the end-point was liver cirrhosis morbidity and/or mortality as defined above; (iv) three or more categories of alcohol consumption were reported, one of them being abstention; and (v) clinical assessment of morbidity and mortality (the latter via death certificates). Studies were excluded if: they were not published as full reports, such as conference abstracts and letters to editors; a crosssectional design was used; a continuous measure or only two categories of alcohol consumption were included. If multiple published reports from the same study cohort were available, the one with the most comprehensive data on alcohol consumption was included. Data extraction Information from the identified studies was independently extracted by two investigators. A third investigator extracted data from a random sample of 15 articles to assess inter-rater reliability of the process. Of the 710 data points extracted for the reliability, 95.6% agreement was obtained. Using a standardised spreadsheet, the following data were extracted from each study: title, authors, year of publication, year of study, sample size, country, region, ethnicity, age, sex, end-points, adjustments, design of the study baseline, methods of interview, time period of alcohol consumption, number of beverages, patterns of drinking and the RR with corresponding 95% CIs for each category of alcohol consumption. Throughout this paper, the term RR is used to describe risk estimates, including odds ratios, RRs or hazard ratios. If a study only reported RRs relative to current abstention, the RR and its 95% CI was adjusted to lifetime abstention. To do this, all studies reporting RRs of former drinkers compared with lifetime abstainers were grouped together, and a pooled RR for former drinkers was estimated. In addition, the ratio of former drinkers to lifetime abstainers in this pooled estimate was calculated. For those studies reporting only current abstention, the proportion of lifetime abstainers and former drinkers was estimated based on the ratio of former drinkers previously calculated and the RRs in each article were adjusted based on the pooled RR.This was done to avoid the ‘sick-quitter effect’ [17], a situation where previously heavy drinkers who have stopped drinking because of health reasons confound the true RR for abstainers, thus artificially increasing lifetime abstainers’ risk for liver cirrhosis. If studies only reported results for both sexes combined, the same results were applied to both female and male datasets. Similarly, if combined results were
439
reported for mortality and morbidity studies, the same results were applied to both mortality and morbidity datasets. Sensitivity analyses were conducted to check if this procedure resulted in biased results compared with restricting the analyses to ‘pure’ categories, and excluding studies with ‘mixed’ categories. Standardisation of alcohol consumption Information on alcohol consumption was extracted and, if not already presented as such, converted to grams per day based on the type of alcohol and the size of a standard drink in the study’s country of origin. In cases where the standard drink was not 12 g of pure alcohol, we used the International Monitoring Guide for Monitoring Alcohol Consumption and Related Harm [18] and another overview http://www.icap.org/PolicyTools/ ICAPBlueBook/BlueBookModules/20StandardDrinks/ tabid/161/Default.aspx, last accessed 13 August 2009) to obtain the country-specific standard drink size.When ranges of alcohol were given to define a category, the midpoint was taken. In cases where no upper bound for the highest category existed, 75% of the width of the previous category’s range was added to the lower bound and this measure was used. This procedure had already been used in previous meta-analyses [19,20]. Statistical analysis The dose–response relationship was assessed with random effects meta-regression models. We used the method proposed by Greenland et al. [21,22] to backcalculate and pool study-specific trend estimates. We completed several meta-analyses. In the first, alcohol consumption was first modelled as a continuous variable using the fractional polynomial method [23] to demonstrate the relationship between alcohol consumption and the logarithmised RR of liver cirrhosis. Only first-degree fractional polynomials with powers of -2, -1, -0.5, 0, 0.5, 1, 2 and 3 were evaluated as a monotonic dose–response relationship between alcohol consumption and risk of liver cirrhosis was assumed on biological grounds. Fractional polynomial models are potentially unstable at both tails [24]. Because we were interested in examining whether a threshold exists, that is, a level of alcohol consumption under which the risk of liver cirrhosis was not increased, the potential instability of fractional polynomial models in the left tail was of concern. Therefore, we also did a second meta-analysis in which we modelled alcohol intake using the categories: 0 (reference group), >0–12, >12–24, >24–36, >36– 48, >48–60 and >60 g day-1. We assigned the level of alcohol consumption from each study to these categories based on the calculated midpoint of alcohol © 2010 Australasian Professional Society on Alcohol and other Drugs
440
J. Rehm et al.
Table 1. Characteristics of the included studies (starting with earliest study) Author
Country
Sex
Study design
n cases
Total sample size
End-point
Blackwelder et al. (1980) [29] Klatsky et al. (1981) [30] Gordon and Kannel (1984) [31] Kono et al. (1986) [32] Gordon and Doyle (1987) [33] Garfinkel et al. (1988) [34] Boffetta and Garfinkel (1990) [35] Corrao et al. (1991) [36] Corrao et al. (1993) [37] Fuchs et al. (1995) [38] Becker et al. (1996) [39] Bellentani et al. (1997) [40] Corrao et al. (1997) [41] Thun et al. (1997) [42] Yuan et al. (1997) [43] Becker et al. (2002) [44] Klatsky et al. (2003) [45]
USA USA USA Japan USA USA USA Italy Italy USA Denmark Italy Italy USA China Denmark USA
Male Both Both Male Male Female Male Both Both Female Both Both Both Both Male Both Both
Cohort Cohort Cohort Cohort Cohort Cohort Cohort Case–control Case–control Cohort Cohort Cohort Case–control Cohort Cohort Cohort Cohort
16 51 25 40 15 568 611 184 320 52 124 35 462 446 35 292 108
7 888 8 060 4 747 4 639 1 762 548 185 239 462 369 640 85 709 13 236 6 442 1 113 457 423 18 244 30 630 49 338
Mortality Mortality Mortality Mortality Mortality Mortality Mortality Morbidity Morbidity Mortality Both Morbidity Morbidity Mortality Mortality Both Mortality
consumption. Dummy variables were generated, and they were included in the categorical meta-analysis to assess the possibility of threshold effects across alcohol consumption categories. Statistical heterogeneity among studies was assessed using both the Cochrane Q-test and the I2 statistic [25]. We assessed publication bias using the tests of Egger et al. [26] and Begg and Mazumdar [27]. All statistical analyses were completed for women and men separately using the GLST procedure in stata Version 10.1 (StataCorp LP, College Station, Texas, USA) [28]. Results Characteristics of the included studies We identified 17 studies that met the inclusion criteria outlined in Figure 1. Table 1 summarises the characteristics of the included studies. The majority of studies were conducted in the USA (n = 9), followed by Italy (n = 4) and Denmark (n = 2). China and Japan each contributed one study. Of the 17 studies, seven [31,37,39,41,42,44,45] reported liver cirrhosis endpoints separately for women and men. Only two studies [34,38] reported data for women only, while five studies [29,32,33,35,43] included men only. Three studies [30,36,40] reported undifferentiated by sex. Collectively, the 17 studies provided 12 datasets for women and 15 datasets for men for a total of 1477 887 individuals with 3384 cases of liver cirrhosis. Assessing publication bias Figure 1. Flow diagram describing selection of articles. © 2010 Australasian Professional Society on Alcohol and other Drugs
Among women, the Begg and Mazumdar’s [27] (P = 0.837) and Egger et al.’s [26] tests (P = 0.215)
Alcohol and liver cirrhosis
441
suggested no significant asymmetry of the funnel plot, indicating no evidence of substantial publication bias. A similar finding was also observed among men. The Begg and Mazumdar’s and Egger et al.’s values were 0.767 and 0.196, respectively. Dose–response meta-analyses and potential threshold effects A total of eight first-degree fractional polynomial models were examined separately for women and men. Among women, the best first-degree model with power (0.5) and function b1x0.5 fitted the data significantly better than the linear model with a deviance decrease of 76.12. For men, the linear model (P = 1) and function b1x1 fitted the data best. However, the goodness-of-fit test suggested that marked heterogeneity was still present in the models for both women [Q = 200.59, P ⱕ 0.001, I2 = 72%, 95% CI (63%, 78%)] and men [Q = 305.22, P ⱕ 0.001, I2 = 78%, 95% CI (72%, 82%)]. To assess whether the heterogeneity of study results might be as a result of differences in the study’s endpoint (mortality or morbidity), we created an interaction (product) term between the type of end-point under consideration (1 for mortality and 0 for morbidity) and the dose variable, and we included it in the model for each sex.The interaction term was significant for women (X2 = 7.89, d.f. = 1, P = 0.005) and for men (X2 = 17.06, d.f. = 1, P ⱕ 0.001), suggesting that the effect of alcohol consumption on the risk of liver cirrhosis was different in mortality compared with morbidity studies. Figure 2 displays the results of these analyses for both sexes. Results indicate that, for both sexes, there was a continuous dose–response relationship between alcohol consumption and risk of liver cirrhosis in both mortality and morbidity studies. However, the effect of alcohol consumption was greater for mortality in comparison with morbidity studies for both sexes. In mortality studies, compared with women who were lifetime abstainers, the RRs of liver cirrhosis were 4.9 (95% CI 4.0, 6.2) and 12.5 (95% CI 8.8, 17.7) for those who consumed 24 and 60 g of alcohol per day, respectively. In morbidity studies, relative to women who were lifetime abstainers, those who consumed 24 and 60 g of alcohol per day had RRs of 3.2 (95% CI 2.6, 3.9) and 6.2 (95% CI 4.4, 8.7). Although less pronounced, a similar pattern of effect was observed among men. Table 2 presents the results of the second assessment of the association between the risk of liver cirrhosis and alcohol consumption, with consumption modelled in categories. The results were similar to those reported in Figure 2 in that alcohol consumption had a greater impact on the risk of liver cirrhosis in studies that had mortality as compared with those that reported mor-
Figure 2. Relative risk and meta-regression curve of liver cirrhosis associated with alcohol consumption by sex and end-point.
bidity as the end-point. For all levels of consumption and for both genders the RR for mortality was significantly higher than the RR for morbidity. With respect to sex differences, results in Table 2 revealed that women had higher RRs than men for the same amount of drinking in both mortality and morbidity studies. Whereas in mortality studies the RR for five drinks per day for women was 14.7 (95% CI 11.0, 19.6), for men the RR was 7.0 (95% CI 5.8, 8.5). In morbidity studies, the differences between the sexes were not as pronounced as they were in mortality studies (Table 2). There were, however, some noteworthy differences in the results displayed in Figure 2 and Table 2. For women in morbidity studies, one to two drinks daily had virtually the identical risk as lifetime abstention [RR = 1.0, 95% CI (0.5, 1.9), P = 0.981], while two to three drinks was the lowest category associated with a significantly higher risk [RR = 2.4, 95% CI (1.8, 3.2), P ⱕ 0.001]. Among men in mortality studies, a similar dose threshold was observed. Another difference of note was that, relative to lifetime abstainers, men who consumed one to two drinks per day in morbidity studies had lower risk of liver cirrhosis (RR = 0.3, 95% CI 0.2, 0.4). © 2010 Australasian Professional Society on Alcohol and other Drugs
442
J. Rehm et al.
Table 2. Relative risk (95% CI) of liver cirrhosis associated with alcohol consumption derived from the random effects categorical models by sex and end-point Mortality Alcohol consumption (pure alcohol g day-1) Women >0–12b >12–24b >24–36b >36–48b >48–60b >60b Men >0–12c >12–24c >24–36c >36–48c >48–60c >60c
Morbidity
RR
P-value
(95% CI)
RR
P-value
(95% CI)
Interaction P-valuea
1.9 5.6 7.7 10.1 14.7 22.7
0.013
