European Journal of Epidemiology 10: 657-664, 1994. © 1994 Kluwer Academic Publishers. Printed in the Netherlands.

The effect of drinking coffee and smoking cigarettes on the risk of c irrhosis associated with alcohol consumpt ion A case - con t ro l s t u d y

G. Corrao 1, A.R. Lepore 2, P. Torchio 2, M. Valenti 2, G. Galatola 3, A. D'Amicis 4, S. Aric6 3, F. di Orio 5 & The Provincial Group for the Study of Chronic Liver Disease* ~ lnstitute of Statistical and Mathematical Sciences, University of Milano, Italy; ~ Epidemiological Research Center, University of L'Aquila, Italy; 3 Division of Gastroenterology, Mauriziano Umberto I Hospital, Torino, Italy; 4 National Nutrition Institute, Rome, Italy; 5 Chair of Clinical Epidemiology, University of L'Aquita, Italy

* Provincial Group for the Study of Chronic Liver Disease: A. Attili (Cattedra di Fisiopatologia Digestiva, Universitd di L'Aquila), S. Santini, F. Bruccoteri & E. Zepponi (Laboratorio Analisi Cliniche, Ospedale Civile Tagliacozzo), G. Tullio & G. Tonietti (Cattedra di Clinica Medica, Universitd di L'Aquila), V. Festuccia, G. Giandomenico & G. Natali (Cattedra di Patologia Medica, Universit~ di L'Aquila), M. Pozone, A. Giusti & F. Caione (Divisione di Geriatria, Ospedale Civile L'Aquila), M. Mariani, A. Grimaldi & A. Iannessi (Divisione di Malattie Infettive, Ospedale Civile L'Aquila), F. Marchionni, G. Del Bove Orlandi & G. Rabitti (Divisione di Medicina Generale, Ospedale Civile Avezzano), G. Sgr6 & S. Cercone (Divisione di Medicina Generale, Ospedale Civile Sulmona), E. Bernardini & P. Capobianchi (Divisione di Medicina Generale, Ospedale Civile Tagliacozzo), M. Giovannone, M. Cincis & P. Caracciolo (Divisione di Medicina Generale, Ospedale Civile Castel di Sangro), L. Colitti & A. Biocca (Divisione di Medicina Generale, Ospedale Civile Pescina), C. Ercole, C. Miccoli & C. Rapone (Scuola di Specializzazione in Patologia Clinica, Universitg~ di L'Aquila), S. Necozione & G. Pantaleo (Centro Interdipartimentate di Epidemiologia, Universitd di L'Aquila)

Accepted in revised form 14 September t994

Abstract. In order to assess the interaction between alcohol intake, tobacco smoking and coffee con- sumption in determining the risk of liver cirrhosis we carried out a hospital-based case-control study involving 115 patients at their first diagnosis of cirrhosis and 167 control patients consecutively enrolled in the General Hospitals of the Province of L'Aquila (Central Italy). The mean life-time daily alcohol intake (as g ethanol consumed daily) was measured by direct patient interviews, whose repro- ducibility was > 0.80 and similar for cases and controls, as checked by interviewing the relatives of a sample of 50 cases and 73 controls. During the same patient's interview we also measured the mean consumption of coffee (daily number of cups of filtered coffee) and tobacco (life-time daily number of cigarettes smoked). A dose-effect relationship on the risk of cirrhosis was present both for alcohol intake - for which the risk was significantly increased above 100 g of daily intake - and for cigarette consumption. The latter did not however

improve the goodness-of-fit of a logistic regression model including alcohol intake as covariate. By contrast, coffee consumption had a protective effect on the risk of cirrhosis and significantly improved the goodness-of-fit of such a model. Abstaining from coffee consumption determined both a significantly increased risk of cirrhosis, even for daily alcohol intake below 100 g, and a multiplicative effect with alcohol intake on this risk. In patients drinking /> 101 g ethanol daily the relative risk increased from 5.5 (95% confidence interval: 1.4-22.0) for coffee consumers to 10.8 (95% confidence interval: 1.3-58.1) for coffee abstainers. We conclude that: (t) tobacco smoking is likely to be a faint risk factor for cirrhosis, and studies on wider patients series should be performed for confirmation; (2) coffee drinking is associated with a reduced risk of cir- rhosis. Whether coffee contains some hitherto unknown protective substances, or is just a marker of other life-style or dietary protective factors, deserves further clarification.

Key words: Alcohol consumption, Cirrhosis, Coffee consumption, Tobacco smoking

Introduction

Results from studies in animal models have originally supported the view that alcoholic liver cirrhosis

directly develops on the basis of a dose-dependent toxicity of ethanol [17, 27]. Subsequently, this hypothesis has been questioned [1] and evidence has been given for the existence of other associated

658

factors which could modulate the risk of developing liver cirrhosis due to alcohol intake [29]. In this respect, chronic infections due to hepatitis B virus (HBV) [8, 21] and hepatitis C virus (HCV) [6, 20] as well as nutritional factors [22] have been high- lighted as risk factors for alcoholic cirrhosis, although their actual role in modifying the effect of alcohol is still controversial [3, 9, 11].

Less evidence is available for the existence of a significant risk associated with tobacco smoking [4, 16] and for a protective effect of coffee con- sumption, which has recently been hypothesized [ 16].

Thus, there is a consensus on the need of more extensive and accurate studies which should clarify the role of factors interacting with alcohol con- sumption on the causation of liver cirrhosis [15, 29].

The aim of the present investigation was therefore to assess the interaction between alcohol consump- tion, tobacco smoking and coffee consumption in determining the risk of liver cirrhosis, using a hospital-based case-control study in Italian patients.

Subjects and methods

Selection of cirrhotic cases. We enrolled 135 patients consecutively admitted to the Medical Departments of six District hospitals in the province of L'Aquila (Central Italyy during the period November 1989- May 1990, who received for the first time the diagnosis of liver cirrhosis. All patients presented signs of liver decompensation (ascites, oedema, jaundice) or bleeding from oesophageal varices. Twenty patients were excluded from the study, respectively because of the presence of severe dis- abling hepatic encephalopathy (n -- 7), hepatocellular carcinoma (n -- 8) and primary biliary cirrhosis (n -- 5). In 75 of the remaining 115 patients (65.2%), the diagnosis of cirrhosis was confirmed by percuta- neous liver biopsy. The latter was not performed in 40 patients because of specific medical contraindi- cations.

Selection of control patients. We randomly enrolled 167 patients admitted during the same period and in the same hospitals considered for the enrollment of cases. We excluded patients diagnosed as having diseases known to be associated with hepatitic virus infections, alcohol abuse and cigarette consumption, or chronic diseases likely to alter dietary or life-style habits. Thus the specific exclusion criteria were as follows: (1) admission to orthopaedic, infectious disease, psychiatric or gynaecology/obstetric wards; (2) presence of gastrointestinal, metabolic, tumoral, pulmonary and cardiovascular diseases; (3) abnormal liver function tests (serum albumin, aminotrans- ferases, ~#glutamyl-transpeptidase, bilirubin, alkaline phosphatase, and prothrombin time). The reasons for hospital admission were as follows: minor surgical

interventions (48.5%); acute urinary tract infections (21.0%); ocular or dental non-traumatic diseases (17.9%); ear-nose-and-throat diseases (12.6%).

Data collection. All 282 patients enrolled in this study were interviewed on admission by one of 4 trained interviewers blinded with respect to the aim of the study. The interviews were performed only when reproducible answers were obtained during simulated questionnaires, and were presented as part of the process of recording the patient's medical history, focusing on the life and alimentary habits. A structured questionnaire was used to obtain, infor- mation on social and demographic factors, general characteristics of the patient and his or her life-style habits, including alcohol and coffee intake and tobacco smoking.

Alcohol intake data were obtained after the patient's life was divided in 10-year periods; for each period, the average intake of wine, beer, aperitifs, digestive alcoholic drinks and spirits was separately investigated and expressed as the average daily consumption of ethanol in grams, considering the different concentration of alcohol in each type of alcoholic beverage (12, 5, 18, 30 and 40% respec- tively). The date of beginning and stopping alcohol intake was also recorded. From this information we calculated for each patient the life-time daily alcohol intake (LDAI). Reproducibility of such questionnaire was assessed by interviewing the relatives of the first 50 cirrhotic cases and 73 control patients.

A similar approach was used for measuring tobacco smoking. After dividing the patient's life into 10-year periods, we recorded the average daily number of cigars or cigarettes smoked, or the daily grams of tobacco used by the pipe smokers. Due to the negligible number of cigar- or pipe-smoking subjects, information from these subjects was not used and tobacco smoking was quantified as the average life-time daily number of cigarettes smoked (LDCS).

Coffee intake was determined as the weekly consumption of cups of coffee - which invariably was represented by espresso and moka coffee - during the period immediately preceding hospital admission. Changes in the coffee drinking habit were recorded back for 20 years before the onset of health problems that brought the patient to hospital admis- sion; however, after collection of the data, it was evident that these changes were negligible and unfrequent, being also similar between casks and controls. Thus only information on the most recent coffee intake was used in this study.

The serological patterns of HBV and hepatitis delta virus (HDV) infections were obtained by commercial immune-enzymatic assays; serum anti- HCV antibodies were detected by a commercial second generation enzyme-linked immunosorbent assay (ELISA; Ortho-Diagnostic, New Jersey, USA).

659

Data analysis. The frequencies of LDAI, LDCS and daily coffee intake were divided into ordinal cate- gories (0, 1-20, 21-100 and /> 101 g/ethanol for LDAI; 0, 1-20 and 1> 21 cigarettes for LDCS; 0, 1 and/> 2 cups/day for coffee intake).

The reproducibility of alcohol intake was measured by the estimation of the intraclass correla- tion coefficient of reliability (R) using a random effect model of the analysis of variance [13]. R indicates the proportion of the variance of alcohol categories attributable to the between-patients variance; 1-R indicates the proportion attributable to other sources (i.e., alcohol category disagreement on the alcohol intake category between the patient and his or her relative).

The risk of cirrhosis associated with each category of the three variables considered and with sero- positivity for HBV surface antigen (HBsAg) and anti-HCV was estimated according to the classical method of unmatched data and expressed as odds ratio (OR) and the corresponding 95% confidence interval (CI). The null hypothesis of the absence of linear trend for the risk of underlying population was tested according to Mantel when appropriate [18].

Subsequently we used various logistic regression models for estimating the association between the risk of cirrhosis and the factors considered. When analysing LDAI, LDCS and daily coffee intake, the 'dummy' variables obtained from the category 'zero' of each factor were considered as the independent variables; from these models we derived the estimates of the maximum likelihood of the OR and the corresponding 95% CI.

The goodness-of-fit of the logistic regression models was measured by the 'G' statistic and the model couplets were compared using the chi-square statistics, which was obtained by subtracting the goodness-of-fit 'G' values of the two models [5]. This approach tests the null hypothesis that the addition of a covariate to a regression model does not improve the goodness-of-fit of the latter.

Finally, we used a logistic regression model where the 'dummy' variables were obtained by crossing the alcohol and coffee intake categories. The corre- sponding OR and 95% CI were referred to the alcohol abstainers consuming coffee. Thus the main effects for LDAI and daily coffee intake were assessed using a multiplicative model of interaction between the two variables.

The observed ORs for drinkers of various amounts of alcohol in the coffee abstainers were compared with those expected on the basis of the additive model. The S index and its 95% CI were used to evaluate the empirical deviation from this model and therefore from the theoretical joint effect on the assumption of independence of the effects of the two factors [26].

Results

Table 1 shows the comparison between the general characteristics of cases and controls. No significant difference was observed for age and sex, whereas the prevalence of HBsAg seropositivity was significantly higher in the cases (OR --- 23.0; CI --- 4.9-107.8). Three anti-HDV positive and one HBeAg positive patients were present in the HBsAg positive cases. The prevalence of anti-HCV positivity was higher in the cases (OR = 8.7; CI -- 4.3-17.6), in whom six of the 38 anti-HCV positive cases were also HBsAg positive.

All subjects agreed to answer to the alcohol, smoke and coffee questionnaire. The intraclass correlation coefficient between the alcohol data col- lected from the relatives and the respective patients was 0.85 (95% CI -- 0.77-1.00) for cases, and 0.82 (95% CI = 0.76-1.00) for controls.

Both the mean LDAI and LDCS were significantly higher in cases, among whom there were also fewer abstainers from the respective habits. The propor- tion of patients who stopped smoking or drinking alcohol before their present hospital admission was similar in the two groups. By contrast, the same mean daily coffee intake was observed in cases and controls and no difference was appreciable between the two groups with respect to those who abstained from coffee drinking before admission.

Table 2 shows the relationship between the risk of cirrhosis and the life-style variables considered. A significant positive trend was observed for LDAI and LDCS: the OR values reached the conventional statistical significance in the last categories (>~ 101 g/day for LDAI and t> 21 cigarettes/day for LDCS). A faint and not significant negative trend was observed for the daily coffee intake, with decreased risk at increased coffee intake values.

Table 3 shows a dose-dependent relationship between the alcohol intake categories and the risk of liver cirrhosis, whether or not an adjustment for HBsAg or anti-HCV seropositivity was made. The comparison of the 'G' statistics showed that the inclusion of HBsAg and anti-HCV significantly improved the goodness-of-fit (AG = 51.21; p < 0.05), suggesting that seropositivity for these markers modifies the risk of cirrhosis associated with alcohol intake.

Table 4 shows the results of the analysis of the risk associated with the three life-style habits according to the logistic regression models. The first model assumes that alcohol intake and cigarette smoking are the covariates, and shows that even if the effect of these two factors is separately considered, a positive association still exists between the risk of cirrhosis and LDCS although in none of the LDCS category the OR has significantly increased. However, by comparing the 'G' statistics obtained from the second model of Table 3 with this model (AG -- 2.06;

660

Table 1. General characteristics of the sample

Cases Controls Test (n = 115) (n = 167) p value

Age (years) Mean + sd 58.1 _+ 15.7 60.4 + 15.4 t = 1.20 p = 0.23

Sex Males N (%) 78 (67.8) 100 (59.9) Z~ = 1.85 Females N (%) 37 (37.2) 67 (40.1) p = 0.17

HBsAg Positive N (%) 14 (12.2) 1 (0.6) X2 = 18.12 Negative N (%) 101 (87.8) 166 (99.4) p < 0.001

Anti-HCV Positive N (%) 38 (33.0) 9 (5.4) X 2 = 37.50 Negative N (%) 77 (67.0) 158 (94.6) p < 0.001

LDAI (g/day) a Mean + sd 95.4 + 108.2 50.3 +_ 69.3 t = 3~.95 p = 0.0001

Abstainers N (%) 9 (7.8) 25 (15.0) ~2 = 3.31 Previous drinkers N (%) 5 (4.3) 6 (3.6) p = 0.19 Present drinkers N (%) 101 (87.8) 136 (81.4)

LDCS b M e a n + s d 11.8+12.3 8 .7+ 11.7

No smokers N (%) 45 (39.1) 81 (48.5) Previous smokers N (%) 5 (4.3) 7 (4,2) Present smokers N (%) 65 (56.5) 79 (47.3)

Coffee (cups/day) Mean + sd 1.5 + 2.2 1.5 + 1.7

No coffee consumers N (%) Previous coffee consumers N (%) Coffee consumers N (%)

28 (24.3) 32 (19.2) 2 (1.7) 2 (1.2)

85 (73.9) 133 (79.6)

t = 2.16 p = 0.03

~z = 2.48 p = 0.29

t = 0.30 p = 0.77

X ~= 1.29 p -- 0.52

a LDAI = Lifetime daily alcohol intake. b LDCS = Lifetime daily cigarettes smoked.

Table 2. Dose-effect relationship between alcohol intake, cigarette smoking, coffee consumption and risk of liver cir- rhosis

Cases Controls OR a CI b N (%) N (%)

LDAP 0 ~ 9 (7.8) 25 (15.0) 1.0 1-20 19 (16.5) 41 (24.6) 1.3 21-100 45 (39.1) 76 (45.5) 1.6 1> 101 42 (36.5) 25 (15.0) 4.7*

LDCS ~ 0 r 45 (39.1) 81 (48.5) 1.0 1-20 38 (33.0) 56 (33.5) 1.2 21-100 32 (27.8) 30 (18.0) 1.9"

Coffee 0 c 28 (24.4) 32 (19.2) q .0 (cups/day) 1 49 (42.6) 75 (44.9) 0.7

t> 2 38 (33.0) 60 (35.9) 0.7

0.5 - 3.3 X z = 14.76 0.7 - 3.8 p = 0.001 1 . 0 - 1 1 . 3

0.7 - 2 . 1 X 2 ~ n a = 4.08 1.0 - 3.6 p = 0.043

0.5 - 1.6 ~2 t r en d = 0.82 0.4 - 1.4 p -- 0.367

a Crude odds ratio. b 95% confidence interval of OR. c Reference category. d Lifetime daily alcohol intake.

Lifetime daily cigarettes smoked. * p < 0.05.

Table 3. Risk of cirrhosis associated with alcohol intake

LDAI b Models ~ (g/day)

1 2

OR 95% CI OR 95% CI

0 c 1.0 - 1.0 - 1-20 1.3 (0.5-3.3) 1.9 (0.6-5.8) 21-100 1.8 (0.8-4.4) 2.5 (0.9-7.2) /> 101 5.7* (2.1-15.4) 7.8* (2.4-25.1)

G statistic -359.753 -308.548

a The Odds ratio (OR) and the corresponding 95% confidence interval (95% CI) from two models of multiple logistic regression were adjusted for sex and age. The two models considered (1) alcohol intake and (2) alcohol intake, HBsAg and anti-HCV status. b Lifetime daily alcohol intake.

Reference category. * p < 0.05.

p > 0.05) it is evident that LDCS does not improve the goodness-of-fi t o f the model, implying that cigarette smoking has no significant effect on the risk of cirrhosis and does not modify the effect of LDAI on this risk.

The second logistic regression model considers alcohol and coffee intake as covariates and shows that correcting the effect of the latter by the former

661

results in a negative association between coffee intake and risk of cirrhosis; in the last coffee intake category the OR was significantly lower than the unity. Furthermore the comparison of the 'G ' statis- tics of the second model in Table 3 and this model (AG = 7.88; p < 0.05) shows that the inclusion of the coffee intake categories significantly improves the goodness-of-fit of the model.

The third model considers all three variables as covariates and confirms that the addition of coffee intake to a model including alcohol intake and cigarette smoking determines an improvement of the goodness-of-fit of the model (G statistic of the third model subtracted from that of the first model: AG = 7.19; p < 0.05). Conversely the addition of cigarette smoking to a model including alcohol and coffee intake does not modify the goodness-of-fi t of the model (G statistic of the third model subtracted from that of the second model: AG = 1.37; p > 0.05).

Finally, Table 5 shows the combined risk associ- ated with LDAI and daily coffee intake: in this analysis the second and third LDAI categories have been grouped together and coffee intake was dichotomised (yes vs. no) due to the resulting fewer number of patients in each category. A dose-depen- dent positive association was found between LDAI and the risk of cirrhosis, both in coffee drinkers and coffee abstainers. With the exception of teetotallers, the OR values were higher in patients who did not drink coffee than in those who did. Furthermore, in

Table 4. Risk of cirrhosis associated with alcohol intake, cigarette smoking and coffee consumption

Models a

1 2 3

OR 95% CI OR 95% CI OR 95% CI

LDAI b 0 d 1.0 - 1.0 - 1.0 - (g/day) 1-20 1.8 (0.6-5.6) 2.2 (0.7-6.8) 2.t (0.7-6.5)

21-100 2.5 (0.9-7.1) 2.9* (1.0-8.6) 2,8* (1.0-8.4) /> 101 7.0* (2.1-22.8) 9.5* (2.8-31.9) 9.5* (2.5-28.7)

LDCS ~ 0 a 1.0 - 1.0 - 1-20 1.1 (0.5-2.3) 1.2 (0.5-2.5) t> 21 1.8 (0.7-4.2) 1.9 (0.8-4.6)

Coffee 0 d 1.0 - 1.0 - (cups/day) 1 0.6 (0.3-1.2) 0.6 (0.2-1.1 )

/> 2 0.6* (0.3-1.0) 0.5* (0.3-1.0)

G statistic -306.486 -300.668 -299.296

The Odds ratio (OR) and the corresponding 95% confidence interval (95% CI) from three models of multiple logistic regression were adjusted for sex, age, and HBsAg and anti-HCV status. The three models considered (1) alcohol intake and cigarette smoking, (2) alcohol intake and coffee consumption, (3) alcohol intake, cigarette smoking and coffee consumption. b Lifetime daily alcohol intake.

Lifetime daily cigarettes smoked. d Reference category. * p < 0.05.

662

Table 5. Interaction between alcohol intake and coffee consumption on the risk of liver cirrhosis

LDAI ~ Coffee Cases Controls Observed Expected ~ (g/day) consumption N (%) N (%) OR

OR b CI ~

S f

(95% CI)

0 d Yes 6 (5.2) 15 (9.0) 1.0 d No 3 (2.6) 10 (6.0) 0.7 (0.1-3.6)

1-100 Yes 47 (40.9) 98 (58.7) 1.6 (0.5-5.6) No 17 (14.8) 19 (11.4) 3.7* (1.0-14.7)

101 Yes 34 (29.6) 22 (13.2) 5.5* (1.4-22.0) No 8 (7.0) 3 (1.8) 10 .8" (1.3-58.1)

1.3 9.0 (0.9-27.2)

5.2 2.3* (t.2-12.3)

a Lifetime daily alcohol intake. b Odds Ratio (OR) estimated from the multiple logistic regression analysis with adjustment for age, sex, smoking and HBsAg and anti-HCV status.

95% confidence intervals of ORs. d Reference category.

Odds Ratio expected according to the additive model. f S index according to Rothman [26]. * p < 0.05.

the intermediate LDAI category (1-100 g) the OR reached the statistical significance only in patients who did not drink coffee. In the latter patients the OR values were higher than expected according to the additive model, as also supported by the S indices that were above 1. However, only for the category of LDAI I> 101 g and coffee abstainers the null hypothesis of the additive structure of the joint effect of the two factors can be rejected.

Discussion

The accurate collection of information for imple- menting case-control studies on the risk of cirrhosis due to alcohol intake depends on the reliability of patient interviews on their life-time alcohol habits. This may be poor, due to the fallibility of memory or to a tendency to underestimate, whether purposely or not, their alcohol intake [24]. Our results have shown a good agreement on the patient's alcohol intake measured blindly from information obtained by the direct patient interview and one of his or her relatives, without differences between cirrhotic cases and control patients, this suggests that our study is scarcely affected by a differential misclassification of alcohol intake.

As expected, there was an evident dose-effect relationship between alcohol intake and risk of cir- rhosis, which was in the similar order of magnitude to that already reported in other case-control studies from our group [9, 10], but far lower than that reported by Norton et al. [23] and Pequignot et al. [25]. This may be due to different methods of com- puting daily alcohol intake, since these authors considered an approximate estimate of alcohol consumption in a period just preceeding hospital

admission. This approach was likely to measure the effect of alcohol in determining the symptoms that brought the patient to hospital admission, rather than the effect of alcohol on causing cirrhosis [29]. Results similar to our have been reported in a cohort study on liver patients originally free from cirrhosis [28].

The main aim of our study was to assess whether tobacco smoking and coffee drinking interact with the effect of alcohol intake in determining cirrhosis, considering HBV and HCV infection as possible confounding variables. In this respect, our data confirmed that both viral infections have a role in causing cirrhosis, which was independent of alcohol intake.

To our knowledge, only two epidemiological studies have attempted to measure the risk of cir- rhosis due to tobacco smoking. Bouliere et al. could not show any association [4], whereas Klatsky & Armstrong found a positive dose-effect relationship similar to that observed in our study. We have however shown that this relationship was not signif- icant nor independent of alcohol consumption: this observation may however be biased by a scarce potency of this study with respect to data on smoking habits and needs to be confirmed by investigation of wider scale.

Conversely, we observed a strong inverse rela- tionship between coffee intake and risk of cirrhosis, similar to that reported by Klatsky & Armstrong [16]: although this effect did not appear to be independent of alcohol intake, our data suggest the existence of a protective role of coffee on the liver damage due to alcohol intake. We performed our study using an accurate questionnaire on life-style habits in patients free from symptoms due to their liver disease until observed for enrolment in this study, thus our results

663

are unlikely to be conditioned by modification of alcohol and coffee intake as an effect of the symp- tomatic progression of liver disease. However alcohol and cigarette consumption was measured throughout the patients' life, whereas for coffee consumption we only considered data from the most recent period. This may have determined a bias, had changes in coffee consumption recently occurred in cirrhotic patients due to the presence of abdominal symptoms, such as dyspepsia. According to the way we collected data, we cannot exclude the occurrence of such a bias. However, we found no differences between the proportion of cases and controls who had decreased coffee consumption both recently and during the last 20 years. This should reasonably minimize such a bias.

Caution should be however used in interpreting our data, due to the fact that the number of coffee abstainers in both the cirrhotic and control groups at the high alcohol consumption level was small.

An explanation of the protective effect of coffee consumption on the risk of cirrhosis due to alcohol intake may be that coffee intake is a marker of other life-style or alimentary habits, which act themselves as protective factors. However, this hypothesis is difficult to reconcile with the observation that our results in Italian patients are similar to those obtained in American patients, who have a considerably dif- ferent cultural and dietary pattern [16]. More data on the effect of the life-style pattern should be however available in this respect.

Alternatively our results imply the presence among the many substances contained in coffee [32] of one or more component(s) that may have a still unknown protective effect on the liver against the alcoholic damage. Caffein is the most important and pharmacological active substance contained in coffee, but it has no apparent direct hepato-protection effect. It is metabolized within the hepatocyte by the micro- somal enzyme oxydative system and its elimination could be impaired in patients with liver disease [12, 30]. Thus, the hypothesis that higher blood caffeine levels in cirrhotics might reduce the necessity of drinking coffee and therefore explain our findings should be tested. Alcohol, as well as other dietary components, have been shown to modify caffein elimination [14, 19]. Aromatic compounds such as phenolic acids may interact with metabolic pathways of the liver, similarly to what could be the case for other substances, such as the still unknown factor(s) determining an increase in the serum cholesterol levels, contained in the lipid fraction of coffee boiled according to the Northern European brewing fashion, but not in filtered coffee [2, 31 ].

The possibility that substances contained in coffee may exert a protective effect on the hepatocyte has recently been underlined by the observation that liver enzimes tended to decrease at increased coffee consumption [7].

In conclusion, our data suggest the existence of a protective effect of drinking Italian style coffee on the risk of cirrhosis due to alcohol intake. We believe that this finding deserves further confirmation and accurate epidemiological studies with a larger sample size performed. The probability that factors protec- tive against liver damage are contained in coffee should stimulate the research in this field. For the time being, there is not enough evidence for justi- fying the prescription of coffee to alcoholic patients in order to protect them against a severe progression of chronic liver disease.

References

1. Ainley CC, Senapati A, Brown IMH, Iles CA, et al. Is alcohol hepatotoxic in the haboon? J Hepatol 1988; 7: 85-92.

2. Bak AAA, Grobee DE. The effect on serum choles- terol levels of coffee brewed by filtering or boiling. N Engl J Med 1989; 321: 1432-1437.

3. Batey RG, Burns T, Benson RJ, et al. Alcohol con- sumption and the risk of cirrhosis. Med J Aust 1992; 156: 413-416.

4. Bouliere M, Barthet M, Berthezene P, et al. Is tobacco a risk factor for chronic pancreatitis and alcoholic cirrhosis? Gut 1991; 32: 1392-1395.

5. Breslow NE, Day NE. Statistical methods in cancer research, Vol. 1: The analysis of case-control studies. IARC Scientific Publication No. 32. Lyon, 1980.

6. Brillanti S, Barbara L, Miglioli M, Bonino F. Hepatitis C virus: A possible cause of chronic hepatitis in alcoholics. Lancet 1989 (ii): 1390-1391.

7. Casiglia E, Spolaore P, Ginocchi G, Ambrosio GB. Unexpected effect of coffee consumption on liver enzymes. Eur J Epidemiot 1993; 9: 293-297.

8. Chung HT, Lai CL, Wu PC, et al. Synergism of chronic alcoholism and hepatitis B infection in liver disease. J Gastroenterol Hepatol 1989; 4: 11-16.

9. Corrao G, Aric6 S, Carle F, et al. A case-control study on alcohol consumption and the risk of chronic liver disease. Rev Epidemiol Sant6 Publ 1991; 39: 333-343.

10. Corrao G, Hric6 S, Lepore AR, Valenti M, Torchio P, Galatota G, Tabone M, di Orio F. Amount and duration of alcohol intake as risk factors of sympto- matic liver cirrhosis: A case-control study. J Clin Epidemiol 1993; 46: 601-607.

11. Corrao G, Carle F, Lepore AR, et at. Interaction between alcohol consumption and positivity for hepatitis C virus on the risk of liver cirrhosis: A case- control study. Eur J Epidemiol 1992; 8: 634-639.

12. Desmond PV, Patwardhan RV, Johnson RF, Schenker S. Impaired elimination of caffeine in cirrhosis. Dig Dis Sci 1980; 25: t93-197.

13. Fleiss JL. The design and analysis of clinical experi- ments. New York: John Wiley and Sons, 1986.

14. George J, Murphy T, Roberts R, Cookstey WGE, Halliday JW, Powelt LW. Influence of alcohol and caffeine consumption on caffeine elimination. Clin Exp Pharmac Physiol 1986; 13:731-736.

15. Grant BF, Dufour MC, Hartford TC. Epidemiology of

664

alcoholic liver disease. Semin Liver Dis 1988; 8: 12-23.

16. Klatsky AL, Armstrong MA. Alcohol, smoking, coffee and cirrhosis. Am J Epidemiol 1992; 136: 1248-1257.

17. Leiber CS, Decarli LM, Rubin E. Sequential produc- tion of fatty liver, hepatitis, and cirrhosis in subhuman primates fed hetanol with adequate diets. Proc Nat Acad Sci USA 1975;72: 437-441.

18. Mantel N. Chi-square tests with one degree of freedom: Extensions of the Mantel-Haenszel proce- dure. J Am Statist Ass 1963; 58: 690-700.

t 9. Mitchell MC, Hoyumpa AM, Schenker S, Johnson RF, Nichols S, Patwardhan RV. Inibition of caffeine elimination by short-term ethanol administration. J Lab Clin Med 1983; 101: 826-834.

20. Nalpas B, Driss F, Pol S, et al. Association between HCV infection in hepatocellular carcinoma and alcoholic liver disease. J Hepatol 1991; 12: 70-74.

21. Namura H, Kashiwagi S, Hayashi J, et al. An epi- demiologic study of effects of alcohol in the liver in hepatitis B surface antigen carriers. Am J Epidemiol 1988; 128: 277-284.

22. Nanji AA, French SW. Relationship between pork consumption and cirrhosis. Lancet 1985 (i): 681-683.

23. Norton R, Batey R, Dwyer T. Alcohol consumption and the risk of alcohol related cirrhosis in women. Br Med J 1987; 295: 80-82.

24. Orrego H, Blendis LM, Blake JE. Unreliability of assessment of alcohol intake based on personal inter- views in a liver clinic. Lancet 1979 (ii): 1354-1356.

25. Pequignot GU, Tuyns AJ, Berta JL. Ascitic cirrhosis in relation to alcohol consumption. Int J Epidemiol 1978; 7:113-120.

26. Rothman KJ. The estimation of synergy or anta- gonism. Am J Epidemiol 1976; 103:506-51 t.

27. Rubin E, Lieber CS. Fatty liver, alcoholic hepatitis and cirrhosis produced by alcohol in primates. N Engl J Med 1974; 290: 128-135.

28. Sorensen TIA, Orholm M, Benzen KD. Prospective evaluation of alcohol abuse and alcoholic liver injury in man as predictors of development of cirrhosis. Lancet 1984 (ii): 241-244.

29. Sorensen TIA. Alcohol and liver injury: Dose-related or permissive effect? Liver 1989: 189-197.

30. Statland BE, Demas TJ. Serum caffeine half-lives. Healthy subjects vs patients having alcoholic hepatic disease. Am J Clin Path 1980; 73: 390-393.

31. van Dusseldorp M, Katan MB, Demacker PNM. Effect of decaffeinated versus regular coffee on serum lipoproteins. Am J Epidemiol 1990; 132: 33-34.

32. Viani R. The composition of coffee. In: Garattini S, ed. Caffeine, coffee and health. New York: Raven Press, 1993: 17-41.

Address for correspondence: Dr Anna Raffaella Lepore, Centro Interdipatimentale di Epidemiologia, Facolt~ di Medicina, Via Giuseppe Verdi 28, 1-67100 L'Aquila, Italy Phone: +39 862 433 484; Fax: +39 862 433 482