GASTROENTEROLOGY 2007;132:1740 –1745

CLINICAL–LIVER, PANCREAS, AND BILIARY

TRACT

Coffee Consumption and Risk of Liver Cancer: A Meta-Analysis

SUSANNA C. LARSSON and ALICJA WOLK

Division of Nutritional Epidemiology, The National Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden
PANCREAS, AND
CLINICAL–LIVER,
BILIARY TRACT

Background & Aims: Mounting evidence indicates coffee on liver carcinogenesis.17 Emerging epidemiologic
that coffee drinking may protect against liver injury evidence also indicates that coffee drinking may reduce
and lower the risk of liver cancer. We quantitatively the risk of primary liver cancer and hepatocellular carci-
assessed the relation between coffee consumption noma (HCC), the dominant form of primary liver cancer.
and the risk of liver cancer in a meta-analysis of Because the epidemiologic evidence on the association
epidemiologic studies. Methods: Relevant studies between coffee consumption and liver cancer risk has not
were identified by searching MEDLINE (from 1966 to yet been summarized, we conducted a meta-analysis to
February 2007) and the reference lists of retrieved quantitatively summarize the results from cohort and
articles. We included cohort and case– control studies case– control studies on this issue. We also investigated
that reported relative risk (RR) estimates with 95% whether the association between coffee drinking and liver
confidence intervals (CIs) of primary liver cancer or cancer differed by history of liver disease.
hepatocellular carcinoma by quantitative categories
of coffee consumption. Study-specific RRs were Materials and Methods
pooled using a random-effects model. Results: Four Study Selection
cohort and 5 case– control studies, involving 2260
Pertinent studies were identified by a computer-
cases and 239,146 noncases, met the inclusion crite-
ized MEDLINE search from 1966 to February 2007 using
ria. All studies observed an inverse relation between
the search term coffee combined with hepatocellular carci-
coffee consumption and risk of liver cancer, and in 6
noma, liver cancer, or liver neoplasm. Furthermore, we re-
studies the association was statistically significant.
viewed citations from retrieved articles to search for more
Overall, an increase in consumption of 2 cups of
studies. No language restrictions were imposed.
coffee per day was associated with a 43% reduced risk Studies were included in the meta-analysis if (1) they
of liver cancer (RR, 0.57; 95% CI, 0.49 – 0.67). There had a cohort or case– control design; (2) the exposure of
was no statistically significant heterogeneity among interest was coffee consumption; (3) the outcome of
studies (P ⴝ .17). In stratified analysis, the summary interest was primary liver cancer or HCC; and (4) relative
RRs of liver cancer for an increase in consumption of risk (RR) estimates (odds ratios in case– control studies)
2 cups of coffee per day were 0.69 (95% CI, 0.55– 0.87) with their 95% confidence intervals (CIs) (or data to
for persons without a history of liver disease and 0.56 calculate them) were reported. If data were duplicated in
(95% CI, 0.35– 0.91) for those with a history of liver more than 1 study, the most recent study was included in
disease. Conclusions: Findings from this meta-anal- the analysis.
ysis suggest that an increased consumption of coffee We identified 11 potentially relevant articles18 –28 con-
may reduce the risk of liver cancer. cerning coffee consumption and liver cancer risk. Three
publications18 –20 were excluded because of duplicate
reports from the same study population. The remaining
D ata on potential beneficial effects of coffee on liver
function and liver diseases have accrued over the
last 2 decades. Several epidemiologic studies have re-
publications, consisting of 4 cohort studies21–23 (1 article
presented results from 2 separate cohorts) and 5 case–
control studies,24 –28 were included in the meta-analysis.
ported inverse associations of coffee drinking with levels
of liver enzymes, including ␥-glutamyltransferase (an in-
dicator of cirrhosis risk)1– 8 and alanine aminotransferase Abbreviations used in this paper: CI, confidence interval; HCC, he-
patocellular carcinoma; RR, relative risk.
(a marker of liver injury),2,6,9,10 as well as with risk of © 2007 by the AGA Institute
chronic liver disease11 and liver cirrhosis.12–16 Moreover, 0016-5085/07/$32.00
studies in animals have shown an inhibitory effect of doi:10.1053/j.gastro.2007.03.044
May 2007 COFFEE AND LIVER CANCER 1741

Data Extraction test.33 All statistical analyses were performed with Stata
The following data were extracted from each (version 9.0; StataCorp, College Station, TX). P values
study: the first author’s last name, publication year, ⬍ .1 were considered statistically significant.
country where the study was performed, study design,
type of controls for case– control studies (patients with Results
other diseases or population-based controls), sample size The 4 cohort studies21–23 and 5 case– control stud-
(cases and controls or cohort size), type of outcome ies
24 –28 that were included in this meta-analysis were
(primary liver cancer or HCC), variables adjusted for in published between 2002 and 2007 (Table 1) and involved
the analysis, and the RR estimates with corresponding a total of 2260 cases and 239,146 noncases. Of these
95% CIs for each category of coffee consumption. From studies, 6 were conducted in Japan and 3 in southern Eu-
each study, we extracted the risk estimates that reflected rope (Italy and Greece). All studies consisted of both men
the greatest degree of control for potential confounders. and women. The outcome was incidence of primary liver
cancer in 2 studies,22 incidence of HCC in 6 studies,21,24 –28

PANCREAS, AND
CLINICAL–LIVER,
For 1 case– control study27 that employed 3 different

BILIARY TRACT
control groups (population-based controls [n ⫽ 1253], and mortality from HCC in 1 study.23 Among case– control
hospital-based controls [n ⫽ 275], and chronic liver dis- studies, 4 used hospital-based controls24 –26,28 and 1 used
ease patients [n ⫽ 381]), we used the results based on population-based controls.27 In the cohort studies, par-
comparison with population-based controls in the main ticipants were asked about their coffee consumption dur-
analysis and chronic liver disease patients in sensitivity ing the past month21 or year23 before baseline, or about
analysis. their recent or usual coffee consumption at baseline.22 In
the case– control studies, participants were inquired
Statistical Analysis about their coffee consumption 1 year,24 2 years,28 or 10
years before.25,27 In 1 case– control study that consisted of
The measure of effect of interest is the relative
persons chronically infected with hepatitis C virus, par-
risk for cohort studies, approximated by the odds ratio
ticipants were asked about their coffee consumption
in case– control studies, and the corresponding statis-
both before and after first identification of liver disease26;
tical significance (95% CI). We attempted to place the
we used the results based on coffee consumption before
studies on a common scale by estimating the RR per 2
identification of liver disease.
cups per day increase of coffee consumption (eg, from The estimated RRs for each study and all studies
no coffee consumption to 2 cups per day). For each combined for an increase of 2 cups of coffee per day
study, we estimated the median coffee consumption are presented in Figure 1. All studies reported an
for each category by assigning the midpoint of the inverse association between coffee consumption and
upper and lower boundary in each category as the risk of liver cancer, and in 6 studies21,23–25,27,28 the
average consumption. The highest, open-ended cate- association was statistically significant. Meta-analysis
gory was assumed to have the same amplitude of of all studies found a 43% reduced risk of liver cancer
consumption as the preceding category. We used the (RR, 0.57; 95% CI, 0.49 – 0.67) for an increment of 2
covariance-corrected method of Greenland and Long- cups of coffee per day. There was no statistically sig-
necker29,30 to model the log RRs for liver cancer as a nificant heterogeneity among the results of individual
linear function of coffee consumption. This provided studies (Q ⫽ 11.56; P ⫽ .17; I2 ⫽ 30.8%). In a sensitivity
an estimate of the regression coefficient and its stan- analysis in which 1 study at a time was removed and
dard error for each study. Study-specific RR estimates the rest analyzed, the summary RR ranged from 0.54
were combined using a random-effects model, which (when excluding the study by Gallus et al24) to 0.63
incorporates both within- and between-study variabil- (when excluding the study by Tanaka et al27). Exclud-
ity.31 We checked for nonlinearity of the dose–response ing the study that consisted of persons chronically
relationship between coffee consumption and liver infected with hepatitis C virus26 did not alter the
cancer by estimating polynomial models. This was results essentially (RR, 0.58; 95% CI, 0.49 – 0.68). One
done using the “pool-first” method described by case– control study employed different control groups.27
Greenland and Longnecker.29 We found that the best- In a sensitivity analysis, the summary RR did not change
fitting model was a linear model. materially when we used results based on comparison
Statistical heterogeneity among studies was evaluated with chronic liver disease patients (RR, 0.63; 95% CI,
using the Q and I2 statistics.32 We did a sensitivity anal- 0.56 – 0.71). Restricting the analysis to 6 studies in which
ysis in which 1 study at a time was removed and the rest the outcome was HCC incidence21,24 –28 yielded a sum-
analyzed to evaluate whether the results could have been mary RR of 0.59 (95% CI, 0.49 – 0.71). Summary relative
affected markedly by a single study. We also conducted risks were similar for cohort and case– control studies
analyses stratified by study design, geographic area (Japan (Figure 1). Stratifying by geographic region, the summary
vs. Europe), and history of liver disease. To assess the RRs were 0.52 (95% CI, 0.43– 0.61) for studies conducted
potential for publication bias, we used Egger’s regression in Japan and 0.68 (95% CI, 0.58 – 0.80) for studies con-
 1742 LARSSON AND WOLK GASTROENTEROLOGY Vol. 132, No. 5

Table 1. Characteristics of Cohort and Case–Control Studies of Coffee Consumption and Liver Cancer
Country; Age, No. of No. in cohort Coffee Relative risk
Study study period years Outcome cases or controls consumption (95% CI) Adjustments

Cohort studies
Inoue et al, 200521 Japan; 1990–2001 40–69 HCC incidence 334 90,452 Almost never 1.00 (reference) Age, sex, study area,
(cohort) 1–2 days/week 0.75 (0.56 –1.01) smoking, and
3– 4 days/week 0.79 (0.55–1.14) intakes of alcohol,
1–2 cups/day 0.52 (0.38 – 0.73) green tea, and
3– 4 cups/day 0.48 (0.28 – 0.83) green vegetables
ⱖ5 cups/day 0.24 (0.08 – 0.77)
Shimazu et al, Japan; 1984–1992 ⱖ40 Primary liver cancer 70 22,404 Nondrinkers 1.00 (reference) Age, sex, history of
200522; Cohort 1 incidence (cohort) Occasionally 0.56 (0.33– 0.97) liver disease,
ⱖ1 cup/day 0.53 (0.28 –1.00) smoking, and
alcohol intake
Shimazu et al, Japan; 1990–1997 40–64 Primary liver cancer 47 38,703 Nondrinkers 1.00 (reference) Age, sex, history of
200522; Cohort 2 incidence (cohort) Occasionally 1.05 (0.52–2.16) liver disease,
ⱖ1 cup/day 0.68 (0.31–1.51) smoking, and
PANCREAS, AND
CLINICAL–LIVER,

alcohol intake
BILIARY TRACT

Kurozawa et al, Japan; 1988–1999 40–79 HCC mortality 258 83,966 Nondrinkers 1.00 (reference) Age, sex, education,
200523 (cohort) ⬍1 cup/day 0.83 (0.54 –1.25) history of diabetes
ⱖ1 cup/day 0.50 (0.31– 0.79) or liver disease,
smoking, and
alcohol intake
Case–control studies
Gallus et al, 200224 Greece; 1995–1998 18–79 HCC incidence 834 1912 Nondrinkers 1.0 (reference) Age, sex, education,
and Italy; 1984– 1 cup/day 1.2 (0.9 –1.6) history of diabetes
1997 2 cups/day 1.0 (0.7–1.3) or hepatitis, body
ⱖ3 cups/day 0.7 (0.5–1.0) mass index,
smoking, and
alcohol intake
Gelatti et al, 200525 Italy; 1994–2003 ⬍80 HCC incidence 250 500 Nondrinkers 1.0 (reference) Age, sex, HBV and/or
1–2 cups/day 0.8 (0.4 –1.3) HCV infection, and
3– 4 cups/day 0.4 (0.2– 0.8) alcohol intake
ⱖ5 cups/day 0.3 (0.1– 0.7)
Ohfuji et al, 200626 Japan; 2001–2002 NA HCC incidence 73a 253a Nondrinkers 1.00 (reference)b Age, sex, time since
⬍1 cup/day 0.61 (0.18 –2.03)b first identification
ⱖ1 cup/day 0.38 (0.13–1.12)b of liver disease,
BMI, disease
severity, family
history of liver
disease, interferon
therapy, smoking,
and alcohol intake
Tanaka et al, 200727 Japan; 2001–2004 40–79 HCC incidence 209 1253 Nondrinkers 1.00 (reference) Age, sex, smoking,
Occasionally 0.33 (0.22– 0.48)c and alcohol intake
1–2 cups/day 0.27 (0.15– 0.48)c
ⱖ3 cups/day 0.22 (0.11– 0.43)c
Montella et al, Italy; 1999–2002 43–84 HCC incidence 185 412 Nondrinkers 2.28 (0.99 –5.24) Age, sex, center, HBV
200728 ⬍14 cups/week 1.00 (reference) and/or HCV
14 –20 cups/week 0.54 (0.27–1.07) infection,
21–27 cups/week 0.57 (0.25–1.32) education,
ⱖ28 cups/week 0.43 (0.16 –1.13) smoking, and
alcohol intake

BMI, body mass index; CI, confidence interval; HCC, hepatocellular carcinoma; HBV, hepatitis B virus; HCV, hepatitis C virus; NA, not available.
aPatients chronically infected with hepatitis C virus.
bOdds ratios for hepatocellular carcinoma calculated before first identification of liver disease; the corresponding odds ratios calculated after first identification of liver disease were 1.00
(reference), 0.57 (95% CI, 0.20 –1.67), and 0.19 (95% CI, 0.05– 0.71).
cOdds ratios for hepatocellular carcinoma for coffee consumption 10 years before by comparing cases with population-based controls; the corresponding odds ratios for coffee consumption

during the last 1–2 years were 1.00 (reference), 0.31 (95% CI, 0.21– 0.46), 0.11 (95% CI, 0.06 – 0.21), and 0.10 (95% CI, 0.04 – 0.24).

ducted in Europe. We found no evidence of publication Discussion

bias for cohort studies (P ⫽ .24) or case– control studies The findings from the present meta-analysis of
(P ⫽ .22). observational studies indicate that an increased coffee
Four studies presented results stratified by history of consumption is associated with a reduced risk of liver
liver disease.21–24 In meta-analysis of these studies, the cancer, both among individuals with and without a his-
summary RRs of liver cancer for a 2 cups per day increase tory of liver disease. Overall, the risk of liver cancer
in coffee consumption were 0.69 (95% CI, 0.55– 0.87) for decreased by 43% for an increment of 2 cups of coffee per
persons without a history of liver disease and 0.56 (95% day.
CI, 0.35– 0.91) for those with a history of liver disease Our study has several potential limitations. First, as in
(Figure 2). The difference in summary RR by strata of all observational studies of diet and disease, the possibil-
history liver disease was not statistically significant ity of bias and confounding cannot be excluded. How-
(P ⫽ .44). ever, cohort studies, which are less susceptible to bias
May 2007 COFFEE AND LIVER CANCER 1743

studies, publication bias could be of concern because

small studies with null results tend not to be published.
Because of the relatively small number of studies, we had
limited statistical power to conclusively reject the null
hypothesis of no publication bias. The presence of pos-
sible publication bias could have led to an overestimation
of the relation between coffee consumption and risk of
liver cancer.
A protective effect of coffee consumption on liver
cancer is biologically plausible. Coffee contains large
amounts of antioxidants, such as chlorogenic acids, and
experimental studies in animals have demonstrated an
Figure 1. Relative risks of liver cancer associated with coffee con- inhibitory effect of coffee and chlorogenic acids on liver

PANCREAS, AND
CLINICAL–LIVER,

BILIARY TRACT
sumption (per 2 cups/day increment). Squares represent study-specific carcinogenesis.17 Caffeine is another major component of
relative risk estimates (size of the square reflects the study-specific
coffee. In one animal study, caffeine levels of coffee ex-
statistical weight, that is, the inverse of the variance); horizontal lines
represent 95% CIs; diamonds represent summary relative risk esti- tracts were inversely related to liver injury.37 A popula-
mates with corresponding 95% CIs. Tests for heterogeneity: all studies, tion-based study in the United States showed that higher
Q ⫽ 11.56; P ⫽ .17; I2 ⫽ 30.8%; cohort studies, Q ⫽ 1.74; P ⫽ .63; intake of coffee, and especially caffeine, was associated
I2 ⫽ 0%; case– control studies, Q ⫽ 9.28; P ⫽ .05; I2 ⫽ 36.9%. with a lower prevalence of abnormal alanine aminotrans-
ferase activity (a marker of liver injury).36 The protective
because of the prospective design, also showed an inverse relationships of coffee and caffeine were consistent across
association between coffee consumption and risk of liver subgroups at risk for liver injury and remained in analysis
cancer, suggesting that the finding is not likely attribut- limited to persons without impaired liver function.36
able to recall and selection bias. Individual studies may Several studies in Europe and Japan have also observed
have failed to adjust for potential known or unknown inverse relations between coffee consumption and serum
confounders. For example, only 5 studies controlled for levels of aminotransferases2,6,9,10 and ␥-glutamyltrans-
liver disease22,23,26 or hepatitis,24 and only 3 case– control ferase1– 8 (a sensitive indicator of several liver diseases). In
studies adjusted for hepatitis virus infection.25,26,28 Caf- addition, prospective cohort studies in the United
feine metabolism is impaired in persons with chronic States15 and Norway16 and case– control studies in
liver disease.34,35 Hence, if persons with liver disease or Italy12–14 have reported an inverse association between
hepatitis virus infection who are at high risk of liver coffee consumption and risk of liver cirrhosis, which is
cancer consume less coffee (eg, to avoid the side effects of strongly related to HCC.38 Therefore, the observed asso-
caffeine) compared with healthy persons, it could lead to ciation of coffee consumption with liver cancer could
a spurious protective association between coffee con- potentially represent an association with liver disease.
sumption and liver cancer. Arguing against this possibil-
ity, in 3 cohort studies with data on liver disease, coffee
consumption was not significantly related to history of
liver disease at baseline.22,23 In addition, in a U.S. popu-
lation-based study, intakes of coffee and caffeine were
not significantly associated with the prevalence of risk
indicators for liver injury, including viral hepatitis and
elevated transferrin saturation.36 A second limitation is
that our results are likely to be affected by some misclas-
sification of coffee consumption. In cohort studies, such
misclassification is probably nondifferential, and would
most likely lead to an underestimation of the relation-
ship. The influence of misclassification on the results in
case– control studies is less predictable. Third, heteroge-
neity may have been introduced by methodologic differ- Figure 2. Relative risks of liver cancer associated with coffee con-
ences among studies, such as differences in type of coffee sumption (per 2 cups/day increment), stratified by history of liver dis-
consumed (eg, filtered vs instant coffee) in the studied ease. Squares represent study-specific relative risk estimates (size of
populations and differences in outcome (primary liver the square reflects the study-specific statistical weight, that is, the in-
verse of the variance); horizontal lines represent 95% CIs; diamonds
cancer vs HCC). Fourth, all studies in this meta-analysis
represent summary relative risk estimates with corresponding 95% CIs.
were conducted in Japan or southern Europe; therefore, Tests for heterogeneity: without a history of liver disease, Q ⫽ 4.58; P ⫽
the observed finding may not be generalizable to other .21; I2 ⫽ 34.6%; with a history of liver disease, Q ⫽ 7.00; P ⫽ .07; I2 ⫽
populations. Finally, in a meta-analysis of published 57.1%.
 1744 LARSSON AND WOLK GASTROENTEROLOGY Vol. 132, No. 5

Nevertheless, in a stratified analysis by history of liver Katsurada M, Nakagawa K, Yoshizawa N. Coffee drinking and
disease, coffee consumption was inversely associated with serum gamma-glutamyltransferase: an extended study of
Self-Defense Officials of Japan. Ann Epidemiol 1999;9:325–331.
risk of liver cancer both among those with and without a 8. Nakanishi N, Nakamura K, Nakajima K, Suzuki K, Tatara K.
history of liver disease. This finding suggests that coffee Coffee consumption and decreased serum gamma-glutamyltrans-
drinking may lower the risk of liver cancer even after the ferase: a study of middle-aged Japanese men. Eur J Epidemiol
acquisition of liver disease. A recent meta-analysis indi- 2000;16:419 – 423.
cated that coffee consumption may reduce the risk of 9. Nakanishi N, Nakamura K, Suzuki K, Tatara K. Effects of coffee
consumption against the development of liver dysfunction: a
type 2 diabetes,39 and another meta-analysis showed a
4-year follow-up study of middle-aged Japanese male office work-
strong positive association between type 2 diabetes and ers. Ind Health 2000;38:99 –102.
risk of HCC.40 Thus, a potential protective effect of coffee 10. Honjo S, Kono S, Coleman MP, Shinchi K, Sakurai Y, Todoroki I,
consumption against liver cancer may also, in part, be Umeda T, Wakabayashi K, Imanishi K, Nishikawa H, Ogawa S,
mediated through a reduced risk of type 2 diabetes. Katsurada M, Nakagawa K, Yoshizawa N. Coffee consumption
and serum aminotransferases in middle-aged Japanese men.
One of the case– control studies included in this meta-
PANCREAS, AND
CLINICAL–LIVER,

J Clin Epidemiol 2001;54:823– 829.

BILIARY TRACT

analysis employed 3 different control groups.27 In that 11. Ruhl CE, Everhart JE. Coffee and tea consumption are associated
study, results based on community controls and a con- with a lower incidence of chronic liver disease in the United
trol group with chronic liver disease (chronic hepatitis or States. Gastroenterology 2005;129:1928 –1936.
cirrhosis), but not hospital-based controls, showed that 12. Corrao G, Zambon A, Bagnardi V, D’Amicis A, Klatsky A. Coffee,
coffee drinking during the last 1–2 years and 10 years caffeine, and the risk of liver cirrhosis. Ann Epidemiol 2001;11:
458 – 465.
before was significantly inversely associated with the risk 13. Corrao G, Lepore AR, Torchio P, Valenti M, Galatola G, D’Amicis
of HCC, even after adjustment for hepatitis virus markers A, Arico S, di Orio F. The effect of drinking coffee and smoking
(in analysis with liver disease patients as control group).27 cigarettes on the risk of cirrhosis associated with alcohol con-
In summary, findings from this meta-analysis indicate sumption. A case– control study. Provincial Group for the Study of
that coffee consumption may reduce the risk of liver Chronic Liver Disease. Eur J Epidemiol 1994;10:657– 664.
14. Gallus S, Tavani A, Negri E, La Vecchia C. Does coffee protect
cancer. The mechanisms involved and the substances in against liver cirrhosis? Ann Epidemiol 2002;12:202–205.
coffee that may be responsible for the relation remain to 15. Klatsky AL, Armstrong MA. Alcohol, smoking, coffee, and cirrho-
be elucidated. Cohort studies with information on coffee sis. Am J Epidemiol 1992;136:1248 –1257.
consumption throughout life and type of coffee con- 16. Tverdal A, Skurtveit S. Coffee intake and mortality from liver
sumed (eg, caffeinated vs decaffeinated) and that take cirrhosis. Ann Epidemiol 2003;13:419 – 423.
17. Tanaka T, Nishikawa A, Shima H, Sugie S, Shinoda T, Yoshimi N,
into account potential confounders (such as the presence
Iwata H, Mori H. Inhibitory effects of chlorogenic acid, reserpine,
and severity of liver disease and hepatitis virus infections) polyprenoic acid (E-5166), or coffee on hepatocarcinogenesis in
and, ideally, intervention studies among persons at high rats and hamsters. Basic Life Sci 1990;52:429 – 440.
risk for liver cancer are warranted to clarify the potential 18. La Vecchia C, Ferraroni M, Negri E, D’Avanzo B, Decarli A, Levi F,
protective effect of coffee drinking on liver cancer. Such Franceschi S. Coffee consumption and digestive tract cancers.
Cancer Res 1989;49:1049 –1051.
studies may also establish the temporal relationship be-
19. Kuper H, Tzonou A, Kaklamani E, Hsieh CC, Lagiou P, Adami HO,
tween coffee use, liver disease, and liver cancer. Trichopoulos D, Stuver SO. Tobacco smoking, alcohol consump-
tion and their interaction in the causation of hepatocellular car-
References cinoma. Int J Cancer 2000;85:498 –502.
1. Nilssen O, Forde OH, Brenn T. The Tromso Study. Distribution and 20. Kurozawa Y, Ogimoto I, Shibata A, Nose T, Yoshimura T, Suzuki
population determinants of gamma-glutamyltransferase. Am J H, Sakata R, Fujita Y, Ichikawa S, Iwai N, Fukuda K, Tamakoshi A.
Epidemiol 1990;132:318 –326. Dietary habits and risk of death due to hepatocellular carcinoma
2. Casiglia E, Spolaore P, Ginocchio G, Ambrosio GB. Unexpected in a large scale cohort study in Japan. Univariate analysis of JACC
effects of coffee consumption on liver enzymes. Eur J Epidemiol study data. Kurume Med J 2004;51:141–149.
1993;9:293–297. 21. Inoue M, Yoshimi I, Sobue T, Tsugane S. Influence of coffee
3. Kono S, Shinchi K, Imanishi K, Todoroki I, Hatsuse K. Coffee and drinking on subsequent risk of hepatocellular carcinoma: a pro-
serum gamma-glutamyltransferase: a study of self-defense offi- spective study in Japan. J Natl Cancer Inst 2005;97:293–300.
cials in Japan. Am J Epidemiol 1994;139:723–727. 22. Shimazu T, Tsubono Y, Kuriyama S, Ohmori K, Koizumi Y, Nishino
4. Pintus F, Mascia P. Distribution and population determinants of Y, Shibuya D, Tsuji I. Coffee consumption and the risk of primary
gamma-glutamyltransferase in a random sample of Sardinian liver cancer: pooled analysis of two prospective studies in Japan.
inhabitants. “ATS-SARDEGNA” Research Group. Eur J Epidemiol Int J Cancer 2005;116:150 –154.
1996;12:71–76. 23. Kurozawa Y, Ogimoto I, Shibata A, Nose T, Yoshimura T, Suzuki
5. Poikolainen K, Vartiainen E. Determinants of gamma-glutamyltrans- H, Sakata R, Fujita Y, Ichikawa S, Iwai N, Tamakoshi A. Coffee
ferase: positive interaction with alcohol and body mass index, neg- and risk of death from hepatocellular carcinoma in a large cohort
ative association with coffee. Am J Epidemiol 1997;146:1019 – study in Japan. Br J Cancer 2005;93:607– 610.
1024. 24. Gallus S, Bertuzzi M, Tavani A, Bosetti C, Negri E, La Vecchia C,
6. Tanaka K, Tokunaga S, Kono S, Tokudome S, Akamatsu T, Lagiou P, Trichopoulos D. Does coffee protect against hepatocel-
Moriyama T, Zakouji H. Coffee consumption and decreased serum lular carcinoma? Br J Cancer 2002;87:956 –959.
gamma-glutamyltransferase and aminotransferase activities among 25. Gelatti U, Covolo L, Franceschini M, Pirali F, Tagger A, Ribero ML,
male alcohol drinkers. Int J Epidemiol 1998;27:438 – 443. Trevisi P, Martelli C, Nardi G, Donato F. Coffee consumption
7. Honjo S, Kono S, Coleman MP, Shinchi K, Sakurai Y, Todoroki I, reduces the risk of hepatocellular carcinoma independently of its
Umeda T, Wakabayashi K, Imanishi K, Nishikawa H, Ogawa S, aetiology: a case– control study. J Hepatol 2005;42:528 –534.
May 2007 COFFEE AND LIVER CANCER 1745

26. Ohfuji S, Fukushima W, Tanaka T, Habu D, Tamori A, Sakaguchi 35. Wahllander A, Renner E, Preisig R. Fasting plasma caffeine con-
H, Takeda T, Kawada N, Seki S, Nishiguchi S, Shiomi S, Hirota Y. centration. A guide to the severity of chronic liver disease. Scand
Coffee consumption and reduced risk of hepatocellular carci- J Gastroenterol 1985;20:1133–1141.
noma among patients with chronic type C liver disease: A case– 36. Ruhl CE, Everhart JE. Coffee and caffeine consumption reduce
control study. Hepatol Res 2006;36:201–208. the risk of elevated serum alanine aminotransferase activity in
27. Tanaka K, Hara M, Sakamoto T, Higaki Y, Mizuta T, Eguchi Y, the United States. Gastroenterology 2005;128:24 –32.
Yasutake T, Ozaki I, Yamamoto K, Onohara S, Kawazoe S, Shige- 37. He P, Noda Y, Sugiyama K. Suppressive effect of coffee on
matsu H, Koizumi S. Inverse association between coffee drinking lipopolysaccharide-induced hepatitis in D-galactosamine-sensi-
and the risk of hepatocellular carcinoma: a case– control study in tized rats. Biosci Biotechnol Biochem 2001;65:1924 –1927.
Japan. Cancer Sci 2007;98:214 –218. 38. La Vecchia C, Negri E, Cavalieri d’Oro L, Franceschi S. Liver
28. Montella M, Polesel J, La Vecchia C, Maso LD, Crispo A, Crovatto M, cirrhosis and the risk of primary liver cancer. Eur J Cancer Prev
Casarin P, Izzo F, Tommasi LG, Talamini R, Franceschi S. Coffee and 1998;7:315–320.
tea consumption and risk of hepatocellular carcinoma in Italy. Int J 39. van Dam RM, Hu FB. Coffee consumption and risk of type 2
Cancer [published online Jan 4, 2006, ahead of print] 2007. diabetes: a systematic review. JAMA 2005;294:97–104.
29. Greenland S, Longnecker MP. Methods for trend estimation from 40. El-Serag HB, Hampel H, Javadi F. The association between dia-

PANCREAS, AND
CLINICAL–LIVER,
summarized dose–response data, with applications to meta-anal-

BILIARY TRACT
betes and hepatocellular carcinoma: a systematic review of epi-
ysis. Am J Epidemiol 1992;135:1301–1309. demiologic evidence. Clin Gastroenterol Hepatol 2006;4:
30. Orsini N, Bellocco R, Greenland S. Generalized least squares for 369 –380.
trend estimation of summarized dose–response data. Stata J
2006;6:40 –57.
31. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control
Clin Trials 1986;7:177–188. Received January 26, 2007. Accepted March 8, 2007.
32. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta- Address requests for reprints to: Susanna C. Larsson, PhD, Division
analysis. Stat Med 2002;21:1539 –1558. of Nutritional Epidemiology, The National Institute of Environmental
33. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta- Medicine, Karolinska Institutet, P.O. Box 210, SE-171 77 Stockholm,
analysis detected by a simple, graphical test. BMJ 1997;315: Sweden. e-mail: susanna.larsson@ki.se; fax: (46) 8-304571.
629 – 634. The sponsor had no role in the study design or in the collection,
34. Hasegawa M, Yamada S, Hirayama C. Fasting plasma caffeine analysis, and interpretation of the data.
level in cirrhotic patients: relation to plasma levels of cat- No conflicts of interest exist.
echolamines and renin activity. Hepatology 1989;10:973–977. Supported by research grants from the Swedish Cancer Society.