For citation purposes: Li J, Xu M, Cheng Z, Wang P, Wei S, Ma X, Li G. Coffee consumption and risk of hepatocellular carcinoma: A meta-analysis of cohort studies. OA Evidence-Based Medicine 2014 Jan 18;2(1):1.

Systematic review

 
Systematic Reviews

Coffee consumption and risk of hepatocellular carcinoma: a meta-analysis of cohort studies

J Li1, M Xu2, Z Cheng3, P Wang4, S Wei1, X Ma1, G Li5*
 

Authors affiliations

1 Department of Clinical Medicine, Xingtai Medical College, 618 Gangtie Road Xingtai, Hebei 054000, China

2 Department of Occupational & Environmental Health Sciences, School of Public Health, Peking University, 38 Xueyuan Road, Haidian District, Beijing 100191, China

3 Faculty of Education, Brock University, 500 Glenridge Avenue, St. Catharines, ON, L2S 3A1, Canada

4 Department of Nutrition & Food Hygiene, School of Public Health, Peking University, 38 Xueyuan Road, Haidian District, Beijing 100191, China

5 Department of Clinical Epidemiology & Biostatistics, McMaster University, 1280 Main Street West, Hamilton, ON, L8S 4L8, Canada

* Corresponding author Email: lig28@mcmaster.ca

Abstract

Introduction

Quantities of studies investigating coffee consumption and risk of hepatocellular carcinoma have reported different findings. The aim of this study was to perform a meta-analysis of cohort studies to clarify the effect of coffee intake on risk of hepatocellular carcinoma and to conduct an exhaustive stratified and dose–response analyses.

Materials and Methods

Electronic reference databases including MEDLINE, the Cochrane Controlled Trials Register, EMBASE, and Science Citation Index and PubMed (up to May 2013) were searched to identify eligible cohort studies investigating relationship between coffee consumption and risk of hepatocellular carcinoma. Study-specific relative risk estimates were pooled using a random-effects model.

Results

Ten cohort studies (including 553,088 participants and 1649 hepatocellular carcinoma cases) were included in the meta-analysis. The summary relative risk for coffee drinkers versus non-drinkers was 0.64 (95% confidence interval: 0.55, 0.75), while relative risk for lowest and highest drinkers was 0.75 (95% confidence interval: 0.65, 0.87) and 0.48 (95% confidence interval: 0.39, 0.60), respectively. An increment of one cup of coffee per day was significantly related to decreased risk of hepatocellular carcinoma (relative risk: 0.82, 95% confidence interval: 0.78, 0.87). Subgroup, sensitivity and detailed dose–response analyses indicated the robustness and insensitiveness of the relationship between coffee intake and reduced risk of hepatocellular carcinoma.

Conclusion

Based on the evidence of cohort studies, the meta-analysis confirms that coffee consumption is associated with reduced risk of hepatocellular carcinoma. The inverse relationship between coffee and risk of hepatocellular carcinoma was consistent across different populations and settings.

Introduction

As one of the most widely consumed beverages in the world, coffee has many potential beneficial health effects. For example, coffee consumption has been marked to be associated with reduced risks of several chronic diseases including type 2 diabetes mellitus, Parkinson’s disease, hepatocellular diseases and gout[1,2,3,4,5,6]. The relationship between coffee intake and reduced risks of cancers is a matter of concern, because coffee contains high level of anti-cancer compounds such as chlorogenic acids, cafestol and kahweal[7]. Moreover, emerging epidemiological studies have shown promising effects of coffee on several cancers including oral cavity[8], colorectal[9], prostate[10], leukemic, pancreatic cancer[11], etc.

Liver cancer (LC) is the most common digestive tumour nowadays. For men, it is the fifth most frequently diagnosed cancer worldwide and the second cause of death due to cancer; however for women, it is the seventh most commonly diagnosed cancer and the sixth leading cause of death[12]. Hepatocellular carcinoma (HCC) is the major histological subtype, accounting for 70–85% of the total LC burden worldwide[13] and with a moderately increasing rate in North America and North Europe over the last few decades[14,15]. Till date, liver cirrhosis and chronic infection with hepatitis B or C viruses are considered the most important risk factors for HCC[16,17]. Some lifestyle-related factors such as alcohol consumption, obesity and diabetes are also identified to be independently associated with the risk of HCC, while no significant relation between smoking and HCC was reported[14,17,18,19].

Inverse association between coffee consumption and the risk of HCC has recently been investigated in observational studies[17,18,19,20,21,22,23,24,25,26]. Moreover, several meta-analyses summarising the evidence of observational studies clarified the significant relationship between the risk of HCC and coffee intake consistently[11,27,28,29,30,31]. However, all the existing meta-analyses lacked an exhaustive stratified analysis, for example, none of them investigated association between HCC and smoking, and only one study included data of alcohol drinking. Moreover, even though there were three meta-analyses reported results of dose–response analyses[28,29,31], little was known about the relationship between an increment of one cup of coffee per day and risk of HCC stratified by gender, region, history of liver disease, etc. Besides, majority of the included studies in the meta-analyses were conducted in Japan, while more observational studies in other countries had been published recently. Thus, a meta-analysis with comprehensive analysis is imperative to summarise the latest evidence till date. Provided cohort studies are considered to be hierarchically superior to case–control studies mainly in avoiding recall and selection bias[31,32] in this study we carried out a meta-analysis of prospective cohort studies to clarify the association between coffee consumption and risk of HCC.

Materials and Methods

Search strategy

MEDLINE, the Cochrane Controlled Trials Register, EMBASE, Science Citation Index and PubMed were used to search for articles (in English, up to May 2013), which described cohort studies investigating the relationship between coffee consumption and risk of HCC. In our searches, we used descriptors that included synonyms for coffee and HCC in various combinations, for example, ‘coffee or caffeine’, ‘hepatocelluar or liver’ and ‘carcinoma or neoplasm’. Titles and abstracts of trials identified from the search were independently reviewed and pooled for further screening by two authors (GL and JL). Later each reviewer examined the full text of all studies independently that were identified from the title and abstract screens. Disagreement was resolved by discussion and consensus among the reviewers and a third author (PW) was available to help if consensus failed to be reached. All references related to reviews and papers retrieved by the search were also examined. Additionally, authors were tried to contact who included studies to obtain unpublished data.

Eligibility criteria

Studies were selected for analysis if they met the following criteria: (1) prospective cohort studies or nested case–control studies; (2) the exposure studied was coffee consumption or caffeine; (3) the outcome of interest was HCC, primary liver cancer (PLC) or LC; (4) RRs, odds ratios (ORs) or hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) (or any relevant data able to calculate these statistics) were presented.

Data extraction

Two reviewers (G.L. and J.L.) extracted data independently. The items extracted included first author’s last name, year of publication, country where the study was performed, years of study/follow-up period, number of subjects (cases, controls or cohort size), frequency of coffee consumption or caffeine intake, measures of the association (RR, HR or OR) between coffee consumption and HCC incidence and the corresponding 95% CI and confounding variables allowed in the analyses. Estimates adjusted for multiple potential confounding variables were used in all possibilities. If RRs or the corresponding 95% CIs were not provided, they were derived from available tabular data[29] or by contacting the authors of included studies. However, if data were only available in graphic format, we imputed approximations of the statistics of interest. The study quality was assessed using the 9-star Newcastle–Ottawa Quality Assessment Scale[33], with which each included study was rated based on the selection of study groups, the comparability of groups and the ascertainment of exposure and outcome.

Statistical analysis

We used the generic inverse variance method to pool the summary RR for coffee drinkers versus non-drinkers and for highest/lowest coffee drinkers versus non-drinkers. The highest and lowest coffee drinkers were defined as the last and the second stratum of coffee consumption, respectively in each included study. If the study reported only data with two strata (e.g. heavy drinkers versus light drinkers), then the second stratum (i.e. heavy drinkers) would be used as both the highest and the lowest coffee data for meta-analyses.

A random effects model that did not assume homogeneity of RRs across studies was used to synthesise data by pooling the results of the included studies. Heterogeneity was evaluated through the I2 statistic and the Q-test[34], with a value of I2 > 50% or P value <0.1 taken as statistically significant heterogeneity[35].

If statistical heterogeneity was found to investigate the heterogeneity, we conducted subgroup analyses stratified by the following variables: geographic region, sex, smoking status, alcohol consumption, history of liver disease, follow-up period (using 10 years as a cut-off according to the mean of follow-up periods of included studies) and quality scores (considering studies of scores >7 in the Newcastle–Ottawa Quality Assessment Scale as of high quality).

A sensitivity analysis was performed by applying a fixed effects model. Furthermore, to make the probability statement of a beneficial risk of HCC related to coffee, another sensitivity analysis based on a hierarchical Bayesian random effects model using a ‘non-informative’ prior distribution was conducted. The summary statistics were acquired from the posterior distribution of the Bayesian analysis, presenting as a pooled RR with 95% associated credible interval (CrI)[36]. We fitted the models in WinBUGS using 100,000 Markov chain Monte Carlo cycles with two chains of simulations, a burn-in of 10,000 and thin of 10. Convergence was assessed using the Gelman Rubin statistic, and the autocorrelation was assessed based on the autocorrelation function plots. We applied two different distributions for the between-study variance or standard deviation (SD) (i.e. gamma distribution for the between-study variance and uniform distribution for the between-study SD) to test the robustness of the results of Bayesian analyses.

For dose–response analysis, we used the method proposed by Greenland[37] to provide a continuous summary RR for an increment of one cup of coffee per day. Because each study presented coffee consumption in different units (cups/week, cups/day, days/week, daily drinks, times/week), we converted them into cups/day to facilitate calculation and comparison. In each study, we estimated the average coffee consumption for each category by assigning the midpoint of the upper and lower boundary. If the highest category of consumption was not provided, we considered it of the same amplitude as the preceding category[31].

The funnel plots with Begg’s rank correlation and Egger regression tests were performed to detect publication bias. All statistical analyses were conducted using the software STATA 11.0 (Stata Corporation, College Station,TX, USA) and Review Manager 5.2 for windows (the Nordic Cochrane Center, the Cochrane Collaboration, Copenhagen, Denmark), while the Bayesian random effects model was performed using the software WinBUGS 1.4 (MRC Biostatistics Unit, Cambridge, UK). All statistical tests were 2-sided.

Results

Characteristics of included studies

Ten cohort studies were included in the meta-analysis, with 8 cohort studies[17,18,20,21,22,24,38] and 2 nested case–control studies[23,25]. Shimazu presented data in 2 prospective cohorts in a study[24], thus we considered it as 2 different studies (cohort 1 and cohort 2). Of the included studies, 9 were conducted in Asia[17,18,20,21,22,24,25], and 1 in Europe (Finland)[38]. A total of 553,088 participants were included for analyses, whose ages ranged 40–79 years. There were 1,649 HCC cases identified during the follow-up. The periods of follow-up varied from 7[24] to 24[23] years, with an average of 10 years approximately. All the studies were published between 2005 and 2012.

Characteristics of the studies included are shown in Table 1. The outcome was incidence of HCC in five studies[18,20,21,23,25], incidence of PLC in two cohorts[24], incidence of LC in two studies[17,38] and mortality from HCC in one study[22]. The quality scores ranged from 7 to 9 stars on the Newcastle–Ottawa Quality Assessment Scale.

Table 1

Characteristics of studies included in the meta-analysis

All 10 studies provided RR estimates adjusted for alcohol drinking, all but one study[18] reported RR estimates adjusted for tobacco smoking, all but two[23,25]. studies presented RR estimates adjusted for age and sex, all but two[20,21] studies demonstrated results further adjusted for history of liver disease, five[17,18,21,23,38] for body mass index (BMI), six[17,18,21,22,25,38] for history of diabetes mellitus, three[21,22,38] for education, three[17,20,21] for tea drinking, one[20][] for vegetables consumption and one[17] for serum alanine aminotransferase (ALT) level. There were five studies providing RRs for HCC in men and women, respectively[17,20,22,24,38], five offering information on history of liver disease[17,20,22,24,25], three reporting data on tobacco smoking[20,24,38] and three on alcohol history[20,24,38].

Drinkers and lowest/highest drinkers versus non-drinkers

The study-specific and summary RRs of HCC for coffee drinkers versus non-drinkers are shown in Figure 1. There was a significant relationship between coffee intake and HCC (RR: 0.64, 95% CI: 0.55, 0.75). Significant heterogeneity was found between studies (χ2 = 20.78, P = 0.01, I2 = 57%).

The point estimate and the total meta-analysis result for association between HCC and lowest/highest coffee drinkers versus non-drinkers are shown in Figures 2 and 3, respectively. The summary RR for lowest drinkers was 0.75 (95% CI: 0.65, 0.87), while for highest drinkers was 0.48 (95% CI: 0.39, 0.60). No significant heterogeneity was observed between studies for highest (χ2 = 3.73, P = 0.93, I2 = 0%) or lowest coffee consumption (χ2 = 5.73, P = 0.77, I2 = 0%).

Subgroup analyses

Table 2 shows the pooled RRs for coffee drinkers versus non-drinkers in subgroup analyses. Compared with male coffee drinkers, women tended to show a lower risk of HCC. The overall RR was 0.58 (95% CI: 0.47, 0.71) for men[17,20,22,24,38], and 0.64 (95% CI: 0.50, 0.81) for women[17,20,22,24,38]. A significant RR of 0.63 was found after pooling results from 8 Japan studies[17,20,22,23,24,25])], while no meta-analyses was conducted because of only one study included from Singapore[21] and Finland[38], respectively. With regard to duration of follow-up, there were six[17,18,21,23,25,38] and four[20,22,24] studies of which the follow-up periods were over and <10 years, respectively. The overall RR was 0.62 for studies with long duration, and 0.64 for studies with short periods of follow-up.

Table 2

Subgroup analysis for associations between HCC and coffee drinkers versus non-drinkers

There was a significant relationship found between HCC and coffee drinkers versus non-drinkers stratified by alcohol consumption, with a pooled RR of 0.57 (95% CI: 0.45, 0.71) for non-alcohol drinkers[20,24,38], and 0.57 (95% CI: 0.40, 0.80) for alcohol drinkers[20,24,38]. The summary RR was 0.59 for subjects with self-reported history of hepatitis B and/or C or other liver diseases[17,20,22,24,25], and 0.74 for subjects without history of liver diseases[17,20,22,24,25]. For smoking status, the combined RR was 0.49 (95% CI: 0.30, 0.82) for smokers[20,24,38], which tended to be lower than that for non-smokers[20,24,38] (RR=0.59, 95% CI: 0.39, 0.89). The pooled RR was 0.65 from 7 studies of high quality[17,18,21,22,24,38], and 0.62 from 3 studies of low quality[20,23,25].

Results of assessment of heterogeneity among studies are also shown in Table 2. There was marginally significant heterogeneity for males (χ2 = 7.86, P = 0.10, I2 = 49%) and for alcohol drinkers (χ2 = 4.21, P = 0.12, I2 = 52%). Nevertheless, significant heterogeneity was found between studies with long periods of follow-up and of high quality. When data were stratified by smoking status, there was significant heterogeneity between studies for both smokers and non-smokers (Table 2).

Sensitivity analyses

A sensitivity analysis applying a fixed effects model indicated the robustness of the results of relationship between coffee consumption and HCC (RR = 0.64, 95% CI: 0.55, 0.75 for drinkers; RR = 0.75, 95% CI: 0.65, 0.87 for lowest coffee drinkers;RR = 0.48, 95% CI: 0.39, 0.60 for highest coffee drinkers).

When Bayesian meta-analyses were applied using a ‘non-informative’ prior distribution (gamma distribution for the between-study variance), the pooled RR was 0.65 (95% CrI: 0.54, 0.75) for drinkers, 0.75 (95% CrI: 0.63, 0.88) for lowest drinkers and 0.48 (95% CrI: 0.38, 0.61) for highest drinkers versus non-drinkers, which were consistent with results of the classical approach (Table 3). The posterior probabilities of a beneficial risk of HCC associated with coffee intake (i.e. RR < 1) was very approximately 1, with all the probabilities >0.99. Similar results were found when the uniform distribution of the between-study SD was used, indicating the robustness of the results of Bayesian analyses.

Table 3

Sensitivity analyses for relationship between HCC and coffee using Bayesian meta-analyses

Dose–response analyses

The dose–response analysis showed a significant association, with the summary RR of 0.82 (95% CI: 0.78, 0.87) for an increment of one cup of coffee per day after pooling all the studies. When stratified by region, the dose–response analysis presented a significant association between risk of HCC and each increased cup of coffee in Japan[17,18,20,22,23,24,25] (RR=0.72, 95% CI: 0.65, 0.79). The RR for participants without history of liver diseases was marginally significant[22,25] (RR = 0.61, 95% CI: 0.37, 1.00, P = 0.051), while the RR was 0.68 (95% CI: 0.53, 0.87)[17,22,25] for subjects with history of liver diseases. With regard to gender-specific dose–response analysis, results from three studies[20,22,38] showed a significant relationship with decreased risk of HCC for males (RR=0.81, 95% CI: 0.76, 0.87), whereas the association was not significant for females (RR=0.90, 95% CI: 0.80, 1.02).

Publication bias

There was no evidence of publication bias in the studies of coffee consumption and HCC for drinkers (Figure 4; Begg P = 0.93, Egger P = 0.28), for lowest drinkers (Figure 5; Begg P = 0.25, Egger P=0.32) and for highest drinkers (Figure 6; Begg P = 0.13, Egger P = 0.15) versus non-drinkers.

Funnel plot for coffee drinkers versus non-drinkers.

Funnel plot for coffee lowest drinkers versus non-drinkers (the lowest coffee drinkers defined as the second stratum in each included study).

Funnel plot for coffee highest drinkers versus non-drinkers (the highest coffee drinkers were defined as the last stratum in each included study).

Discussion

The authors have referenced some of their own studies in this systematic review. These referenced studies have been conducted in accordance with the Declaration of Helsinki (1964) and the protocols of these studies have been approved by the relevant ethics committees related to the institution in which they were performed. All human subjects, in these referenced studies, gave informed consent to participate in these studies.

The findings of current meta-analyses from 10 cohort studies demonstrated the risk of HCC for the coffee drinkers, highest drinkers and lowest drinkers was approximately 36%, 52%, 25% lower than the non-drinkers, respectively. These results were similar to other meta-analyses that included both cohort and case–control studies[27,29]. No significant heterogeneity was found between studies for highest/lowest drinkers versus non-drinkers, while results for drinkers versus non-drinkers implied significant heterogeneity (χ2=20.78, P=0.01, I2=57%). In a dose–response analysis, the pooled RR for an increment of one cup of coffee per day was 0.82 (95% CI: 0.78, 0.87).

With regard to subgroup analyses to explore heterogeneity, a significant association was found in Finland and Japan, but not in Singapore (Table 2). Only highest drinkers experienced a marginally 44% decreased risk of HCC for the Singapore Chinese (HR = 0.56, 95% CI: 0.31, 1.00, P = 0.05)[21]. The discrepancy may be due to racial differences or differences in coffee drinking habits. Inverse relation between coffee and HCC was consistently found across other subgroup analyses. However, there was significant heterogeneity among studies for males, alcohol drinkers and smoking status with long periods of follow-up and of high quality (Table 2).

Bayesian meta-analyses can calculate the posterior probability that coffee intake provides a reduced risk of HCC, and explore the robustness of the analyses by comparing results from a classical approach and under different assumptions (i.e. with different distributions for between-study variance or SD)[36]. The significant results were consistent with the classical meta-analyses. Moreover, all the posterior probabilities of a beneficial risk of HCC in coffee drinkers and lowest/highest drinkers were approximately 1 (i.e. >0.99), which could further corroborate the positive relationship between coffee consumption and decreased risk of HCC.

The outcome in this study was HCC; however, there were four studies included providing data on LC. Nevertheless, the pooled results of coffee drinkers versus non-drinkers stratified by HCC (RR = 0.69, 95% CI: 0.57, 0.83) and LC (RR = 0.57, 95% CI: 0.45, 0.71) were very similar to the result using all the included studies (Figure 1). There were three[17,20,21] studies providing data on tea drinking. No significant associations were found between coffee and risk of HCC after adjusting for tea consumption[20,21], or between tea drinking and risk of HCC after adjusting for coffee[17]. It may indicate that the relationship between decreased risk of HCC and coffee consumption would be biased by tea intake to some extent. However, in this study no meta-analysis taking into tea intake account was conducted for further clarification, because no sufficient information could be extracted.

Summary RRs of HCC for coffee drinkers versus non-drinkers.HCC, hepatocellular carcinoma.

Summary RRs of HCC for coffee lowest drinkers versus non-drinkers (the lowest coffee drinkers defined as the second stratum in each included study). HCC, hepatocellular carcinoma.

Summary RRs of HCC for coffee highest drinkers versus non-drinkers (the highest coffee drinkers defined as the last stratum in each included study).HCC, hepatocellular carcinoma.

It remained unclear which ingredient of coffee could protect coffee drinkers against HCC[24]. Biologically, there were three major components of coffee which had been considered to contribute to the beneficial effects against liver carcinogenesis: chlorogenic acids, diterpenes cafestol and kahweol and caffeine[39,40]. However, several studies argued that caffeine may not be related to reduce risk of HCC[17,20,24,41], while Johnson’s study[21] attested and concluded the inverse association between caffeine consumption and risk of HCC. Owing to limited information retrieved in this study, however no meta-analysis could be performed to examine the relationship between caffeine and risk of HCC.

The significant result of dose–response analysis was similar to the findings from three previous meta-analyses[28,29,31], which reported the RR of 0.57 related to two cups of coffee per day[31] and the approximate RR of 0.80 associated with one cup of coffee per day[28,29]. In this study, further stratification of dose–response analyses found that males and drinkers without history of liver diseases tend to show a lower risk of HCC compared with females and drinkers with history of liver diseases, respectively. It may imply different beneficial effect by incremental coffee consumed per day on different gender and subjects with or without liver diseases[31]. Nevertheless, taking into account the sample size, more research is needed to clarify the stratified dose–response analyses.

Critical appraisal of the validity of relevant articles

Meta-analysis had some advantages. Firstly, a comprehensive search of the literature could warrant summarising all the available evidence till date of relationship between coffee intake and risk of HCC. Secondly, the included studies were of high quality with good quality scores, thus leading to convincing results. Thirdly, detailed analyses including subgroup, sensitivity and dose–response analyses were conducted to utilise the existing evidence thoroughly.

There were also several limitations in this study. Firstly, the possibility of bias and confounding cannot be fully controlled or excluded in nutritional observational studies. For instance, it was postulated that tea drinking may bias the association between coffee and reduced risk of HCC based on the individual findings from the included cohort studies[17,20,21]. Secondly, the results are likely to be influenced by misclassification of coffee consumption, because each study presented coffee consumption in different units (cups/week, cups/day, days/week, daily drinks, times/week). Furthermore, the majority of included studies were still from Japan, with only one study from Singapore and one from Finland. Therefore, the findings may not be generalised to other populations. Moreover, detailed information on coffee consumption (e.g. what was the method of preparation, which brewing method was chosen, it was decaffeinated or caffeinated, etc.) was not available, which may account for, at least in part the significant heterogeneity among the included studies.

Conclusion

The current meta-analysis confirms that coffee consumption is associated with reduced risk of HCC. The inverse relationship between coffee and decreased risk of HCC was consistent across different populations and settings.

Clinical applicability

Nevertheless, given the observational design and the potential confounding, the causality remains open to discussion and more better-controlled studies are warranted.

Acknowledgement

The authors thank Mr Yunzheng Mo for his help in writing and editing the manuscript.

Abbreviations list

ALT, alanine aminotransferase; BMI, body mass index; CIs, confidence intervals; CrI, credible interval; HRs, hazard ratios; HCC, hepatocellular carcinoma; LC, Liver cancer; ORs, odds ratios; PLC, primary liver cancer; RR, relative risk; SD, standard deviation.

Authors contribution

All authors contributed to the conception, design, and preparation of the manuscript, as well as read and approved the final manuscript.

Competing interests

None declared.

Conflict of interests

None declared.

A.M.E

All authors abide by the Association for Medical Ethics (AME) ethical rules of disclosure.

References

  • 1. Cano-Marquina A, Tarin JJ, Cano A. The impact of coffee on health. Maturitas 2013 May;75(1):7-21.
  • 2. Choi HK, Curhan G. Coffee consumption and risk of incident gout in women: the Nurses’ Health Study. Am J Clin Nutr 2010 Oct;92(4):922-7.
  • 3. Choi HK, Willett W, Curhan G. Coffee consumption and risk of incident gout in men: a prospective study. Arthritis Rheum 2007 Jun;56(6):2049-55.
  • 4. Huxley R, Lee CM, Barzi F, Timmermeister L, Czernichow S, Perkovic V. Coffee, decaffeinated coffee, and tea consumption in relation to incident type 2 diabetes mellitus: a systematic review with meta-analysis. Arch Intern Med 2009 Dec;169(22):2053-63.
  • 5. Modi AA, Feld JJ, Park Y, Kleiner DE, Everhart JE, Liang TJ. Increased caffeine consumption is associated with reduced hepatic fibrosis. Hepatology 2010 Jan;51(1):201-9.
  • 6. Saaksjarvi K, Knekt P, Rissanen H, Laaksonen MA, Reunanen A, Mannisto S. Prospective study of coffee consumption and risk of Parkinson’s disease. Eur J Clin Nutr 2008 Jul;62(7):908-15.
  • 7. Tanaka T, Nishikawa A, Shima H, Sugie S, Shinoda T, Yoshimi N. Inhibitory effects of chlorogenic acid, reserpine, polyprenoic acid (E-5166), or coffee on hepatocarcinogenesis in rats and hamsters. Basic Life Sci 1990 Jan;52(1):429-40.
  • 8. Radoi L, Paget-Bailly S, Menvielle G, Cyr D, Schmaus A, Carton M. Tea and coffee consumption and risk of oral cavity cancer: results of a large population–based case–control study, the ICARE study. Cancer Epidemiol 2013 Jun;37(3):284-9.
  • 9. Li G, Ma D, Zhang Y, Zheng W, Wang P. Coffee consumption and risk of colorectal cancer: a meta-analysis of observational studies. Public Health Nutr 2012 Jun;16(2):346.
  • 10. Wilson KM, Balter K, Moller E, Adami HO, Andren O, Andersson SO. Coffee and risk of prostate cancer incidence and mortality in the Cancer of the Prostate in Sweden Study. Cancer Causes Contr 2013 Aug;24(8):1575-81.
  • 11. Yu X, Bao Z, Zou J, Dong J. Coffee consumption and risk of cancers: a meta-analysis of cohort studies. BMC Cancer 2011 Mar;1196-106.
  • 12. Jemal A, Bray F, Center MM, Ferlay J, Ward E, Forman D. Global cancer statistics. CA Cancer J Clin 2011(Mar–Apr);61(2):69-90.
  • 13. Perz JF, Armstrong GL, Farrington LA, Hutin YJ, Bell BP. The contributions of hepatitis B virus and hepatitis C virus infections to cirrhosis and primary liver cancer worldwide. J Hepatol 2006 Oct;45(4):529-38.
  • 14. Chuang SC, La Vecchia C, Boffetta P. Liver cancer: descriptive epidemiology and risk factors other than HBV and HCV infection. Cancer Lett 2009 Dec;286(1):9-14.
  • 15. Bosetti C, Levi F, Boffetta P, Lucchini F, Negri E, La Vecchia C. Trends in mortality from hepatocellular carcinoma in Europe, 1980–2004. Hepatology 2008 Jul;48(1):137-45.
  • 16. Fares N, Peron JM. Epidemiology, natural history, and risk factors of hepatocellular carcinoma. Rev Prat 2013 Feb;63(2):216-17, 220–2.
  • 17. Inoue M, Kurahashi N, Iwasaki M, Shimazu T, Tanaka Y, Mizokami M. Effect of coffee and green tea consumption on the risk of liver cancer: cohort analysis by hepatitis virus infection status. Cancer Epidemiol. Biomarkers Prev 2009 Jun;18(6):1746-53.
  • 18. Michikawa T, Inoue M, Sawada N, Iwasaki M, Tanaka Y, Shimazu T. Development of a prediction model for 10-year risk of hepatocellular carcinoma in middle-aged Japanese: the Japan Public Health Center-based Prospective Study Cohort II. Prev Med 2012 Jun;55(2):137-43.
  • 19. Hamed MA, Ali SA. Non-viral factors contributing to hepatocellular carcinoma. World J Hepatol 2013 Jun;5(6):311-22.
  • 20. Inoue M, Yoshimi I, Sobue T, Tsugane S. Influence of coffee drinking on subsequent risk of hepatocellular carcinoma: a prospective study in Japan. J Natl Cancer Inst 2005 Feb;97(4):293-300.
  • 21. Johnson S, Koh WP, Wang R, Govindarajan S, Yu MC, Yuan J. Coffee consumption and reduced risk of hepatocellular carcinoma: findings from the Singapore Chinese Health Study. Cancer Causes Control 2011 Mar;22(3):503-10.
  • 22. Kurozawa Y, Ogimoto I, Shibata A, Nose T, Yoshimura T, Suzuki H. Coffee and risk of death from hepatocellular carcinoma in a large cohort study in Japan. Br J Cancer 2005 Aug;93(5):607-10.
  • 23. Ohishi W, Fujiwara S, Cologne JB, Suzuki G, Akahoshi M, Nishi N. Risk factors for hepatocellular carcinoma in a Japanese population: a nestedcase-control study. Cancer Epidemiol Biomarkers Prev 2008 Apr;17(4):846-54.
  • 24. Shimazu T . Coffee consumption and the risk of primary liver cancer Pooled analysis of two prospective studies in Japan. Int J Cancer 2005 Aug;116(1):150-4.
  • 25. Wakai K, Kurozawa Y, Shibata A, Fujita Y, Kotani K, Ogimoto I. Liver cancer risk, coffee, and hepatitis C virus infection: a nested case–control study in Japan. Br J Cancer 2007 Jul;97(3):426-28.
  • 26. Ohfuji S, Fukushima W, Tanaka T, Habu D, Tamori A, Sakaguchi H. Coffee consumption and reduced risk of hepatocellular carcinoma among patients with chronic type C liver disease: a case-control study. Hepatol Res 2006 Nov;36(3):201-8.
  • 27. Sang LX, Chang B, Li XH, Jiang M. Consumption of coffee associated with reduced risk of liver cancer: a meta-analysis. BMC Gastroenterol 2013 Feb;1334.
  • 28. Bravi F, Bosetti C, Tavani A, Bagnardi V, Gallus S, Negri E. Coffee drinking and hepatocellular carcinoma risk: A meta-analysis. Hepatology 2007 Mar;46(2):430-35.
  • 29. Bravi F, Bosetti C, Tavani A, Gallus S, La Vecchia C. Coffee reduces risk for hepatocellular carcinoma: an updated meta-analysis. Clin Gastroenterol Hepatol 2013 Nov;11(11):1413-21.
  • 30. Bravi F, Bosetti C, Tavani A, La Vecchia C. Coffee drinking and hepatocellular carcinoma: an update. Hepatology 2009 Oct;50(4):1317-8.
  • 31. Larsson SC, Wolk A. Coffee consumption and risk of liver cancer: a meta-analysis. Gastroenterology 2007 May;132(5):1740-5.
  • 32. Devereaux P, Yusuf S. The evolution of the randomized controlled trial and its role in evidence-based decision making. J Intern Med 2003 Aug;254(2):105-13.
  • 33. Wells GA, Shea B, O’Connell D, Peterson J, Welch V, Losos M. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analyses. http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp [Accessed 1 Sep 2013] .
  • 34. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002 Jun;21(11):1539-58.
  • 35. Hedges LV, Pigott TD. The power of statistical tests in meta-analysis. Psychol Methods 2001 Sep;6(3):203-17.
  • 36. Spiegelhalter DAKR, Myles JP. Bayesian approaches to clinical trials and health care evaluation, statistics in practice; variation: statistics in practice. Chichester, England: John Wiley & Sons .
  • 37. Greenland S . Quantitative methods in the review of epidemiologic literature. Epidemiol Rev 1987;91-30.
  • 38. Hu G, Tuomilehto J, Pukkala E, Hakulinen T, Antikainen R, Vartiainen E. Joint effects of coffee consumption and serum gamma-glutamyltransferase on the risk of liver cancer. Hepatology 2008 Jul;48(1):129-36.
  • 39. Tao KS, Wang W, Wang L, Cao DY, Li YQ, Wu SX. The multifaceted mechanisms for coffee’s anti-tumorigenic effect on liver. Med Hypotheses 2008 Nov;71(5):730-36.
  • 40. La Vecchia C . Coffee, liver enzymes, cirrhosis and liver cancer. J Hepatol 2005 Apr;42(4):444-6.
  • 41. Kiyosawa K, Umemura T, Ichijo T, Matsumoto A, Yoshizawa K, Gad A. Hepatocellular carcinoma: recent trends in Japan. Gastroenterology 2004 Nov;127(5 Suppl 1):S17-26.
Licensee to OAPL (UK) 2014. Creative Commons Attribution License (CC-BY)

Characteristics of studies included in the meta-analysis

Author, year Country No. of cases/cohort size or controls Study period Coffee consumption RR/HR/OR(95%CI) Adjustment factors Quality score
Inoue 2005 Japan 334/90,452 1990–2001 almost never Reference Age, sex, study centre, tobacco smoking, alcohol drinking, vegetables consumption, tea drinking 7
1~2days/week 0.75 (0.56–1.01) M 0.74 (0.52–1.05) F 0.77 (0.43–1.37)
3~4days/week 0.79 (0.55–1.14) M 0.76 (0.50–1.16) F 0.89 (0.43–1.84)
1~2cups/day 0.52 (0.38–0.73) M 0.55 (0.38–0.80) F 0.43 (0.20–0.90)
3~4cups/day 0.48 (0.28–0.83) M 0.41 (0.21–0.77) F 0.89 (0.31–2.59)
>=5cups/day 0.24 (0.08–0.77) M 0.27 (0.09–0.87) F (N/A)*
Kurozawa 2005 Japan 258/83,966 1988–1999 almost never Reference Age, sex, education, history of diabetes and liver disease, tobacco smoking, alcohol drinking 8
<1cups/day 0.83 (0.54–0.25) M 0.91 (0.57–1.45) F 0.64 (0.27–1.51)
>=1cups/day 0.50 (0.31–0.79) M 0.49 (0.28–0.85) F 0.51 (0.20–1.31)
Shimazu 2005 Cohort 1 Japan 70/22,404 1984–1992 never Reference Age, sex, history of liver disease, tobacco smoking, alcohol drinking 9
occasionally 0.56 (0.33–0.97)
>=1cups/day 0.53 (0.28–1.00)
Shimazu 2005 Cohort 2 Japan 47/38,703 1990–1997 never Reference Age, sex, history of liver disease, tobacco smoking, alcohol drinking 8
occasionally 1.05 (0.52–2.16)
>=1cups/day 0.68 (0.31–1.51)
Shimazu 2005 combin never Reference Age, sex, history of liver disease, tobacco smoking, alcohol drinking 9
occasionally 0.71 (0.46–1.09) M 0.73 (0.44–1.21) F 0.66 (0.28–1.57)
>=1cups/day 0.58 (0.36–0.96) M 0.64 (0.37–1.12) F 0.54 (0.14–2.07)
Wakai 2007 Japan 96/110,792 1988–1999 non-drinker Reference Area, smoking and drinking habits, history of diabetes mellitus and liver diseases 6
<1cups/day 0.77 (0.45 – 1.32)
>=1cups/day 0.49 (0.25 – 0.96)
Hu 2008 Finland 128/60,323 1997 – 2002 <1cups/day Reference Age, sex, study year, alcohol drinking, tobacco smoking, education, diabetes and chronic liver disease, BMI 9
2~3cups/day 0.66 (0.37–1.16) M 0.68 (0.35–1.31) F 0.62 (0.19–2.04)
4~5cups/day 0.44 (0.25–0.77) M 0.35 (0.18–0.71) F 0.60 (0.20–1.82)
6~7cups/day 0.38 (0.21–0.69) M 0.31 (0.15–0.63) F 0.58 (0.19–1.82)
>=8cups/day 0.32 (0.12–0.62) M 0.28 (0.13–0.61) F 0.41 (0.10–1.70)
Ohishi 2008 Japan 139/472 1999–2002 never Reference Hepatitis virus infection, alcohol consumption, smoking habit, BMI, diabetes mellitus, and radiation dose to the liver. 7
daily 0.40 (0.16–1.02)
Inoue 2009 Japan 110 /18,815 1994–2006 almost never Reference Age, sex, study centre, tobacco smoking, alcohol drinking, BMI, history of diabetes mellitus, tea drinking, serum ALT level, HCV infection, HBV infection 8
<1cups/day 0.67 (0.42–1.07) M 0.79 (0.46–1.37) F 0.39 (0.15–1.03)
1~2cups/day 0.49 (0.27–0.91) M 0.37 (0.17–0.81) F 0.92 (0.36–2.38)
>=3cups/day 0.54 (0.21–1.39) M 0.32 (0.10–1.10) F 0.69 (0.11–4.22)
Johnson 2011 Singapore 362/63,257 1993–2006 never Reference Age, sex, dialect group, study year, BMI, education, alcohol drinking, tobacco smoking, tea drinking, history of diabetes 8
<1cups/day 0.94 (0.63–1.40)
1~2cups/day 1.17 (0.87–1.56)
2~3cups/day 0.78 (0.56–1.07)
>=3cups/day 0.56 (0.31–1.00)
Michikawa 2012 Japan 104/17,654 1993–2006 almost never Reference Age, sex, area, alcohol consumption, body mass index, diabetes, coffee consumption, hepatitis B surface antigen, and anti-hepatitis C virus antibody 8
<1cups/day 0.77 (0.48–1.21)
>=1cups/day 0.42 (0.23–0.72)

M: male; F: female; ALT: alanine aminotransferase; BMI: body mass index; HBV: hepatitis B virus; HCV: hepatitis C virus

* N/A: no data could be extracted

Subgroup analysis for associations between HCC and coffee drinkers versus non-drinkers

Subgroup analysis No. of cases RR (95% CI)a Heterogeneity
τ2b P-value 12 (%)
Gender Male 674 0.58 (0.47, 0.71) 0.03 0.10 49
Female 273 0.64 (0.50, 0.81) 0.00 0.96 0
Geographic region Japan 1159 0.63 (0.56, 0.70) 0.00 0.88 0
Singapore 362 0.92 (0.76, 1.10) c - -
Finland 128 0.46 (0.34, 0.63) c - -
Follow-up period Shortd 709 0.64 (0.56, 0.73) 0.00 0.61 0
Longd 940 0.62 (0.47, 0.82) 0.08 <0.01 73
Alcohol consumption Yes e 0.57 (0.40, 0.80) 0.05 0.12 52
No e 0.57 (0.45, 0.71) 0.00 0.37 0
History of liver disease Yes 580 0.59 (0.50, 0.70) 0.00 0.88 0
No 250 0.74 (0.58, 0.94) 0.00 0.87 0
Smoking status Yes e 0.49 (0.30, 0.82) 0.15 0.01 76
No e 0.59 (0.39, 0.89) 0.09 0.06 63
Study quality Lowf 569 0.62 (0.53, 0.73) 0.00 0.62 0
Highf 1080 0.65 (0.52, 0.81) 0.06 <0.01 68

aRR, relative risk; CI, confidence interval.

bτ2, between-study variance.

cNot applicable because of only one study included.

dShort period of follow-up, < 10 years; long period, < 10 years.

eNo data could be extracted.

fLow study quality: scores of Newcastle–Ottawa Quality Assessment Scale ≤ 7; high study quality: scores > 7.

Sensitivity analyses for relationship between HCC and coffee using Bayesian meta-analyses

Meta-analysis approach Drinkers versus non-drinkers Lowest versus non-drinkers Highest versus non-drinkers
RR (95% CrI/CI)a Probability of RR<1 τ2b RR (95% CrI/CI) Probability of RR<1 τ2 RR (95% CrI/CI) Probability of RR<1 τ2
Bayesian analysis
‘Non-informative’ prior distributionc 0.65(0.54–0.75) >0.99 0.04 0.75(0.63–0.88) >0.99 0.01 0.48(0.38–0.61) >0.99 0.02
‘Non-informative’ prior distributiond 0.64(0.53–0.77) >0.99 0.06 0.75(0.63–0.89) >0.99 0.02 0.48(0.37–0.61) >0.99 0.03
Classical analysis 0.64(0.55–0.75) 0.03 0.75(0.65–0.87) 0.00 0.48(0.39–0.60) 0.00

aCrI, credible interval; CI, confidence interval.

bτ2, between-study variance.

cUsing gamma distribution for the between-study variance.

dUsing uniform distribution for the between-study standard deviation.

Keywords