Coffee and the Risk of Hepatocellular Carcinoma: A Systematic Review and Meta-Analysis of Mendelian Randomization Studies.

INTRODUCTION
Liver cancer is one of the most frequently diagnosed malignancies and ranks third in global cancer-related deaths.1 Hepatocellular carcinoma (HCC), the primary form of liver cancer, poses a growing public health challenge worldwide. Given its substantial and increasing global burden, identifying modifiable risk or protective factors for HCC remains a critical public health priority.
Coffee is among the most widely consumed beverages globally, following tea.2 As its popularity has grown, so has interest in its potential health benefits, including a possible role in lowering liver cancer risk.
Numerous epidemiological studies have explored this association, and several meta-analyses based on observational studies have reported an inverse association between coffee intake and HCC development.2-5 However, these studies often struggle to establish causality because they are prone to confounding and reverse causation.
Mendelian randomization (MR) offers a powerful alternative to observational study designs by using genetic variants associated with exposures of interest as instrumental variables. MR studies can accurately estimate causality by consolidating temporal causality using genome analysis and can minimize the influence of confounding variables.6 Therefore, to more robustly evaluate the causal relationship between coffee consumption and the risk of HCC, we conducted a systematic review and meta-analysis including only MR studies.

METHODS
Systematic literature searches were conducted on February 20, 2025, using PubMed, EMBASE, Web of Science, and the Cochrane Database. The search terms included “coffee,” “caffeine,” and a range of liver-related conditions such as “HCC,” “fatty liver,” and “cirrhosis.” The full search strategy is detailed in the Supplementary Materials.
Identified studies were collected and imported into EndNote X9, and duplicate studies were removed. Two reviewers (H.B.C. and H.K.) independently reviewed titles and abstracts, followed by full-text screening for inclusion. Studies were eligible if they assessed the relationship between coffee intake and HCC using MR approaches. Five studies met all inclusion criteria. A Preferred Reporting Items for Systematic review and Meta-Analyses (PRISMA) flow diagram summarizing the selection process is provided in Supplementary Fig. 1.
Two independent researchers (H.B.C. and H.K.) extracted data from the included articles. Extracted data included the name of the primary author, publication year, country and database where the study was conducted, inverse variance weighted (IVW), MR-Egger, weighted median, and weighted mode values. The accuracy of the data was additionally checked by another researcher (J.J.Y.).
This study was registered with the PROSPERO database (Registration ID: CRD420251026479). Statistical analysis was conducted using R version 4.4.3, and pooled odds ratios (ORs) with 95% confidence intervals (CIs) were calculated.

RESULTS
Table 1 outlines the demographic and methodological characteristics of the five MR studies included. Among the five MR studies included, four (Deng et al.,7 Zhao et al.,8 Carter et al.,9 and Hu et al.10) applied three MR analytical approaches: IVW, MR-Egger, and weighted median. One study (Zhang et al.11) utilized only the IVW method. All single-nucleotide polymorphisms (SNPs) used in the included MR studies were strongly associated with coffee consumption at genome-wide significance (p<5×10-8) and were clumped using linkage disequilibrium pruning (R2<0.01) to ensure independence. Furthermore, all SNPs had F-statistics greater than 10, indicating a low risk of weak instrument bias.
Three MR methods were employed in this meta-analysis. IVW was designated as the primary analytical approach due to its superior statistical power in establishing causality in MR studies.12 MR-Egger and weighted median methods were also considered, as they provide robust estimates even when assumptions about instrumental variable validity are partially violated.13
Moderate to substantial heterogeneity was observed across the three methods: IVW (I2=62.6%, τ2=0.058) (Fig. 1A), MR-Egger (I2=44.1%, τ2=0.091) (Supplementary Fig. 2A), and weighted median (I2=70.6%, τ2=0.051) (Supplementary Fig. 3A).14,15 Accordingly, a random-effects model was applied in all analyses.
In the IVW analysis, coffee consumption was not significantly associated with the risk of HCC (OR, 0.92; 95% CI, 0.58 to 1.47; pheterogeneity=0.030) (Fig. 1A). The results of the MR-Egger (OR, 0.84; 95% CI, 0.49 to 1.44; pheterogeneity=0.147) (Supplementary Fig. 2A) and weighted median (OR, 0.91; 95% CI, 0.58 to 1.44; pheterogeneity=0.017) (Supplementary Fig. 3A) analyses were consistent with the IVW method and likewise indicated no statistically significant association between coffee intake and HCC risk.
No evidence of publication bias was detected in all three methods (IVW: p=0.624, MR-Egger: p=0.497; weighted median: p=0.497) (Fig. 1B, Supplementary Figs 2B and 3B). Funnel plots were also visually symmetrical, further supporting the absence of publication bias. The quality of each included study was independently assessed by two reviewers using the Risk of Bias Assessment tool for Non-randomized Studies (RoBANS), as described in the Supplementary Materials and Methods.

DISCUSSION
Previous meta-analyses based solely on observational studies and have concluded that coffee consumption has a protective effect against HCC. However, these designs are inherently limited in their ability to establish causality due to potential confounding and reverse causality. For instance, individuals with impaired liver function may intentionally avoid coffee intake, potentially leading to a biased inverse relationship.
In contrast, MR studies utilize genetic variants as instrumental variables, allowing for more robust causal inference by minimizing the impact of confounding variables and reverse causation. As genetic variants are randomly assigned at conception and remain unchanged throughout life, MR studies preserve temporal ordering and are less affected by behavioral or environmental factors. Based on these methodological strengths, we conducted a meta-analysis including only MR studies to assess the causal relationship between coffee consumption and HCC incidence. Given the nature of MR studies, the findings of our analysis may offer more reliable causal estimates than those derived from observational evidence.
An important limitation of our meta-analysis is the potential impact of ethnic heterogeneity across the included study populations. Among the five MR studies analyzed, four were conducted using data from individuals of European ancestry (UK Biobank and FinnGen), while only one study utilized an East Asian cohort from Biobank Japan.7-11 Interestingly, the Japanese study reported a significant protective effect of coffee consumption on HCC risk, whereas the studies based on European populations did not identify such an association. This divergence may be attributed to genetic, metabolic, or environmental differences between populations that could modify the biological effects of coffee. Although the limited number of MR studies currently precludes formal subgroup analyses by ethnicity, these findings underscore the importance of including ethnically diverse cohorts in future MR investigations to better understand potential gene–environment interactions and population-specific effects.
In addition, we also observed moderate to substantial heterogeneity across the included studies, particularly in the IVW and weighted median analyses. To address this, we employed a random-effects model to generate more conservative and generalizable effect estimates. Although the MR-Egger intercept test is commonly used to assess horizontal pleiotropy, its application requires SNP-level summary data. Because our meta-analysis was based on aggregated effect estimates extracted from published MR studies, it was methodologically infeasible to perform or synthesize MR-Egger intercept tests across studies.
In conclusion, our findings suggest that the inverse association between coffee intake and HCC development reported in previous observational meta-analyses may have been overstated due to inherent methodological limitations. In contrast, our MR-based meta-analysis did not identify a statistically significant association, suggesting that the relationship between coffee consumption and HCC incidence may be weaker than previously suggested and possibly non-causal.