Liver research is where coffee's numbers get dramatic. In most of the coffee literature, associations are modest: risk reductions in the range of ten to twenty percent, easily explained away by confounding. In liver disease, the reported effect sizes are several times larger, they show up consistently, and they persist in populations where you would expect other factors to overwhelm them. This article covers why that is interesting, and why the researchers who produced the largest of those numbers simultaneously rated their own evidence as very low quality.
The meta-analysis
The central paper is by Kennedy and colleagues, published in BMJ Open in 2017[1], examining coffee and hepatocellular carcinoma, the most common form of primary liver cancer. They pooled 18 cohort studies involving 2,272,642 participants and 2,905 cases, plus 8 case-control studies involving 1,825 cases and 4,652 controls.
The headline result was that an extra two cups of coffee per day was associated with a 35% reduction in the risk of hepatocellular carcinoma, a relative risk of 0.65 with a confidence interval of 0.59 to 0.72. When they separated study types, the association was weaker among the cohort studies, at a relative risk of 0.71, than among the case-control studies, at 0.53. That difference matters and is discussed below, because cohort studies are generally the more reliable design here.
The detail that makes this hard to dismiss
The most striking finding is not the size of the association but its stubbornness. The researchers tested whether the effect was altered by the stage of pre-existing liver disease, by high alcohol consumption, by high body mass index, by type 2 diabetes, by smoking, and by hepatitis B and C infection. It was not significantly altered by any of them.
That is unusual. Those factors are the dominant causes of liver disease worldwide. If coffee's apparent benefit were simply a marker of people who drink less alcohol or who are otherwise healthier, you would expect the association to shrink or vanish once you separated out the heavy drinkers and the hepatitis patients. Instead it persisted across all of those strata, including among people who already had liver disease. That pattern is more consistent with something coffee itself is doing than with coffee marking out a healthier kind of person.
Caffeinated versus decaffeinated
The same paper looked at both. An extra two cups of caffeinated coffee was associated with a 27% reduction in risk, a relative risk of 0.73. An extra two cups of decaffeinated coffee was associated with a 14% reduction, a relative risk of 0.86, with a confidence interval touching 1.00, meaning the decaf finding sat right at the boundary of statistical significance and rested on only three cohort studies.
So decaf points the same way but more weakly, and on far thinner data. The cautious reading is that caffeine may contribute something here that it does not appear to contribute in the diabetes research, but that the decaf evidence is too sparse to build on. This is a good example of why it is worth reading how many studies sit behind a subgroup number rather than just the number itself.
It is also worth registering how few cases sit behind the cohort estimates. Across 2,272,642 participants there were 2,905 cases of hepatocellular carcinoma. That is a very large population producing a small number of events, which is normal for a relatively uncommon cancer but does mean the estimates rest on far fewer observations than the headline participant count implies.
Why the authors graded their own evidence as very low
This is the part that most coverage of coffee and liver health leaves out, and it is the most important sentence in the paper. Kennedy and colleagues applied the GRADE framework, a standard system for rating how much confidence a body of evidence deserves, and concluded that the quality of evidence was very low.
Three reasons drove that grade. There are no randomised controlled trials, so everything rests on observational data. Publication bias is likely, meaning studies finding an effect are more likely to have been published than studies finding nothing, which inflates pooled estimates. And there is no accepted standard definition of what a cup of coffee is across these studies, so the dose axis is genuinely fuzzy. The gap between the case-control and cohort estimates, 0.53 versus 0.71, is itself a warning sign, since case-control studies are more vulnerable to recall bias, where people who have been diagnosed with cancer remember their past habits differently.
It is entirely coherent to hold both facts at once: this is the largest consistent effect size in the coffee literature, and it is supported by evidence its own authors classify as very low quality. Those are not contradictory statements. They describe a strong signal in a weak type of data.
What the broader literature says
The liver findings are not isolated to cancer. The umbrella review by Poole and colleagues[2] looked across 201 meta-analyses and found coffee consumption associated with lower risk of several liver conditions alongside metabolic and neurological ones, with the liver outcomes among the more consistent in their survey. The mortality studies point the same way from a different angle: Gunter and colleagues in the European EPIC cohort[3] examined serum biomarkers in a subgroup of 14,800 participants and found coffee drinking associated with a more favourable profile of liver function markers, which is a mechanistically relevant observation rather than just another risk-ratio.
Beyond cancer: fibrosis and liver enzymes
Hepatocellular carcinoma is the endpoint with the cleanest meta-analysis, but it sits at the far end of a long disease process. Chronic liver injury, from alcohol, viral hepatitis or metabolic causes, produces inflammation, which produces fibrosis, which in some people becomes cirrhosis, which is the substrate on which most liver cancer develops. If coffee were acting anywhere along that chain, you would expect signals at the earlier stages too, and researchers generally report that they find them.
The most commonly reported are changes in liver enzyme levels, particularly gamma-glutamyl transferase and the aminotransferases, which are the standard blood markers used to detect liver injury. Coffee consumption is repeatedly associated with lower levels of these markers, and the association tends to be strongest in exactly the people whose levels are elevated to begin with. Studies using non-invasive measures of liver stiffness, a proxy for fibrosis, have pointed the same way.
This coherence across stages is part of why liver researchers take the coffee association more seriously than a single cancer meta-analysis would warrant on its own. A finding that appears in enzymes, in fibrosis measures and in cancer incidence, all in the same direction, is harder to attribute to one methodological quirk than a finding that appears at only one point in the process.
It also connects to something visible in the mortality data. Gunter and colleagues[3] examined biomarkers in a 14,800-person subgroup of the European cohort and reported coffee drinking associated with a more favourable profile of liver function markers. That is an independent dataset, using blood chemistry rather than disease diagnoses, arriving at a compatible answer.
None of this identifies a mechanism. Coffee's diterpenes, its chlorogenic acids and caffeine itself have all been proposed as acting on hepatic processes, and the honest position is that several plausible pathways exist without any of them being established as the operative one.
Proposed mechanisms include coffee's effects on liver enzyme levels, on hepatic fat accumulation, and on the fibrotic processes that turn chronic liver injury into cirrhosis. Chlorogenic acids and the diterpenes cafestol and kahweol have all been proposed as active compounds. None of this is settled, and the honest position is that researchers can describe several plausible pathways without being able to say which, if any, produces the association seen in the population data.
What this does not mean
It needs saying directly, because this research gets misrepresented more than any other coffee finding. None of it shows that coffee treats liver disease, protects a liver from alcohol, or offsets any other risk factor. The studies observed that people who drank more coffee developed hepatocellular carcinoma less often. They did not test coffee as an intervention in anybody.
The finding that the association held among people who drink heavily is frequently reported as though coffee were a countermeasure for drinking. It is not, and the study does not support that reading. What it supports is the narrower claim that the association did not disappear when researchers looked at heavy drinkers separately, which is a statement about the robustness of a statistical relationship, not a licence. Liver disease risk is driven overwhelmingly by alcohol, viral hepatitis and metabolic factors, and nothing in this literature changes that ordering.
Similarly, the finding that decaffeinated coffee showed a weaker association is sometimes reported as proof that caffeine is the protective agent. Three cohort studies is not enough to support that, and the confidence interval on the decaf estimate touched 1.00. The correct reading is that the decaf evidence here is too thin to conclude anything either way.
How to read a very low GRADE rating without dismissing the finding
It is worth spending a moment on what GRADE actually does, because this article turns on it. GRADE is a structured system for rating confidence in a body of evidence, and it starts observational research at low confidence by default before adjusting up or down for factors such as effect size, dose-response and risk of bias. A very low rating does not mean the researchers think the finding is wrong. It means that if you asked how confident anyone should be that the true effect is close to the estimated one, the answer is not very.
That framing is useful across this whole section. Most nutrition epidemiology would rate low or very low under GRADE, including findings that later turned out to be correct. The rating is a statement about the strength of the inference, not a verdict on the claim. What it should change is how you act on it: a very low rating is a reason to find a question interesting and a reason not to make decisions on it, which is exactly the posture this article recommends.
The honest summary
Coffee shows the largest and most persistent protective associations in the liver literature of anywhere in coffee research, the associations survive adjustment for the major competing causes of liver disease, and they appear across more than two million people. At the same time, the definitive meta-analysis rates its own evidence base as very low quality because it is entirely observational, likely affected by publication bias, and built on inconsistent definitions of a cup.
That is an unusual combination, and it is worth naming plainly rather than resolving it in either direction.
Two things follow for a reader. If you drink coffee and have wondered whether it is doing your liver any harm, this literature is uniformly reassuring on that narrow question. And if you have seen coffee described as protective against liver disease, that description outruns evidence its own authors graded as very low quality, however large the numbers attached to it look.
This is also the article in the section where the distance between what the headlines say and what the paper says is widest, which is why it is worth reading the methods rather than the abstract.
The gap between those two statements is where most reporting on this topic goes wrong, and holding both is the whole skill of reading nutrition research.
It is a habit worth practising, because this same gap opens up in almost every nutrition story you will read.
For a reader, that combination justifies interest and not action. If you already drink coffee, this is a reason not to worry about your liver on that account. If you have liver disease, the relevant expertise sits with your hepatologist, not with an epidemiological association. And for the practical side of coffee at home, our health guide and the cold brew guide, which covers acidity and stomach comfort, are the more useful reading.