“Liver support” is the vaguest phrase in the supplement aisle. It covers milk thistle, N-acetylcysteine, caffeine, and anything else that ends up in a pen with a liver icon on it. The reason it stays vague is not marketing laziness. It is that the underlying literature is genuinely split, and the split is not about whether these things work in principle — it is about which species the trials were run in, and about how much the word “significant” is worth when two meta-analyses of the same exposure land in different places.
The one with real human trials
The silymarin meta-analysis of 2024 is the most substantial file in this area: 26 randomised controlled trials, 2,375 patients, all of them human, all of them in non-alcoholic fatty liver disease or steatohepatitis. The pooled effects are not subtle. Total cholesterol down by a standardised mean difference of −0.85 (95 % CI −1.23 to −0.47), triglycerides −0.62, LDL cholesterol −0.81, HDL cholesterol up by 0.46. On the liver side, ALT fell by an SMD of −12.39 (−19.69 to −5.08) and AST by −10.97 (−15.51 to −6.43), and where biopsies were available the intervention groups showed significantly improved steatosis with an odds ratio of 3.25 (1.80 to 5.87).
Then read the last sentence of the abstract, because it is the one that matters: the effects of silymarin will need to be confirmed by further research. A meta-analysis of 26 trials that ends by asking to be confirmed is telling you something about heterogeneity, about trial quality, or about publication bias. It is not telling you the effect is zero. It is telling you the effect is not yet yours to bank on.
The one with a clean mechanism and no humans
N-acetylcysteine has the tidier story. It is an antioxidant; fatty liver disease is partly an oxidative problem; the two should connect. The 2023 meta-analysis connects them properly — and every one of its thirteen studies is an animal study. In NAFLD-induced animals, the substance improved systemic and hepatic lipid metabolism, inflammation-related liver injury, glucose intolerance and hepatic steatosis, by restoring hepatic glutathione and glutathione reductase levels, and transcriptomic data from mice and patients pointed at the same pathways. The quality assessment uses the SYRCLE guideline for laboratory animal experimentation, which is the giveaway: this is preclinical evidence, assembled with preclinical methods.
The authors are honest about where that leaves the question. Their conclusion is that N-acetylcysteine has therapeutic potential for NAFLD and should be considered for future clinical trials. That is the correct tense. It is a candidate for trials, not a result from them.
Coffee: the same question, two answers
Here is the part worth reading twice. Two meta-analyses, both careful, both about coffee and fatty liver, disagree.
The first, from 2021, pooled 11 of 321 collected epidemiological articles and found the risk of NAFLD lower among coffee drinkers, pooled relative risk 0.77 (95 % CI 0.60 to 0.98), and among patients who already had NAFLD, a lower risk of developing fibrosis, relative risk 0.68 (0.68 to 0.79).
The second, also 2021, pooled 11 observational articles — two case-control, eight cross-sectional, one prospective cohort — with 92,075 subjects in the incidence analysis, 9,558 in prevalence and 4,303 in fibrosis, and found no association with NAFLD incidence (RR 0.88, 95 % CI 0.63 to 1.25, p = 0.48) and none with prevalence (RR 0.88, 95 % CI 0.76 to 1.02, p = 0.09). It did find the fibrosis signal: 35 percent lower odds of significant fibrosis, RR 0.65 (0.54 to 0.78, p < 0.00001), with no heterogeneity at I² = 11 percent and no evidence of publication bias.
So the two analyses agree on fibrosis and disagree on whether coffee changes whether you get fatty liver in the first place. One of them is pooling mostly cross-sectional data, where people who drink coffee differ from people who do not in ways nobody measured. Both end with the same sentence in different words: more prospective work is needed, and nobody knows the threshold.
What the data do not show
They do not show that a supplement labelled “liver support” does anything a person can feel or measure. They do not show that lowering ALT by an SMD of −12 is the same as reversing fatty liver disease: liver enzymes are a surrogate, and a supplement can move a surrogate without changing histology, and a drug can change histology without moving enzymes much.
They also do not show that the mechanism transfers. N-acetylcysteine has the most elegant pathway story in this file and not one human trial in it; silymarin has 2,375 patients and still ends in a request for confirmation. Species and design are doing more work here than the effect sizes.
The practical reading rule that survives all four papers: check the species before the effect size, check the design before the odds ratio, and treat any liver claim that comes with a pen and no trial registration as marketing copy rather than as evidence.
