Meta-analysis: antioxidant supplements for liver diseases – the Cochrane Hepato-Biliary Group


  • G. Bjelakovic,

    1. The Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark.
    2. Department of Internal Medicine, Medical Faculty, University of Nis, Nis, Serbia.
    Search for more papers by this author
  • L. L. Gluud,

    1. The Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark.
    2. Department of Internal Medicine, Copenhagen University Hospital Gentofte, Hellerup, Denmark.
    Search for more papers by this author
  • D. Nikolova,

    1. The Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark.
    Search for more papers by this author
  • M. Bjelakovic,

    1. Institute of Anatomy, Medical Faculty, University of Nis, Nis, Serbia.
    Search for more papers by this author
  • A. Nagorni,

    1. Department of Internal Medicine, Medical Faculty, University of Nis, Nis, Serbia.
    Search for more papers by this author
  • C. Gluud

    1. The Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark.
    Search for more papers by this author

Prof. G. Bjelakovic, Department of Internal Medicine – Gastroenterology and Hepatology, Medical Faculty, University of Nis, Boulevard Dr Zorana Djindjica 81, 18000 Nis, Serbia.


Aliment Pharmacol Ther 2010; 32: 356–367


Background  Several liver diseases have been associated with oxidative stress. Accordingly, antioxidants have been suggested as potential therapeutics for various liver diseases. The evidence supporting these suggestions is equivocal.

Aim  To assess the benefits and harms of antioxidant supplements for patients with liver diseases.

Methods  We identified trials through electronic and manual searches until August 2009. We included randomized trials comparing antioxidant supplements (beta-carotene, vitamin A, C, E and selenium) vs. placebo or no intervention for autoimmune liver diseases, viral hepatitis, alcoholic liver disease and cirrhosis (any aetiology). Random-effects and fixed-effect meta-analyses were conducted. Results were presented as relative risks (RR), or mean difference (MD), both with 95% confidence intervals (CI).

Results  Twenty randomized trials with 1225 participants were included. The trials assessed beta-carotene (3 trials), vitamin A (2 trials), vitamin C (9 trials), vitamin E (15 trials) and selenium (8 trials). The majority of the trials had high risk of bias and showed heterogeneity. Overall, the assessed antioxidant supplements had no significant effect on all-cause mortality [relative risk (RR) 0.84, 95% confidence interval (CI) 0.60–1.19, I2 = 0%] or liver-related mortality (RR 0.89, 95% CI 0.39–2.05, I2 = 37%). Stratification according to the type of liver disease assessed did not affect the conclusions. Antioxidant supplements significantly increased the activity of gamma glutamyl transpeptidase (MD 24.21 IU/L, 95% CI 6.67–41.75, I2 = 0%).

Conclusions  We found no evidence to support or refute antioxidant supplements in patients with liver disease. Antioxidant supplements may increase liver enzymes.


Oxidative stress may damage cells and possibly lead to serious damage of the tissues, including cancer.1–4 Experimental models suggest that oxidative stress is involved in the pathogenesis of liver diseases. Therefore, antioxidant supplements have been suggested as potential chemopreventive agents.5–8 Previous systematic reviews have assessed supplements with antioxidant effects for a number of specific liver diseases.9–13 The results are inconclusive. No apparent effect was found on N-acetylcysteine, vitamin C, vitamin E and S-adenosyl-l-methionine for non-alcoholic fatty liver diseases,9 milk thistle for alcoholic liver disease or hepatitis B or hepatitis C virus liver disease10 or S-adenosyl-l-methionine for alcoholic liver diseases.11 Beta-carotene, vitamin A, vitamin C and vitamin E had no significant influence on the incidence of liver cancer.12 Systematic reviews of randomized trials on selenium for prevention of gastrointestinal cancers not only found a potential positive effect on mortality but also identified a number of random and systematic errors in the included trials which could confound this effect.13, 14 Accordingly, we conducted a Cochrane Hepato-Biliary Group systematic review to assess the beneficial and harmful effects of commonly used antioxidant supplements (beta-carotene, vitamin A, vitamin C, vitamin E and selenium) for liver diseases not previously included in Cochrane reviews.


We included randomized clinical trials, irrespective of blinding, publication status or language.15 To avoid overlap with previous systematic reviews on antioxidant supplements, we included trials with participants who had autoimmune liver diseases, viral hepatitis, alcoholic liver disease and cirrhosis of any aetiology.15 We excluded trials with participants who had special dietary needs including children, pregnant or lactating women, patients with non-alcoholic steatohepatitis and malignant liver diseases.15

Antioxidant supplements, given alone or in combination with other vitamins or trace elements and compared with placebo or no treatment, were considered eligible for inclusion.15 The studied antioxidant supplements were beta-carotene, vitamin A, vitamin C, vitamin E, and selenium. We did not put limitations on the dose or treatment duration of these supplements.

Our outcome measures were all-cause mortality, liver-related mortality, liver-related morbidity (gastrointestinal bleeding, hepatic encephalopathy, hepato-renal syndrome, ascites, jaundice, liver cancer), biochemical indices [alanine aminotransferase, (ALT), aspartate aminotransferase (AST), gamma glutamyl transpeptidase, bilirubin, and albumin] at maximum follow-up in the individual trials as well as adverse events, quality-of-life measures and cost-effectiveness. In addition, for patients with hepatitis B or hepatitis C, we also considered the end of treatment and sustained virological response (number of patients without hepatitis B virus DNA or hepatitis C virus RNA at the end of treatment and 6 months after treatment).15

Search strategy for identification of trials

As described in Figure 1, the electronic searches were performed in The Cochrane Hepato-Biliary Group Controlled Trials Register,16 The Cochrane Library (Issue 3, 2009), MEDLINE (1966 to August 2009), EMBASE (1966 August 2009), Science Citation Index Expanded (1945 to August 2009) and Conference Proceedings Citation Index-Science (1990 to August 2009).17

Figure 1.

 Flow diagram of identification of randomized trials for inclusion.

Manual searches included reading through reference lists of relevant papers, conference proceedings and correspondence with experts and manufacturers of antioxidant supplements.

Methods of the review

Trial selection.  Four authors (GB, LLG, DN, and MB) participated in the selection of trials and extraction of data. Disagreements among authors were resolved through discussion before analyses. Excluded trials were listed with the reasons for exclusion. The extracted data included publication date, location of trial, funding, bias risk of trials, length of follow-up, publication status, patient inclusion and exclusion criteria, mean age of patients, proportion of women, and dose and duration of treatment. One author (GB) wrote to authors of the included trials to obtain information that the published trial reports lacked.

Assessment of bias risk.  Due to the risk of overestimation of beneficial intervention effects in randomized trials with unclear or inadequate methodological quality,18–21 we assessed the influence of the risk of bias on our results. We used the following domains: allocation sequence generation, allocation concealment, blinding, complete outcome data reporting, selective outcome reporting, baseline imbalance, early stopping and other apparent biases. The allocation sequence generation was classified as adequate if based on a table of random numbers or a computer. The allocation sequence concealment was classified as adequate if the allocation of patients involved a central independent unit, on-site locked computer, or identically appearing, numbered, drug bottles prepared by an independent investigator. Sealed envelopes were considered adequate if they were opaque and serially numbered. Blinding was classified as adequate, if the trial was described as double blind and the method of blinding involved identical placebo and active drugs. The description of incomplete outcome data was classified as adequate if the numbers and reasons for dropouts and withdrawals in all intervention groups were described, or if it was specified that there were no dropouts or withdrawals. Selective outcome reporting was classified as adequate, if reported outcomes were predefined, or clinically relevant and expected. We also registered whether there were baseline imbalances in important characteristics, if a sample size calculation was reported and achieved, or the trial was stopped early by formal stopping rules at a point where the likelihood of observing an extreme intervention effect due to chance was low, and whether there were any other apparent biases.

Trials with adequate assessments in all of the above mentioned bias risk domains were considered as having the lowest risk of bias.20, 22

Statistical analyses

We performed this review and meta-analyses according to the recommendations of The Cochrane Handbook for Systematic Reviews of Interventions,23 and The Cochrane Hepato-Biliary Group.16 For the statistical analyses, we used RevMan 5,24stata 8.2 (STATA Corp, College Station, TX, USA), sigma stat 3.0 (SPSS Inc., Chicago, IL, USA), and Stats-Direct (StatsDirect Ltd, Altrincham, UK). For dichotomous outcomes, we calculated the relative risks (RR), and for continuous outcomes we calculated the mean differences (MD). To account for trials with zero events, meta-analyses of dichotomous data were repeated using risk differences (RD). For all association measures, 95% confidence intervals (CI) were used. The I2 statistic was presented as a measure of the percentage of variation due to heterogeneity rather than chance.25 We analysed the data using both fixed-effect26 and random-effect27 models. In case there was no difference in statistical significance between the results obtained with the two models, we presented the results of the random-effects model analyses. The random-effects model was chosen due to expected clinical heterogeneity. Otherwise, we presented the results of both analyses. The analyses were performed using the intention-to-treat principle including all randomized participants irrespective of completeness of data. Patients with missing data were included in the analyses using carry-forward of the last observed response. Accordingly, patients who had been lost to follow-up were counted as survivors. For trials using parallel group design with more than two groups and additional therapy, we compared only the antioxidant intervention group with placebo or no intervention group. Only data from the first period of cross-over trials were included.23 We compared the intervention effects in subgroups of trials with the test of interaction.28 We performed adjusted rank correlation29 and regression asymmetry test30 for detection of bias.


Description of studies

We identified a total of 1542 references of possible interest through searching The Cochrane Hepato-Biliary Group Controlled Trials Register (n = 58), the Cochrane Central Register of Controlled Trials in The Cochrane Library (n = 729), MEDLINE (n = 482), EMBASE (n = 48), Science Citation Index Expanded (n = 196), Conference Proceedings Citation Index-Science (n = 14) and reference lists (n = 15). We excluded 805 duplicates and 695 clearly irrelevant references through reading the abstracts. Accordingly, 42 references were retrieved for further assessment. Of these, we excluded 20 references because they were not randomized trials. In total, 20 randomized trials31–50 described in 22 references fulfilled our inclusion criteria (Figure 1).

Eighteen trials used trial designs with parallel groups. Two trials34, 42 used a cross-over design. The majority of the trials (n = 18) were conducted in a single centre. Two trials44, 46 were multicentre trials. The single centre trials were conducted in Europe (10 trials), Asia (4 trials), North America (3 trials) and South America (1 trial). The multicentre trials were conducted in Europe and the United States of America.


A total of 1225 participants were randomized in the 20 trials. The number of participants in each trial ranged from 16 to 226 (median 48). We were not able to extract relevant data on the sex of the participants in three trials34, 36, 37 and on the mean age of the participants in two trials.34, 36 The percentage of women was 38%. The mean age of participants was 49 years (Table 1).

Table 1.   Characteristics of included trials
TrialDesignArmsBias riskBlindingNo of participantsWomen %Mean age yearsTreatment yearsFollow-up yearsBC mg*Vit A IU*Vit C mg*Vit E IU*Se μg*
  1. PL, placebo; NI, no intervention; n/a, not available; BC, beta-carotene; Vit, vitamin; Se, selenium. Blank cells indicate that the supplement was not part of the study.

  2. * Dosages are per day.

Worner et al., 1988Parallel2HighPL30049.50.330.33 10000   
Butcher et al., 1993Parallel2HighPL2839.345.50.020.0240 1000400100
de la Maza et al., 1993Parallel2HighPL7414.949.811   500 
van Gossum et al. 1995Parallel2HighPL16n/an/a0.330.33    100
von Herbay et al., 1997Cross-over2HighPL2647.8550.230.31   800 
Yu et al., 1997Parallel2HighPL226n/an/a48    200
Ideo et al., 1999Parallel2HighNI120n/a47.60.50.5   600 
Look et al., 1999Parallel3HighNI2441.736.10.460.46   544400
Andreone et al., 2001Parallel2HighNI322539.50.251.25   660 
Nishiguchi et al., 2001Parallel2HighNI622950.50.250.25  600  
Gueguen et al., 2003Parallel2HighPL11824.540.60.0570.0576 12030100
Prince et al., 2003Cross-over2LowPL6192580.230.2312 600298300
Takagi et al., 2003Parallel2HighNI9355.462.555   600 
Mezey et al., 2004Parallel2LowPL5133.347.50.251   1000 
Saeian et al., 2004Parallel2HighNI4719.144.80.460.46–0.92   1600 
Groenbaek et al., 2006Parallel2LowPL2352.2450.50.5  500945200
Hernandez-G et al., 2006Parallel2HighPL3745.960.40.0030.003  3000  
Hino et al., 2006Parallel2HighNI3046.752.50.150.15  750500 
Gabbay et al., 2007Parallel2LowPL503257.40.460.92  6000800 
Stewart et al., 2007Parallel2LowPL7749.344.20.50.5 5000600400187.5

There were seven trials on patients with alcoholic liver disease (alcoholics;31, 32, 41 alcoholic hepatitis;44, 50 and alcoholic cirrhosis;33, 34), nine trials in patients with chronic hepatitis C,35, 37, 38, 43, 45, 46, 48, 49 two trials in patients with chronic hepatitis B,36, 39 one trial in patients with primary biliary cirrhosis,42 and one trial in patients with liver cirrhosis47 (Table 2).

Table 2.   Participants and outcome measures of included trials
TrialParticipantsOutcome measures
Worner et al., 1988AlcoholicsSexual functioning
Butcher et al., 1993AlcoholicsFree radical markers and the enzymatic markers of liver damage
de la Maza et al., 1993Alcoholic cirrhosisOverall mortality, liver-related mortality, liver-related morbidity
van Gossum et al. 1995Alcoholic cirrhosisLiver function
von Herbay et al., 1997Chronic hepatitis CLiver function
Yu et al., 1997HBsAg carriersPrimary liver cancer
Ideo et al., 1999Chronic hepatitis CEnd-of-treatment and sustained biochemical and virological response
Look et al., 1999Chronic hepatitis CEnd-of-treatment biochemical and virological response
Andreone et al., 2001Chronic hepatitis BEnd-of-treatment and sustained biochemical and virological response
Nishiguchi et al., 2001Chronic hepatitis COccurrence of retinopathy during IFN therapy
Gueguen et al., 2003AlcoholicsEnd-of-treatment biochemical response
Prince et al., 2003Primary biliary cirrhosisFatigue
Takagi et al., 2003Posthepatitis C cirrhosisPrimary liver cancer
Mezey et al., 2004Alcoholic hepatitisOverall mortality
Saeian et al., 2004Chronic hepatitis CRibavirin-associated haemolysis
Groenbaek et al., 2006Chronic hepatitis CEnd-of-treatment biochemical and virological response
Hernandez-Guerra et al., 2006Liver cirrhosisHepatic endothelial dysfunction, portal pressure
Hino et al., 2006Chronic hepatitis CClinical observations
Gabbay et al., 2007Chronic hepatitis CLiver enzymes, HCV-RNA levels, and histology
Stewart et al., 2007Alcoholic hepatitisOverall mortality

Experimental interventions

The route of antioxidant administration was oral in 19 trials and parenteral in one trial.47 Antioxidants were administered either singly, or in combination with other antioxidants, or with or without other minerals or other interventions. The types, doses, dose regimens and duration of supplementation with antioxidants were as follows: beta-carotene 6 mg to 40 mg (3 trials), vitamin A 5000 μg to 10 000 μg (2 trials), vitamin C 120 mg to 6000 mg (9 trials), vitamin E 30 IU to 1600 IU (15 trials), and selenium 100 μg to 400 μg (8 trials), daily (Table 1). The duration of supplementation was from 1 day to 5 years (median 0.33 year) and the duration of follow-up period was from 1 day to 8 years (median 0.48 years) (Table 1).

Vitamin A, vitamin E and selenium were administered singly in 12 trials. Beta-carotene, vitamin A, vitamin C, vitamin E and selenium formed different combinations of antioxidant supplements only or together with non-antioxidant supplements in the other eight trials. The combinations of antioxidant supplements were vitamin C and vitamin E; vitamin C, vitamin E, and selenium; beta-carotene, vitamin C, vitamin E, and selenium; and vitamin A, vitamin C, vitamin E, and selenium (Table 1).

Three trials added vitamins like B group vitamins,34, 50 or coenzyme Q42 to the experimental intervention group. Three trials tested N-acetylcysteine together with antioxidants in the experimental intervention group.37, 38, 50 Three trials treated all participants with interferon37, 38, 40 and two trials treated all participants with interferon and ribavirin.45, 48

Control interventions

Thirteen trials used placebo as control intervention. Seven trials had no intervention in the control group.37–40, 43, 45, 48

Bias control in included trials

Five trials (25%) were considered as having the lowest risk of bias42, 44, 46, 49, 50 (Table 1). The remaining trials had unclear or inadequate bias control in one or more of the components assessed.

Outcome measures

All 20 trials examined severity and progression of liver disease. From two trials, however, we were not able to extract relevant data for our meta-analyses.40, 47

Seven of the trials provided data on all-cause mortality,33, 37, 42–44, 49, 50 three trials on liver-related mortality,33, 43, 44 three trials on liver-related morbidity34, 36, 50 and five trials on adverse events.31, 37, 39, 42, 48

Ten (50%) of the trials mentioned that antioxidant supplements were provided by pharmaceutical companies. This information was not available in the remaining ten trials.31, 33, 34, 36, 39–41, 43, 44, 50

All-cause mortality

The antioxidant supplements had no significant effect on all-cause mortality (RR 0.84, 95% CI 0.60–1.19, I2 = 0%) (Figure 2). The inclusion of the zero event trials in the meta-analysis increased the intertrial heterogeneity, but had no influence on the overall conclusions (RD −0.00, 95% CI −0.05–0.04, I2 = 35%). Likewise, no significant effects on mortality was seen when analysing the trials with lowest risk of bias (RR 0.98, 95% CI 0.64–1.49, I2 = 0%) or the remaining trials (RR 0.69, 95% CI 0.30–1.56, I2 = 35%) (Figure 2). The difference between the estimates of antioxidant effect in the trials stratified by the bias risk was not significant by test of interaction (z = −0.74, P = 0.46).

Figure 2.

 All-cause mortality.

Liver-related mortality

Overall, the effect of the antioxidant supplements on liver-related mortality was not significant (RR 0.89, 95% CI 0.39–2.05, I2 = 37%) (Figure 3). The antioxidant supplements had no significant effect on liver-related mortality caused by gastrointestinal bleeding (RR 1.08, 95% CI 0.26–4.44, I2 = 5%), liver failure (RR 1.36, 95% CI 0.50–3.71, I2 = 0%), liver cancer (RR 0.31, 95% CI 0.09–1.09) or sepsis (RR 1.04, 95% CI 0.07–15.74).

Figure 3.

 Liver-related mortality.

Liver-related morbidity

The antioxidant supplements had no significant effect on gastrointestinal bleeding (RR 1.18, 95% CI 0.38–3.63, I2 = 0%), hepatic encephalopathy (RR 0.90, 95% CI 0.65–1.25), sepsis (RR 0.90, 95% CI 0.47–1.71), or liver cancer (RR 0.36, 95% CI 0.12–1.11).

Biochemical indices

In fixed-effect meta-analyses, the antioxidant supplements significantly increased the activity of aspartate aminotransferase (MD 7.64 IU/L, 95% CI 3.78–11.51, I2 = 80%) and gamma glutamyl transpeptidase (MD 24.21 IU/L, 95% CI 6.67–41.75, I2 = 0%), but had no significant effect on alanine aminotransferase (MD 3.86 IU/L, 95% CI −0.17–7.89, I2 = 74%), serum concentration of albumin (MD 0.05 g/L, 95% CI −0.90–1.00, I2 = 8%) and bilirubin (MD −0.04 micromol/L, 95% CI −0.38–0.29, I2 = 53%) (Figure 4). In random-effects meta-analysis, the antioxidant supplements had no significant effect on activity of aspartate aminotransferase (MD 3.80 IU/L, 95% CI −6.07–13.67).

Figure 4.

 Biochemical indices.

Adverse events

Several adverse events were reported in the experimental group of the included trials (liver function test elevation;31 dyspepsia and fatigue, leucocytopaenia, thrombocytopaenia, autoimmune thyroiditis, depression;37 general fatigue, anorexia, and/or depression;48 arthralgia, diarrhoea, dyspepsia, dysgeusia, low mood, menorrhagia, shingles, coryza, dizziness, dysuria42). However, the antioxidant supplements were without significant effects on these as well as other adverse events.

Quality-of-life measures

Only one trial49 provided data on quality-of-life. The average baseline quality of life values in the treatment and placebo group were not significantly different compared to post-treatment values.


We did not find any data on cost-effectiveness in the randomized trials included in this review.

End-of-treatment and sustained virological response

The antioxidant supplements had no significant effect on end-of-treatment virological response in patients with chronic hepatitis B (RR 0.57, 95% CI 0.32–1.02) or chronic hepatitis C (RR 0.85, 95% CI 0.55–1.32, I2 = 95%). The antioxidant supplements had no significant effect on the sustained virological response in patients with chronic hepatitis C (RR 0.89, 95% CI 0.54–1.45, I2 = 79%).


The assessed antioxidant supplements, i.e. beta-carotene, vitamin A, vitamin C, vitamin E and selenium given singly, or in combinations do not seem to significantly benefit patients with liver disease. In fact, the present review suggests that these antioxidant supplements may be associated with biochemical markers of liver damage. Although we found no clinical effect of this association, the evidence does underline the importance of retrieving a complete history including supplement use when assessing reasons for changes in liver enzyme activity of patients with liver disease. None of these antioxidant supplements seems to affect survival, liver-related mortality, liver-related morbidity, or adverse events significantly. Nor were the antioxidant supplements beneficial in improving the number of patients with end-of-treatment or sustained virological response.

Adverse events were insufficiently reported. Adverse events are very often neglected in randomized trials.51 In a number of trials in patients with chronic hepatitis B or hepatitis C, the antioxidant supplements were administered in combination with interferon or with other antioxidants like N-acetylcysteine that made it difficult to judge their beneficial or harmful effects or assign any of the observed adverse events.

The results of our systematic review should be interpreted with caution because a majority of the included trials (75%) had high risk of bias. Despite the relatively high number of included trials in the majority of the comparisons, we were not able to conduct subgroup analyses according to the risk of bias. Furthermore, our results are based on few patients and outcomes which raise the risks of random errors.52

Although oxidative stress is considered a common pathogenetic mechanism in a variety of liver diseases,53 we could not find convincing evidence that the studied antioxidant supplements are beneficial in these patients. Several antioxidants, like beta-carotene, vitamin A, vitamin C, vitamin E, and selenium, have been tested in patients with liver diseases. Only some of these antioxidants have shown promising results in observational studies as well as in some of the randomized clinical trials with high risk of bias; however, the results of randomized trials with low risk of bias and meta-analyses did not confirm the presumed benefits to the patients. Other Cochrane systematic reviews dealing with the topic, similarly, could not identify beneficial effects of these and other antioxidants.9–14 Extensive efforts in evaluating the benefits and harms of antioxidant supplements have resulted in neutral or detrimental results. It is likely that oxidative stress is not the only pathogenetic mechanism contributing to liver damage. Presumptions that antioxidants could be beneficial as adjuvants to other drugs like interferon or ribavirin53 were not confirmed or refuted in our review.


Different types of bias can influence the results of our meta-analyses including selective reporting of some results in trial publications.54–58 Selective outcome reporting represents a significant threat to the validity of a meta-analysis. The outcome reporting in the included trials was insufficient and inconsistent. There are several possible explanations for selective reporting of outcomes in randomized trials. Trials in which the outcome was not reported may not have measured outcomes of interest. Researchers may not have reported unexpected results or results may not have satisfied funders.59 Pharmaceutical companies provided antioxidant supplements in 10 of 20 included trials (50%). This number could be higher as information was not available for the remaining ten trials. It could be that researchers have selectively omitted to report some of the outcomes. We are well aware of the difficulties in collecting data on the outcomes in clinical trials that focus on safety and efficacy of the interventions. The worst result of outcome reporting bias and suppression of some significant or non-significant findings could be the use of harmful interventions. The results of meta-analyses may overestimate the true effects of interventions when there is exaggerated outcome reporting bias. One would wish that the results of the randomized clinical trials are reported in greater details. In some of the trials, instead of full reporting, we found partial or qualitative reporting. The huge human efforts of investigators and the high cost of randomized clinical trials should be justified with more rigour in their reporting. In spite of the large investment in the reviewed trials, a number of questions remain unanswered.

Other types of bias, like bias from trials with deficiencies in the trial design,18–20 small trials,60 trials with early stopping for benefit61 and trials with publication bias62, 63 could probably influence our results. Meta-analysis of randomized trials increases the power and precision of the estimated intervention effect, but this effect may be influenced by systematic errors or random errors and can lead to a report of false significant results.52, 64 It is likely that the results of our meta-analyses are influenced by random errors and systematic errors. The majority of the trials included in our meta-analyses are with small number of participants and of short follow-up period without the possibility to assess clinical outcomes with slow progression of a disease.

We have compared antioxidants with different properties, given at different doses and duration, singly or in combination vs. placebo or no intervention. We are well aware of the potential risks in assessing the effects of different types of antioxidants together with different mechanisms of action, biotransformation and bioavailability. Furthermore, many of the trials assessed multiple antioxidants given together in mixtures. Therefore, we have been unable to assess the individual antioxidants, which should have been examined in randomized clinical trials evaluating multiple different doses of the individual types of antioxidants.

The diseases included in the present review also represent a heterogeneous group of participants. We cannot exclude that antioxidant stress may vary among the different disease categories as well as within each disease category. In consequence, the therapeutic potential of the antioxidant supplements could considerably vary. When looking at subgroups of patients, we did not see indications that the examined antioxidant supplements benefited or harmed a certain diagnostic group. Furthermore, including the results of previously published Cochrane Hepato-Biliary Group reviews9–12 would not affect our conclusions. Only future trials tailoring the antioxidant supplementation to the perceived need of the individual patient can ultimately answer the question we pose in this systematic review.

In the majority of the trials, the authors have used non-validated surrogate outcomes such as blood levels of certain antioxidants, assuming that higher levels are beneficial for patients. Surrogate outcomes, especially laboratory indices, are very often unreliable substitutes for clinical outcomes.64 The ideal primary outcome in randomized clinical trials is one that is relevant to the patient’s quality of life or course of disease.64 Significant correlation between a surrogate and a clinical outcome does not explicitly mean that the observed beneficial effect of an intervention on the surrogate outcome will be the same on the clinical outcome. Relying on nonvalidated surrogate outcomes is potentially dangerous when assessing new therapies.64 We lack validated surrogate outcome measures in hepatology. No trials included in this review examined sustained virological response as a surrogate outcome for successful treatment. Improved sustained virological response does not definitely mean significant improvement in clinical outcomes.64, 65 The use of new interventions in hepatology should not be justified until these have been proven to be beneficial on clinical outcomes.66, 67 The increase in the number of hepato-biliary randomized trials will never be considered a sufficient valuable source for data if aspects of the trial design, like sample size, completeness of data reporting, duration of follow-up and bias risk, are not improved, and trial performance and reporting do not follow the CONSORT guidelines.68

Our results are in accordance with the results of other recently published meta-analyses and systematic reviews9–14 as well as recommendations of the United States Preventive Services Task Force and British Nutrition Foundation for the use of vitamin supplementation.69–71


We found no convincing evidence to support or refute beta-carotene, vitamin A, vitamin C, vitamin E and selenium for patients with autoimmune, alcoholic, acute and chronic hepatitis B or hepatitis C virus liver diseases, or liver cirrhosis. They could lead to increased liver damage. Therefore, we cannot recommend the use of these antioxidant supplements for the examined liver diseases outside randomized clinical trials.


Declaration of personal interests: None. We extend our gratitude to all participants and investigators in the randomized clinical trials. We are grateful to Pietro Andreone, Karin Groenbaek, Stephen Stewart, Christopher P. Day, and Yaron Ilan for the information on the trials they were involved in. Declaration of funding interests: The present study was supported by Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen, Denmark, and Ministry of Science and Technological Development, Republic of Serbia, project number 145081B.