Silymarin for treatment of adults with nonalcoholic fatty liver disease

Objectives

This is a protocol for a Cochrane Review (intervention). The objectives are as follows:

To evaluate the benefits and harms of silymarin in adults with nonalcoholic fatty liver disease (NAFLD).

Background

Description of the condition

Nonalcoholic fatty liver disease (NAFLD) is defined as the presence of excessive fat accumulation in the liver parenchyma (> 5% of hepatocytes) in people without any history of significant alcohol consumption or any other aetiology of hepatic steatosis (Abenavoli 2021; EASL 2016; Tokushige 2021). According to the aetiology of NAFLD, it can be divided into primary NAFLD, often associated with metabolic syndrome, and secondary NAFLD, caused by rare hereditary hepatic and metabolic diseases, intestinal diseases, endocrine disorders, and drugs (Liebe 2021). NAFLD is a continuum of various stages from nonalcoholic fatty liver (NAFL) to nonalcoholic steatohepatitis (NASH). NAFL exhibits only excessive hepatic fat accumulation, whilst NASH is characterised by hepatic lobular inflammation and hepatocyte ballooning, and may progress to advanced liver disease, including liver fibrosis, cirrhosis, and even hepatocellular carcinoma (HCC) (Abenavoli 2021; Friedman 2018). Approximately one‐third of people with NAFLD have concurrent NASH (Younossi 2016). The progression to liver fibrosis takes an average of 14.3 years from NAFL and 7.1 years from NASH (Singh 2015). Traditionally, the 'two hits' theory, which comprises the first 'hit' causing steatosis and the second 'hit' initiating lipid peroxidation, has been used to explain the progression of NAFLD (Day 1998). This has been replaced by the 'multiple parallel hits' model, where various factors, such as genetic, dietary, and environmental factors, act simultaneously, resulting in insulin resistance, adipocytes proliferation and dysfunction, and alteration of intestinal microbiota, inducing secretion and activation of inflammatory cytokines, thereby promoting the progression of NAFLD (Gariani 2021).

People with metabolic syndrome, which includes obesity, diabetes mellitus, dyslipidaemia, insulin resistance, and hypertension, are more prone to developing NAFLD (Friedman 2018; Tanase 2020). Firstly, the obesity epidemic worldwide is associated with a rising morbidity of NAFLD (Vilalta 2022; Yaqub 2021). Secondly, about 75% of people with type 2 diabetes mellitus are accompanied by NAFLD (Kim 2021; Stefan 2022; Younossi 2019). Thirdly, insulin resistance, a prerequisite for the progression of NAFLD, disturbs glucose and lipid metabolism, further resulting in the accumulation of free fatty acids and triglycerides, thereby exacerbating inflammation, fibrosis, and carcinogenesis (Sakurai 2021; Tanase 2020). NAFLD has been considered the manifestation of metabolic syndrome in the liver (Sakurai 2021; Sharma 2022). A recent study suggests that the term 'NAFLD' should be replaced by 'metabolic (dysfunction) associated with fatty liver disease (MAFLD)' to more accurately demonstrate the correlation of NAFLD with metabolic disorders (Eslam 2020a).

Liver enzymes are routinely tested in people with NAFLD, and ultrasound is considered the first‐line diagnostic approach for NAFLD (Amernia 2021; Anstee 2022; Eslam 2020b). Though liver biopsy is the gold standard for diagnosing NAFLD and assessing its severity, it is not readily accepted because of its invasiveness and potential risk (Tokushige 2021). Some noninvasive approaches based on imaging methods and laboratory and anthropometric parameters have thus been developed to reduce the requirement of liver biopsy (Friedman 2018). Transient elastography can measure liver stiffness to noninvasively assess the degree of liver fibrosis and measure the controlled attenuation parameter (CAP) to detect hepatic steatosis, but its efficiency is limited by obesity (Amernia 2021; Lee 2022; Miele 2020). NAFLD fibrosis score (NFS), Fibrosis‐4 (FIB‐4) score, and AST/Platelet Ratio Index (APRI) are noninvasive scoring systems based on laboratory and anthropometric parameters to detect fibrosis in people with NAFLD (Amernia 2021; Younes 2021). However, they have some limitations. First, age is a component of FIB‐4, which may overestimate the grade of liver fibrosis in the elderly (Amernia 2021; Tokushige 2021). Second, body mass index (BMI) is a component of NFS, but its interpretation varies across different ethnic groups (Hanif 2020).

Due to an increase in the prevalence of sedentary lifestyle and fat‐rich diet, and popularisation of effective vaccines and antiviral treatments, NAFLD has surpassed hepatitis B virus as the leading cause of chronic liver disease, and has become the fastest growing contributor to hepatic complications since 2009 (Golabi 2021; Hanif 2020). The estimated prevalence of NAFLD in adults is 32.4% worldwide, and is 31.6% in Asia, 32.6% in Europe, 47.8% in North America, and 56.8% in Africa (Riazi 2022). Generally, men have a higher prevalence of NAFLD than women (Alqahtani 2021; Burra 2021), but postmenopausal women have a comparable prevalence to men (Burra 2021). The prevalence of NAFL is up to 80% in obese children (Suri 2021). People with NAFLD with some concomitant risk factors (e.g. diabetes mellitus, obesity, and smoking) are at higher risk of developing HCC (Huang 2021). Using data from 2016, the incidence of NAFLD‐related HCC is expected to increase dramatically by 2030, by 82% in China, 117% in France, and 122% in the USA (Huang 2021). NAFLD‐related HCC is projected to cause 110,900 deaths between 2015 and 2030 in the USA (Huang 2021). Screening for HCC has been recommended in people with advanced fibrosis and cirrhosis (Huang 2021; Tokushige 2021). Additionally, NASH has become the most rapidly growing indication for liver transplantation in people without HCC, especially in females and the elderly (≥ 65 years) (Stepanova 2022; Younossi 2021). Taken together, NAFLD and its related complications place tremendous psychological, economic, and medical burdens on individuals, families, and public health.

To date, recommendations for the management of NAFL have not been well‐established, and pharmacological treatments are only limited to people with biopsy‐proven NASH and fibrosis (Anstee 2022; Blazina 2019; Chalasani 2018; EASL 2016). Lifestyle modifications, such as diet, exercise, and weight loss, remain the first‐line and cornerstone treatment options for NAFLD (Eslam 2020b). Several pharmacotherapies for NAFLD have been explored, but most of them are limited to specific populations or are associated with significant serious adverse events. For example, the use of statins is associated with a lower incidence of HCC and mortality by decreasing free cholesterol levels or exerting direct anti‐inflammatory activity, but are only recommended for people with NAFLD/NASH with hypercholesterolaemia (German 2020; Thomson 2022; Tokushige 2021). Pioglitazone can improve liver histology and insulin sensitivity, but is only recommended for people with NASH with insulin resistance (Chalasani 2018; Tokushige 2021), and its use is limited due to increased risk of hypoglycaemia, lower limb oedema, weight gain, and atypical bone fractures (Blazina 2019; Lian 2021). Vitamin E can exert the effect of antioxidative stress by suppressing the production of reactive oxygen species, and thus improving both biochemical and histological characteristics of people with NAFLD (Chalasani 2018; Tokushige 2021). However, of concern is a potential risk of prostate cancer, hepatotoxicity, and haemorrhagic stroke (Nagashimada 2019; Vadarlis 2021). Hence, silymarin seems to be a promising option for improving NAFLD with a good safety profile and tolerance (Abenavoli 2021; Kalopitas 2021).

Description of the intervention

For centuries, silymarin, a complex crude extract from Silybum marianum fruit, has been used as a herbal agent for multiple diseases, including liver diseases, ulcerative colitis, allergic rhinitis, and various cancers (Devi 2017; Kaur 2007; Lynch 2021; Soleimani 2019). It is a mixture of mainly multiple flavonolignans (e.g. silybin, isosilybin, and silychristin) (Abenavoli 2018; Milosevic 2014; Soleimani 2019). Silybin is considered the most important constituent, but it has low oral bioavailability because it dissolves poorly in water and in the gastrointestinal tract lumen (Tvrdy 2021). A major approach to overcoming this drawback is the complexation with solubilising substances, such as vitamin E and phosphatidylcholine (Bijak 2017; Tvrdy 2021).

Silymarin may be beneficial for various liver diseases, such as alcoholic liver disease, NAFLD, viral hepatitis, drug‐induced liver injury, and cirrhosis (CMA 2018; Gillessen 2022; Niu 2021; Wei 2013; Xu 2020; Yang 2022). It is generally considered safe, and the observed adverse effects, such as nausea and diarrhoea, even at a high dose of 2100 mg per day for 24 weeks, are only transient (Loguercio 2012; Soleimani 2019; Wah 2017).

How the intervention might work

The development of NAFLD is due to excessive accumulation of fatty acid and significant lipid peroxidation (Abenavoli 2021; Day 1998). Silymarin exerts an antioxidant effect by increasing the level of endogen radical scavengers and mitigating lipid peroxidation. It achieves anti‐inflammatory effect by suppressing the release of cytokines (e.g. tumour necrosis factor‐alpha), antifibrotic effect by inhibiting the conversion of hepatic stellate cells to myofibroblasts, and exerts metabolic effect on insulin resistance and hyperlipidaemia (Abenavoli 2018; Milosevic 2014; Navarro 2019; Soleimani 2019). Based on these multiple pharmacological effects, silymarin may have potential benefits for treating NAFLD.

Why it is important to do this review

To date, a large number of randomised clinical trials have been published exploring the efficacy of silymarin in people with NAFLD. However, their findings are controversial. We found one systematic review with meta‐analysis, published in 2021 (Kalopitas 2021). It included eight randomised clinical trials assessing the efficacy of silymarin in people with NAFLD, but all of them only evaluated silymarin as a monotherapy. Silymarin is reported to be one of the top‐selling herbal dietary supplements in the USA, with retail sales reaching USD 2.6 million in the mainstream multi‐outlet channel in 2015 (Andrew 2017). Despite this, only a Chinese guideline and an international consensus, but not other practice guidelines or consensuses, suggest that silymarin is potentially efficacious in people with NAFLD, in whom a higher risk of progression to NASH and cirrhosis is suspected (CMA 2018; Gillessen 2022). Additionally, the optimal dose and duration of silymarin have as yet to be identified. This Cochrane Review with meta‐analyses of randomised clinical trials aims to assess the benefits and harms of silymarin as monotherapy and in combination with vitamin E or nutrients in people with NAFLD.

Objectives

To evaluate the benefits and harms of silymarin in adults with nonalcoholic fatty liver disease (NAFLD).

Methods

Criteria for considering studies for this review

Types of studies

We will include randomised clinical trials, with any trial design, assessing the benefits and harms of silymarin in adults with NAFLD. We will include relevant trials irrespective of publication status, country, year, publication language, and outcomes assessed. We will also include trials with unpublished data.

We will exclude quasi‐randomised studies (as the generation of allocation sequence can be predicted by alternation, date of birth, or day of admission) as well as other observational studies.

Types of participants

We will include adults (18 years of age or older) diagnosed with NAFLD with or without metabolic diseases, regardless of diagnostic methods, disease stage, and ethnic origin. We will exclude participants with other concomitant causes of liver diseases, including hepatitis B, hepatitis C, autoimmune hepatitis, drug‐induced liver injury, and genetic liver diseases, such as Wilson's disease and haemochromatosis.

If a trial includes a subset of participants diagnosed with NAFLD, and data for the subset are not provided separately and cannot be obtained from the trial authors, we will exclude the trial. We will also exclude trials in which the subset of participants with NAFLD is less than 10%, as small trials tend to overestimate intervention effects (McKenzie 2022a).

Types of interventions

Experimental intervention

• Silymarin as monotherapy

• Silymarin combined with vitamin E or nutrients (e.g. L‐carnitine).

Silymarin may be administered as tablets, at any dose, for any duration, frequency, administration time, or setting.

Control intervention

• No intervention, placebo

• Another intervention different from the experimental intervention.

We will allow co‐interventions if they are administered equally to participants in both the experimental and control groups.

Types of outcome measures

We will analyse the primary and secondary outcomes below at the end of treatment and at the longest follow‐up. We will use the longest time points in the trials for our main analyses and conclusions. Considering that NAFLD is a metabolic dysfunction associated with fatty liver disease, we have chosen the outcomes below as they are of high importance to people with NALFD and clinicians. We will include trials fulfilling the inclusion criteria of our review, irrespective of their reported outcomes.

Primary outcomes

• Proportion of people with serious adverse events. We will use the International Conference on Harmonisation (ICH) Guidelines for Good Clinical Practice's definition of a serious adverse event (ICH‐GCP 2016), which is a congenital anomaly or birth defect, or any untoward medical occurrence at any dose that:

  • results in death;

  • is life‐threatening;

  • requires inpatient hospitalisation or prolongation of existing hospitalisation;

  • results in persistent or significant disability or incapacity.

We will contact trial authors to enquire about serious adverse events if these are not reported or are reported insufficiently in the identified trial publications. We will consider any other adverse events as nonserious.

• All‐cause mortality.

• Health‐related quality of life. We will accept any validated scale used by the trialists, e.g. World Health Organization Quality of Life, WHOQOL, and 36‐item Short Form Health Survey, SF-36. If a trial uses several scales in their outcome reporting, we will select the most frequently used one across the included trials.

Secondary outcomes

• Proportion of participants without improvement of histological features (e.g. steatosis, lobular inflammation, and fibrosis) based on validated scoring systems employed in the included studies.

• Proportion of participants without normalisation of liver enzymes, i.e. alanine aminotransferase, aspartate aminotransferase, and gamma‐glutamyl‐transpeptidase.

• Proportion of participants without improvement of ultrasonographical indices, i.e. CAP, liver stiffness measurement (LSM), and ultrasound score.

• Liver enzymes, i.e. alanine aminotransferase, aspartate aminotransferase, and gamma‐glutamyl‐transpeptidase.

• Lipid‐related biomarkers, i.e. cholesterol, triglycerides, high‐density lipoprotein, and low‐density lipoprotein.

• Blood glucose‐related biomarkers, i.e. fasting blood glucose, homeostasis model assessment of insulin resistance, and insulin.

• Anthropometric indices, i.e. weight, BMI, and waist circumference.

• Ultrasonographical indices, i.e. CAP, LSM, and ultrasound score.

• Proportion of participants with nonserious adverse events. See Primary outcomes. We will also accept the authors' definitions as employed in the trials.

We will contact study authors if the outcomes of interest have not been sufficiently reported or have been measured but not reported.

Search methods for identification of studies

To minimise bias in our search results, we will follow the guidance in Chapter 4 of the Cochrane Handbook for Systematic Reviews of Interventions,Lefebvre 2022a, and in the PRISMA‐S, Rethlefsen 2021, to plan and describe the search process for the review. We will prepare the search strategies and complete the literature retrieval under the guidance of the Cochrane Hepato‐Biliary Group Information Specialist. We will impose no restrictions on publication language, date, or status.

Electronic searches

We will search the Cochrane Hepato‐Biliary Group Controlled Trials Register, which will be searched internally by the Cochrane Hepato‐Biliary Group Information Specialist via the Cochrane Register of Studies Web. We will also search the Cochrane Central Register of Controlled Trials in the Cochrane Library, MEDLINE Ovid, Embase Ovid (Excerpta Medica Database), LILACS (BIREME) (Latin American and Caribbean Health Science Information database), Science Citation Index Expanded, and Conference Proceedings Citation Index ‐ Science. The last two will be searched simultaneously through the Web of Science.

The search strategies for the respective databases, with the expected date range of the searches, are provided in Appendix 1. We will provide the actual date of the electronic searches at the review stage.

Searching other resources

We will search online trial registries such as ClinicalTrials.gov (clinicaltrials.gov), the World Health Organization International Clinical Trials Registry Platform (www.who.int/ictrp), European Medicines Agency (EMA; www.ema.europa.eu), and the US Food and Drug Administration (FDA; www.fda.gov), as well as pharmaceutical company sources for ongoing or unpublished trials and for study information. We will also contact relevant individuals and organisations for information about unpublished or ongoing studies.

We will search Chinese regional bibliographic databases such as China National Knowledge Infrastructure (CNKI), WanFang, and WeiPu databases for eligible studies.

We will search for relevant grey literature sources such as reports, dissertations, theses, and conference abstracts, for example in Google Scholar.

We will use the PubMed/MEDLINE 'similar articles search' tool on all included studies. We will manually check citations and reference lists of the included studies and any relevant systematic reviews identified.

We will search for and examine any relevant retraction statements through the Retraction Watch Database (retractionwatch.com/retraction-watch-database-user-guide/) and errata for information as errata can reveal important limitations or even fatal flaws in the included trials (Lefebvre 2022b).

We will contact the authors of identified trials for additional published or unpublished trials.

We will provide the actual date of searching other resources as described in this section at the review stage. We will also use items from the PRISMA‐S checklist that relevant to our review to ensure that we have reported and documented our searches as advised (PRISMA-S Checklist; Rethlefsen 2021).

Data collection and analysis

We will follow the instructions in the Cochrane Handbook for Systematic Reviews of Interventions for data collection and analysis (Higgins 2022a). We will use Review Manager Web software to perform the meta‐analyses (RevMan Web 2022).

Selection of studies

Two review authors (CW and YS) will independently screen the titles and abstracts of all retrieved studies and code them as ‘included’ (eligible, potentially eligible, or unclear) or ‘excluded’.

Two review authors (CW and YS) will independently screen the full texts of potentially eligible studies for inclusion or exclusion, recording the reasons for exclusion of ineligible studies. Any disagreements will be resolved by discussion with a third review author (LC), and a consensus will be reached with the corresponding author (XQ).

We will examine any relevant retraction statements and erratum for the included trials to avoid their potential impact on the overall estimates of our systematic review (Lefebvre 2022a).

We will sufficiently document the selection process to complete a flow diagram, Page 2021a; Page 2021b, and ‘Characteristics of excluded studies’ table.

We will inspect multiple reports of the same trial to ensure that each trial, rather than each report, is the unit of interest in the review. We will use EndNote software to manage and achieve de‐duplication of studies (EndNote x9).

Data extraction and management

We will develop a data collection form for study characteristics and outcome data that will be piloted on at least two trials in the review. Two review authors (CW and YS) will independently extract the following data from the included trials.

• Methods: study design, duration, setting, number of study centres and location, date of enrolment, intention‐to‐treat (ITT), sample size, date of publication, and place of trial publication.

• Participants: total number of participants randomised and followed, mean age, sex, type of condition, diagnostic criteria, inclusion and exclusion criteria, baseline metabolic diseases (e.g. overweight and insulin resistance), number and reasons for dropouts, and follow‐up duration.

• Interventions: type of intervention (in the case of a compound, its composition will be given), dose, frequency, duration of interventions, concurrent medications, and lifestyle.

• Comparators: type (placebo, no intervention, or other interventions), dose, frequency, and duration of comparators, concurrent medications, and lifestyle.

• Outcomes: primary and secondary outcomes as specified and time points. Any other outcomes planned, measured, and reported.

• Notes: trial registration, ethics committee approval, trial protocol published, funding, and declarations of interest of study authors.

We will synthesise the characteristics of included trials and present the information in the ‘Characteristics of included studies’ table. If study authors reported outcome data in an unusable way, we will note this in the ‘Characteristics of included studies’ table. Any disagreements will be resolved through discussion with a third review author (LC), and a consensus will be reached with the corresponding author (XQ). We will contact study authors for clarification as needed. One review author (CW) will enter the data from the data extraction form into Review Manager Web (RevMan Web 2022). We will double‐check and confirm that the data are entered correctly by comparing the data presented in the Cochrane Review with the primary study reports. One review author (YS) will check the accuracy of the data regarding study characteristics based on the primary study reports. If the data are presented based on both ITT and per‐protocol analyses, we will only extract ITT data.

If the data are only presented in figures, we will use WebPlotDigitizer to extract data from the figures (WebPlotDigitizer 2022).

Assessment of risk of bias in included studies

Two review authors (CW and YS) will independently assess risk of bias using Cochrane RoB 2 (current version of RoB 2) in Review Manager Web, RevMan Web 2022, following the guidance in Chapters 8 and 23 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022a; Higgins 2022b).

We plan to assess the effect of assignment to the intervention based on the ITT principle, which includes all randomised participants regardless of the intervention they actually received.

There are five required domains in the RoB 2 tool:

• randomisation process;

• deviations from intended interventions;

• missing outcome data;

• measurement of the outcome;

• selection of the reported results.

We will use the signalling questions in the RoB 2 tool to rate each domain as 'low risk of bias', 'some concerns', or 'high risk of bias' (Higgins 2022a; Sterne 2019). The response options for the signalling questions are: 'Yes'; 'Probably yes'; 'Probably no'; 'No'; and 'No information'.

We will use the most recent RoB 2 Excel tool (Sterne 2019). An algorithm in Excel maps the responses to the signalling questions per outcome and proposes a risk of bias judgement for each domain.

We will use the overall risk of bias for a study result, rather than specific domains. The overall risk of bias for the result is the least favourable assessment across the risk of bias domains.

We will consider the overall risk of bias to be 'low' when one trial is judged to be at low risk of bias for all domains for its result. We will consider the overall risk of bias to be with 'some concerns' when one trial is judged to raise some concerns in at least one domain for its result, but not to be at high risk of bias for any domain. We will consider the overall risk of bias to be 'high' when one trial is judged to be at high risk of bias in at least one domain for its result or has some concerns for multiple domains such that confidence in the result is substantially lowered.

We will justify our judgements by providing direct quotes from the study reports in the risk of bias table. We will contact study authors to obtain any incompletely reported information related to the risk of bias assessment and note this in the risk of bias table. Any discrepancies will be resolved through discussion with a third review author (LC), and a consensus will be reached with the corresponding author (XQ).

We will present the risk of bias assessments to the right of the forest plot results of each trial included in a meta‐analysis, which will give a visual impression of each study's contribution at different levels of risk of bias, especially when considered in combination with each study's weight (Boutron 2022). All information relevant to our risk of bias assessments will be disclosed at the review stage as supplementary material in the Open Science Framework platform (www.osf.io).

We will assess risk of bias in cross‐over trials as in trials with parallel‐group design because we will only use data from the first period (i.e. before cross‐over) (Higgins 2022a).

RoB 2 for cluster‐randomised trials includes one additional domain, that is bias arising from the timing of identification and recruitment of participants. We will assess risk of bias in any cluster‐randomised trials found to fulfil the inclusion criteria of our review.

The risk of bias assessment will feed into one domain of the GRADE approach for assessing the certainty of a body of evidence (Schünemann 2022a). We will included the following outcomes in the summary of findings tables: serious adverse events, all‐cause mortality, and health‐related quality of life (see 'Summary of findings and assessment of the certainty of the evidence' section below).

Further information on RoB 2 is available at www.riskofbias.info and Risk of bias tools - RoB 2 for cluster-randomized trials.

Our primary analysis will include data from trials at any overall risk of bias.

Assessment of bias in conducting the Cochrane Review

We will conduct the Cochrane Review according to this published protocol and document any deviations from it in the 'Differences between protocol and review' section of the final review.

Measures of treatment effect

We will enter the outcome data for each study into the data tables in Review Manager Web to calculate the treatment effects (RevMan Web 2022). We will follow the guidance in Chapters 6 and 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2022; Higgins 2022c). For dichotomous data, we will calculate risk ratios (RR) with 95% confidence intervals (CIs). For continuous data, such as quality of life, we will use the mean difference (MD) with 95% CIs if trials used the same tool; we will use the standardised mean difference (SMD) with 95% CIs if trials used different scales to measure quality of life. We will interpret the SMD as follows: SMD less than 0.40 for small intervention effects; SMD between 0.40 and 0.70 for moderate intervention effects; and SMD greater than 0.70 for large intervention effects (Schünemann 2022b). It is important to note that the SMD method does not correct for differences in the direction of the scale. In some scales, the score increases with disease severity (e.g. a higher score indicates a more severe condition), whilst in others, the score behaves in the opposite direction (e.g. a higher score indicates a less severe condition). We will multiply the mean values from one set of studies by −1 (or subtract the mean from the maximum possible value for the scale) to ensure that all the scales point in the same direction before standardisation (Higgins 2022c). If trials report on survival, then we will use hazard ratios (Higgins 2022c). We will present medians and interquartile ranges for continuous data that are not normally distributed (skewed data), in a narrative format. We will present a forest plot that displays effect estimates and CIs for individual trials. We will undertake meta‐analyses only when a group of trials is sufficiently homogeneous (Deeks 2022).

We will deal with counts data as continuous data if they present common events, such as nausea, diarrhoea, or vomiting. We will consider counts of rare events as Poisson data and calculate the rate ratio (Deeks 2022).

We will convert data in an appropriate format if they are reported in a way that is difficult to use for meta‐analysis (Higgins 2022c). We will deal with skewed data according to Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2022).

Unit of analysis issues

Where possible, we will consider the randomised participant with NAFLD as a unit of analysis in a simple parallel‐group design. However, there are a variety of possible study designs, which we will address separately following Chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022b). For example, for multiple intervention groups, we will separately compare each of the relevant experimental group with each half of the control group if used within the same comparison to avoid double‐counting. Regarding adverse events, we will record if the trial measures adverse events in relation to the frequency of a participant with an adverse event (e.g. five participants reported an adverse event), or to multiple adverse events in the same participant (e.g. one participant had more than one episode of an adverse event). If multiple adverse events developed in the same participant, we will tabulate each type of adverse events separately. For all adverse events, we will try to take into account the interdependence of the adverse events when we extract data for analysis. Where the number of participants appears to be equal to that of events, we will treat the participants as the unit of analysis as described in Chapter 6 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022c). For cluster‐randomised clinical trials, we will consider the cluster (e.g. schools, villages, medical practices, or families) as the unit of analysis. For cross‐over randomised clinical trials, we will include only data from the first period of the trial (i.e. before the cross‐over) to avoid carry‐over effects (Higgins 2022b).

Dealing with missing data

We will not exclude trials from our review based on the statistical approach used. Whenever possible, we will contact study authors by email to obtain additional data that are needed but are lacking or unclear in the original trials. We will also consider several approaches to compute the missing data according to Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2022). For example, standard deviations can sometimes be computed with other reported statistics, such as standard error, interquartile range, or P values. We plan to analyse our data using the ITT principle. If study authors have not used ITT and do not respond to our enquiries about missing data, and we cannot assess the percentage of dropouts for each included trial or other information of relevance to the analysis is not reported in the trial, we will use the trial data as available to us (available‐case analysis). For sensitivity analyses, we will include missing data by considering participants as treatment failures or successes by imputing them according to the following two scenarios (Hollis 1999).

• Extreme‐case analysis favouring the experimental intervention ('best‐worse' case scenario): none of the dropouts/participants lost from the experimental group, but all the dropouts/participants lost from the control group experienced the outcome, including all randomised participants in the denominator.

• Extreme‐case analysis favouring the control intervention ('worst‐best' case scenario): all dropouts/participants lost from the experimental group, but none from the control group experienced the outcome, including all randomised participants in the denominator.

By performing these two sensitivity analyses, we will assess the impact of attrition bias on our findings. If the CIs and P value of the results of the primary meta‐analysis and the results of the sensitivity analysis are similar, the validity of the results is increased (Jakobsen 2014). If they differ substantially, this would suggest a risk of attrition bias.

We will perform sensitivity analyses to investigate the impact of imputed data on the overall assessment of results.

We will address the potential impact of missing data on our findings in the Discussion section of the review (Deeks 2022).

Assessment of heterogeneity

We will consider sources of heterogeneity, that is clinical, methodological, and statistical heterogeneity. We will assess clinical heterogeneity by analysing resemblances amongst included participants, interventions, comparators, and outcomes. We will assess methodological heterogeneity by examining the methodological characteristics (i.e. variation in study design and outcome measurement tools) and conducting risk of bias assessments with RoB 2 (Higgins 2022a). We will present a summary of the PICOT (participants, interventions, comparisons, outcomes, and time (duration of follow‐up)) elements and the risk of bias assessments in the beginning of the Results section. We will use the I² statistic alongside the Chi² test P value to assess statistical heterogeneity. We will use a P value of 0.10 to determine statistical significance, as it has low power in the (common) situation of a meta‐analysis when few studies are included, or the sample size of included studies is small. As recommended in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions, we will interpret the I² statistic as follows (Deeks 2022):

• 0% to 40%: might not be important;

• 30% to 60%: may represent moderate heterogeneity;*

• 50% to 90%: may represent substantial heterogeneity;*

• 75% to 100%: considerable heterogeneity.*

*The importance of the I² statistic depends on (1) magnitude and direction of effects and (2) the strength of evidence for heterogeneity, for example P value from the Chi² test, or a confidence interval for I²: uncertainty in the I² statistic is substantial when the number of trials is small.

We will employ forest plots to report heterogeneity visually, and describe the direction and magnitude of effects and the degree of overlap between CIs. In the case of moderate to substantial heterogeneity, we will explore the causes of heterogeneity by conducting subgroup or meta‐regression analyses (Deeks 2022). In the case of substantial heterogeneity, we will recheck whether the data have been extracted or entered correctly, or both. If the data are correct, we will not proceed with meta‐analysis (see Data synthesis).

Assessment of reporting biases

We will detect the source of selective reporting bias by comparing the protocols of included studies, published from 2006 and onwards, with their published reports, if possible, according to Chapter 13 of the Cochrane Handbook for Systematic Reviews of Interventions (Page 2022). We will attempt to contact investigators and authors of trial registrations or protocols, or abstracts of unpublished studies, to determine the status of these unpublished studies if identified. We will use a funnel plot to identify nonreporting bias when more than 10 trials are included in a meta‐analysis. Asymmetry in a funnel plot may be observed if study effect sizes are small, and may or may not be due to publication bias. We will comprehensively consider possible sources of funnel plot asymmetry, such as nonreporting bias, poor methodological quality, true heterogeneity, artefactual, and chance. We will perform Egger's test to assess the asymmetry (Egger 1997).

Data synthesis

Irrespective of the number of included trials, we will perform meta‐analyses using Review Manager Web (RevMan Web 2022). However, participants, interventions, comparisons, and outcomes should be sufficiently similar in the included trials to ensure that the conclusion is meaningful and of clinical significance. We will perform quantitative synthesis and illustrate the results of the meta‐analyses using forest plots according to Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2022). We will perform meta‐analyses using a random‐effects model as our main analysis and a fixed‐effect model as a sensitivity analysis.

If meta‐analysis is not possible (e.g. incompletely reported outcome/effect estimates or different effect measures used across studies), we will consult a statistician, according to Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (McKenzie 2022b), and describe the results in a narrative format based on the guidance on Synthesis Without Meta‐analysis (SWiM) (Campbell 2020).

Subgroup analysis and investigation of heterogeneity

Based on what we have presented in the Background and Methods sections so far, we plan to conduct the following subgroup analyses for the outcome all‐cause mortality. The following variables employed in subgroup analyses may be associated with the treatment response.

• Risk of bias: trials at 'low risk of bias' compared to those having 'some concerns' or at 'high risk of bias'.

• NAFLD stages: NAFL compared to NASH.

• Silymarin monotherapy compared to other molecules of silymarin (e.g. silymarin with phosphatidylcholine or choline), or in combination with vitamin E, or with a nutrient.

• Doses of silymarin: low dose (defined as less than 420 mg per day) compared to conventional dose (defined as 420 mg per day), or compared to high‐dose (defined as more than 420 mg per day).

• Treatment duration: short term (defined as less than three months) compared to long term (defined as three months or longer).

• Participant characteristics (e.g. overweight and insulin resistance).

• Trials with lifestyle modification compared to those without lifestyle modification.

• Trials without risk of vested interests compared to trials at risk of vested interests (Lundh 2017).

If necessary, we will consult a statistician to perform meta‐regression analyses to investigate the source of heterogeneity. We will employ a formal statistical method (e.g. the Z‐test, the Q‐test for heterogeneity, or the Q‐test based on one‐way analysis of variance (ANOVA)) to examine subgroup differences in Review Manager Web (RevMan Web 2022; Spineli 2020).

Sensitivity analysis

We will undertake the following sensitivity analyses for all outcomes to assess the robustness of the results.

• Repeating the analysis with the fixed‐effect model.

• Trials at 'low risk of bias' following the overall outcome assessment.

• 'Best‐worse' case scenario analysis (as described in Dealing with missing data).

• 'Worse‐best' case scenario analysis (as described in Dealing with missing data).

• Excluding trials with imputed data.

• Assessing imprecision with Trial Sequential Analysis (TSA).

Trial Sequential Analysis

We will use TSA as a sensitivity analysis to assess the imprecision for our three primary outcomes (i.e. serious adverse events, all‐cause mortality, and quality of life) (Castellini 2018; Gartlehner 2019; Jakobsen 2014). The underlying assumption of TSA is that testing for statistical significance may be performed each time a new trial is added to the meta‐analysis. We will add the trials at any overall risk of bias, according to the year of publication, and, if more than one trial was published in a year, we will add the trials alphabetically according to the last name of the first author. We will use the random‐effects model for our analyses. We will also calculate the diversity‐adjusted required information size (DARIS), that is the number of participants needed in a meta‐analysis to detect or reject a certain intervention effect (Brok 2008; Brok 2009; Thorlund 2010; Wetterslev 2008; Wetterslev 2009; Wetterslev 2017). We will also calculate the TSA‐adjusted CIs (Thorlund 2017; Wetterslev 2017). Based on the DARIS, we will construct the trial sequential monitoring boundaries for benefit, harm, and futility (Thorlund 2017; Wetterslev 2008; Wetterslev 2009; Wetterslev 2017). These boundaries determine the statistical inference one may draw regarding the cumulative meta‐analysis that has not reached the DARIS; if the trial sequential monitoring boundary for benefit or harm is crossed before the DARIS is reached, firm evidence may be established, and further trials may be superfluous. However, if the boundaries for benefit or harm are not crossed, it is probably necessary to continue conducting trials in order to detect or reject a certain intervention effect. If the cumulative Z‐curve crosses the trial sequential monitoring boundaries for futility, no more trials are needed.

For the dichotomous primary outcomes of our review, we will base the DARIS on the event proportion in the control group and use the following parameters to calculate its value: assuming an anticipated relative risk reduction of 20% based on the intervention effect suggested by trials at low risk of bias; a risk of type I error of 2.5% due to three primary outcomes (Jakobsen 2014); and a risk of type II error of 10% (Castellini 2018). For continuous outcomes, we will use a minimal relevant difference equal to standard deviation (SD)/2, where SD is the SD of the control group; type I error of 2.0%; and type II error of 10% for calculating the DARIS.

In TSA, we will downgrade our assessment of imprecision by two levels if the accrued number of participants is below 50% of the DARIS, and by one level if it is between 50% and 100% of the DARIS. We will not downgrade if futility or DARIS is reached. A more detailed description of TSA, and the software program, can be found at www.ctu.dk/tsa/ (Thorlund 2017).

Summary of findings and assessment of the certainty of the evidence

We will use GRADEpro GDT software to construct four summary of findings tables presenting the following comparisons (GRADEpro GDT).

• Silymarin monotherapy versus no intervention or placebo.

• Silymarin monotherapy versus another intervention different from the experimental intervention.

• Silymarin combined with vitamin E or nutrients (e.g. L‐carnitine) versus no intervention or placebo.

• Silymarin combined with vitamin E or nutrients (e.g. L‐carnitine) versus another intervention different from the experimental intervention.

We will present the following outcomes in the summary of findings tables: serious adverse events, all‐cause mortality, and quality of life. We will provide outcome results at the longest follow‐up (long term), with mean or median of the longest follow‐up in the trials and the range of follow‐up, for each outcome if available.

Summary of findings tables provide information on comparative risk, relative risk, number of participants, number of studies, and certainty of the evidence for the outcomes in the review comparisons.

We will use the GRADE approach to interpret the findings according to Chapter 15 of the Cochrane Handbook for Systematic Reviews of Interventions, Higgins 2022a; Schünemann 2022b, and the GRADE Handbook (Schünemann 2013).

Five factors reduce the certainty of evidence in randomised clinical trials: risk of bias, inconsistency of results, indirectness of evidence, imprecision, and publication bias. Based on defined criteria for these five factors, we will downgrade the evidence by one level for serious or two levels for very serious limitations.

Regarding risk of bias, we will use the overall judgement for an outcome result. 'Low risk of bias' will indicate 'no limitation' (no downgrading of the certainty of the evidence); 'some concerns' will indicate either 'no limitation' or 'serious limitation' (certainty of the evidence will be downgraded by one level); and 'high risk of bias' will indicate either 'serious limitation' or 'very serious limitation' (certainty of the evidence will be downgraded by two levels).

The GRADE approach categorises the certainty of evidence into one of four levels.

• High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.

• Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

• Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.

• Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

Two review authors (CW and YS) will independently assess the certainty of the evidence for each outcome and categorise the certainty as above. Any discrepancies will be resolved through discussion with a third review author (LC), and a consensus will be reached with the corresponding author (XQ). We will use footnotes to justify and document all decisions to downgrade the certainty of evidence. If necessary, we will provide comments to aid the reader’s understanding of the Cochrane Review.

Appendices

Appendix 1. Search strategies

Database	Time span	Search strategy
Cochrane Hepato‐Biliary Group Controlled Trials Register (via the Cochrane Register of Studies Web)	Date of search will be given at review stage	(Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu* or Marisil or Milk thistle or Realsil or Silibinin or Silimar* or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar* or Silyrin or Sivylar) and (((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction*))) and fat*) or NAFL* or steatohepat* or steato‐hep* or steatos* or NASH* or MAFL* or cirrho* or fibro*)
Cochrane Central Register of Controlled Trials in the Cochrane Library	Latest issue	#1 MeSH descriptor: [Milk Thistle] explode all trees #2 MeSH descriptor: [Silymarin] explode all trees #3 (Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu* or Marisil or Milk thistle or Realsil or Silibinin or Silimar* or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar* or Silyrin or Sivylar) #4 #1 or #2 or #3 #5 MeSH descriptor: [Fatty Liver] explode all trees #6 MeSH descriptor: [Liver Cirrhosis] explode all trees #7 MeSH descriptor: [Fibrosis] explode all trees #8 (((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction*))) NEAR fat*) or NAFL* or steatohepat* or steato‐hep* or steatos* or NASH* or MAFL* or cirrho* or fibro*) #9 #5 or #6 or #7 or #8 #10 #4 and #9
MEDLINE Ovid	1946 to date of search	1. exp Milk Thistle/ 2. exp Silymarin/ 3. (Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu* or Marisil or Milk thistle or Realsil or Silibinin or Silimar* or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar* or Silyrin or Sivylar).mp. [mp=title, book title, abstract, original title, name of substance word, subject heading word, floating sub‐heading word, keyword heading word, organism supplementary concept word, protocol supplementary concept word, rare disease supplementary concept word, unique identifier, synonyms] 4. 1 or 2 or 3 5. exp Fatty Liver/ 6. exp Liver Cirrhosis/ 7. exp Fibrosis/ 8. (((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction*))) adj fat*) or NAFL* or steatohepat* or steato‐hep* or steatos* or NASH* or MAFL* or cirrho* or fibro*).mp. [mp=title, book title, abstract, original title, name of substance word, subject heading word, floating sub‐heading word, keyword heading word, organism supplementary concept word, protocol supplementary concept word, rare disease supplementary concept word, unique identifier, synonyms] 9. 5 or 6 or 7 or 8 10. 4 and 9 11. (randomized controlled trial or controlled clinical trial or retracted publication or retraction of publication).pt. 12. clinical trials as topic.sh. 13. (random* or placebo*).ab. or trial.ti. 14. 11 or 12 or 13 15. exp animals/ not humans.sh. 16. 14 not 15 17. 10 and 16
Embase Ovid	1974 to date of search	1. exp Silybum marianum/ 2. exp silymarin/ 3. (Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu* or Marisil or Milk thistle or Realsil or Silibinin or Silimar* or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar* or Silyrin or Sivylar).mp. [mp=title, abstract, heading word, drug trade name, original title, device manufacturer, drug manufacturer, device trade name, keyword heading word, floating subheading word, candidate term word] 4. 1 or 2 or 3 5. exp fatty liver/ 6. exp liver cirrhosis/ 7. exp fibrosis/ 8. (((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction*))) adj fat*) or NAFL* or steatohepat* or steato‐hep* or steatos* or NASH* or MAFL* or cirrho* or fibro*).mp. [mp=title, abstract, heading word, drug trade name, original title, device manufacturer, drug manufacturer, device trade name, keyword heading word, floating subheading word, candidate term word] 9. 5 or 6 or 7 or 8 10. 4 and 9 11. Randomized controlled trial/ or Controlled clinical study/ or randomization/ or intermethod comparison/ or double blind procedure/ or human experiment/ or retracted article/ 12. (random$ or placebo or parallel group$1 or crossover or cross over or assigned or allocated or volunteer or volunteers).ti,ab. 13. (compare or compared or comparison or trial).ti. 14. ((evaluated or evaluate or evaluating or assessed or assess) and (compare or compared or comparing or comparison)).ab. 15. (open adj label).ti,ab. 16. ((double or single or doubly or singly) adj (blind or blinded or blindly)).ti,ab. 17. ((assign$ or match or matched or allocation) adj5 (alternate or group$1 or intervention$1 or patient$1 or subject$1 or participant$1)).ti,ab. 18. (controlled adj7 (study or design or trial)).ti,ab. 19. (erratum or tombstone).pt. or yes.ne. 20. or/11‐19 21. (random$ adj sampl$ adj7 ('cross section$' or questionnaire$ or survey$ or database$1)).ti,ab. not (comparative study/ or controlled study/ or randomi?ed controlled.ti,ab. or randomly assigned.ti,ab.) 22. Cross‐sectional study/ not (randomized controlled trial/ or controlled clinical study/ or controlled study/ or randomi?ed controlled.ti,ab. or control group$1.ti,ab.) 23. (((case adj control$) and random$) not randomi?ed controlled).ti,ab. 24. (Systematic review not (trial or study)).ti. 25. (nonrandom$ not random$).ti,ab. 26. 'Random field$'.ti,ab. 27. (random cluster adj3 sampl$).ti,ab. 28. (review.ab. and review.pt.) not trial.ti. 29. 'we searched'.ab. and (review.ti. or review.pt.) 30. 'update review'.ab. 31. (databases adj4 searched).ab. 32. (rat or rats or mouse or mice or swine or porcine or murine or sheep or lambs or pigs or piglets or rabbit or rabbits or cat or cats or dog or dogs or cattle or bovine or monkey or monkeys or trout or marmoset$1).ti. and animal experiment/ 33. Animal experiment/ not (human experiment/ or human/) 34. or/21‐33 35. 20 not 34 36. 10 and 35
LILACS (BIREME)	1982 to date of search	(Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu$ or Marisil or Milk thistle or Realsil or Silibinin or Silimar$ or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar$ or Silyrin or Sivylar) [Words] and (((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction$))) and fat$) or NAFL$ or steatohepat$ or steato‐hep$ or steatos$ or NASH$ or MAFL$ or cirrho$ or fibro$) [Words] and (random$ or blind$ or placebo$ or meta‐analys$) [Words]
Science Citation Index Expanded (Web of Science)	1900 to date of search	#5 #4 AND #3 #4 TI=(random* or blind* or placebo* or meta‐analys* or trial*) OR TS=(random* or blind* or placebo* or meta‐analys*) #3 #2 AND #1 #2 TS=(((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction*))) and fat*) or NAFL* or steatohepat* or steato‐hep* or steatos* or NASH* or MAFL* or cirrho* or fibro*) #1 TS=(Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu* or Marisil or Milk thistle or Realsil or Silibinin or Silimar* or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar* or Silyrin or Sivylar)
Conference Proceedings Citation Index – Science (Web of Science)	1990 to date of search	#5 #4 AND #3 #4 TI=(random* or blind* or placebo* or meta‐analys* or trial*) OR TS=(random* or blind* or placebo* or meta‐analys*) #3 #2 AND #1 #2 TS=(((nonalcoholic or non alcoholic or non‐alcoholic or (metabolic and (associated or dysfunction*))) and fat*) or NAFL* or steatohepat* or steato‐hep* or steatos* or NASH* or MAFL* or cirrho* or fibro*) #1 TS=(Carsil or Higado or Karsil or Legalon or Levalon or Limarin or Livarin or Liveril or Livosil or Lyma or Marianu* or Marisil or Milk thistle or Realsil or Silibinin or Silimar* or Siliphos or Silipide or Siliu or Silvia or Silybin or Silybon or Silybum or Silycristin or Silydianin or Silymar* or Silyrin or Sivylar)

Contributions of authors

Conceived the protocol: Xingshun Qi
Drafted the protocol: Cai'e Wang, Xingshun Qi
Revised the protocol: Cai'e Wang, Ghid Kanaan, Yiyang Shang, Lu Chai, Hui Li, Xingshun Qi

All authors agreed on the current protocol version for publication.

Sources of support

Internal sources

• None, Other

  Not applicable

External sources

• None, Other

  Not applicable

• The Cochrane Hepato‐Biliary Editorial Team, Denmark

  Provided advice during the protocol preparation and conducted the editorial process.

Declarations of interest

CW: none
GK: none
YS: none
LC: none
HL: none
XQ: none

New