Magnetic resonance imaging performed with gadoxetate disodium for the diagnosis of hepatocellular carcinoma in cirrhotic and non‐cirrhotic patients

Abstract

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

To determine the per‐lesion and per‐person diagnostic accuracy of magnetic resonance imaging (MRI) performed with gadoxetate disodium for the diagnosis of HCC in people with chronic liver disease.

To identify factors that influence the reported diagnostic performance of hepatobiliary contrast agents.

Background

Hepatocellular carcinoma (HCC) is the most common primary liver cancer. It is the fifth most commonly diagnosed cancer in men and seventh in women worldwide (Jemal 2011). In the United States, the overall five‐year survival rate is less than 12%, making HCC the fastest rising cause of cancer‐related death (Mittal 2013). Worldwide, HCC constitutes the second leading cause of cancer‐related mortality (WHO 2012).

HCC is occasionally diagnosed with an independent reference standard such as histopathology, whether by biopsy, resection or explant (Bruix 2011; Costa 2015). However, biopsy is not warranted for the diagnosis of HCC because of concerns for tumour seeding on the needle track, bleeding, and rate of false‐negative biopsies (Silva 2008; Pomfret 2010; Bruix 2011). Nowadays, biopsy is reserved for lesions with an atypical appearance and imaging features raising concerns about malignancy (Bruix 2011).

In a clinical setting, unequivocal imaging features on dynamic contrast‐enhanced computed tomography (CT) and magnetic resonance imaging (MRI) are accepted for definite diagnosis of HCC. Among people eligible for liver transplantation, the final diagnosis of HCC is based on biopsy in fewer than 5% of cases and on imaging in the remainder of cases (Wald 2013). Hence, imaging plays a critical role for the diagnosis of HCC. Many non‐invasive imaging tests have been proposed for the diagnosis of HCC (ultrasound, CT, MRI) and previous meta‐analyses have compared the diagnostic accuracy between these imaging modalities (Colli 2006; Chou 2015). In a meta‐analysis of noninvasive methods for the diagnosis of HCC, MRI had the highest pooled sensitivity estimates (81%) when compared with ultrasound (60%) and CT (68%) (Colli 2006). Yet the MRI results reported in this meta‐analysis were based on extracellular contrast agents. A recent meta‐analysis confirmed higher per‐lesion sensitivity with MRI (80%) than multidetector CT (68%), based on the pooled data of paired comparison between these two imaging modalities (Lee 2015).

The intravenous administration of gadolinium‐based contrast agents allows more specific diagnoses. The recent advances of hepatobiliary contrast agents has improved the ability to detect and characterise liver lesions (Semelka 2001; Reimer 2004; Cruite 2010). Two such contrast agents, gadobenate dimeglumine (Gd‐BOPTA) and gadoxetate disodium (Gd‐EOB‐DTPA), are commercially available. Multiphasic dynamic contrast‐enhanced phases acquired during the first minutes after injection of hepatobiliary contrast agents provide contrast properties similar to extracellular contrast agents (Vogl 1996). In addition to these early dynamic or vascular phases, a delayed hepatobiliary phase that peaks at between one and three hours (for gadobenate dimeglumine) (Petersein 2000; Reimer 2004) and at around 20 minutes (for gadoxetate disodium) (Hamm 1995; Reimer 1996) after contrast injection provides information on hepatocellular function and facilitates detection of small HCCs (Kim 2010; Park 2012a; Park 2012b).

We will not conduct a comparison of the diagnostic test accuracy of hepatobiliary versus extracellular agents because a meta‐analysis has previously demonstrated higher sensitivity (86% versus 78%) and specificity (94% versus 84%) with hepatobiliary than with extracellular agents for the diagnosis of HCC (Hanna 2016). However, these results were based on indirect comparisons in different participant cohorts. Only two studies have performed a direct comparison of extracellular and hepatobiliary agents (Filipone 2010; Park 2010), although the first study only assessed the degree of enhancement without evaluating the diagnostic accuracy (Filipone 2010), whereas the second study only included 18 participants, an insufficient sample size to perform a valid comparison between these contrast agents (Park 2010).

In cirrhosis, the degree of background liver enhancement with hepatobiliary contrast agents decreases with higher Child‐Pugh classes due to decreased liver function (Nakamura 2012). Furthermore, the ability to differentiate HCCs with the hepatobiliary phase of gadoxetate‐enhanced‐MRI may only apply to Child‐Pugh class A cirrhosis and not to Child‐Pugh B and C (Kim 2012a,Kim 2012b).

Hence, the clinical decision to adopt hepatobiliary contrast agents depends on several factors, including the performance characteristics in different grades of chronic liver disease severity, the additional time required to acquire the hepatobiliary phase images, the diagnostic accuracy compared to other contrast agents, and the incremental cost of the contrast agent.

For hepatobiliary contrast agents to replace extracellular agents as the preferred contrast agents for liver MRI in people with chronic liver disease, it must not only demonstrate the presence of HCC but also provide diagnostic information for adequate liver imaging in cirrhotic people with decreased liver function. This is critical, because 80% of people undergoing diagnostic imaging for HCC have underlying cirrhosis (McGlynn 2005).

Therefore, the focus of this review will be to determine whether the diagnostic accuracy of MRI performed with gadoxetate disodium for the diagnosis of HCC is affected by liver function in people with chronic liver disease (cirrhotic and non‐cirrhotic).

Target condition being diagnosed

Hepatocellular carcinoma

Hepatocellular carcinoma is the most common type of primary liver cancer. Most cases occur in people with chronic liver disease. The incidence of HCC increases in those who are hepatitis B carriers, those with a family history of HCC, and those with cirrhosis of various causes (including hepatitis C, primary biliary cirrhosis, haemochromatosis, alpha‐1‐antitrypsin deficiency, and other causes) (Bruix 2011). The target condition is 'hepatocellular carcinoma'. While there is no definite threshold for what constitutes 'small' or 'large', the imaging literature tends to consider 'small' hepatocellular carcinomas to be those with a diameter less than 2 cm (Kim 2010; Hwang 2011; Park 2012a).

Index test(s)

Gadoxetate disodium‐enhanced MRI

Gadoxetate disodium is a contrast agent used in liver MRI examinations that behaves like an extracellular contrast agent during the dynamic phase of contrast injection, from the arterial phase to the portal venous phase (Choi 2014). Additionally, gadoxetate disodium provides a hepatobiliary phase at about 20 minutes during which normal liver parenchyma enhances strongly, whereas cells that do not contain the transporters expressed only in functioning hepatocytes do not uptake this contrast agent. Hence, hypointensity with hepatobiliary contrast agents is a strong predictor of premalignancy or malignancy (Choi 2014).

Clinical pathway

Standard diagnostic practice:

HCC surveillance in at‐risk people is based on ultrasound at a six‐month interval for detection of nodules. If a distinctive nodule of 10 mm or more is found on surveillance, ultrasound, dynamic contrast‐enhanced CT or MRI are warranted as diagnostic tests (Bruix 2011). The choice of diagnostic imaging modality (CT or MRI) depends on local expertise and preference. However, MRI is often preferred over CT at liver transplant centres because of additional tissue contrast mechanisms. A negative MRI result leads to a return to six‐month surveillance, whereas a positive MRI (i.e. probable or definite HCC) leads to multidisciplinary tumour board discussion for patient‐specific management (Mitchell 2015).

Standard treatment practice:

Treatment of HCC depends on the tumour stage found at imaging (Bruix 2011). Curative treatments such as resection, transplantation and radiofrequency ablation may be offered for early‐stage HCC. Palliative treatments such as transarterial chemoembolisation or systemic chemotherapy with sorafenib are respectively offered for intermediate‐ and advanced‐stage HCC. Symptomatic treatment is reserved for end‐stage disease.

Prior test(s)

For screening and surveillance of HCC in at‐risk people, ultrasound performed at six‐month intervals is currently recommended (Trevisani 2002;Santagostino 2003;Bruix 2011).

For diagnosis of HCC, dynamic CT or dynamic MRI are recommended as first‐line imaging modalities in diagnostic systems in North America by the American Association for the Study of Liver Diseases (AASLD), the United Network for Organ Sharing (UNOS) and the Organ Procurement Transplantation Network (OPTN) and the Liver Imaging Reporting and Data System (LI‐RADS); in Europe by the European Association for the Study of the Liver (EASL) and the European Organization for Research and Treatment of Cancer (EORTC); and in Asia by the Japan Society of Hepatology (JSH) and the Asian Pacific Association for the Study of the Liver (APASL).

Role of index test(s)

Dynamic MRI performed with gadoxetate disodium, a hepatobiliary contrast agent, may improve the detection and characterisation of liver lesions over extracellular agents. While initially advocated as a second‐line (i.e. add‐on) imaging technique because of its additional tumour characterisation ability, multiphasic dynamic contrast‐enhanced phases acquired during the first minutes after injection of hepatobiliary contrast agents provide contrast properties similar to extracellular contrast agents. For this reason, MRI performed with gadoxetate disodium is increasingly used as a replacement for MRI performed with extracellular agents.

Alternative test(s)

We will focus on the diagnostic accuracy of MRI performed with gadoxetate disodium, and will not investigate alternative contrast agents or tests. Although gadobenate dimeglumine (Gd‐BOPTA) has a hepatobiliary phase, this is seldom used in practice because it occurs one to three hours after contrast injection, making this inconvenient for characterisation of liver lesions in clinical practice.

Rationale

Liver MRI has traditionally been performed with gadolinium‐based extracellular contras agents (Schneider 2005). The advent of gadolinium‐based contrast agents with hepatobiliary uptake, gadobenate dimeglumine (Gd‐BOPTA) and gadoxetate disodium (Gd‐EOB‐DTPA) has added a new contrast mechanism for detection and characterisation of liver lesions. Recently, four meta‐analyses have assessed the diagnostic performance of gadoxetate disodium for the detection of hepatocellular carcinoma (Liu 2013; Wu 2013; Chen 2014; Junqiang 2014). However, these meta‐analyses have included a limited number of studies (respectively 10, 10, 18, and 11) and have not assessed all confounders of interest. This field of research is rapidly evolving and it remains to be seen whether more recent studies confirm the results of early studies performed in centres of excellence where these hepatobiliary agents were first introduced.

While the Japan Society of Hepatology (Kudo 2011) and the North American LI‐RADS diagnostic systems (LI‐RADS 2014) include guidance on the inclusion and interpretation of hepatobiliary contrast agents, these guidelines do not advocate the use of these contrast agents over that of extracellular contrast agents.

Furthermore, hepatobiliary contrast agents are not currently included in consensus guideline recommendations from major scientific societies involved in the diagnosis of HCC, including those of the APASL, AASLD, European Association for the Study of the Liver and European Organization for Research and Treatment of Cancer (EASL/EORTC), and Organ Procurement and Transplantation Network and United Network for Organ Sharing (OPTN/UNOS) (Omata 2010; Bruix 2011; EASL‐EORTC 2012; OPTN/UNOS 2015).

A systematic review of the diagnostic performance of gadoxetate disodium for liver function is required to determine whether hepatobiliary agents should be included in future guidelines.

We will focus on per‐lesion diagnosis because management of HCC is dictated by the diagnosis and staging (according to the size and number) of individual tumours (OPTN/UNOS 2015). We will also report per‐person diagnosis whenever possible.

Four meta‐analyses (Liu 2013; Wu 2013; Chen 2014; Junqiang 2014) have focused on gadoxetate disodium‐enhanced MRI for the diagnosis of HCC. They have reported sensitivity in the range of 91% to 92% and specificity in the range of 93% to 95%. One meta‐analysis (Lee 2015) has shown higher per‐lesion sensitivity with gadoxetate disodium than with extra‐cellular agents in lesions of 2 cm or less.

Previous meta‐analyses (Liu 2013; Wu 2013; Chen 2014; Junqiang 2014) have computed summary receiver operating characteristic (ROC) curves for showing the trade‐off between sensitivity and specificity across the included studies. However, one study (Chen 2014) has used the Moses‐Littenberg model, which does not account for imprecision in individual study estimates or estimates of between‐study heterogeneity (random effects). For this study, as we expect that all the studies provide data using a similar set of diagnostic criteria for HCC, we will use a bivariate model. This approach can be used to compute estimates of the average sensitivity and specificity.

There is also a lack of cost‐effectiveness studies on hepatobiliary contrast agents versus extracellular agents. Depending on countries and institutions, MRI performed with hepatobiliary contrast agents may be used as an add‐on diagnostic test or as a replacement for extracellular contrast agents as the preferred noninvasive test for the diagnosis of hepatocellular carcinoma.

In view of the higher cost of gadoxetate disodium, a narrow estimate of its diagnostic performance as a noninvasive test is highly desirable. A high sensitivity and specificity for the diagnosis of hepatocellular carcinoma may encourage transition from MRI performed with extracellular agents to hepatobiliary contrast agents as the preferred imaging technique in people with chronic liver disease.

Objectives

To determine the per‐lesion and per‐person diagnostic accuracy of magnetic resonance imaging (MRI) performed with gadoxetate disodium for the diagnosis of HCC in people with chronic liver disease.

Secondary objectives

To identify factors that influence the reported diagnostic performance of hepatobiliary contrast agents.

Methods

Criteria for considering studies for this review

Types of studies

We will include studies performed in humans that report the diagnostic accuracy of gadoxetate disodium‐enhanced MRI for the diagnosis of hepatocellular carcinoma. We will accept all study designs, except patient‐control studies (formerly known as case‐control studies), as these are prone to spectrum bias. We will also exclude studies that have MR imaging as both the index test and the reference standard, because these are prone to incorporation bias and would lead to overestimation of diagnostic accuracy.

We will include studies irrespective of publication status, language, or publication date.

We will document both per‐lesion and per‐person analyses for studies that report these results. We will include phase IIA studies (Colli 2014) that estimate the accuracy (sensitivity and specificity) of an index test (gadoxetate disodium‐enhanced MRI) in discriminating between diseased (with HCC) and non‐diseased people.

Participants

Participants will include adults of any age with chronic liver disease, irrespective of the aetiology, in whom the diagnosis of hepatocellular carcinoma is confirmed by pathology (based on biopsy of focal liver lesion, surgical resection, explant) or a composite reference standard that also includes typical imaging findings with a follow‐up period of at least six months to confirm a negative result. The review focuses on the diagnosis of hepatocellular carcinoma. People with a known diagnosis of hepatocellular carcinoma or prior treatment by percutaneous ethanol injection (PEI), radiofrequency ablation (RFA), microwave ablation, transarterial chemoembolisation (TACE), transarterial beads embolisation (TABE), transarterial radioembolisation (TARE), selective radiation through intra‐arterial injection of lipiodol‐I‐131 or Yttrium‐90 labelled microspheres, sorafenib, or systemic chemotherapy form a distinct group. They are not the focus of this review, and therefore, we will exclude studies that include such participants, unless data are presented in such a way as to allow this group to be isolated from other included participants.

Index tests

Gadoxetate disodium‐enhanced MRI. Minimal field strength must be 1.5 T. Hepatobiliary phase delay must be of at least 20 minutes for gadoxetate disodium. Studies must contain a hepatobiliary phase to be included in this meta‐analysis.

Target conditions

Hepatocellular carcinoma.

Reference standards

Presence or absence of hepatocellular carcinoma should be confirmed by pathology (based on core needle biopsy of focal liver lesion, surgical resection, explant) or a composite reference standard that also includes imaging. To avoid incorporation bias, follow‐up with a different imaging modality demonstrating size stability or tumour growth may be used to confirm the diagnosis noninvasively, but only using imaging techniques other than MRI (such as contrast‐enhanced ultrasonography, CT or lipiodol).

The imaging findings must be typical for hepatocellular carcinoma to confirm a positive result or require a follow‐up period of at least six months to confirm a negative result.

A composite reference standard implies the following combinations:

Positive result (true positive or false positive): diagnosis of hepatocellular carcinoma confirmed by pathology or typical imaging findings. Typical imaging findings are defined as a mass equal to or larger than 1 cm that presents the hallmark enhancement pattern, defined as arterial phase hyperenhancement and washout appearance on dynamic contrast‐enhanced studies (LI‐RADS 2014).

Negative result (true negative or false negative): absence of tumour at pathology or absence of tumour growth during follow‐up period of at least six months to confirm benign nature of nodule.

There are two grading systems of hepatocellular carcinoma (Desmet 2009):

• The International Working Party (IWP) of the World Congress of Gastroenterology classified the nodular lesions encountered in chronic liver disease into large regenerative nodule, low‐grade dysplastic nodule, high‐grade dysplastic nodule, and hepatocellular carcinoma. Histological grading of hepatocellular carcinoma is based on the degree of tumour differentiation: well‐differentiated, moderately‐differentiated, poorly‐differentiated, and undifferentiated types. Small HCC was defined as a tumour measuring less than 2 cm (IWP 1995).

• The International Consensus Group for Hepatocellular Neoplasia (ICGHN) has subdivided small HCC into two clinical‐pathological groups named early HCC and progressed HCC (ICGHN 2009). The presence of stromal invasion differentiates early HCC from high‐grade dysplastic nodules. Early HCC has a well‐differentiated and nodular appearance. Progressed HCC has a well‐defined nodular pattern and is usually moderately differentiated.

In this meta‐analysis, we will consider any type of hepatocellular carcinomas as reference standard‐positive, regardless of the degree of tumour differentiation.

Search methods for identification of studies

Electronic searches

We will conduct searches in the Cochrane Hepato‐Biliary Group Controlled Trials Register, the Cochrane Hepato‐Biliary Group Diagnostic Test of Accuracy Studies Register, the Cochrane Library, MEDLINE (OvidSP), Embase (OvidSP), ACP Journal Club (OvidSP), Database of Abstracts of Reviews of Effects (DARE) (OvidSP), Health Technology Assessments (HTA) (OvidSP), NHS Economic Evaluation Database (NHSEED) (OvidSP), and Science Citation Index EXPANDED (SCI‐EXP) (ISI Web of Knowledge) (Royle 2003). We will search for abstracts posted online on Radiology Society websites (RSNA, ARRS, ECR). We will also search online trial registries such as ClinicalTrial.gov (clinicaltrials.gov/), the European Medicines Agency (EMA) (www.ema.europa.eu/ema/), the WHO International Clinical Trial Registry Platform (www.who.int/ictrp), the Food and Drug Administration (FDA) (www.fda.gov), as well as pharmaceutical company sources for ongoing or unpublished trials.

We will restrict the searches to studies performed on humans.

We will not restrict searches according to publication language.

We will not restrict searches by publication date.

We will use the multipurpose search command for the OvidSP interface (.mp.) and the topic search command for the ISI Web of Knowledge interface (TS=) to search both text and database subject heading fields. To capture variations in suffix endings, we will use the unlimited truncation symbol ‘*’ in both interfaces. The preliminary search strategies with the expected time spans of the searches are given in Appendix 1.

Searching other resources

We will seek additional references by handsearching the reference lists of articles retrieved from the computerised databases and relevant review articles.

Data collection and analysis

We will follow the available guidelines provided in the Cochrane Handbook for Diagnostic Test Accuracy Reviews (DTA Handbook 2013). We will use a prevalidated study‐report form (data abstraction form) to abstract data from the papers retained in the meta‐analysis. We will ask the contact author for papers where the reporting of data is unclear or where some data are missing.

Selection of studies

Two independent review authors (AT, KV) will review the publications independently for eligibility. We will do the selection of papers in three steps:

1. The review authors will read the title of all papers found;

2. The review authors will read the abstract of all papers retained after step 1;

3. The review authors will read the full text of all articles retained after step 3.

We will calculate the concordance and the Kappa score describing the inter‐rater reproducibility after each step. In case of disagreement, both review authors will re‐read the title, the abstract or the paper, and will reassess their decision independently. If the disagreement persists, we will ask a third review author (MM) to read the paper, and we will take the decision made by two out of three review authors.

Inclusion and exclusion criteria

We will include studies if they meet the following inclusion criteria:

1. Population: adults (18 years of older) with chronic liver disease or cirrhosis. 2. Pathology: undergoing imaging for diagnosis of HCC. 3. Imaging: MRI performed with gadoxetate disodium. 4. Reference standard: either pathology based on biopsy of focal liver lesion, surgical resection, explant; or a composite reference standard that also includes typical imaging findings with a follow‐up period of at least six months to confirm a negative result.

We will exclude studies if they meet any of the following criteria:

1. Participant number: fewer than 10 study participants, regardless of the number with HCC. 2. MRI field strength lower than 1.5 T. 3. Animal studies. 4. Incorporation bias: Studies that include MR imaging as both the index test and as part of the composite reference standard.

Data extraction and management

Two review authors (AT and MM) will independently complete a data extraction form for all included studies. We will retrieve the following data:

1. Study details: author, journal, year of publication, dates that the study was conducted, PDF availability, language, study population, setting (academic or community hospital), age range, inclusion criteria, exclusion criteria, country, region, maximum time interval between index test and reference standard, reference standard.

2. Sample size: number of participants meeting the criteria and total number screened.

3. Baseline characteristics: baseline aetiology of chronic liver disease, age, sex, race, disease severity. Severity of liver disease of the studied population may be considered using the Child‐Pugh score (Pugh 1973).

4. Study flow: number of people eligible, number of participants included, number of people excluded, number of lesions included, number of lesions excluded, number of participants allocated to each group.

5. Imaging technique: field strength, contrast agent, dosage, concentration, inclusion of hepatobiliary phase, hepatobiliary phase imaging delay.

6. Imaging interpretation: lesion size, T1 signal characteristics, T2 signal characteristics, number of readers (if more than one reader is reported we will use the ‘average’ results across readers).

7. The index test with all cut‐offs tested (sets of diagnostic criteria).

8. Clinical reference standard test: pathology only (biopsy, surgical resection, explant) or composite reference standard (combination of pathology and follow‐up imaging).

9. Number of true positive (TP), true negative (TN), false positive (FP), and false negative (FN). We will extract these data for each cut‐off presented and for the target condition.

Missing data


We will contact primary authors for missing or unclear data.

Assessment of methodological quality

We will assess eligible studies using Quality Assessment of Diagnostic Accuracy Studies‐2 (QUADAS‐2), an evidence‐based tool for the assessment of quality in systematic reviews of diagnostic accuracy studies (Whiting 2011). The tool is based on items that cover a wide range of methodological issues in diagnostic test studies. We will adopt the QUADAS‐2 items to examine the four domains: participant selection, index test, reference standard, and flow and timing. We will assess all four domains in terms of risk of bias, and the first three domains for concerns about applicability. The QUADAS‐2 signalling questions are listed in Appendix 2. We will classify studies at low risk of bias if all answers to the signalling questions for a domain are 'yes'. We will flag for potential bias if the answer to any signalling question is 'no'. We will classify risks as 'unclear' only when insufficient data are reported to permit a judgement.

AT and TH will independently perform quality assessment of studies. AT, TH, and MM will resolve any disagreement by consensus. 
We will address the impact on results of the seven quality items individually in sensitivity analyses.

Statistical analysis and data synthesis

We will summarise data from each study in several 2 x 2 tables of FP, FN, TP, TN (absolute counts) according to the different cut‐offs, the target condition and subpopulations, and we will enter the data into Review Manager 5 (RevMan) software.

Because we expect that all the studies will provide data using a similar set of diagnostic criteria for hepatocellular carcinoma, we will use a bivariate random‐effects model. We will assume that logit‐transformed sensitivities and specificities have an approximately bivariate normal distribution where sensitivities and specificities are potentially correlated.

We will combine extracted data from all included studies to compute the sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio, along with 95% confidence intervals, by using a bivariate random‐effects model (Reitsma 2005).

We will use bivariate models to pool the results of studies that share a common set of diagnostic criteria. We will conduct all analyses and plots using RevMan and SAS software. We will use the SAS macro 'metadas' to perform the analysis of the bivariate model.

Investigations of heterogeneity

We plan a priori subgroup analyses as follows, according to clinical or exploratory sources of heterogeneity:


Clinical heterogeneity subgroups: liver disease severity (non‐cirrhosis versus cirrhosis; Child‐Pugh A, B, or C).

Exploratory sources of heterogeneity:

• No hepatobiliary phase imaging performed or hepatobiliary phase imaging performed;

• Participant enrolment in academic or community hospital;


• Geographic region (Asia, Europe, North America);

• MRI field strength (1.5 T, 3.0 T);

• Dose of contrast agent;

• Type of image analysis (qualitative versus quantitative);

• Time of publication (before median year of included studies versus after median year of included studies);

• Number of readers (single versus multiple);

• Reference standard (pathology based on biopsy of focal liver lesion, surgical resection, explant; or a composite reference standard that also includes typical imaging findings with a follow‐up period of at least six months to confirm a negative result);

• Size threshold (< 1 cm, > 1 cm, 1 ‐ 2 cm, ≤ 2 cm, > 2 cm).

Investigation of heterogeneity will explore to what extent diagnostic accuracy is influenced by covariates.

We will use coupled forest plots of sensitivity and specificity to observe and describe the extent of heterogeneity. We will also use the parameter estimates from the bivariate model to produce the summary ROC curve, the summary values for sensitivity and specificity, a 95% confidence region around sensitivity and specificity, and a 95% prediction region. The latter will describe the extent of statistical heterogeneity.

We will perform subgroup analyses comparing sensitivity and specificity for subgroups listed in the potential sources of heterogeneity, adding factors one by one. We will use the Chi² statistic, computed as the change in the ‐2Log likelihood when a covariate is added (or removed) from the bivariate model, to assess if a subgroup factor should be added (or removed). If the statistic is significant, we will further examine the parameter estimates (and confidence regions) to ascertain whether the subgroup factor is associated with sensitivity, specificity, or both. Finally, we will present separately significant subgroup results.

Sensitivity analyses

In order to assess the robustness of the eligibility criteria, we will undertake sensitivity analyses to explore the effect of risk of bias and eligibility criteria.

1. Risk of bias: low risk of bias only.

2. Study design: prospective studies only.


3. Reference standard: pathology only.

4. Sample size: studies with 30 or more participants only.

Assessment of reporting bias

Reporting bias may be due to publication bias, i.e. studies with positive results are more likely to be published than those with negative or less favourable results. Tests to detect publication bias are currently used for systematic reviews of clinical trials. However, these tests are not widely used in diagnostic test accuracy, due to lack of power or consensus on their use. Therefore, we will not use these tests to explore publication bias in our review.

Appendices

Appendix 1. Search strategies

Database	Time span	Search strategy
The Cochrane Hepato‐Biliary Group Controlled Trials Register	Date will be given at review stage	((Magnetic AND resonance) OR MR*) AND (gadoxet* OR gadolinium* OR primovist OR eovist OR Gd‐EOB* OR EOB*) AND (((liver OR hepato*) AND (carcinom* OR cancer* OR neoplasm* OR malign* OR tumo*)) OR hepatoma OR HCC)
The Cochrane Hepato‐Biliary Group Diagnostic Test of Accuracy Studies Register	Date will be given at review stage	((Magnetic AND resonance) OR MR*) AND (gadoxet* OR gadolinium* OR primovist OR eovist OR Gd‐EOB* OR EOB*) AND (((liver OR hepato*) AND (carcinom* OR cancer* OR neoplasm* OR malign* OR tumo*)) OR hepatoma OR HCC)
The Cochrane Library	Latest issue	#1 MeSH descriptor: [Magnetic Resonance Imaging] explode all trees #2 (Magnetic and resonance) or MR* #3 #1 or #2 #4 MeSH descriptor: [Gadolinium DTPA] explode all trees #5 gadoxet* or gadolinium or primovist or eovist or Gd‐EOB* or EOB* #6 #4 or #5 #7 MeSH descriptor: [Liver Neoplasms] explode all trees #8 ((liver or hepato*) and (carcinom* or cancer* or neoplasm* or malign* or tumo*)) or hepatoma or HCC #9 #7 or #8 #10 #3 and #6 and #9
MEDLINE (OvidSP)	1946 to the date of search	1. exp Magnetic Resonance Imaging/ 2. ((Magnetic and resonance) or MR*).mp. [mp=title, abstract, original title, name of substance word, subject heading word, keyword heading word, protocol supplementary concept word, rare disease supplementary concept word, unique identifier] 3. 1 or 2 4. exp Gadolinium DTPA/ 5. (gadoxet* or gadolinium or primovist or eovist or Gd‐EOB* or EOB*).mp. [mp=title, abstract, original title, name of substance word, subject heading word, keyword heading word, protocol supplementary concept word, rare disease supplementary concept word, unique identifier] 6. 4 or 5 7. exp Liver Neoplasms/ 8. (((liver or hepato*) and (carcinom* or cancer* or neoplasm* or malign* or tumo*)) or hepatoma or HCC).mp. [mp=title, abstract, original title, name of substance word, subject heading word, keyword heading word, protocol supplementary concept word, rare disease supplementary concept word, unique identifier] 9. 7 or 8 10. 3 and 6 and 9
Embase (OvidSP)	1974 to the date of search	1. exp nuclear magnetic resonance imaging/ 2. ((Magnetic and resonance) or MR*).mp. [mp=title, abstract, heading word, drug trade name, original title, device manufacturer, drug manufacturer, device trade name, keyword] 3. 1 or 2 4. exp gadoxetic acid/ 5. (gadoxet* or gadolinium* or primovist or eovist or Gd‐EOB* or EOB*).mp. [mp=title, abstract, heading word, drug trade name, original title, device manufacturer, drug manufacturer, device trade name, keyword] 6. 4 or 5 7. exp liver tumor/ 8. (((liver or hepato*) and (carcinom* or cancer* or neoplasm* or malign* or tumo*)) or hepatoma or HCC).mp. [mp=title, abstract, heading word, drug trade name, original title, device manufacturer, drug manufacturer, device trade name, keyword] 9. 7 or 8 10. 3 and 6 and 9
Science Citation Index EXPANDED	1900 to the date of search	#4 #3 AND #2 AND #1 #3 TS=(((liver or hepato*) and (carcinom* or cancer* or neoplasm* or malign* or tumo*)) or hepatoma or HCC) #2 TS=(gadoxet* OR gadolinium* OR primovist OR eovist OR Gd‐EOB* OR EOB*) #1 TS=((Magnetic and resonance) or MR OR MRI)

Appendix 2. Risk of bias and applicability judgements in QUADAS‐2

Domain	Participant selection	Index test	Reference standard	Flow and timing
Description	Describe methods of participant selection Describe included participants (prior testing, presentation, intended use of index test, and setting) Studies that meet the inclusion criteria of this meta‐analysis should have a population of adults (18 years or older) with chronic liver disease or cirrhosis. The participants have clinical suspicion of HCC (defined broadly as elevated alpha‐fetoprotein (AFP), discovery on surveillance imaging or on nonsurveillance imaging). Participants must undergo MRI performed with gadoxetate disodium for diagnosis of HCC We will exclude studies with fewer than 10 participants, MRI field strength lower than 1.5 T, if they include animals, and if MRI is both part of the index test and the composite reference standard	Describe the index test and how it was conducted and interpreted MRI performed with gadoxetate disodium, conducted with field strength of 1.5 or 3.0 T and interpreted by at least one radiologist	Describe the reference standard and how it was conducted and interpreted Reference standard: either pathology based on biopsy of focal liver lesion, surgical resection, explant; or a composite reference standard that also includes typical imaging findings with a follow‐up period of ≥ 6 months to confirm a negative result	Describe any people who did not receive the index test(s) or reference standard (or both) or who were excluded from the 2 x 2 table (refer to flow diagram) Describe the time interval and any interventions between index test(s) and reference standard. If pathology is the reference standard, it should be performed within ≤ 3 months the index test If imaging is the reference standard, follow‐up period should be ≥ 6 months to confirm a negative result
Signalling questions: yes/no/unclear	Was a consecutive or random sample of participants enrolled? Yes: all consecutive participants or a random sample of people with clinical suspicion of HCC were enrolled in the study No: inclusion of normal volunteers or exclusion of eligible participants Unclear: insufficient data were reported to permit a judgement	Were the index test results interpreted without knowledge of the results of the reference standard? Yes: MRI results were interpreted without knowledge of the pathology results No: MRI results were interpreted with knowledge of the results of pathology results Unclear: insufficient data were reported to permit a judgement	Is the reference standard likely to classify the target condition correctly? Yes: if participants have undergone pathology analysis of tissue specimen or if follow‐up imaging was at least ≥ 6 months after index test to permit confirmation of negative diagnosis No: pathology was not deemed adequate for HCC diagnosis or if follow‐up imaging was performed too soon (< 6 months) to confirm a true negative Unclear: insufficient data were reported to permit a judgement	Was there an appropriate interval between index test(s) and reference standard? Yes: If pathology is the reference standard, the interval between pathology and MRI was ≤ 6 months If imaging is the reference standard, follow‐up period was ≥ 6 months. No: If pathology is the reference standard, the interval between pathology and MRI was > 6 months If imaging is the reference standard, follow‐up period was < 6 months Unclear: insufficient data were reported to permit a judgement
	Was a patient‐control design avoided? This question is not relevant because we exclude studies using patient‐control design. Hence, none of the included studies will have this design	If a threshold was used, was it prespecified? Yes: sets of imaging criteria for diagnosis of HCC were prespecified No: sets of imaging criteria for diagnosis of HCC were not prespecified Unclear: it is not reported or not clearly described	Were the reference standard results interpreted without knowledge of the results of the index test? Yes: the reference standard (pathology or follow‐up imaging) were interpreted without knowledge of the results of the MRI No: the reference standard (pathology or follow‐up imaging) were interpreted with knowledge of the results of the MRI Unclear: insufficient data were reported to permit a judgement	Did all participants receive the reference standard? Yes: all participants underwent the reference standard (pathology or follow‐up imaging) No: not all participants underwent pathology or follow‐up imaging Unclear: insufficient data were reported to permit a judgement
	Did the study avoid inappropriate exclusions? Yes: the study avoided inappropriate exclusions (e.g. small tumours, large number of tumours, infiltrative HCC, MRI difficult to interpret) No: the study excluded participants inappropriately Unclear: insufficient data were reported to permit a judgement			Did all participants receive the same reference standard? Yes: all participants received the same reference standard, i.e. pathology or follow‐up imaging No: not all participants received the same reference standard, i.e. pathology or follow‐up imaging Unclear: insufficient data were reported to permit a judgement
				Were all participants included in the analysis? Yes: all participants meeting the selection criteria (selected participants) were included in the analysis, or data on all the selected participants were available so that a 2 x 2 table including all selected participants could be constructed No: not all participants meeting the selection criteria were included in the analysis, or the 2 x 2 table could not be constructed using data on all selected participants Unclear: insufficient data were reported to permit a judgement
Risk of bias: high/low/unclear	Could the selection of participants have introduced bias? High risk of bias: yes, if the selection of participants could have introduced bias Low risk of bias: no, if the selection of participants would not have introduced bias Unclear risk of bias: insufficient data on participants selection were reported to permit a judgement on the risk of bias	Could the conduct or interpretation of the index test have introduced bias? High risk of bias: if the answer to the signalling questions on the conduct or interpretation of the index test is 'no' Low risk of bias: if the answer to the signalling questions on the conduct or interpretation of the index test is 'yes' Unclear risk of bias: if the answers to the 2 signalling questions on the conduct or interpretation of the index test is either 'unclear' or any combination of 'unclear' with 'yes' or 'no'	Could the reference standard, its conduct, or its interpretation have introduced bias? High risk of bias: if the answer to the signalling questions on the reference standard, its conduct, or its interpretation is 'no' Low risk of bias: if the answer to the signalling questions on the reference standard, its conduct, or its interpretation is 'yes' Unclear risk of bias: if the answers to the 3 signalling questions on the reference standard, its conduct, or its interpretation is either 'unclear' or any combination of 'unclear' with 'yes' or 'no'	Could the participant flow have introduced bias? High risk of bias: if the answer to the signalling questions on flow and timing is 'no' Low risk of bias: if the answer to the signalling questions on flow and timing is 'yes' Unclear risk of bias: if the answers to the 4 signalling questions on flow and timing is either 'unclear' or any combination of 'unclear' with 'yes' or 'no'
Concerns regarding applicability: high/low/unclear	Are there concerns that the included participants do not match the review question? High concern: there is high concern that the included participants do not match the review question Low concern: there is low concern that the included participants do not match the review question Unclear concern: if it is unclear	Are there concerns that the index test, its conduct, or interpretation differ from the review question? High concern: there is high concern that the conduct or interpretation of the MRI test differs from the way it is likely to be used in clinical practice Low concern: there is low concern that the conduct or interpretation of the MRI test differs from the way it is likely to be used in clinical practice Unclear concern: if it is unclear	Are there concerns that the target condition as defined by the reference standard does not match the review question? High concern: Not all participants underwent pathology or follow‐up imaging for diagnosis of HCC Low concern: all participants underwent pathology or follow‐up imaging for diagnosis of HCC	‐‐

Contributions of authors

Formulated the research question: AT, MM, CBS. Drafted the protocol: AT. Develop and run the search strategy: AT, CR. Involved in decision‐making: AT, MM, TH, KNV. Provided methodological opinion: AT, MM, CBS. Provided statistical expert opinion: TW, BM. Reviewed the protocol: MM, TH, KNV, ADK, TW, BM, CBS.

Sources of support

Internal sources

• No sources of support supplied

External sources

• FRQS‐ARQ, Canada.

  An Tang was supported by a Chercheur‐Boursier Junior 1 Research Award from the Fonds de Recherche du Québec en Santé and Fondation de l'association des radiologistes du Québec (FRQS‐ARQ #26993).

Declarations of interest

An Tang: none known. Matthew DF McInnes: none known. Tom Hope is on the advisory committee and receives a research grant from Guerbet SA. Grant funding, consultant and speakers bureau for GE Healthcare. Kim‐Nhien Vu: none known. Devendra K Amre: none known. Tanya Wolfson: none known. Benoît Mâsse: none known. Chantal Roy: none known. Claude B Sirlin receives research grants from GE, Siemens, and Guerbet.

New