Practice changing RCT design and rationale: Abbreviated MRI plus AFP vs. ultrasound plus AFP for HCC surveillance in cirrhosis (PREMIUM study)

Summary

Abdominal ultrasound every 6 months with or without serum alpha-fetoprotein (AFP) is recommended for hepatocellular carcinoma (HCC) screening in patients with cirrhosis. However, high-quality evidence demonstrating that this screening strategy reduces HCC-related mortality in cirrhosis is lacking. Dynamic contrast-enhanced abbreviated MRI (DCE aMRI) protocols for HCC screening have nearly identical performance to full multiphasic MRI (the gold standard for HCC diagnosis) and can be completed in under 15 min. PREMIUM is a multicentre randomised controlled trial comparing HCC screening by ultrasound+AFP every 6 months vs. DCE aMRI+AFP every 6 months for up to 8 years among patients with cirrhosis. Sponsored by and executed within the Department of Veterans Affairs (VA), participant recruitment began in November 2023. Eligible participants are Veterans aged 18-75 years with cirrhosis, high HCC risk, Child-Turcotte-Pugh (CTP) score ≤9, model for end-stage liver disease (MELD) score ≤20, and no MRI contraindications or life-threatening comorbidities. The DCE aMRI protocol consists of T1-weighted axial pre-contrast and DCE sequences (arterial, portal, and 5-minute delayed) following administration of extracellular gadolinium-based contrast agent, plus a T2-weighted sequence between portal and delayed phases. PREMIUM aims to randomise 4,700 participants (2,350 in each arm), who will undergo per-protocol imaging and follow-up for up to 8 years. The primary outcome is HCC-related mortality. Secondary outcomes include HCC stage at diagnosis, receipt of potentially curative HCC treatment, and all-cause mortality. The study is powered to detect at least a 35% relative reduction in HCC-related mortality in the aMRI+AFP arm vs. the ultrasound+AFP arm. If PREMIUM demonstrates reduced HCC-related mortality in the aMRI+AFP arm, it could provide the necessary evidence to recommend aMRI+AFP for HCC screening in patients with cirrhosis (ClinicalTrials.gov identifier: NCT05486572).

The PREMIUM Study is a registered clinical trial: NCT05486572.

Graphical abstract

Keypoints.

• •Specialty societies endorse ultrasound for HCC screening in individuals with cirrhosis.
• •However, the sensitivity and specificity of ultrasound to detect HCC at early stages is poor, and studies suggest there may be little survival benefit.
• •Abbreviated MRI protocols have been developed for HCC screening that have a similarly high sensitivity and specificity to complete MRI but can be completed in much less time (∼15 min), making them more suitable for screening.
• •Abbreviated MRI protocols that include dynamic contrast-enhanced images are more sensitive and specific than ultrasound and almost as sensitive for HCC as complete MRI.
• •However, it is not known whether screening for HCC in patients with cirrhosis by abbreviated MRI vs. ultrasound results in a reduction in HCC-related mortality.
• •Demonstration of a reduction in HCC-related mortality in an adequately designed and powered randomised controlled trial such as PREMIUM is necessary to change the standard of care.

Introduction

Professional hepatology associations recommend abdominal ultrasound every 6 months, with or without serum alpha-fetoprotein (AFP), for hepatocellular carcinoma (HCC) screening in patients with cirrhosis.[1], [2], [3] However, the US Preventive Services Task Force and the American Cancer Society do not provide specific recommendations for HCC screening in cirrhosis and high-quality evidence demonstrating that it reduces HCC-related mortality is lacking.[4], [5], [6]

Ultrasound ± AFP has inherent limitations. A meta-analysis suggested that the sensitivity of ultrasound for detecting T1 or T2 HCC was only 47% (95% CI 33%-61%) with a specificity of 91% (95% CI 86%-94%).⁷ The addition of AFP increased sensitivity to 63% (95% CI 48%-75%) but decreased specificity to 84% (95% CI 77%-89%). A prospective study found that ultrasound sensitivity for “very early-stage” HCC was only 27%, compared to 85% for MRI.⁸ Ultrasound and AFP are also limited by low specificity and high false positive rates, which can lead to screening harms.⁹ Additionally, ultrasound is limited by operator variability and performs less well in patients with obesity or hepatic steatosis,^2,10 who now represent a growing proportion of the population with cirrhosis due to rising MASLD (metabolic dysfunction-associated steatotic liver disease) rates.

In cirrhosis, HCC can be diagnosed definitively as a Liver Imaging Reporting and Data System (LI-RADS) 5 observation on a multiphasic, liver-protocol abdominal MRI (Table 1), which is considered a gold-standard diagnostic test[11], [12], [13], obviating the need for a diagnostic biopsy. HCCs as small as 1 cm in diameter can be reliably diagnosed. When prospectively evaluated as a screening test for HCC in patients with cirrhosis,^8,14 liver-protocol MRI had significantly higher sensitivity for early HCC detection than ultrasound (86.0% vs. 27.9%) and greater specificity (97.0% vs. 94.4%).⁸ Furthermore, 74.4% of HCCs in this MRI screening study were diagnosed at a very early stage (T1) and 23.3% at an early stage (T2), and 67.4% received potentially curative treatments. Only one out of 38 patients diagnosed with HCC died due to HCC progression.⁸

Table 1.

Side-by-side comparison of the sequences included in complete liver-protocol MRI vs. the DCE aMRI protocol for HCC screening.

Complete MRI: Liver/HCC protocol	Time (min)	DCE aMRI: HCC screening protocol	Time (min)
GETTING PATIENT ON SCANNER	3	GETTING PATIENT ON SCANNER	3
Coronal T2-weighted localizer	1	Coronal T2-weighted localizer	1
Coronal SSFSE T2 weighted	2	N/A	N/A
Axial FSE T2 weighted with fat saturation	8	N/A	N/A
Axial diffusion weighted	4	N/A	N/A
Axial in-phase and out-of-phase T1 weighted	2	N/A	N/A
Axial non-enhanced T1-weighted with fat saturation∗	1	Axial non-enhanced T1-weighted with fat saturation∗	1
CONTRAST INJECTION	2	CONTRAST INJECTION	2
Axial contrast-enhanced T1 weighted 3D GRE with fat saturation (arterial, portal, and 3-min and 5-minute delayed phases)∗	5	Axial contrast-enhanced T1 weighted 3D GRE with fat saturation (arterial, portal, and 5-minute delayed phases)∗ + Axial T2 weighted 3-minute delayed phase	5
Coronal contrast-enhanced T1 weighted with fat saturation (delayed phase)	1	N/A	N/A
GETTING PATIENT OFF SCANNER	3	GETTING PATIENT OFF SCANNER	3
Acquisition time
TOTAL SCANNER TIME	32	TOTAL SCANNER TIME	15

This illustrates that the sequences of the aMRI protocol (in bold) are a subset of the sequences normally included in a complete protocol (adapted from Canellas et al.⁵¹). The table shows that all the LI-RADS major features of HCC are characterised based on the sequences included in the aMRI protocol. aMRI, abbreviated MRI; DCE, dynamic contrast-enhanced; FSE, fast spin echo; GRE, gradient-echo; HCC, hepatocellular carcinoma; LI-RADS, Liver Imaging Reporting and Data System; SSFSE, single-shot fast spin echo.

^∗Sequences that can determine the LI-RADS major criteria for HCC: Arterial phase hyperenhancement, non-peripheral washout, enhancing capsule, threshold growth.

However, complete multiphasic MRI is time-consuming (30-45 min scanner time) and costly. For this reason, abbreviated MRI (aMRI) protocols have been developed specifically for the purpose of HCC screening,[15], [16], [17], [18] which reduce scanner time, thus reducing costs and improving patient convenience, with minimal reduction in sensitivity and specificity. To achieve this, dynamic contrast-enhanced (DCE) aMRI protocols include only T1-weighted pre-contrast and DCE sequences obtained after gadolinium-based contrast agent (GBCA) injection and a quick, single-shot fast spin echo T2-weighted sequence (Table 1). These sequences are necessary and sufficient to determine all the major LI-RADS criteria used for the diagnosis of a LI-RADS 5 “definite HCC” observation, i.e. arterial phase hyperenhancement, non-peripheral washout, enhancing capsule, and threshold growth. DCE aMRI protocols have scanner times (defined as acquisition time + table on/off time) of ∼15 min or less but have almost identical sensitivity and specificity as complete MRI for LI-RADS 5 HCC,¹⁵ and higher sensitivity/specificity than ultrasound or non-contrast-enhanced aMRI protocols.^19,20 Modelling studies suggest that aMRI may be a cost-effective HCC screening strategy,^21,22 while patients reported that they would prefer aMRI over ultrasound or biomarkers for HCC screening based on prioritising test performance over convenience or cost.²³

Randomised controlled trials (RCTs) with a no-screening control arm are not feasible,²⁴ since ultrasound ± AFP is the clinical standard of care in the US. Cancer screening tests ultimately need to demonstrate a reduction in cancer-related mortality, not just detection of cancer at earlier stages or more frequent receipt of potentially curative treatments. The US Department of Veterans Affairs (VA) Cooperative Studies Program (CSP) is conducting an RCT comparing ultrasound+AFP every 6 months with DCE aMRI+AFP every 6 months for up to 8 years in high-risk patients with cirrhosis, which is powered to detect a moderate difference in HCC-related mortality. This paper describes the rationale and methods of this study (CSP #2023), known as the Preventing Liver Cancer Mortality through Imaging with Ultrasound versus MRI (PREMIUM) study.

Methods

Overview of study design, primary outcome, and secondary outcomes

PREMIUM is an RCT comparing ultrasound+AFP every 6 months vs. DCE aMRI+AFP every 6 months for up to 8 years, among patients with cirrhosis at high risk of HCC, with planned enrolment of 2,350 participants per arm. Enrolment at each site will occur in Years 1-3, with screening and follow-up continuing through Year 8 (Fig. 1). The primary outcome is cumulative HCC-related mortality. Secondary outcomes include HCC stage at diagnosis, receipt of potentially curative treatment, and all-cause mortality. A biorepository of plasma collected at biannual screening intervals and an image repository consisting of the digital files of all per-protocol aMRI and ultrasound tests will also be established.

Fig. 1.

Schematic representation of enrolment, screening and follow-up periods in PREMIUM.

HCC, hepatocellular carcinoma.

Study setting

PREMIUM plans to recruit participants at up to 47 VA medical centres (Table S1), aiming to randomise on average 100 participants/site. Participating sites were selected based on having an adequate number of patients with cirrhosis eligible for HCC screening, a qualified gastroenterologist/hepatologist and radiologist willing to serve as co-local site investigators (LSIs), adequate MRI and ultrasound capacity to perform the per-protocol studies in a timely manner, and access to a multidisciplinary liver tumour board (MLTB) for appropriate and timely treatment of study participants with newly diagnosed HCC. PREMIUM began enrolment in November 2023. There are currently 17 actively enrolling sites with plans to expand the number of sites.

Study team

Each site’s team includes a gastroenterologist/hepatologist-LSI, a radiologist-LSI, and a study coordinator. The gastroenterologist/hepatologist-LSI works with and oversees the study coordinator to ensure local regulatory approvals are in place, recruit eligible participants, execute study procedures, and collect study data. The radiologist-LSI ensures that the correct imaging protocol and reporting templates are used. The radiology-LSI and one to two other local radiologists interpret each imaging study.

The responsibilities of PREMIUM’s two study Chairs, national study coordinators, the CSP Coordinating Center in West Haven (which includes the Study Biostatistician, Project Manager, and Data Monitoring team), the CSP Site Monitoring, Auditing and Resource Team (SMART), the CSP Clinical Research Pharmacy and PREMIUM’s Executive Committee and Data Monitoring Committee are outlined in the supplementary appendix.

Eligibility criteria

PREMIUM’s inclusion criteria (Box 1) include age 18-75 years, having a diagnosis of cirrhosis (any aetiology) and high HCC risk, based on at least one of the following high-risk criteria, which were shown to be associated with an HCC incidence of ∼2.5 per 100 patient-years or more in patients with cirrhosis: active HCV,[25], [26], [27], [28] fibrosis-4 (FIB-4) ≥3.25,[29], [30], [31] estimated annual risk >2.5% using the HCC risk calculator^27,31 or non-invasive criteria for clinically significant portal hypertension (CSPH) based on liver stiffness ± platelet count.^32,33 These high-risk criteria were included to ensure sufficient incident HCCs for the study to be adequately powered to detect a moderate reduction in HCC-related mortality associated with receipt of DCE aMRI+AFP screening. PREMIUM has 15 exclusion criteria (Box 1), including contraindications to HCC screening. Overall, about 50% of patients with documented cirrhosis in the VA are expected to fulfil at least one of PREMIUM’s high-risk criteria.

Box 1. PREMIUM study eligibility criteria.

Alt-text: Box 1

CirCom, Cirrhosis comorbidity; CSPH, clinically significant portal hypertension; FIB-4, fibrosis-4; HCC, hepatocellular carcinoma; INR, international normalised ratio; LS, liver stiffness; LSI, local site investigator; VA, Veterans Affairs; VCTE, vibration-controlled transient elastography.

Screening and identification of potentially eligible participants

Multiple approaches are used to identify potentially eligible participants (Fig. 2), taking advantage of VHA’s comprehensive electronic health records and the availability of an “Advanced Liver Disease dashboard”, a national population health management tool for patients with cirrhosis that is updated in real time and includes many of PREMIUM’s eligibility criteria. Study coordinators at each participating site review patients on the Advanced Liver Disease dashboard, patients with cirrhosis and upcoming GI/hepatology clinic appointments, as well as patients referred to PREMIUM, and perform an initial chart review for eligibility criteria. A VA Central Institutional Review Board-approved invitation letter is sent to potentially eligible participants followed by a telephone call 2 weeks later to describe the study, establish eligibility, and confirm interest before scheduling a baseline visit.

Fig. 2.

PREMIUM screening and identification of potentially eligible study participants.

ALD Dashboard, Advanced Liver Disease Dashboard; CirCom, Cirrhosis comorbidity score; CTP, Child-Turcotte-Pugh; Cub-I, sub-investigator; FIB-4, fibrosis 4; GFR, glomerular filtration rate; HCC, hepatocellular carcinoma; LS, liver stiffness; MELD, model for end-stage liver disease.

Baseline visit, informed consent, and randomisation

At a face-to-face baseline visit, informed consent is obtained, the eligibility form is completed, laboratory tests are performed, and baseline forms are completed (Fig. 3). These forms capture demographics; medical history (including alcohol use, smoking, and physical activity); laboratory results; medications; physical examination findings (including the liver frailty index); as well as CTP and MELD scores. Information is obtained from VA electronic health records, outside medical records, participant interviews, and the LSI-conducted history and examination. The participant’s most recent prior ultrasound, contrast-enhanced abdominal CT, and contrast-enhanced abdominal MRI are captured at the time of study enrolment. Once eligibility is confirmed by the LSI, the participant is randomised via Interactive Web Response System software to either the ultrasound+AFP or the aMRI+AFP arm. Randomisation is stratified by participating site and MELD score (MELD 6-10 vs. MELD 11-20). The first per-protocol screening (ultrasound+AFP or aMRI+AFP) is scheduled 6 months (±30 days) from the participant’s most recent ultrasound or complete, multiphasic CT/MRI, or, if none had been performed in the 6 months prior to enrolment, within 30 days after enrolment. Participants are to be re-imbursed $45 for the baseline visit and $25 for each subsequent per-protocol screening visit.

Fig. 3.

Baseline visit, informed consent, and randomisation.

aMRI, abbreviated MRI; CTP, Child-Turcotte-Pugh.

Overview of study procedures: ultrasound+AFP vs. aMRI+AFP every 6 months ± 30 days

Eligible participants are randomised in a 1:1 ratio to screening aMRI+AFP every 6 months vs. screening ultrasound+AFP every 6 months, continuing for up to 8 years after randomisation or until study termination or participant withdrawal (Fig. 4). Additionally, every 6 months blood is drawn for laboratory tests (serum AFP, complete blood count, prothrombin time/international normalised ratio, and complete metabolic panel) and biorepository blood collection (for participants who consented to the optional biorepository collection).

Fig. 4.

Cadence of PREMIUM 6-monthly procedures with per-protocol ultrasound + AFP vs. aMRI+AFP performed every 6 months ± 30 days.

AE, adverse event; aMRI, abbreviated MRI; CTP, Child-Turcotte-Pugh; HCC, hepatocellular carcinoma; MELD, model for end-stage liver disease; PCP, primary care provider; SAE, serious adverse event; US, ultrasound.

Screening ultrasound, aMRI and serum AFP tests are ordered and scheduled by the study team. Imaging is performed by the radiology departments of each participating site and interpreted by local radiologists at each site. Three days following the per-protocol screening, study coordinators contact participants to document any adverse events (aMRI arm only) or serious adverse events (both arms). Within 14 days following per-protocol screening, the study coordinator and LSI ensure that the participant and their gastroenterologist/hepatologist and primary care physician are alerted to the findings. Appropriate follow-up and treatment of any “positive” or “abnormal” findings on screening ultrasound, aMRI, or serum AFP testing are determined by the participant’s gastroenterologist/hepatologist and primary care provider in consultation with the local MLTB.

Each subsequent per-protocol screening episode (i.e. ultrasound+AFP or aMRI+AFP) is scheduled by the study team at 6 months ± 30 days from the prior screening episode; failure to do so constitutes a protocol deviation. If per-protocol screening is missed, the study team tries to schedule it as soon as possible and continues to do so every 2 months for up to 12 months, after which the participant is declared lost to follow-up. If the participant undergoes a complete multiphasic abdominal CT or MRI as part of routine clinical care after their last per-protocol screening, the next screening is rescheduled for 6 months after that clinical CT or MRI, and subsequent screening studies continue every 6 months thereafter. Rescheduling of the per-protocol imaging when diagnostic quality imaging is performed for any clinical reason (such as rising AFP) avoids redundant studies, onerous travel for the participant, and interference with clinical decision-making. Prior to each per-protocol screening episode, study coordinators assess if the participant is still eligible for screening (i.e. no new diagnosis of HCC, MELD score ≤20, CTP score ≤9, no life-threatening comorbidities, no liver transplantation).

Participants randomised to the aMRI arm who develop a permanent contraindication to MRI (e.g. permanent pacemaker placement or severe claustrophobia) are offered the option to cross over to the ultrasound arm. Patients with known poor visualisation of the liver by ultrasound are excluded from the study (Box 1); participants randomised to the ultrasound arm who have an ultrasound that documents poor visualisation (LI-RADS C) are offered multiphasic CT or MRI.

Description of DCE aMRI screening protocol and rationale

The DCE aMRI protocol includes T1-weighted axial pre-contrast and T1-weighted axial DCE phases obtained after administration of intravenous GBCA, i.e. the arterial, portal venous, and 5-minute delayed phases (Table 2). An additional T2-weighted, axial sequence is included before the 5-minute delayed T1-weighted sequence, because it improves specificity by diagnosing benign cysts and haemangiomas and reduces the number of call-backs for complete MRIs due to indeterminate lesions without prolonging the total scan time. Any of the available American College of Radiology (ACR) group II extracellular GBCA are acceptable. We selected the regimen (DCE aMRI + AFP) that would be most likely to result in a reduction in HCC-related mortality.

Table 2.

Dynamic contrast-enhanced abbreviated MRI protocol.

Sequence	Plane	Type	Slice/gap [mm]	Minimum resolution: image matrix [mm]	Fat saturation	Field of view	Comments
Coronal and axial T2-weighted localizer
T1-weighted pre-contrast	Axial	3D GRE∗	3-5/0	∼1.5 × 1.5 mm	Yes	Cover entire liver	Check for artifacts or motion prior to contrast administration.
Gadolinium-based contrast agent IV injection Bolus tracking: Start dynamic sequence when contrast reaches SMA
T1-weighted post-contrast dynamic	Axial	3D GRE∗	3-5/0	∼1.5 × 1.5 mm	Yes	Cover entire liver	Post-contrast timing: Arterial Portal venous 5-minute delay
T2-weighted	Axial	2D SSFSE†	6-8/2	∼1.5 × 1.5 mm	No	Cover entire liver	3-min delay post-contrast injection (prior to 5 min delay T1w)

SMA, superior mesenteric artery.

^∗3D GRE: 3D gradient-echo sequences (includes GRE protocols like GE’s LAVA, Siemens’ VIBE and Philips’ THRIVE).

^†SSFSE: GE’s single-shot fast spin echo (or Philips’ Single-shot turbo spin echo, Siemens’ HASTE).

Non-contrast and hepatobiliary contrast agent aMRI protocols are also available and have strengths and weaknesses compared to DCE aMRI.³⁴ This DCE aMRI protocol was chosen because it has sensitivity and specificity for HCC almost identical to that of complete MRI³⁵ (the gold-standard diagnostic test), is more compatible with the LI-RADS system, can result in the definitive diagnosis of a LI-RADS 5 observation by determining all the major LI-RADS criteria for HCC and is expected to result in fewer call-backs for complete MRI to evaluate “positive” or “indeterminate” lesions. A DCE aMRI can be executed using any of the four ACR group II extracellular GBCAs, which are cheaper and more widely available than gadoxetate, the GBCA used for hepatobiliary aMRI. The DCE sequences are very familiar to radiologists and technologists because they are a routine component of all liver-protocol MRIs and can still be executed in <15 min of scanner time (including <10 min of acquisition time).

Participants randomised to the aMRI arm who develop renal impairment with estimated glomerular filtration rate <30 ml/min, ongoing acute kidney injury, or hypersensitivity reaction after exposure to contrast during the study will be unable to receive GBCAs and therefore cannot undergo DCE aMRI. These participants will instead undergo a non-contrast abbreviated MRI protocol (Table 3), which has lower sensitivity/specificity than the DCE aMRI protocol,¹⁹ but higher sensitivity/specificity than ultrasound screening.²⁰

Table 3.

Non-enhanced abbreviated MRI protocol: for participants who develop GFR <30 ml/min during the study.

Sequence	Plane	Type	Slice/gap [mm]	Minimum resolution: image matrix [mm]	Fat saturation	Field of view
Coronal and axial T2-weighted localizer
T2-weighted	Axial	2D SSFSE†	6-8/2	∼1.5 × 1.5 mm	Optional	Cover entire liver
T2-weighted	Coronal	2D SSFSE†	6-8/2	∼1.5 × 1.5 mm	Optional	Cover entire liver
T1-weighted	Axial	3D GRE∗	3-5/0	∼1.5 × 1.5 mm	Yes (Dixon optional)	Cover entire liver
Diffusion weighted	Axial	2D EPI∗∗	6-8/2	∼1.5 × 1.5 mm	Yes	Cover entire liver

GFR, glomerular filtration rate.

^†2D SSFSE: 2D single-shot fast spin echo.

^∗3D GRE: 3D gradient-echo sequences (includes GRE protocols like GE’s LAVA, Siemens’ VIBE and Philips’ THRIVE).

^∗∗2D EPI: 2D echo planar imaging.

Description of ultrasound screening protocol and rationale

The PREMIUM screening ultrasound examination includes all the views recommended by ultrasound LI-RADS v2017 (and subsequently by ultrasound LI-RADS v2024) for an ultrasound performed for HCC screening.³⁶ Additional views of focal observations are obtained as needed.

Ultrasound and aMRI reporting

PREMIUM established a radiology workgroup that developed standardised reporting templates (see supplementary appendix) to ensure consistent documentation of findings and visualisation scores based on the LI-RADS system, and to support future studies utilising the image repository. The aMRI templates prompt documentation of the size, segmental location and CT/MRI LI-RADS v2018 categories (LR 1-5, LR-M, LR-TIV, LR-NC) of each observation and the subjective visualisation score (categorised as A [unlikely to have reduced sensitivity for observations <10 mm], B [likely reduced sensitivity for observations <10 mm but not observations ≥10 mm], or C [likely reduced sensitivity for observations ≥10 mm]). Important incidental findings, either in the liver or other organs, that require further evaluation are also documented. Ultrasound templates prompt documentation of the ultrasound LI-RADS v2017 categories (negative [US-1] if there are no focal abnormalities or if only definitely benign lesions such as cysts are identified; subthreshold [US-2] if there are lesions measuring <10 mm that are not definitely benign and no lesions that would define a “positive” examination; or positive [US-3] if there are lesions measuring ≥10 mm that are not definitely benign) and the visualisation score (A [no or minimal limitations], B [moderate limitations], or C [severe limitations]).

Radiologist and technologist training and quality control

The PREMIUM radiology workgroup provides training and ensures adequate quality control. At study kick-off, the radiologist-LSIs underwent training on the ultrasound and aMRI protocols and reporting templates, which are also available at the PREMIUM SharePoint site. Prior to initiating the study, each radiologist-LSI ensured that these protocols and templates were uploaded locally, that their technologists were adequately trained to execute them, and that any additional local radiologists interpreting PREMIUM imaging studies were aware of them.

The digital files of PREMIUM screening aMRIs and ultrasounds and their reports are uploaded to a central image repository and de-identified. For quality control, the PREMIUM radiology workgroup reviews the images files and reports of the first three study aMRIs and study ultrasounds from each site, and then one out of every 20 studies for the first 2 years and one out of every 40 thereafter, using dedicated review forms. These forms document whether PREMIUM imaging and reporting protocols were executed correctly, with emphasis on LI-RADS reporting of findings, visualisation score and scan time. Problems, discrepancies, or deviations are discussed with participating site radiologists.

Ascertainment of primary outcome: HCC-related death

All study participants who die and have a diagnosis of HCC (based on a LI-RADS 5 observation, a biopsy proven HCC, or documented consensus of MLTB) will be reviewed by the Endpoint Attribution Committee (EAC). Not all deaths in patients with HCC are related to the presence of HCC. PREMIUM will use validated criteria to determine whether HCC contributed to the patient’s death.^6,37 In patients who do not have an obvious non-liver-related cause of death (including myocardial infarction, cerebrovascular accident, pulmonary embolism, and non-liver-related malignancy), the presence of one or more of the criteria listed in Box 2 will be considered sufficient for the EAC to classify the death as HCC-related.

Box 2. Criteria used by the PREMIUM Endpoint Attribution Committee to determine if hepatocellular carcinoma contributed to a participant’s death.

Alt-text: Box 2

∗These criteria refer to the maximum tumour burden recorded – not the tumour burden at diagnosis or at death. CVA, cerebrovascular accident; EAC, Endpoint Attribution Committee; HCC, hepatocellular carcinoma; MI, myocardial infarction; PE, pulmonary embolism.

EAC members will be blinded to the randomisation arm. All necessary records required to make a determination of cause of death will be provided after redaction to remove any mention of the randomisation strategy or how HCC was detected. Each of the three EAC members will perform their initial review independently and submit their determination whether the cause of death was definitely HCC, probably HCC, probably not HCC, or definitely not HCC. Disagreements will be resolved by a meeting for full review and vote.

Ascertainment of secondary outcomes: stage at HCC diagnosis, receipt of curative treatments and all-cause mortality

HCC is defined by the presence of at least one of the three criteria listed above and the stage at diagnosis is confirmed and documented by the gastroenterologist/hepatologist-LSI, including size and location of each LI-RADS 5 lesion and presence/absence of vascular invasion, infiltration, lymph node metastasis, or distant metastasis.

All HCC-directed treatments, imaging, and other diagnostic studies are captured by dedicated electronic case report forms every 6 months from the time of diagnosis of HCC until the end of the study or participant death, based on participant interview, chart abstraction and review of all imaging and other diagnostic tests. Potentially curative treatments for HCC are defined as any of the following: liver transplantation, surgical resection of liver tumour, ablation by any modality (not in combination with other locoregional treatments for concomitant HCCs).

Deaths are identified during 6-monthly structured chart extractions and telephone surveys as well as by electronic queries of the VA Corporate Data Warehouse, which ascertains deaths from both VA and non-VA sources.

PREMIUM biorepository and image repository

Among participants who provide additional consent, 10 ml of blood is collected every 6 months in an EDTA tube and shipped at 4 °C to the VA Central Biorepository in Boston, MA, where it is processed robotically into four 1.0 ml aliquots of plasma and two 1.0 ml aliquots of buffy coat (for DNA) in 2D-barcoded Matrix™ tubes which are stored at -80 °C at VA Central Biorepository. It is anticipated that these specimens will facilitate future phase III and IV studies of HCC biomarker validation.³⁸

The digital image files and their associated reports, for participants who provide consent, are electronically transferred to the VA Precision Oncology Data Repository in Boston, MA, where they are de-identified and stored. It is anticipated that this image repository will facilitate future image analytics for liver cancer screening and prediction.

Addressing potential harms of contrast-enhanced aMRI and HCC screening

Patients with a glomerular filtration rate <30 at baseline are excluded from the study. We are using only ACR group II GBCAs. Macrocyclic (group 2) agents, characterised by higher kinetic and thermodynamic stability, are now the preferred choice due to their lower propensity for gadolinium release and tissue deposition. With these measures in place almost no cases of nephrogenic systemic fibrosis or other serious organ damage, including in the brain, have been reported.

To adequately capture harms and costs, we are collecting data on all additional imaging performed in each 6-month screening period, i.e. not just the per-protocol ultrasound or aMRI, but any additional imaging (CT or MRI with or without contrast and ultrasound). We are capturing any liver biopsies. We are also prospectively capturing all treatment administered for any liver lesions identified in study participants. Adverse events and serious adverse events will be captured within 3 days of aMRI.

We expect the aMRI to detect incidental findings, including both hepatic findings (such as intrahepatic biliary cancer), and extrahepatic findings such as abdominal malignancies (stomach, pancreas, kidney, intestine, and soft tissue), abdominal aortic aneurysm, or other clinically significant vascular abnormalities. All associated tests and outcomes for clinically significant incidental findings are being collected.

The psychological consequences of screening are difficult to assess and would require repeated administration of dedicated, time-consuming assessment instruments. Although undoubtedly useful, we reasoned that this would place great burden on study participants and ultimately reduce our enrolment and subsequent adherence to 6-monthly screening visits for up to 8 years, thus jeopardising our ability to test our primary outcome.

Statistical analysis

Primary outcome

The primary analysis is the intent-to-screen comparison of time to HCC death in the two study arms. The Kaplan-Meier product limit method will be used to estimate cumulative HCC mortality rates. A stratified log-rank statistic will be used to test differences in HCC mortality rates between the two arms to account for the stratified randomisation by site and MELD category. All time-to-event data will be censored at the date of death from causes other than HCC or at the end of the 8-year follow-up period for survivors (i.e. administrative censoring).

Secondary outcomes

Analyses of stage at HCC diagnosis (secondary outcome #1) between screening arms will be assessed using Pearson chi-square test at an overall significance level of 0.05, adjusting for multiple comparisons. Additional analyses will be conducted using logistic models to adjust for other clinical factors, such as: age, sex, race/ethnicity, FIB-4 score, MELD score, and aetiology of cirrhosis (active HCV, cured HCV, other). We will analyse receipt of potentially curative treatment (secondary outcome #2) using time-to-event methods. For this statistical analysis, follow-up will extend from the time of randomisation until the time of first receipt of curative treatment and an unadjusted log-rank test comparing the two arms will be performed. An intent-to-screen comparison of time to all-cause death (secondary outcome #3) between the two arms will be performed using methodology similar to that used for the primary outcome of HCC-related death.

Sample size and power calculations

The sample size was determined by the method of Lakatos^39,40 for the log-rank test with a two-sided significance level of 5%, based on the time-to-event analysis of the primary outcome of 8-year cumulative HCC-related mortality. We assumed that each participant would be followed for a mean of 6.75 years and those who do not experience the event (HCC-related death) would be censored at the end of Year 8 of the study. We estimated the adjusted cumulative 8-year HCC-related mortality in the ultrasound+AFP arm to be 7.1 per 100 participants, based on an annual HCC incidence of 3.0 per 100 patient-years and an HCC case fatality rate of 43%, accounting for a drop-out rate of 4% per year due to non-HCC-related mortality and a 1% per year withdrawal rate. We employed modelling to determine the relative risk reduction in HCC-related mortality that would be expected as a result of screening by aMRI+AFP vs. ultrasound+AFP, by combining different estimates of HCC stage migration that would be expected together with estimates of case fatality according to stage. Inputs were informed by the published sensitivity and specificity of imaging modalities^7,8 and HCC-related case fatality rates.⁶ These calculations yielded a conservative estimate of a 35% relative reduction in HCC-related mortality in the aMRI+AFP arm vs. the ultrasound+AFP arm, corresponding to a reduction in cumulative 8-year HCC-related mortality from 7.1% in the ultrasound+AFP arm to 4.6% in the aMRI+AFP arm (or a 2.5% absolute reduction in HCC-related mortality through Year 8). The target sample size of 4,700 was required to detect this hypothesised effect size with 88% power and a two-sided type I error of 5%, also accounting for a crossover rate from aMRI to ultrasound of 10% and losses to follow-up of 5%. We chose a higher power of 88% rather than the more conventional 80%, given the uncertainties in some of our estimates of expected cumulative HCC-related mortality and for added protection against the possibility of higher rates of loss to follow-up or crossover.

Monitoring efficacy and futility

An external, independent Data Monitoring Committee, composed of subject-matter experts in HCC screening and biostatistics, reviews the study semiannually and may recommend continuation or early termination based on safety concerns (due to adverse events) or futility (due to inadequate recruitment). Two interim analyses of the primary outcome (HCC-related mortality) are planned when the study reaches the 50% (approximately 115 events) and 75% (approximately 172 events) information level. If the event rate of the primary endpoint is found to be substantially lower than expected, then consideration will be given to either stopping the trial for futility or planning for study extension, as appropriate.

Discussion

The PREMIUM study is designed to determine whether screening with aMRI+AFP every 6 months is superior to standard of care ultrasound+AFP in reducing HCC-related mortality in individuals with cirrhosis of any aetiology who are at high risk of HCC. We hypothesise that screening by aMRI+AFP will result in HCC diagnosis at an earlier stage, leading to administration of HCC treatment that is more likely to be curative or to prolong survival, thereby reducing HCC-related mortality. aMRI can be executed in under 15 min at any MRI facility without requiring additional resources, equipment, or expertise. Therefore, if PREMIUM demonstrates a reduction in HCC-related mortality, aMRI+AFP may be recommended as the new standard of care for HCC screening in patients with cirrhosis.

Several HCC surveillance trials are under way that may inform how to improve and optimise HCC screening. FASTRAK is a French, multicentre RCT of ultrasound plus non-contrast aMRI vs. ultrasound alone every 6 months for 3 years in high-risk patients (incidence >3 per 100 patient-years) with the outcome being cost/quality-adjusted life year and cost/patient detected with a BCLC stage 0 HCC.⁴¹ The FAST-MRI study is an National Cancer Institute-sponsored, one-time screening study comparing the performance and cost-effectiveness of several types of aMRI with ultrasound for HCC detection, with and without AFP.⁴² The TRACER study is a longitudinal, phase IV RCT sponsored by the National Cancer Institute comparing GALAD every 6 months vs. usual care for up to 5.5 years in patients with cirrhosis or HBV, designed to demonstrate a reduction in late-stage HCC.^43,44 GALAD is a biomarker panel based on a Fujifilm assay that combines three protein-based biomarkers (AFP, AFP-L3% and des-gamma-carboxy prothrombin) sex, and age in a linear logistic regression formula. The CLiMB study⁴⁵ is a recently completed cross-sectional screening study in patients with cirrhosis that included concurrent HelioLiver testing, ultrasound, AFP, and MRI. It was designed to demonstrate the superiority of HelioLiver over ultrasound for sensitivity, and its non-inferiority for specificity, in detecting HCC.⁴⁶ HelioLiver, developed by Helio Genomics, is a multiparametric algorithm that incorporates information from many circulating tumour DNA methylation markers, protein-based biomarkers (AFP, AFP-L3% and des-gamma-carboxy prothrombin) sex and age. The ongoing ALTUS study is also a cross-sectional screening study with concomitant Oncoguard, ultrasound and MRI/CT scans in patients with cirrhosis or HBV, designed to determine sensitivity and specificity for HCC detection.⁴⁷ Oncoguard, developed by Exact Sciences, is a multiparametric algorithm that incorporates information from three circulating tumour DNA methylation markers (HOXA1, TSPYL5 and B3GALT6), AFP and sex.

PREMIUM is to our knowledge the only study that is designed and powered to detect a reduction in HCC-related mortality, which is the appropriate outcome to support a cancer screening recommendation.⁴⁸ Earlier detection or receipt of potentially curative treatments should not be used by themselves to support cancer screening recommendations because they do not necessarily result in a reduction in cancer mortality, and, in isolation, may even increase costs and harms, e.g. due to overdiagnosis and the detection of small or slow-growing tumours that may never cause symptoms or death.

PREMIUM does not have a “usual care” arm in which participants undergo screening only if it is ordered by their clinical providers. Having a usual care arm creates an imbalance between the two arms in the propensity to undergo screening and ultimately in screening rates. Instead, participants in each arm of PREMIUM undergo per-protocol screening every 6 months ± 30 days for up to 8 years according to their randomised arm (i.e. aMRI+AFP vs. ultrasound+AFP), and failure to do so represents a protocol deviation. PREMIUM is therefore designed to determine differences in efficacy between the two screening protocols, i.e. to answer the question “if a patient with cirrhosis undergoes screening by aMRI+AFP are they less likely to die of liver cancer than if they undergo screening by ultrasound+AFP?”

The PREMIUM DCE aMRI protocol has sensitivity and specificity for HCC almost identical to that of complete MRI, which is the gold-standard diagnostic test. It is unlikely that any other screening modality (e.g. a new biomarker) can exceed the effectiveness of a DCE aMRI-based protocol in reducing HCC-related mortality. This is because a “true-positive” biomarker screening test cannot result in treatment unless the lesion is also visible on MRI, and without treatment, there cannot be a reduction in HCC-related mortality. Therefore, the PREMIUM DCE aMRI-based screening protocol will likely determine the upper limit of effectiveness in HCC-related mortality reduction that can be achieved with any HCC screening programme, given currently available treatment modalities and assuming that MRI remains the gold-standard diagnostic test.

Patients with cirrhosis have a high risk of liver cancer, ranging from ∼1.0 to >5 per 100 patient-years.^49,50 PREMIUM applies specific inclusion criteria (FIB-4 >3.25, presence of clinically significant portal hypertension, or active HCV) to select patients at the higher end of this risk spectrum – i.e. those with a risk greater than 2.5 per 100 patient-years – because a high HCC incidence is necessary to ensure sufficient power to detect the primary outcome. Thus, the results of PREMIUM are most directly applicable to a population that matches these criteria. However, it is reasonable to assume that a beneficial effect of screening by aMRI on HCC-related mortality (or any of the secondary outcomes) likely generalises to the broader population with cirrhosis because the study’s “high-risk” inclusion criteria have little effect on the performance characteristics of the screening tests (i.e. whether a person’s FIB-4 is < or >3.25 has little effect on the sensitivity or specificity of screening ultrasound or aMRI, although positive and negative predictive values would be greatly affected) or the potential to undergo curative treatments. If PREMIUM demonstrates that screening with aMRI+AFP reduces HCC-related mortality, decisions about which patients with cirrhosis should undergo screening with aMRI vs. ultrasound (or potentially a biomarker panel) will likely depend on cost-effectiveness estimates, which in turn depend critically on estimates of absolute risk of HCC and HCC-related death. Although studies suggest that aMRI becomes cost-effective when annual HCC risk exceeds 1.8-3.3%,^21,22 such calculations would need to be repeated when the results of our study are known and would also critically depend on the cost of aMRI and advances in HCC treatment. Additionally, receipt of therapies that may improve underlying liver disease aetiology and thereby potentially reduce HCC risk could affect study power and were not factored into this study design. However, randomisation ensures that any potential risk reduction would be equally distributed to each study arm. PREMIUM is collecting data on all imaging, diagnostic tests and treatments related to screening in each arm to facilitate such future cost-effectiveness analyses, as well as analyses of harms caused by screening.

The PREMIUM biorepository in the DCE aMRI arm will facilitate prospective, longitudinal, phase IV biomarker validation studies, using the concomitant DCE aMRI as the gold-standard diagnostic test for establishing the ground truth (presence/absence of HCC) at each screening time point.³⁸ The longitudinal image repository will enable future studies of radiomic, deep learning and other markers of HCC development, early detection, diagnosis, and response to treatment.

Future trends and protocol deviations

PREMIUM may undergo protocol amendments as necessitated by different factors (e.g. changes in clinical practice and recruitment, budgetary, and/or regulatory changes) or because of interim monitoring analyses and results. PREMIUM image and biorepositories may change due to advances in image acquisition or novel biomarker collection protocols. Cost-effectiveness will be evaluated formally in future studies, assuming effectiveness is demonstrated by the study. We are capturing all imaging and laboratory tests in each arm, as well as all treatments of any identified lesions to enable such future cost-effectiveness analyses.

Abbreviations

ACR, American College of Radiology; CSP, Cooperative Studies Program; CSPH, clinically significant portal hypertension; CTP, Child-Turcotte-Pugh; DCE, dynamic contrast-enhanced; EAC, Endpoint Attribution Committee; FIB-4, fibrosis-4; GBCA, gadolinium-based contrast agent; LI-RADS, Liver Imaging Reporting and Data System; LSI, local site investigators; MELD, model for end-stage liver disease; MLTB, multidisciplinary liver tumour board; RCT, randomised controlled trials; VA, Veterans Affairs.

Financial support

This trial, Cooperative Studies Program #2023, Preventing Liver Cancer Mortality through Imaging with Ultrasound versus MRI (PREMIUM), is funded by the Cooperative Studies Program of the VHA Office of Research and Development.

Authors’ contributions

Writing of original draft: George Ioannou, Tamar Taddei. Critical revision for important intellectual content: all authors. Final approval of the article: all authors. Conception and design: all authors.

Disclaimer

The sponsor approved the study and its amendments but was not involved in any aspect of the study. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the US government.

Conflicts of interest

George N. Ioannou: No conflicts of interest. Tamar H. Taddei: No conflicts of interest. Beata M. Planeta: No conflicts of interest. Grant D. Huang: No conflicts of interest. Noel S. Weiss: No conflicts of interest. Timothy R. Morgan: No conflicts of interest. Jason A. Dominitz: No conflicts of interest. Ghassan K. Abou-Alfa: Conflicts of interest. Research support from Abbvie, Agenus, Arcus, Astra Zeneca, Atara, Beigene, BioNtech, BMS, Coherus, Digestive Care, Elicio, Genentech/Roche, Helsinn, J-Pharma, Parker Institute, Pertyze, Yiviva, and consulting support from Abbvie, Ability Pharma, Agenus, Alligator Biosciences, Astellas, Arcus, Astra Zeneca, Autem, Berry Genomics, BioNtech, BMS, Boehringer Ingelheim, Fibrogen, Genentech/Roche, Ipsen, J-Pharma, Merck, Merus, Moma Therapeutics, Neogene, Novartis, Regeneron, Revolution Medicines, Servier, Syros, Tango, Tempus, Vector, Yiviva. Mustafa R. Bashir: Conflicts of interest (specifically relevant to this study): Grant to institution as PI: Bayer Healthcare; Conflicts of interest (general, not specifically relevant to this study): Consulting: United Therapeutics, Rectify Pharma, ICON, Grants to institution as PI: Siemens Healthineers, Madrigal Pharmaceuticals, Corcept Therapeutics, NGM Biopharmaceuticals. Michael V. Beheshti: No conflicts of interest. Amit G. Singal: Conflicts of interest Served as consultant or on advisory boards for Genentech, AstraZeneca, Eisai, Exelixis, Bayer, Merck, Elevar, Boston Scientific, Sirtex, FujiFilm Medical Sciences, Exact Sciences, Helio Genomics, Roche, ImCare, Curve Biosciences, Universal Dx, Glycotest, Mursla, DELFI, and Abbott. Cynthia A. Moylan: Conflicts of interest research grants to institution: Madrigal, GSK, Exact Sciences: consulting: NovoNordisk, Sirtex. Robin J. Boland: No conflicts of interest. Lynn F. Buchwalder: No conflicts of interest. Rajni L. Mehta: No conflicts of interest. Kimberly S. Hoisington: No conflicts of interest. Nhan V. Do: No conflicts of interest. Shari S. Rogal: Conflicts of interest: Grants to institution from VA ORD, PCORI, and CDC. David E. Kaplan: Conflict of Interest Research funding to institution from AstraZeneca, Bausch, Bluejay Therapeutics, Exact Sciences, Gilead. Jihane N. Benhammou: No conflicts of interest. Grace L. Su: No conflicts of interest. Linda M. McDonald: No conflicts of interest. Genta Dani: No conflicts of interest. Dell P. Dunn: No conflicts of interest. Stephanie T. Chang: No conflicts of interest. Ifeyinwa Y. Onyiuke: No conflicts of interest. Ashu Sharma: No conflicts of interest. Tassos C. Kyriakides: No conflicts of interest.

Please refer to the accompanying ICMJE disclosure forms for further details.

Supplementary data

The following are the Supplementary data to this article: