Primary liver cancer is a major cause of cancer-related death and hepatocellular carcinoma (HCC) accounts for most cases. Surveillance aims to find disease earlier in people at risk, yet ultrasound, the main tool for routine monitoring, can miss small or subtle lesions. More intensive imaging can detect disease earlier but is demanding for services and should be directed to those most likely to benefit. A large multicentre analysis of adults with viral hepatitis who had ultrasound surveillance developed and validated a practical prediction model that adds selected ultrasound features to routine clinical information. The result, called SELECT, was designed to help identify patients who could benefit from more intensive surveillance while maintaining feasibility in everyday practice.
Building a Practical Risk Tool
SELECT was built using data from HCC-naïve adults with chronic hepatitis B or C who underwent ultrasound surveillance. Patients were drawn from multiple institutions and followed for incident HCC after an index ultrasound. The development process used variables available at or near the time of that ultrasound to reflect information clinicians routinely have when deciding next steps.
Must Read: Higher Mechanical Index Improves Liver Elastography Quality
The model combines demographic, clinical and laboratory factors with two ultrasound-reported features. Age and sex capture baseline demographic risk. Diabetes provides a metabolic context often linked to liver disease progression. Serum albumin and alanine aminotransferase reflect liver function and inflammation, while platelet count acts as a surrogate for portal hypertension and fibrosis burden. From ultrasound reports, the presence of cirrhosis and the presence of multiple cirrhotic nodules are incorporated in a reproducible, binary manner.
An important feature of SELECT is the interaction between platelet count and ultrasound-defined cirrhosis, acknowledging that the impact of cirrhosis on risk is not uniform across different platelet levels. Inter-observer agreement for the ultrasound features was substantial for cirrhosis and moderate for multiple cirrhotic nodules, supporting their use in a real-world tool that relies on standardised reporting rather than specialised measurements.
Validation and Comparative Performance
SELECT was tested internally and then externally across institutions to assess discrimination and calibration. Risk groups defined by two cut-offs separated patients into low, intermediate and high risk with clear gradients of HCC incidence over time. External validation showed stable performance and agreement between predicted and observed outcomes, indicating that the model generalises beyond the development setting.
Performance was compared with widely used risk scores that rely mainly on demographic and laboratory data. Across the external cohort, SELECT achieved higher discrimination than aMAP, THRI, ADRESS-HCC, the Velazquez score and mPAGE-B. Among patients with hepatitis B infection, SELECT also outperformed mPAGE-B. These comparisons indicate that including readily reported ultrasound features that reflect the liver’s structural status adds meaningful information beyond non-imaging models, improving identification of those more likely to develop HCC during surveillance.
Crucially, SELECT balances sensitivity and specificity in a way that avoids classifying most patients as high risk while still identifying a focused subgroup with materially higher risk. Other tools either cast a very wide net, which reduces efficiency and strains services, or set a narrow threshold that risks missing many future cases. By concentrating attention on a smaller group with higher observed incidence, SELECT aligns with the operational realities of surveillance programmes that must prioritise both diagnostic yield and capacity.
Targeting Intensive Surveillance
A prespecified clinical aim was to find a threshold that identifies patients with an annual HCC risk consistent with recommendations for more intensive imaging. SELECT provides a cut-off that meets this target, designating a minority of the surveillance population as eligible for intensified follow-up with alternative modalities. Across development, internal validation and external cohorts, the proportion meeting this threshold remained similar, reinforcing its practicality for use beyond the original institution.
Patients above the threshold experienced substantially higher cumulative incidence over both medium and longer follow-up, supporting the clinical logic of intensifying surveillance for this group. Those below the threshold demonstrated lower long-term incidence, suggesting that routine ultrasound surveillance remains appropriate for the majority. This stratified approach aims to improve early detection where it is most likely to matter while avoiding unnecessary escalation for lower-risk patients.
When measured against other models using sensitivity, specificity and predictive values for mid-term outcomes, SELECT delivered a more balanced profile. Tools such as aMAP and ADRESS-HCC tended to label a large share of patients as eligible, which may dilute the benefit of intensive strategies. Others, including THRI and the Velazquez score, achieved higher specificity but missed a considerable number of eventual cases. mPAGE-B showed similar sensitivity to SELECT but lower specificity overall and in hepatitis B subgroups. These patterns support the role of SELECT as a triage instrument that matches surveillance intensity to risk without overwhelming service capacity.
By combining demographic and laboratory information with simple ultrasound-reported features, SELECT offers a practical method to focus intensive HCC surveillance on those at higher risk within populations with viral hepatitis. Internal and external validation demonstrated consistent discrimination and calibration, and comparisons with established scores showed added value from incorporating imaging findings that reflect liver structure. A clinically actionable threshold identifies a manageable subgroup for alternative imaging while maintaining routine pathways for others. This approach supports earlier detection where it is most likely to improve outcomes and helps healthcare teams direct resources to patients with the greatest need.
Source: European Radiology
Image Credit: iStock
