NCT07444905

Prospective Evaluation of a Locked Risk-Stratified Surveillance Strategy for Extrahepatic Metastasis in Hepatocellular Carcinoma

• Status: Active Not Recruiting
• Phase: —
• Study type: Observational
• Enrollment: 1,523 (actual)
• Sponsor: Tongji Hospital · Academic / Other
• Sex: All
• Age: 18 Years
• Healthy volunteers: Not accepted

Summary

This is a multicenter prospective observational cohort study in adults with hepatocellular carcinoma (HCC) without baseline extrahepatic metastasis who are receiving routine clinical management and follow-up at participating centers. The study evaluates the real-world implementation and clinical utility of a locked machine learning-guided risk stratification strategy for predicting lung and bone metastasis and supporting risk-stratified surveillance. The locked strategy uses prespecified 12-month risk thresholds and linked care pathways and is implemented without retraining or threshold revision. No study-mandated treatment or surveillance assignment is performed; clinical care remains at physician discretion according to local practice. The study assesses model transportability and calibration, implementation outcomes (including completion of prespecified actions, timeliness of action, and treatment activation), clinically actionable detection outcomes, patient-important outcomes, and longer-term survival. Comparative analyses use centre- and calendar-epoch-aligned usual-care episodes as the primary observational comparator.

Detailed description

Background and Rationale: Hepatocellular carcinoma (HCC) carries a substantial risk of postoperative or post-treatment recurrence and extrahepatic metastasis, which may affect prognosis and management opportunities. Early identification of patients at increased risk of extrahepatic metastasis may support more appropriate surveillance intensity, earlier multidisciplinary review, and more timely treatment planning. Study Objectives: This study prospectively evaluates the real-world transportability, calibration, implementation characteristics, and clinical utility of a previously developed and locked machine learning-guided risk stratification strategy for extrahepatic metastasis in HCC, with a focus on lung and bone metastasis. Study Design: This is a multicenter prospective observational cohort study conducted in routine clinical practice. Participants are enrolled and followed without randomization or study-mandated treatment allocation. The locked strategy includes endpoint-specific lung and bone metastasis risk modules, fixed 12-month risk thresholds, and linked risk-stratified surveillance recommendations. The prospective program includes staged real-world implementation, including an independent new-centre replication phase using the unchanged locked package. Study Population: Adults with HCC without baseline extrahepatic metastasis at the first eligible risk assessment who are receiving routine care at participating centers. Study Procedures and Data Collection: Baseline demographics, liver disease etiology, tumor burden and stage, liver reserve and performance status, biomarkers, imaging, treatment intent, follow-up imaging, occurrence and timing of extrahepatic metastasis, downstream management, implementation metrics, and clinical outcomes are collected prospectively from routine care records and study documentation. Outcomes and Analytic Framework: The study evaluates incident extrahepatic metastasis, overall survival, and prespecified implementation and clinical utility outcomes, including completion of prespecified action, time to action, treatment activation, clinically actionable detection, symptom-driven detection, acute deterioration, and resource-related outcomes where available. Comparative implementation analyses use aligned usual-care episodes from the same centers and prespecified calendar epochs as the primary observational comparator. Any strategy-emulation or comparative-effectiveness analyses will be reported as observational estimates and will not be interpreted as randomized treatment effects. Registry and timing statement (transparency \& status): This study was registered on ClinicalTrials.gov (NCT07444905) in February 2026, after study initiation. At the time of registration participant enrollment had ended and long-term follow-up for secondary endpoints remained ongoing. Primary outcome data collection is anticipated to be completed by 2026-12-30 and full study completion is anticipated by 2026-12-31. Overall Recruitment Status on the registry is recorded as Active, not recruiting. The timing of registration reflects administrative and multi-centre onboarding/IT coordination delays. Background and rationale: Hepatocellular carcinoma (HCC) carries a substantial risk of postoperative or post-treatment recurrence and extrahepatic metastasis, which may affect prognosis and management opportunities. Early identification of patients at increased risk of extrahepatic metastasis may support more appropriate surveillance intensity, earlier multidisciplinary review, and more timely treatment planning. Study objectives and design: This multicentre prospective observational cohort study evaluates the real-world transportability, calibration, implementation characteristics, and clinical utility of a previously developed and locked machine-learning guided risk-stratification strategy for extrahepatic metastasis in HCC (lung and bone modules). Participants are enrolled and followed in routine care without randomization or study-mandated treatment allocation. The locked strategy comprises endpoint-specific risk modules, fixed 12-month risk thresholds, and linked risk-stratified surveillance recommendations. The prospective programme includes staged real-world implementation and a prespecified independent new-centre replication phase that reused the unchanged locked package. Governance and freeze evidence: Crucially, key governance objects (protocol v1.0, prespecified thresholds, SOP, statistical analysis plan, and code release tagged v1.0) were archived and time-stamped prior to the first live implementation. To enable independent verification, the following archived objects and logs are provided in the Supplementary Materials: NCT07444905\_Study\_Protocol;NCT07444905\_Model\_Lock\_and\_Version\_Governance\_Note;NCT07444905\_Statistical\_Analysis\_Plan;NCT07444905\_Endpoint\_Definitions\_and\_Adjudication\_Charter. Wave-2 replication and auditability: Wave 2 was a prespecified independent new-centre prospective replication performed in different health systems and with different governance and deployment teams; it reused the unchanged locked package (v1.0) without retraining, recalibration, threshold revision, or SOP remapping. Supplementary materials include activation logs and time-stamped governance evidence documenting unchanged package use, and an independent recomputation/audit summary demonstrating that the analysis plan and codebase were fixed prior to implementation. Outcomes and analytic framework: The study prospectively collects baseline demographics, liver disease etiology, tumor burden and stage, liver reserve and performance status, biomarkers, imaging, treatment intent, follow-up imaging, occurrence and timing of extrahepatic metastasis, downstream management, implementation metrics, and clinical outcomes. Primary and secondary outcomes, implementation metrics, and the prespecified analytic approach are described in the protocol and SAP (see Supplementary Materials).

Conditions

• HCC - Hepatocellular Carcinoma
• Machine Learning
• Evaluation

Interventions

Type	Name	Description
OTHER	Machine Learning-Based Risk Assessment	Point-of-care use of a locked machine learning-guided risk assessment tool with fixed thresholds and linked risk-stratified surveillance recommendations; no study-mandated treatment assignment is performed

Timeline

• Start date: 2022-02-01
• Primary completion: 2026-12-30
• Completion: 2026-12-31
• First posted: 2026-03-03
• Last updated: 2026-04-13

Locations

1 site across 1 country: China