Machine Learning Predictive Model to Guide Treatment Allocation for Recurrent Hepatocellular Carcinoma After Surgery

Area of Science:

• Hepatobiliary Surgery
• Oncology
• Machine Learning in Medicine

Background:

• Selecting optimal retreatment for recurrent hepatocellular carcinoma (HCC) remains challenging due to a lack of clear guidelines.
• Recurrent HCC significantly impacts patient survival and necessitates personalized treatment strategies.

Purpose of the Study:

• To develop a machine learning (ML) predictive model for survival after HCC recurrence.
• To utilize the ML model to allocate patients to their most effective potential treatment, optimizing outcomes.

Main Methods:

• Retrospective analysis of real-world data from an Italian HCC registry (2008-2019) and external validation cohorts (Italian and Japanese).
• Inclusion of patients with recurrent HCC post-initial surgery, profiling them based on clinical factors and treatment received (reoperative hepatectomy/thermoablation, chemoembolization, sorafenib).
• Development and validation of an ML model to predict survival and identify treatment effect modifiers, fitted to individual patient profiles for treatment allocation.

Main Results:

• The predictive model demonstrated good performance with an AUC of 78.5% at 5 years post-recurrence.
• The model suggested that 87.2% of patients would benefit most from reoperative hepatectomy/thermoablation, 5.2% from sorafenib, and 7.6% from chemoembolization.
• Key predictors for survival and treatment allocation included age, cirrhosis, nodule characteristics (number, size, location), extrahepatic spread, and time to recurrence.

Conclusions:

• The developed algorithm aids in patient-tailored allocation for recurrent HCC treatment based on individual characteristics.
• This approach provides a hierarchical framework for selecting the best potential treatment, potentially improving survival outcomes in recurrent HCC.