In an invited commentary in JAMA Network Open, Dr Varvara A. Kirchner and Dr Timothy L. Pruett discuss the potential of machine learning (ML) to optimise treatment decisions for patients with hepatocellular carcinoma (HCC), particularly when balancing liver transplantation (LT) and surgical resection (SR).
Liver transplantation offers superior outcomes compared with surgical resection for patients with HCC, especially those with underlying synthetic dysfunction, cirrhosis, and portal hypertension. However, limited donor availability, unpredictable waiting times, and long-term morbidity from lifelong immunosuppression remain significant challenges.
Commentary Highlights
Kirchner and Pruett review a study by Kim et al., which developed and validated an ML-based decision model to improve patient selection for LT versus SR. The derivation cohort included 3,915 patients treated between 2008 and 2018 from the Korea Central Cancer Registry, and the external validation cohort included 614 patients from Seoul St Mary’s Hospital, treated between 2009 and 2020.
LISTEN TO THE LATEST EPISODE OF SUPPORTIVE CARE MATTERS WITH PROF. BOGDA KOZCWARA AM 
The ML algorithms incorporated demographics, clinical factors, and tumour characteristics to predict 3-year overall survival (OS). Patients were stratified into high- and low-risk groups for each treatment, allowing identification of LT-favourable and LT-non-favourable subgroups. Counterfactual analysis compared ML-guided treatment with actual clinical decisions.
Key predictors for LT included tumour number, platelet count, protein induced by vitamin K absence-II (PIVKA-II), creatinine, alpha-fetoprotein (AFP), and total bilirubin, while SR outcomes were influenced by maximum tumour size, international normalised ratio (INR), AFP, albumin, PIVKA-II, and sodium levels.
Kirchner and Pruett note that in the derivation cohort, ML recommendations would have suggested SR for 221 of 296 patients (74.7%) originally treated with LT, and LT for 701 of 3,619 patients (19.4%) originally treated with SR. ML-guided predictions were associated with improved OS (hazard ratio 0.46 [95% CI, 0.42–0.50]; P < .001), and external validation confirmed model performance (AUROC 0.75 for LT, 0.80 for SR).
Clinical Insights and Limitations
The commentary emphasises the promise of ML to guide complex transplant oncology decisions, particularly in the context of limited donor resources. However, Kirchner and Pruett caution that the study has limitations: it did not account for prior liver-directed therapies, tumour resectability, patient comorbidities, or detailed tumour biology.
They also highlight that both cohorts reflect the South Korean population, where hepatitis B predominates and living donor transplantation accounts for 70% of LTs, potentially limiting applicability in regions with differing HCC aetiologies, such as metabolic dysfunction–associated steatohepatitis (MASH), and constrained donor availability.
Furthermore, the ML model focused on 3-year OS without assessing long-term HCC-free survival or post-transplant survival beyond three years. Given that over 60% of adult liver recipients in Korea survive more than 20 years post-LT, future ML-guided approaches should integrate long-term outcomes into treatment planning.
Future Perspectives
Kirchner and Pruett conclude that while ML holds promise for personalised decision-making in transplant oncology, predictive models should complement, not replace, clinical judgement. Incorporating additional variables such as tumour location, surgical resectability, comorbidities, and emerging biomarkers like circulating tumour DNA may further enhance treatment recommendations.
They emphasise that even with sophisticated ML guidance, treatment decisions must always respect patient autonomy and informed consent.
Reference: KimHU, HanJW, SungPS ,et al. Machine learning–based selection of resection vs transplant and survival in hepatocellular carcinoma. JAMA NetwOpen. 2025;8(9):e2532353.doi:10.1001/jamanetworkopen.2025.32353. Access online here.
Commentary: Varvara A. Kirchner, MD; Timothy L.Pruett, MD. Can AI Guide the Decision to Transplant or Resect for Hepatocellular Carcinoma? JAMA Network Open. September17,2025. Access online here.
