Development and Validation of a Machine Learning-Based Individualized Model to Predict TKI Benefit in Postoperative
Meng Li1,2, Yumin Jiang1, Lin Gong2
1Department of Hepatobiliary and Pancreatic Surgery, Affiliated Hospital of Qingdao University, Qingdao, Shandong, People's Republic of China.
Background:
Postoperative recurrence is a major contributor to poor prognosis in hepatocellular carcinoma (HCC) after curative resection. Tyrosine kinase inhibitors (TKIs) are commonly used for recurrent HCC, yet substantial interpatient heterogeneity limits their universal benefit, and reliable tools to individualize post-recurrence therapy are lacking.
Methods:
We retrospectively analyzed 454 patients with recurrent HCC following curative resection. Overlap weighting based on propensity scores was applied to balance baseline characteristics between TKI-treated and non-TKI-treated groups and to assess the survival benefit of TKI therapy. Machine learning-based survival models were independently developed for each treatment cohort using multiple algorithms and internally validated. A counterfactual inference framework was implemented to estimate individualized survival outcomes under alternative treatment strategies and to identify the optimal therapy for each patient.
Results:
After overlap weighting, TKI therapy was associated with improved survival compared with non-TKI treatment, including longer recurrence-free survival (median, 24 vs. 12 months; HR = 0.355, P < 0.001) and overall survival (median, 36 vs. 18 months; HR = 0.486, P = 0.001). In the non-TKI cohort, an eight-feature survival support vector machine (Surv-SVM) model achieved strong predictive performance (C-index: 0.796 in training, 0.766 in validation). In the TKI cohort, a four-feature random survival forest (RSF) model showed the highest discrimination (C-index: 0.856 and 0.838). Both models demonstrated good calibration and stable time-dependent performance. Counterfactual analysis revealed potential treatment mismatch in 23.4% of patients, including 20.7% who received non-TKI therapy but were predicted to benefit more from TKI treatment.
Conclusion:
TKI therapy provides significant survival benefit in recurrent HCC after curative resection. Our machine learning-based, counterfactual framework enables individualized estimation of treatment benefit, supporting personalized post-recurrence therapy and facilitating precision management in this high-risk population.
