Adjuvant transarterial chemoembolization (TACE) is widely adopted in China for resectable hepatocellular carcinoma (HCC), yet its efficacy remains inconsistent. We aimed to identify factors influencing individual patient benefit using causal machine learning. To this end, we retrospectively collected HCC patients with high risk factors for tumor recurrence from four centers of China, divided into the discovery cohort and the validation cohort . The primary endpoint was disease-free survival (DFS). The primary endpoint was overall survival (OS).Individual treatment effects (ITEs) were estimated within a causal machine learning framework. An ITE < 0 was considered recommendation for adjuvant TACE , while ITE ≥ 0 indicated active surveillance. The model would be validated in the validation cohort. The contribution of each variable to ITE was assessed using the Shapley Additive Explanations (SHAP). An online calculator would be developed for future use by public.

Inclusion Criteria:

• (1) pathologic diagnosis of HCC; (2) received potentially curative hepatectomy with negative cut margins confirmed by pathology; (3) high risk of recurrence, meeting any of the following criteria: largest tumor >5 cm, multiple lesions, poor histological differentiation (Edmondson-Steiner grade III or IV), presence of microvascular invasion or satellite nodules, (4) received adjuvant TACE or active surveillance as postoperative treatment; (5) at least one year of follow-up.

Exclusion Criteria:

• (1) history of other malignancies; (2) received preoperative anti-tumor treatment; (3) presence of vascular invasion or extrahepatic metastasis; (4) tumor recurred within 4 weeks postoperatively; (5) residual tumor was detected by digital subtraction angiography (DSA); (6) received other types of adjuvant therapy; (7) cases with missing data exceeding 10%.

HCC patients with high-risk factors of tumor relapse, who received adjuvant TACE or active postoperative survelliance as treatment.

For patients' privacy