Predicting complications after hepatectomy for hepatocellular carcinoma: a narrative review focused on liver failure

Hepatectomy is a potentially curative treatment for selected patients with hepatocellular carcinoma (HCC), but postoperative complications remain an important source of morbidity and mortality. This narrative review focuses on prediction models for post-hepatectomy liver failure (PHLF), with additional discussion of selected non-PHLF complications. It emphasizes validation design, transportability, calibration, clinical actionability, and the distinction between evidence from HCC cohorts and cohorts with mixed diagnoses. PHLF has the most mature prediction literature because its risk is closely related to measurable liver reserve, portal hypertension, future liver remnant volume and function, and resection extent. Clinical and serum scores are accessible but generally provide moderate discrimination. Volumetric, functional, and imaging assessments add complementary information, whereas machine learning and radiomics models can capture nonlinear and high-dimensional patterns but often rely on small cohorts, random internal splits, and incomplete calibration reporting. Evidence for bile leakage includes validated clinical scores and a large externally validated machine learning model. Models for postoperative ascites, intra-abdominal infection or abscess, and surgical site infection remain sparse and are often perioperative rather than purely preoperative. Across endpoints, high apparent discrimination has not established clinical benefit. Future studies should prioritize standardized endpoints, adequate effective sample size, independent geographic external validation, calibration, net benefit assessment, and risk updating at successive stages. Current models should be used as decision support tools rather than substitutes for multidisciplinary clinical judgment.