Machine Learning-Driven Risk Prediction Models for Posthepatectomy Liver Failure: A Narrative Review

<i>Background and Objectives</i>: Posthepatectomy liver failure (PHLF) remains a major cause of morbidity and mortality for patients undergoing major liver resections. Recent research highlights the expanding role of machine learning (ML), a crucial subfield of artificial intelligence (AI), in optimizing risk stratification. The aim of the current study was to review, elaborate on and critically analyze the available literature regarding the use of ML-driven risk prediction models for posthepatectomy liver failure. <i>Materials and Methods</i>: A systematic search was conducted in the PubMed/MEDLINE, Scopus and Web of Science databases. Fifteen studies that trained and validated ML models for prediction of PHLF were further included and analyzed. <i>Results</i>: The available literature supports the value of ML-derived models for PHLF prediction. Perioperative clinical, laboratory and imaging features have been combined in a variety of different algorithms to provided interpretable and accurate models for identifying patients at risk of PHLF. The ML-based algorithms have consistently demonstrated high area under the curve and sensitivity values, surpassing traditionally used risk scores in predictive performance. Limitations include the small sample sizes, heterogeneity in populations included, lack of external validation and a reported poor ability to distinguish between true positive and false positive cases in several studies. <i>Conclusions</i>: Despite the constraints, ML-driven tools, in combination with traditional scoring systems and clinical insight, may enable early and accurate PHLF risk detection, personalized surgical planning and optimization of postoperative outcomes in liver surgery.