Machine learning to predict the decision to perform surgery in hepatic echinococcosis.

Background: Cystic echinococcosis (CE) is a significant public health issue, primarily affecting the liver. While several management strategies exist, there is a lack of predictive tools to guide surgical decisions for hepatic CE. This study aimed to develop predictive models to support surgical decision-making in hepatic CE, enhancing the precision of patient allocation to surgical or non-surgical management pathways.

Methods: This retrospective analysis included 406 hepatic CE patients treated at our center (2009-2021). Clinical, imaging, and treatment data were used to develop a Cox regression and a decision tree model to identify factors influencing surgical intervention, with model performance validated using K-fold cross-validation, train/test split, bootstrapping.

Results: Imaging findings and symptomatology emerged as the most significant predictors. The Cox model demonstrated a concordance index of 0.94 and an AUC of 0.96, while the decision tree model identified imaging, cyst stage, and symptoms as critical factors, achieving strong performance across validation techniques (mean AUC 0.950; 95% CI: [0.889, 0.978]).

Conclusion: This study presents validated predictive models for assessing surgical risk in hepatic CE. Integrating these models into clinical practice offers a dynamic tool that surpasses static guidelines, optimizing patient allocation to surgical or non-surgical pathways and potentially improving outcomes.