Machine learning prediction of complications after surgery in cirrhotic patients: Integrating nutritional risk with liver function

Abstract

Liver disease is among the most common medical conditions in the world, leading to high rates of morbidity and death. Early identification of liver diseases enables prompt care, which may prevent many conditions from progressing to more serious stages, like cirrhosis or liver cancer. Traditionally, models often overlook the predictive importance of nutritional status and focus primarily on liver-specific measures. This research aims to propose a machine learning (ML) approach to predict surgical complications in patients with liver cirrhosis by integrating existing liver function metrics with a comprehensive nutritional risk assessment. In this study, we used the MIMIC-IV dataset, which consists of patient medical records. We utilized a 504-patient cohort with features such as age, gender, bilirubin, albumin, INR (International Normalized Ratio), Child-Pugh score, MELD score (Model for End-Stage Liver Disease), NRS (Nutritional Risk Screening), complication, and composite risk. Mortality was used as the primary outcome variable for prediction. ML models were utilized for training, along with correlation analysis and Explainable Artificial Intelligence (XAI) approaches like SHAP and LIME to interpret the model. The stackable ensemble model was employed to enhance the model's performance and robustness. The proposed model achieved an accuracy of 94% and an AUC of 0.97. This methodology emphasizes the critical importance of nutritional assessment in surgical risk classification. It offers a clinically relevant framework to guide preparatory strategies, promote collaborative decision-making, and improve perioperative care pathways for this vulnerable population.