[Construction of a machine learning prognostic prediction model based on psoas muscle index for patients with decompensated liver cirrhosis].

Objective: To explore the effect of psoas muscle index (PMI) and construct a machine learning model to validate the 180-day prognosis in patients with decompensated liver cirrhosis. Methods: Retrospective data were collected from patients with decompensated liver cirrhosis at Henan Provincial People's Hospital from January 2022 to November 2022. The area of the psoas muscle index (PMI) at the level of the third lumbar vertebra was measured and calculated based on the abdominal X-ray computed tomography images stored in the Eastern China Hospital Information System (HIS). Patients were divided into low PMI and normal PMI groups according to the receiver operating characteristic curve. Patients clinical data and complication status were collected.The general conditions of both groups were compared using a t-test, chi-square test, and Mann-Whitney U test. The Kaplan-Meier method was applied for survival analysis. The outcome variable was 180-day mortality, and variables were selected using Cox and LASSO regression. The dataset was divided into training and testing sets in a 7∶3 ratio. Machine learning algorithms were used to build models in the training set, and model performance was validated by the test set. The model for MELD-Na score was compared with the model for End-Stage Liver Disease score. Results: A total of 298 patients with decompensated liver cirrhosis were included.The MELD scores, Child-Pugh classification, and NRS2002 scores, along with the incidence rate of complications such as ascites, hepatic encephalopathy, infections, and gastrointestinal bleeding, were significantly higher in the low PMI than the normal PMI group, with statistically significant differences (P<0.05). The area under a receiver operating characteristic curve for the extreme gradient boosting model was higher than traditional clinical scores (MELD score 0.658, MELD_Na score 0.719) in the machine learning model. Furthermore, the application of SHAP results model indicated that PMI, hemoglobin, NRS2002 score, direct bilirubin, and blood ammonia were important factors in predicting the prognosis of patients with decompensated liver cirrhosis. Conclusion: A low PMI is closely related to poorer survival rates and the development of complication rates in patients with decompensated liver cirrhosis. The machine learning prediction model based on this construction, especially extreme gradient boosting, has favorable predictive performance, which is superior to the traditional clinical scoring system and can provide patients with the most accurate risk assessment and individualized treatment plan.