Machine learning prediction of 30-day all-cause mortality risk factors in HCC rupture
Shixiong Shi1, Canbin Xie1, Lin Long1
1The Interventional Vascular Surgery, Hunan Provincial People's Hospital(Hunan Normal University First Hospital), Changsha, Hunan, China.
Background:
Hepatocellular carcinoma (HCC) rupture is a life-threatening complication with high acute-phase mortality, accounting for 25%-75% of early deaths in HCC patients. While previous studies have identified clinical factors associated with mortality, few focused on the Chinese population or utilized machine learning to capture nonlinear relationships. This study aimed to develop a machine learning model to predict 30-day all-cause mortality in patients with ruptured HCC and identify key predictive factors.
Methods:
A retrospective cohort of 156 patients with first-diagnosed ruptured HCC (2012-2021) from Hunan Provincial People's Hospital was analyzed. Clinical variables included demographics, laboratory parameters, tumor characteristics, and treatment modalities. Seven machine learning models were constructed to predict 30-day mortality, with the decision tree model selected for its highest sensitivity. SHAP (Shapley Additive exPlanations) analysis was used to evaluate feature importance, and restricted cubic splines (RCS) explored nonlinear relationships.
Results:
The decision tree model exhibited optimal performance (sensitivity=87.5.00%, AUC = 0.7901) for 30-day mortality prediction. SHAP analysis identified TBIL (Mean |SHAP|=0.1662) and INR (0.0414) as the top predictors. Nonlinear threshold effects were observed: TBIL>40 μmol/L (OR = 1.45, 95%CI:1.25-1.68) and INR>2.5 (OR = 4.95, 95%CI:3.36-10.11) significantly increased mortality risk. Combined detection of TBIL and INR improved predictive performance (AUC = 0.87, 95%CI:0.79-0.95).
Conclusion:
The decision tree model effectively predicts 30-day mortality in ruptured HCC patients, with TBIL and INR as critical nonlinear predictors. Their combination provides a robust tool for rapid clinical risk stratification, aiding in emergency management.