脑出血早期死亡风险预测模型的开发与验证OA

Objective To construct and validate an interpretable machine learning model for predicting the risk of death within 30 days in patients with spontaneous intracerebral hemorrhage,and to provide a reliable and practical auxiliary tool for early risk stratification and individualized decision-making.Methods The medical information mart for intensive care,(MIMIC-Ⅳ)database was selected to include patients with spontaneous intracerebral hemorrhage admitted to the ICU.The 30 days all-cause mortality was set as the outcome.In the training set(70%),a support vector machine model was constructed after data standardization and LASSO variable selection.The performance of the model,such as AUC,was evaluated on the validation set(30%),and Shapley additive explanation(SHAP)was conducted.Results A total of 1 373 patients with spon-taneous intracerebral hemorrhage were included.LASSO regression identified 18 variables,among which age and Glasgow coma scale(GCS)score had the greatest impact on the 30 days mortality risk.The age,proportion of diabetes mellitus,white blood cell count,urea,creatinine,blood glucose,international normalized ratio(INR),and serum sodium,potassium,and chloride levels of the patients who survived within 30 days were lower than those of the deceased patients,and the differences were statistically significant(P＜0.05).The weight,proportion of hypertension,GCS,hemoglobin,and platelet count were higher than those of the deceased patients,and the differences were statistically significant(P＜0.05).There were no statistically significant differences in gender,anion gap,and activated partial thromboplastin time(PTT)between the 30 days survivors and the deceased patients(P＞0.05).A support vector machine prediction model was constructed,and the hyperparameters were optimized through 5-fold cross-validation.The final model had an AUC value of 0.854(95%CI:0.814-0.891)on the independent test set.At the same time,SHAP analysis was performed on this model to show the influence of each variable on the model output,improving the interpretability of the model.Conclusion The machine learning model that can pre-dict the 30 day all-cause mortality risk of patients with spontaneous intracerebral hemorrhage has excellent discriminative ability and clinical application potential,and it is worthy of promotion.