What question did this study set out to answer?

The research aims to develop a machine learning model that predicts one-year all-cause mortality in patients with previous intracerebral hemorrhage undergoing PCI.

February 8, 2026

An explainable machine learning model for predicting one-year mortality in patients with prior intracerebral hemorrhage undergoing PCI

Key Points

The research aims to develop a machine learning model that predicts one-year all-cause mortality in patients with previous intracerebral hemorrhage undergoing PCI.
Selected a multicenter retrospective cohort from 82 hospitals in China.
Included 1379 patients who received PCI, dividing them into a training set (70%) and testing set (30%).
Applied eight machine learning algorithms, including LightGBM and XGBoost, to create predictive models.
Used the area under the ROC curve to evaluate model performance and SHAP for variable contribution explanation.
The one-year all-cause mortality rate was 3.99% in the study population.
LightGBM exhibited the best predictive performance with a ROC of 0.829 in the training set and 0.831 in the internal validation set.
Key features identified by the SHAP method included D-dimer, hemoglobin, potassium, platelet counts, Killip classification, and neutrophils.

Abstract

Abstract Background and Objective Patients with prior intracerebral hemorrhage(ICH)undergoing percutaneous coronary intervention (PCI) often experienced higher risk of death. Early identification of high-risk patients is important in clinical practice. This study aimed to develop and validate an explainable machine learning (ML) model to predict the one-year all-cause mortality (ACM) in patients with prior ICH after PCI. Methods The data for this study were selected from a multicenter retrospective cohort from 82 hospitals in China, from January 2010 to March 2024. A total of 1379 patients with prior ICH who received PCI successfully were included, with 70% in the training set and 30% in the testing set. The dataset comprised 66 variables covering the demographic, comorbidities and the first laboratory results after hospitalization. Eight ML algorithms including logistic regression, K-Nearest Neighbor, random forest, support vector machine, decision tree, eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and deep neural networks were used to develop predictive models, and the area under the receiver operating characteristic curve (ROC) was used to evaluate the predictive performance. The SHapley Additive exPlanations (SHAP) method was applied to explain the result of the final model by calculating the contribution of each variable to the prediction. Results The incidence of one year ACM was 3.99% in the study population. Among the 8 models, the LightGBM algorithm showed the optimal predictive performance. SHAP method identified D-dimer (DD), hemoglobin (HB), potassium (K), platelet counts (PLT), Killip classification and neutrophils (Neu) as key features. The LightGBM model in training set predicts the one-year ACM in patients with prior ICH undergoing PCI with a ROC of 0.829 (95% CI 0.769–0.890). And in the internal validation set, the LightGBM model achieved an ROC of 0.831 (95% CI 0.722–0.939), with sensitivity of 0.750, specificity of 0.723, accuracy of 0.724, and F1 score of 0.740. Conclusions We developed an interpretable ML model to predict one-year ACM in patients with prior ICH undergoing PCI. The LightGBM model showed considerable potential in predicting prognosis, providing a valuable insight for optimizing patient management.ROC curves SHAP plot

اسأل الذكاء الاصطناعي

Bookmark