Development and validation of nomogram and neural network prediction models for stroke-associated pneumonia in patients with acute stroke
10.3760/cma.j.issn.1673-4165.2025.03.003
- VernacularTitle:急性卒中患者卒中相关性肺炎列线图和神经网络预测模型的构建和验证
- Author:
Fengchen GAO
1
;
Haimei SUN
;
Fuqiang ZHOU
;
Weixiang LI
;
Siting HUA
;
Xuejun LONG
;
Ruifei WANG
Author Information
1. 昆明医科大学公共卫生学院，昆明 650500
- Keywords:
Stroke;
Pneumonia;
Nomograms;
Neural networks, computer;
Risk factors
- From:
International Journal of Cerebrovascular Diseases
2025;33(3):173-179
- CountryChina
- Language:Chinese
-
Abstract:
Objectives:To investigate the predictive factors of stroke associated-pneumonia (SAP) in patients with acute stroke, develop nomogram and neural network prediction models and verify their predictive performance.Methods:Patients with acute stroke admitted to the First Affiliated Hospital of Kunming Medical University and Zhenxiong County People's Hospital were included retrospectively. Multivariate logistic regression analysis was used to determine the independent predictive factors of SAP, and develop nomogram and neural network prediction models. Receiver operating characteristic curve (ROC) curves were used to validate and compare the predictive performances. Results:A total of 450 patients with acute stroke were enrolled, including 286 males (63.6%), aged 64.28±13.24 years; 344 patientss (76.4%) had ischemic stroke and 106 (23.6%) had hemorrhagic stroke; 128 patients (28.4%) experienced SAP. According to the random number method, they were divided into a modeling cohort ( n=300) and a validation cohort ( n=150). Multivariate logistic regression analysis in the modeling cohort showed that a higher baseline National Institutes of Health Stroke Scale (NIHSS) score, gastric tube placement, use of proton pump inhibitors, heart failure, and higher neutrophil/lymphocyte ratio (NLR) were the independent predictive factors of SAP. ROC curve analysis showed that the area under the ROC curve of the nomogram model for predicting SAP in the modeling cohort and validation cohort was 0.841 (95% confidence interval [ CI] 0.795-0.880) and 0.863 (95% CI 0.798-0.914), respectively. The sensitivity for predicting SAP were 75.00% and 70.45%, respectively, and the specificity was 81.94% and 92.45%, respectively. The area under the ROC curve of the neural network model for predicting SAP in the modeling cohort and validation cohort was 0.847 (95% CI 0.802-0.866) and 0.862 (95% CI 0.796-0.913), respectively. The sensitivity for predicting SAP were 76.19% and 72.73%, and the specificity was 79.17% and 89.62%, respectively. Conclusions:Higher NIHSS score, gastric tube placement, use of proton pump inhibitors, heart failure, and higher NLR are the independent risk factors for SAP in patients with acute stroke. The nomogram and neural network prediction model developed using the above risk factors have higher predictive value for SAP.
