Construction of a predictive model for stress injury risk in neurocritically ill patients using machine learning algorithms
10.3760/cma.j.cn341190-20240913-01177
- VernacularTitle:基于机器学习算法的神经重症患者压力性损伤风险预测模型的构建
- Author:
Xiaoxia GAO
1
;
Mingya YAO
1
;
Shishi CHEN
1
;
Kaili YE
1
;
Xiaoqing CHEN
1
Author Information
1. 温州医科大学附属第一医院神经外科，温州 325000
- Publication Type:Journal Article
- Keywords:
Neurosurgery;
Intensive care units;
Postoperative complications;
Pressure ulcer;
Machine learning;
Logistic models;
Decision trees;
Neural networks (computer
- From:
Chinese Journal of Primary Medicine and Pharmacy
2025;32(6):835-840
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct logistic regression, decision tree, and neural network models to predict pressure injury in neurocritically ill patients using machine learning algorithms, and compare the predictive performance of the three models.Methods:The clinical data of 341 neurocritically ill patients who received treatment in the Department of Neurosurgery at The First Affiliated Hospital of Wenzhou Medical University from May 2020 to February 2023 were collected retrospectively. The patients were randomly divided into a training set and a testing set in a 7:3 ratio. Univariate and multivariate analyses were conducted based on the clinical data from the training set. According to the results of the multivariate analysis, logistic regression, decision tree, and neural network models were constructed. The predictive performance of the three models was validated and compared using receiver operating characteristic curve analysis.Results:Among the 341 patients, 35 developed pressure injury (a total of 40 occurrences), with an incidence rate of 10.26%. Multivariate analysis indicated that incontinence ( OR = 47.32, 95% CI: 1.360-1 647.700), decreased albumin levels ( OR = 0.56, 95% CI: 0.360-0.870), increased sensory ability ( OR = 0.00, 95% CI: 0.000-0.190), and increased mobility ( OR = 0.03, 95% CI: 0.000-0.390) were independent risk factors for pressure injury in neurocritically ill patients (all P < 0.05). Based on these independent risk factors, logistic regression, decision tree, and neural network models were constructed. Receiver operating characteristic curve analysis revealed that the area under the curve for the three models was 0.987 (95% CI: 0.941-0.999), 0.945 (95% CI: 0.881-0.980), and 0.908 (95% CI: 0.834-0.956), respectively. These results suggest that all three models exhibited high predictive performance for pressure injury in neurocritically ill patients, with the logistic regression model showing a significantly greater area under the curve than the neural network model. Conclusions:The occurrence of pressure injury in neurocritically ill patients is closely related to incontinence, albumin levels, sensory ability, and mobility. Constructing predictive models using machine learning algorithms can provide valuable insights for the early prevention and management of pressure injury in neurocritically ill patients.
