Construction of a web-based calculator of the risk of prolonged length of stay in patients with acute ischemic stroke
10.3760/cma.j.cn211501-20240906-02439
- VernacularTitle:急性缺血性卒中患者住院时间延长风险网络计算器的构建
- Author:
Xiaocong RONG
1
;
Meng RONG
;
Yingguo REN
Author Information
1. 南阳市中心医院神经内科特需病区，南阳 473000
- Publication Type:Journal Article
- Keywords:
Stroke;
Prolonged length of stay;
Boruta algorithm;
SHAP;
XGBoost model;
Network calculator
- From:
Chinese Journal of Practical Nursing
2025;41(13):978-986
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct a network calculator based on interpretable machine learning models to predict the risk of prolonged length of stay in patients with acute ischemic stroke (AIS), and to provide a tool for the development of individualized intervention plans for patients.Methods:Adopting a retrospective analysis method. The 537 patients with AIS admitted to the Nanyang Central Hospital from July 2022 to July 2024 were selected. Length of stay exceeding the median length of stay was defined as length of stay prolongation. Length of stay prolongation risk profile variables were screened by Boruta algorithm. The 537 patients were randomly divided into a training set (322 cases) and a test set (215 cases) in a 3:2 ratio according to the random number table method to construct and train nine machine learning models and perform tenfold cross-validation. The best predictive performance model was assessed using receiver operating characteristic (ROC) curve analysis and calculating the area under the curve (AUC). The predictive accuracy and clinical utility of the model was assessed using calibration curve, clinical impact curve and decision curve analyses. Additional interpretation and visualisation of the machine learning model using Shapley additive explanations (SHAP) bar charts, summary, dependency and force diagrams. A network calculator for predicting the risk of length of stay prolongation in patients with AIS was constructed using the corresponding R package.Results:Among 537 AIS patients, there were 312 males and 225 females with an age of (58.40 ± 9.00) years old. The incidence of length of stay prolongation was 43.0% (231/537). Boruta algorithm screened 10 characteristic variables. The results of ROC curve analysis showed that the AUC of the extreme gradient boosting (XGBoost) model was at the highest among the 9 machine learning models in 10 random samples. In the training and test sets, the calibration curve and clinical impact curve analysis showed that the C-index was 0.815 and 0.816, respectively, indicating high consistency between the predicted results of the XGBoost model and the actual observations. Decision curve analysis showed that the net clinical benefit was>0 when the risk threshold was>0.18 and >0.22, respectively, indicating that the model had a high application value in actual clinical decision-making. The SHAP bar graph showed the order of importance as pneumonia, urinary tract infection, age, myoglobin, triacylglycerol, neuron-specific enolase (NSE), hemoglobin, total cholesterol, homocysteine (HCY), and National Institute of Health Stroke Scale (NIHSS) scores. SHAP summary charts visualised the contribution of the 10 characteristic variables, which showed a 'bipolar distribution' phenomenon. SHAP dependency plots showed the dependency between the observed values of the 10 variables and the SHAP values, with the trend being most significant for patients with pneumonia. SHAP seeks to provide a local interpretation for individual samples, making the XGBoost model more transparent and interpretable. A web-based calculator (https://nomogram1203.shinyapps.io/LOS_web/) based on the interpretable XGBoost model dynamically quantifies the risk of prolonged length of stay in patients with AIS.Conclusions:Pneumonia, urinary tract infection, age, myoglobin, hemoglobin, total cholesterol, NSE, HCY, triacylglycerol, and NIHSS can effect length of stay prolongation in AIS patients, and a network calculator constructed based on the interpretable XGBoost model dynamically predicts the risk of length of stay prolongation in patients with AIS, which can help to achieve an accurate risk assessment for individual patients.
