Related Experiment Video
Updated: May 7, 2025
An R-Based Landscape Validation of a Competing Risk Model
Published on: September 16, 2022
Construction and validation of a readmission risk prediction model for elderly patients with coronary heart disease
Hanyu Luo1, Benlong Wang1, Rui Cao1
1Department of Cardiology of Lu'an People's Hospital, Lu'an Hospital of Anhui Medical University, Lu'an, China.
Insights
Machine learning accurately predicts readmission risk for elderly coronary artery disease patients. Key factors include diabetes, Red blood cell distribution width (RDW), and Triglyceride-glucose body mass index (TyG-BMI).
Area of Science:
- Cardiology
- Medical Informatics
- Machine Learning in Healthcare
Background:
- Elderly patients with coronary artery disease (CAD) face significant readmission risks.
- Identifying predictors for readmission is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To identify risk factors for 3-year readmission in elderly CAD patients.
- To develop and validate a machine learning-based predictive model for readmission risk.
Main Methods:
- Utilized data from 575 elderly CAD patients, categorizing them by 3-year readmission status.
- Employed Lasso and logistic regression for factor identification, and XGBoost, LR, RF, KNN, DT for model development.
- Evaluated model performance using ROC curves, calibration plots, and decision curve analysis, with external validation on 143 patients.
Main Results:
- The XGBoost model achieved the highest predictive accuracy, with an AUC of 0.903 (training) and 0.891 (external validation).
- Identified diabetes mellitus, Red blood cell distribution width (RDW), and Triglyceride-glucose body mass index (TyG-BMI) as significant predictors.
- XGBoost and decision tree models demonstrated strong calibration and clinical utility.
Conclusions:
- Diabetes, RDW, and TyG-BMI are key factors influencing readmission in elderly CAD patients.
- The XGBoost-based predictive model shows excellent efficacy for identifying high-risk patients.
- This model can guide clinical decision-making and inform targeted intervention strategies.
Background:
To investigate the risk factors for readmission of elderly patients with coronary artery disease, and to construct and validate a predictive model for readmission risk of elderly patients with coronary artery disease within 3 years by applying machine learning method.
Methods:
We selected 575 elderly patients with CHD admitted to the Affiliated Lu'an Hospital of Anhui Medical University from January 2020 to January 2023. Based on whether patients were readmitted within 3 years, they were divided into two groups: those readmitted within 3 years (215 patients) and those not readmitted within 3 years (360 patients). Lasso regression and multivariate logistic regression were used to compare the predictive value of these models. XGBoost, LR, RF, KNN and DT algorithms were used to build prediction models for readmission risk. ROC curves and calibration plots were used to evaluate the prediction performance of the model. For external validation, 143 patients who were admitted between February and June 2023 from a different associated hospital in Lu'an City were also used.
Results:
The XGBoost model demonstrated the most accurate prediction performance out of the five machine learning techniques. Diabetes, Red blood cell distribution width (RDW), and Triglyceride glucose-body mass index (TyG-BMI), as determined by Lasso regression and multivariate logistic regression. Calibration plot analysis demonstrated that the XGBoost model maintained strong calibration performance across both training and testing datasets, with calibration curves closely aligning with the ideal curve. This alignment signifies a high level of concordance between predicted probabilities and observed event rates. Additionally, decision curve analysis highlighted that both decision trees and XGBoost models achieved higher net benefits within the majority of threshold ranges, emphasizing their significant potential in clinical decision-making processes. The XGBoost model's area under the ROC curve (AUC) reached 0.903, while the external validation dataset yielded an AUC of 0.891, further validating the model's predictive accuracy and its ability to generalize across different datasets.
Conclusion:
TyG-BMI, RDW, and diabetes mellitus at the time of admission are the factors affecting readmission of elderly patients with coronary artery disease, and the model constructed based on the XGBoost algorithm for readmission risk prediction has good predictive efficacy, which can provide guidance for identifying high-risk patients and timely intervention strategies.
