Development and Validation of a Machine Learning-Based Prediction Model for 30-Day COPD Readmissions: Enhancing Early Risk Stratification and Intervention
Author(s)
Khang Nguyen, PharmD, Mac Townsend, PharmD, MPH, Jiawei Chen, PharmD, Joshua M. Thorpe, PhD, MPH.
University of North Carolina at Chapel Hill, Chapel Hill, NC, USA.
University of North Carolina at Chapel Hill, Chapel Hill, NC, USA.
OBJECTIVES: This study used machine-learning (ML) techniques to generate a chronic obstructive pulmonary disease (COPD) 30-day hospital readmission prediction model to profile patients at the greatest risk. Approximately 20-25% of patients hospitalized for acute COPD exacerbations (AECOPD) are readmitted within 30 days. While pharmacist-led transitions-of-care programs can reduce readmissions, they are costly and only cost-effective under certain circumstances. Limited datasets utilized in traditional models can hinder their ability to detect complex risk factor dynamics. In contrast, ML can enhance efficiency when identifying high-risk subgroups and stratifying readmission risk.
METHODS: The 2019 Nationwide Readmission Database (NRD), an all-payer dataset representative of over 50% of U.S. hospitalizations, was used to identify COPD hospitalizations based on ICD-10 codes. Patients who died during hospitalization, had same-day transfers, or were discharged to non-residential settings were excluded. Several ML algorithms were evaluated, including Classification Trees, Random Forest, Least Absolute Shrinkage and Selection Operator regression, and Extreme Gradient Boosting (XGBoost). The area under the receiver operating characteristic (AUC-ROC) curve helped determine the optimal model. Permutation-based importance identified the key drivers of readmission.
RESULTS: An estimated 304,751 patients survived an index hospitalization for COPD in 2019, at a total cost of $10.2 billion (average cost: $33,499 per stay). Of these, 10.9% (n=33,65) were readmitted for COPD within 30 days; an additional $2.9 billion in potentially avoidable hospital costs. AUC-ROC was highest using the XGBoost algorithm (AUC-ROC=0.63). ML identified 19 unique risk profiles, with 17 involving interactions of three or more factors. The strongest predictors included winter hospitalization, Medicaid insurance, and comorbidities like heart failure. High-risk patients had up to a 23.5% readmission rate, while low-risk groups had 6.4%.
CONCLUSIONS: Machine learning methods applied to national COPD data revealed complex risk profiles undetectable by traditional models. This approach enhances early risk stratification, enabling cost-effective, tailored interventions for patients at greatest risk of readmission.
METHODS: The 2019 Nationwide Readmission Database (NRD), an all-payer dataset representative of over 50% of U.S. hospitalizations, was used to identify COPD hospitalizations based on ICD-10 codes. Patients who died during hospitalization, had same-day transfers, or were discharged to non-residential settings were excluded. Several ML algorithms were evaluated, including Classification Trees, Random Forest, Least Absolute Shrinkage and Selection Operator regression, and Extreme Gradient Boosting (XGBoost). The area under the receiver operating characteristic (AUC-ROC) curve helped determine the optimal model. Permutation-based importance identified the key drivers of readmission.
RESULTS: An estimated 304,751 patients survived an index hospitalization for COPD in 2019, at a total cost of $10.2 billion (average cost: $33,499 per stay). Of these, 10.9% (n=33,65) were readmitted for COPD within 30 days; an additional $2.9 billion in potentially avoidable hospital costs. AUC-ROC was highest using the XGBoost algorithm (AUC-ROC=0.63). ML identified 19 unique risk profiles, with 17 involving interactions of three or more factors. The strongest predictors included winter hospitalization, Medicaid insurance, and comorbidities like heart failure. High-risk patients had up to a 23.5% readmission rate, while low-risk groups had 6.4%.
CONCLUSIONS: Machine learning methods applied to national COPD data revealed complex risk profiles undetectable by traditional models. This approach enhances early risk stratification, enabling cost-effective, tailored interventions for patients at greatest risk of readmission.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
HSD105
Topic
Health Service Delivery & Process of Care
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Injury & Trauma, SDC: Respiratory-Related Disorders (Allergy, Asthma, Smoking, Other Respiratory)