A Neurosurgical Readmissions Reduction Program in an Academic Hospital Leveraging Machine Learning, Workflow...

Area of Science:

• Neurosurgery
• Health Informatics
• Machine Learning

Background:

• Predicting 30-day hospital readmissions is vital for patient care and resource management.
• Existing machine learning (ML) models for neurosurgery readmissions lack clinical implementation details.
• This study addresses the need for practical, implementable ML solutions in neurosurgery readmission prediction.

Purpose of the Study:

• Develop high-performing ML models (AUROC > 0.8) for predicting 30-day neurosurgical readmissions.
• Identify actionable interventions through clinical workflow analysis and interviews.
• Simulate the clinical and financial impact of implementing these predictive models and interventions.

Main Methods:

• Utilized electronic health records and five ML methodologies: gradient boosting, decision tree, random forest, ridge logistic regression, and linear support vector machine.
• Conducted semi-structured interviews to identify intervention points across preoperative, inpatient, discharge, and follow-up phases.
• Applied calibrated agent-based models (ABMs) to simulate readmission rate and cost reductions.

Main Results:

• The random forest model achieved an AUROC of 0.89 for neurosurgical intensive care unit (NSICU) admissions.
• Six interventions were identified, targeting key phases of patient care.
• Simulations indicated significant reductions in readmission rates (e.g., NSICU from 13.13% to 10.12%) and projected savings of over $1.3 million.

Conclusions:

• Successfully developed and simulated an ML-based approach for predicting and reducing 30-day hospital readmissions in neurosurgery.
• The proposed interventions demonstrate feasibility for enhancing patient outcomes and mitigating financial losses.