Validating a Machine Learning Algorithm to Predict 30-Day Re-Admissions in Patients With Heart Failure: Protocol for...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Machine learning (ML) offers real-time data value but faces challenges like overfitting and performance decay.
• Electronic medical record adoption and value-based care drive ML investment in healthcare.
• Independent validation of ML models is crucial before clinical adoption.

Purpose of the Study:

• To prospectively validate an ML-based predictive model for 30-day re-admissions in heart failure patients.
• To compare ML predictions against real-world observed outcomes in an independent cohort.

Main Methods:

• Prospective monitoring of adult heart failure patients discharged alive for 30-day re-admissions.
• Exclusion of training dataset patients to prevent information leakage.
• Assessment of model performance using concordance statistic, sensitivity, specificity, and predictive values.

Main Results:

• Data collection completed in July 2017, with analysis underway.
• First results are anticipated for publication in October 2018.
• Study aims to provide insights into ML model robustness and healthcare gains.

Conclusions:

• This is among the first studies to prospectively evaluate an ML predictive algorithm in a real-world clinical setting.
• Results will inform the reliability of ML predictions and set expectations for healthcare applications.
• Establishes a benchmark for real-world ML performance in predicting heart failure readmissions.