Error and Timeliness Analysis for Using Machine Learning to Predict Asthma Hospital Visits: Retrospective Cohort...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Asthma hospital visits represent a significant healthcare burden.
• Machine learning models can forecast asthma hospitalizations.
• Previous models achieved high accuracy in predicting asthma hospital visits.

Purpose of the Study:

• To determine the advance notice our model provides for asthma hospital visits.
• To assess the likelihood of out-of-system visits or poor outcomes for false-positive predictions.

Main Methods:

• Utilized adult asthma patient data from University of Washington Medicine (UWM) between 2011-2018.
• Applied a machine learning model to predict 2019 asthma hospital visits.
• Calculated prediction lead times and analyzed outcomes for false-positive cases.

Main Results:

• The model provided warnings >= 3 months in advance for 61.9% of patients with future UWM asthma hospital visits.
• Warnings were issued >= 1 day in advance for 84.4% of these patients.
• 29.01% of false-positive predictions indicated out-of-system visits or poor outcomes.

Conclusions:

• The predictive model offers timely risk warnings for asthma patients.
• False-positive predictions identified patients who could benefit from preventive interventions.
• Further model refinement is needed for improved accuracy and timeliness.