Predicting asthma attacks in primary care: protocol for developing a machine learning-based prediction model.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Respiratory Medicine

Background:

• Asthma attacks are unpredictable, potentially fatal events requiring sensitive and specific prediction models.
• Current prediction models for asthma attacks often lack the necessary accuracy for clinical application.
• Electronic health records (EHRs) offer a rich data source for developing predictive models.

Purpose of the Study:

• To develop and validate a risk score for predicting asthma attacks in primary care using routinely collected EHR data.
• To enhance the sensitivity and specificity of asthma attack prediction models to minimize mortality and unnecessary medication.
• To apply statistical learning approaches to identify key predictors of asthma exacerbations.

Main Methods:

• Utilizing machine-learning classifiers including naïve Bayes, support vector machines, and random forests.
• Training models on a large patient registry (Asthma Learning Health System - ALHS) with 500,000 individuals.
• Validating models on reserved, unseen, and external datasets (Seasonal Influenza Vaccination Effectiveness II - SIVE II study).

Main Results:

• The study aims to compare the performance of different machine-learning classifiers.
• Model performance will be rigorously tested on independent and external datasets.
• The developed risk score is expected to provide a reliable prediction of asthma attacks.

Conclusions:

• A validated asthma attack risk score derived from EHR data can significantly aid in primary care management.
• Improved prediction accuracy can lead to earlier interventions and better patient outcomes.
• The study emphasizes the potential of machine learning in proactive respiratory healthcare.