Short-Term Prognosis in Acute Asthma Exacerbations: A Comparative Evaluation of Machine Learning Models Using...

Area of Science:

• Pulmonology
• Medical Informatics
• Emergency Medicine

Background:

• Acute asthma exacerbations are a frequent reason for emergency department (ED) visits, necessitating rapid risk assessment for disposition.
• Spirometry is standard for assessing asthma severity but may not reliably predict outcomes.
• Machine learning (ML) shows promise for enhancing prognostic accuracy in emergency settings.

Purpose of the Study:

• To assess the prognostic value of spirometry, tracheal respiratory sound features, and ML models for short-term outcomes in adults with acute asthma exacerbations.
• To compare the predictive performance of different feature sets and ML models.

Main Methods:

• Prospective cohort study involving adults with acute asthma exacerbations.
• Collected spirometry and tracheal respiratory sound data pre- and post-treatment.
• Defined short-term prognosis by hospitalization or ED re-presentation within seven days.
• Developed and compared ML models using demographic, spirometric, acoustic, and combined features.

Main Results:

• Baseline airflow limitation (FEV1, PEF) was the strongest predictor of short-term outcomes.
• ML models using spirometric data achieved the highest accuracy (up to 84.4%).
• Respiratory sound analysis provided complementary information but did not outperform spirometry-based models.

Conclusions:

• Baseline spirometric impairment is the most reliable predictor of short-term outcomes in acute asthma.
• Tracheal respiratory sound analysis with ML can offer supplementary prognostic insights, especially when spirometry is limited.
• These methods can serve as decision-support tools, not replacements for standard assessment.