Understanding Complex Interactions in Pediatric Diastolic Function Assessment.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Diagnosing pediatric diastolic dysfunction (DD) noninvasively is challenging due to the lack of validated diagnostic algorithms.
• Existing methods struggle to accurately assess DD in children, necessitating novel approaches.

Purpose of the Study:

• To explore the utility of machine learning (ML) in developing a diagnostic model for pediatric DD.
• To correlate echocardiographic measurements with invasively measured markers of DD in pediatric patients.

Main Methods:

• Enrolled children at risk for DD undergoing left heart catheterization.
• Collected simultaneous invasive pressure measurements (Tau, LVEDP, dP/dt max) and echocardiographic data.
• Utilized Random Forest (RF) models and backward stepwise regression to predict invasive markers.

Main Results:

• RF models demonstrated non-inferior performance compared to linear models, with more intuitive feature importance.
• Key echocardiographic predictors identified: propagation velocity for Tau, E/propagation velocity ratio for LVEDP, and systolic global longitudinal strain rate for dP/dt max.
• Pairwise correlations between echocardiographic and invasive markers were generally low.

Conclusions:

• Machine learning models, particularly RF, can elucidate relationships between echocardiographic and invasive DD markers in children.
• This ML approach shows promise for developing pediatric-specific diagnostic algorithms for diastolic dysfunction.
• Improved noninvasive assessment of pediatric DD is achievable through advanced modeling techniques.