Digital phenotyping from heart rate dynamics: Identification of zero-poles models with data-driven evolutionary...

Area of Science:

• Physiology
• Biomedical Engineering
• Digital Health

Background:

• Heart rate response to physical activity is crucial for assessing physical state in clinical and training settings.
• Emerging digital phenotyping requires individualized models for accurate physiological monitoring.
• Existing models may lack the precision needed for personalized health insights.

Purpose of the Study:

• To propose and validate a novel zero-poles dynamic model for heart rate response to exercise.
• To develop a data-driven evolutionary learning method for efficient model parameter identification.
• To compare the proposed model and identification method against existing approaches.

Main Methods:

• A five-phase zero-poles dynamic model was developed to describe heart rate variability.
• Data from 30 healthy participants were collected using treadmills and thoracic sensors under static and dynamic protocols.
• The proposed model and evolutionary learning were compared with first/second-order models and gradient descent methods.

Main Results:

• The zero-poles model demonstrated a strong fit for heart rate response to exercise (Pearson's ρ > .95).
• First and second-order models also showed good suitability (Pearson's ρ > .92).
• The evolutionary learning method significantly outperformed least-squares methods for fast model identification (p < .03).

Conclusions:

• The proposed zero-poles model accurately captures heart rate dynamics during physical activity.
• The evolutionary learning approach offers efficient and rapid parameter identification for dynamic models.
• Investigated linear dynamic model parameters show promise as digital biomarkers for continuous health monitoring.