Cardiorespiratory Markers of Type 2 Diabetes: Machine Learning-Based Analysis.

Area of Science:

• Cardiovascular Physiology
• Autonomic Nervous System Regulation
• Machine Learning in Healthcare

Background:

• Type 2 diabetes mellitus (T2DM) is linked to increased cardiovascular risk and autonomic dysfunction.
• Heart rate variability (HRV) and cardiorespiratory interaction metrics offer insights into autonomic regulation.
• Frequency response function (FRF) and impulse response (IR) metrics capture distinct aspects of cardiorespiratory control.

Purpose of the Study:

• To evaluate the efficacy of HRV, FRF, and IR metrics in distinguishing individuals with and without T2DM.
• To assess the combined predictive value of these physiological features using machine learning classifiers.

Main Methods:

• Derived spectral HRV, FRF, and causal IR features from electrocardiogram and respiratory signals.
• Employed logistic regression and Support Vector Machine (SVM) classifiers.
• Utilized NearMiss-1 (NM) undersampling and Synthetic Minority Oversampling Technique (SMOTE) for data balancing.

Main Results:

• Impulse response (IR) features demonstrated strong standalone performance in distinguishing T2DM.
• The combined HRV+FRF feature set achieved the highest accuracy (0.830) under NM undersampling with SVM RBF.
• Combined HRV+IR showed strong performance under SMOTE, though standalone IR retained superior recall.

Conclusions:

• Systems-based approaches integrating frequency-domain and causal dynamic features provide richer characterization of T2DM-related regulatory differences than HRV alone.
• The findings highlight promising physiological feature domains and sampling strategies for future research.
• Larger datasets are needed to validate generalizability and clinical relevance.