Development of a Heart Rate Variability Based Ambulatory Stress Detection Model for Clinical Populations

Area of Science:

• Biomedical Engineering
• Data Science
• Psychiatry

Background:

• Real-time stress detection via biosensors is crucial for timely interventions in mental health recovery.
• Current stress detection models perform poorly in clinical populations due to reliance on lab data from healthy adults.

Purpose of the Study:

• To develop and test machine learning models for stress detection in individuals recovering from alcohol use disorder (AUD).
• To improve the accuracy of stress detection algorithms for clinical applications.

Main Methods:

• Utilized ambulatory electrocardiogram (ECG) and ecological momentary assessment (EMA) data from 44 individuals in early AUD recovery.
• Applied unsupervised (t-SNE, cluster analysis) and supervised learning models to identify stress-related features.
• Compared model performance with laboratory-derived data from healthy adults.

Main Results:

• Initial model accuracy was 63% in the clinical sample, compared to 94% in healthy adults.
• Accounting for age and body-mass index (BMI) improved accuracy to 80% in the clinical sample.
• Data normalization and stratification by age and BMI enhanced stress prediction.

Conclusions:

• Stress detection in clinical populations is challenging but feasible with advanced methods.
• Machine learning models can be optimized for clinical use by incorporating individual characteristics like age and BMI.
• This research is a step toward developing effective, personalized stress management tools for mental health recovery.