The salivary cortisol classification based on the heart rate variability.

Area of Science:

• Physiology
• Biomarkers
• Machine Learning

Background:

• Stress response involves physiological changes, with heart rate variability (HRV) and cortisol as key indicators.
• Stress system activation is largely subconscious, making objective measurement challenging.
• Current stress assessment often relies on subjective emotional states, which may not accurately reflect physiological stress.

Purpose of the Study:

• To investigate the relationship between stress, cortisol secretion, and electrophysiological biomarkers like HRV.
• To develop machine learning models for predicting cortisol levels using HRV indexes.
• To establish HRV as a reliable, subconscious indicator of stress system activation.

Main Methods:

• Utilized machine learning algorithms (SVM, XGB, MLP) on data from 634 healthy adult men (20-50 years old).
• Employed a trait social stress test to induce a range of cortisol concentrations.
• Analyzed 12 HRV features to classify cortisol levels.

Main Results:

• Machine learning algorithms accurately classified cortisol levels (optimal: 5-15 ng/mL) using HRV indexes.
• The XGB algorithm achieved 99% accuracy and 99% F1 score in classification.
• HRV indexes effectively indicated the individual's stress system state by reflecting cortisol concentration.

Conclusions:

• HRV indexes can be used to classify stress levels in adult men.
• HRV serves as a reliable biomarker for predicting salivary cortisol concentrations.
• This study introduces a novel, objective method for assessing stress through HRV analysis.