From Fitness to Cognition: Machine-Learning Prediction of Cognitive Performance Using Physiological Parameters in...

Area of Science:

• Physiology
• Cognitive Science
• Machine Learning

Background:

• Current cognitive assessments are unsuitable for frequent monitoring in healthy adults.
• Multidimensional relationships between fitness, cardiovascular function, autonomic regulation, and cognition are complex.
• Traditional statistical methods struggle to interpret these complex relationships.

Purpose of the Study:

• To investigate the feasibility of using fitness-related physiological and cardiac autonomic indicators for cognitive performance assessment.
• To apply interpretable machine learning approaches to analyze these relationships.
• To assess relative cognitive performance in healthy adults using physiological data.

Main Methods:

• Cross-sectional study with 240 healthy adults.
• Recorded 39 physiological variables as input features.
• Used Trail Making Test (TMT) completion time as the outcome, dichotomized at the median.
• Employed four feature-selection strategies and grid-tuned classifiers with 5-fold cross-validation.

Main Results:

• A random forest model with ten selected features achieved 70.83% accuracy, outperforming a baseline logistic regression model.
• SHAP interpretation revealed key physiological predictors: older age, higher systemic vascular resistance, and higher resting heart rate predicted longer TMT times.
• Greater stroke volume, cardiac output, high-frequency power, and respiratory sinus arrhythmia predicted shorter TMT times.

Conclusions:

• Physiological parameters related to cardiovascular and autonomic function demonstrate moderate ability to discriminate cognitive performance groups.
• The findings support the feasibility of physiology-based cognitive assessment.
• Identified modifiable features suggest potential for exercise and lifestyle interventions, and future personalized cognitive health applications.