Your Heart Rate and Blood Flow Can Predict How Sharp Your Brain Is
A machine-learning model using 39 cardiovascular and autonomic variables predicted cognitive test performance with ~71% accuracy in healthy adults.
Summary
Researchers trained a machine-learning model on 39 physiological measurements — including heart rate, stroke volume, cardiac output, and heart rate variability — to predict cognitive performance in 240 healthy adults. The model achieved about 71% accuracy in classifying who performed better or worse on the Trail Making Test, a standard measure of processing speed and executive function. Key factors associated with poorer cognition included older age, higher resting heart rate, and higher vascular resistance. Better cognition was linked to greater stroke volume, cardiac output, and autonomic nervous system balance. Importantly, many of these variables are modifiable through exercise and lifestyle changes, suggesting that tracking cardiovascular fitness could serve as a practical, low-burden way to monitor brain health over time.
Detailed Summary
Cognitive decline is notoriously difficult to monitor in healthy people — standard tests are burdensome, require clinical settings, and aren't practical for frequent use. A growing body of research links cardiovascular fitness and autonomic nervous system function to brain health, but teasing apart those complex, multidimensional relationships has challenged traditional statistics. This study asked whether machine learning could bridge that gap.
Researchers at Kaohsiung Medical University enrolled 240 healthy adults and recorded 39 physiological variables spanning cardiovascular function, cardiac output measures, and heart rate variability indices. Participants also completed the Trail Making Test (TMT), a widely used cognitive assessment measuring processing speed and executive function. TMT completion time was split at the median to create high- and low-performance groups, and multiple machine-learning classifiers were tested under rigorous cross-validation.
A random forest model using ten features selected via recursive feature elimination performed best, achieving 70.83% accuracy, a 71.38% F1 score, and an AUC of 71.2%. SHAP analysis — an interpretability tool — revealed which variables drove predictions most strongly. Older age, higher resting heart rate, and elevated systemic vascular resistance pushed predictions toward slower cognitive performance. Conversely, greater stroke volume, higher cardiac output, and stronger parasympathetic markers like high-frequency heart rate variability and respiratory sinus arrhythmia predicted faster TMT completion.
The clinical implications are meaningful. Most of the top predictive features are directly trainable through aerobic exercise, stress reduction, and lifestyle optimization — giving clinicians and individuals concrete targets. The authors also highlight the potential for wearable devices to capture many of these metrics passively, opening doors to continuous, low-burden cognitive health monitoring.
Caveats include the cross-sectional design, which precludes causal conclusions, the modest sample size, and the binary outcome variable that loses nuance. Generalizability beyond the Taiwanese study population remains to be established, and model accuracy, while meaningful, is not yet clinically sufficient on its own.
Key Findings
- A random forest model predicted cognitive test performance with ~71% accuracy using cardiovascular and autonomic variables.
- Higher resting heart rate and vascular resistance were linked to worse cognitive performance.
- Greater stroke volume, cardiac output, and heart rate variability predicted better cognitive scores.
- Most top predictive features are modifiable through aerobic exercise and lifestyle interventions.
- Wearable devices could potentially capture these metrics, enabling passive cognitive health monitoring.
Methodology
Cross-sectional study of 240 healthy adults; 39 physiological variables were used as input features for multiple machine-learning classifiers tested under stratified 5-fold cross-validation. Trail Making Test completion time was dichotomized at the median as the binary outcome; SHAP values provided model interpretability.
Study Limitations
The summary is based on the abstract only, as the full paper is not open access. The cross-sectional design prevents causal inference, and the binary outcome variable reduces sensitivity. The modest sample of 240 adults limits generalizability, and external validation in diverse populations is needed before clinical deployment.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.