Machine learning based prediction of cognitive metrics using major biomarkers in SuperAgers

As populations age, understanding cognitive decline and age-related diseases like dementia has become increasingly important. "SuperAgers," individuals over 65 with cognitive abilities similar to those in their 40s, provide a unique perspective on cognitive reserve. This study analyzed 55 blood biomarkers, including cellular components and metabolism/inflammation-related factors, in 39 SuperAgers and 42 typical agers. While conventional statistical analyses identified significant differences in only four biomarkers, advanced feature selection and machine learning techniques revealed a broader set of 15 key biomarkers associated with SuperAger status. A predictive model built using these biomarkers achieved an accuracy of 76% in cognitive domain prediction. To address the limitation of small sample sizes, data augmentation leveraging large language models improved the model's robustness. Shapley Additive exPlanations (SHAP) provided interpretability, revealing the impact of specific blood factors on cognitive function. These findings suggest that certain blood biomarkers are not only associated with cognitive performance but may also serve as indicators of cognitive reserve. By utilizing simple blood tests, this research presents a clinically significant method for predicting cognitive function and identifying SuperAger status in healthy elderly individuals, offering a foundation for future studies on the biological mechanisms underpinning cognitive resilience.