Cognitive Frailty Classification Models for Older Adults in a Point-of-Care System

Cognitive Frailty (CF) is a prevalent condition among older individuals, combining physical frailty with cognitive impairment, with an increased risk of adverse health outcomes. However, the current diagnostic processes for CF are often hospital-based and not feasible for all patients, leading to a gap in timely and efficient diagnosis. To address this problem, a Point-of-Care (POC) system is proposed by utilizing Lab-on-a-Chip (LOC) technology and Tiny Machine Learning (TinyML) capabilities on Android devices for autonomous blood and urine sample collection and real-time CF assessment. In this paper, the feature selection technique for ML-based CF classification is also proposed by using several statistical test methods to improve the accuracy of the ML model. The results demonstrate significant improvements in the accuracy of CF classification through advanced machine learning models, achieving over 70% accuracy in key models such as Linear Discriminant Analysis (LDA), Gaussian Naive Bayes (GNB), and Support Vector Machines (SVM). This innovative system enhances accessibility and efficiency in diagnosing CF, benefiting older individuals in remote or resource-limited environments.