Predicting Adherence to Computer-Based Cognitive Training Programs Among Older Adults Using Source-Free Domain Adaptation: Algorithm Development and Validation.
Academic Article
Overview
abstract
BACKGROUND: Cognitive decline in the aging population presents an unprecedented challenge worldwide. Recent research has shown the potential of cognitive training programs to mitigate cognitive decline. However, these interventions require sustained adherence to be effective, which can be challenging. OBJECTIVE: In this study, we aim to enhance the accuracy of predicting adherence patterns in cognitive training programs for older adults, with the goal of developing personalized support systems that promote adherence and improve cognitive outcomes. METHODS: A major challenge in developing deep neural networks for predicting adherence patterns is the limited availability of individual participants' training data. Although domain adaptation techniques can address this issue by leveraging training data from other clinical studies, our research considers a more practical scenario where the use of such data from other studies is restricted due to privacy and confidentiality concerns. Therefore, we used source-free domain adaptation (SFDA), which uses models trained on other cognitive studies without requiring access to the corresponding datasets. To the best of our knowledge, this is the first effort to use SFDA to predict older adults' daily adherence to cognitive training programs. RESULTS: Using data from 3 previously conducted cognitive training intervention studies, our results demonstrated the efficacy of deep learning models combined with SFDA to accurately predict adherence lapses while addressing data privacy concerns. CONCLUSIONS: Our findings indicate that deep learning and SFDA techniques can be useful in the development of adherence support systems for computerized cognitive training, aimed at improving the health and well-being of older adults.
