2025/07/18 by Elaheh Moradi, Robert Dahnke, Vandad Imani +3
Medicine · Computer Science · Environmental Science · #Dementia and Cognitive Impairment Research #Machine Learning in Healthcare #Health, Environment, Cognitive Aging
paper · doi:10.1016/j.neurobiolaging.2025.06.010
Plasma biomarkers are associated with cognitive performance and decline in Alzheimer’s disease, making them promising for early detection. This study investigates their predictive value, combined with non-invasive measures, in forecasting cognitive decline in individuals without dementia. We developed a multimodal machine-learning approach incorporating plasma biomarkers (Amyloid β 42/40 (A β 42/40), p-tau181, NfL), MRI, demographics, APOE4, and cognitive assessments to predict the rate of cognitive decline. Various models were designed to predict decline rates across cognitive domains (memory, executive function, language, and visuospatial abilities) and assess their relevance in predicting dementia progression. Cross-validated correlations between predicted and actual cognitive decline rates were 0.50 for memory, 0.49 for language, 0.42 for executive function, and 0.44 for visuospatial ability. MRI showed greater predictive importance than plasma biomarkers. Among plasma biomarkers, NfL and p-tau181 outperformed A β 42/40. Predicting cognitive decline and progression to MCI/dementia was most accurate in the memory domain, where plasma biomarkers (A β 42/40, p-tau181, NfL) added significant value to predictive models, likely due to their AD-specific nature. Plasma biomarkers contributed less to predictions in other cognitive domains. The results indicate that plasma biomarkers, particularly when combined with MRI, demographics, APOE4, and cognitive measures, have significant potential for predicting memory decline and assessing the risk of dementia progression, even in cognitively unimpaired individuals. • Assessed plasma biomarkers with non-invasive measures to predict cognitive decline. • Plasma biomarkers help to predict memory decline, less effective in other domains. • MRI and cognitive scores are top predictors; NfL, p-tau181 outperformed A β 42/40. • Models are more accurate for those with existing cognitive impairments. • Models predict MCI/dementia progression even in cognitively healthy people.