Abstract
About this talk
Background: Mild Cognitive Impairment (MCI) is a critical stage for early cognitive risk identification and timely support in older adults. Nurses play an important role in recognizing cognitive changes, conducting ongoing assessments, and supporting older adults and caregivers. Multimodal digital biomarkers combined with Machine Learning (ML) offer promising non-invasive approaches that may complement traditional cognitive assessment and inform future nursing care.
Objective: To systematically evaluate the diagnostic performance of ML models based on multimodal digital biomarkers for MCI, compare different digital modality and predictor combinations, and examine methodological factors associated with model performance to inform the development of clinically interpretable approaches relevant to nursing assessment.
Methods: We searched pubmed, web of science, cochrane, embase, and IEEE xplore through May 3, 2026. A bivariate meta-analysis with Hartung-Knapp-Sidik-Jonkman adjustment and restricted maximum likelihood estimation was used to pool sensitivity and specificity. SROC analysis, subgroup and sensitivity analyses, and assessments of heterogeneity and small study effects were conducted.
Key Findings: Twenty studies involving 2,244 participants with MCI and 3,291 healthy controls contributed 74 diagnostic models. Models combining eye movement and hand motion showed a mean AUC of 0.92 (SD 0.04) and mean accuracy of 0.86 (SD 0.07). Across 16 studies, pooled sensitivity and specificity were both 0.85, and SROC AUC was 0.93 (95% CI, 0.90–0.95). Predictor composition significantly influenced sensitivity (P < 0.01). Sensitivity analyses indicated little change after excluding studies at high risk of bias.
Conclusions: Multimodal digital biomarkers combined with ML demonstrate promising diagnostic performance for MCI and may inform future nursing-supported cognitive risk assessment. Standardized biomarker definitions, transparent model reporting, and multicenter external validation are required before clinical application.
Speaker
About the speaker
