Substitute or Supplement? The Role of Multimodal Digital Biomarkers in Mobile Cognitive Impairment Assessment Tools

Digital biomarkers (DBMs) have emerged as promising tools for the detection of cognitive impairment (CI). While many studies have examined their diagnostic value individually, few studies have examined how they fare in multimodal settings. To that end, we developed a digital screening tool that integrates three types of DBMs: Neuropsychological, eye movement, and voice. We evaluated classification performance using the area under the receiver operating characteristic curve (AUC) and assessed feature contributions using SHapley Additive exPlanations (SHAP) values. The multimodal model, which incorporated all three DBMs, achieved an AUC of 0.83, outperforming models using only eye movement DBM (AUC = 0.76) and voice DBM (AUC = 0.71). Our feature importance analysis revealed that eye movement features made the greatest contribution in conjunction with the traditional neuropsychological assessment to classification performance across all models. The predictive models using Our results suggest that DBMs are good supplements to enhance the classification performance rather than a substitution of classical means. In other words, using individual DBMs to build predictive models did not achieve the same classification capacity as the model which incorporated all DBMs coupled with the neuropsychological. However, eye movement DBM showed high potential when used as a single input, achieving classification accuracy comparable to the complete model with all three DBMs. These findings warrant further investigation into eye movement as a digital biomarker for detecting cognitive impairment.