Wearable and mobile devices capture cardiac, neural, behavioral, and speech signals that each provide only a partial, noise‑sensitive view of physiological state. BioSync fuses these signals into the BioSync Index (BSI), a continuous composite digital biomarker defined under the BEST framework. The model applies multi‑head self‑attention to modality tokens and adds a linear branch whose hypothesis class includes standard feature concatenation. This architecture is motivated by latent‑variable measurement theory, assuming that joint observations contain information unavailable to any single modality.
We evaluated BioSync on two literature‑informed synthetic cohorts:
- a cognitive‑decline cohort using HRV, EEG, actigraphy, and speech (four modalities);
- a metabolic‑autonomic cohort built around the public AI‑READI wearable schema.
In the cognitive cohort, BioSync achieved an AUC of 0.928 versus 0.926 for concatenation. In the metabolic cohort, BioSync obtained accuracy/F1 of 0.764/0.766, compared with 0.756/0.758 for concatenation. The BSI correlated with latent severity in both cohorts ($r=0.91$ and $r=0.68$). A pure‑attention ablation yielded an AUC of 0.911 on the cognitive cohort, indicating that the combined wide‑and‑deep architecture raised performance to 0.928. With matched modality‑dropout training, BioSync outperformed concatenation at five of six corruption rates in the cognitive cohort and at the highest rate in the metabolic cohort. Its cognitive‑cohort AUC also exceeded five published digital‑biomarker reference values, though dataset and task differences preclude a controlled benchmark claim. Comparison with single‑modality, early‑fusion, and late‑fusion designs across six predefined criteria highlights the model’s computational properties; validation on real cohorts remains necessary.
Review