From Entropy to Outcome: Multimodal Digital Biomarkers in Stroke — NEURO-PERSIST Study Protocol (v0.5)

Abstract

NEURO-PERSIST is a pre-registered clinical study protocol for predicting 90-day functional outcome after ischemic or hemorrhagic stroke using interpretable multimodal digital biomarkers. The proposed biomarker domains include heart-rate-variability entropy, prosody stability, gait and motor microvariability, and baseline imaging/radiomics features. The protocol defines the modified Rankin Scale at 90 days (mRS-90d) as the primary outcome, analysed both ordinally and dichotomously (mRS 0-2 vs. 3-6). It introduces the Persistence Score C (PS_C), a replication-oriented feature robustness framework across centres, devices, time, and missingness. The planned sample size is N=300 with an assumed poor-outcome rate of approximately 35%, yielding about 105 events. The protocol constrains the baseline model to a maximum of 10 pre-specified predictor parameters, reports AUC power, and explicitly treats AUC power as insufficient on its own. Model robustness is addressed through overfitting control, calibration assessment, PR-AUC, decision curve analysis, resampling/bootstrap internal validation, missingness sensitivity analyses, and an optional time- or centre-separated validation layer where data structure permits. XGBoost is included as exploratory add-on modelling with explainability analyses, not as the primary powered model. This document is a study design and priority deposit. It reports no patient data and does not constitute an ethics approval, clinical trial approval, regulatory trial protocol, or completed clinical study. Before data collection, ethics approval will be sought from the responsible ethics committees of the participating centres, and the study will be conducted in accordance with current human-subject research ethics, the Declaration of Helsinki (WMA 2024), ICH E6(R3) principles as applicable, GDPR requirements, and centre-specific data-protection agreements. The protocol is the clinical translation layer of the Holographic Ledger / NEURO-PERSIST series. It cites the Holographic Ledger v2 framework, the Negative Dwell Time preprint, and the Coherence-Modulated Pre-Commit States preprint as theoretical antecedents.

Ähnliche Arbeiten