ORI-CLIN-002 · Deep Research Appendix
An evidence-tiered platform for longitudinal health detection, aging care, and clinical partnership. 62+ peer-reviewed citations. Five-stage validation roadmap. Regulatory pathway mapped.
The Hard Boundary
"Orí is a consent-first longitudinal sensing platform that detects statistically significant changes in voice, physiology, and movement relative to each user's own baseline. It does not diagnose disease. Its purpose is to surface patterns that may warrant user awareness or clinician review, grounded in peer-reviewed evidence."
Tier I · Cleared Precedent
FDA precedent: AliveCor KardiaMobile (2014, 510(k)) and Apple Watch Series 4 (2018, De Novo). 2025 BMC Cardiovascular meta-analysis: 97.4% pooled sensitivity, 96.6% specificity. 2025 npj Digital Medicine multicenter (n=236): 98.5% accuracy.
Sensitivity
97.4%
Specificity
96.6%
Accuracy
98.5%
A principle
Detection without diagnosis.
Honesty as credibility.
Tier II · Strong Research Support
Appropriate for "signal detection" and "change over time" product language. Not appropriate for diagnostic claims until prospective device-specific validation is complete. Honesty is the credibility.
Systematic review of 69 studies (2020-2025). Audio Spectrogram Transformer: 97.14% accuracy. XGBoost: 98.0% accuracy, 0.991 ROC-AUC. 90% of PD patients develop vocal impairment preceding motor symptoms.
AST accuracy
97.14%
XGBoost AUC
0.991
Pre-motor signal
90%
Dem@Care/IBM (2015): HC vs MCI 79%, HC vs AD 87%. Frontiers Aging Neuroscience (2024): 80-97% accuracy. AMYPRED (Novoic, 2021): amyloid-confirmed early AD detectable via smartphone speech. Nature Digital Medicine 2026 confirmed cross-language robustness.
AD detection
87%
Frontiers range
80-97%
Languages
Cross-lingual
Systematic review of 12 studies, n=16,872, AUC 0.71-0.93 (ScienceDirect 2025). Kintsugi Voice n=14,898 (Annals of Family Medicine 2025). FDA published Voice Biomarker AI in Mental Health guidance.
Cohort
16,872
AUC range
0.71-0.93
FDA guidance
Published
Strong evidence for IMU-based gait analysis predicting fall risk. BMJ Open 2025 scoping review: multimodal sensing (gait + voice + physiology) is the most promising direction for comprehensive aging-care monitoring.
Modality
Multimodal
Vertical
Aging care
Precedent
BMJ 2025
Tier III · Promising, Roadmap
Sensor fusion signature is plausible; awaiting consumer device validation.
Hypervigilance physiology is detectable; prospective studies needed.
Real-time voice change detection is credible; regulatory engagement underway.
Voice biomarker literature is early-stage for this population.
Autonomic signatures exist; real-world validation lacking.
80.2% accuracy in controlled study; naturalistic validation pending.
Validation Roadmap
Stage 01
Local-first sensing. No clinical claims. Build longitudinal corpus from consenting users.
Stage 02
University partnerships. Voice + IMU + PPG against gold-standard clinical assessments.
Stage 03
Q-Sub for AF detection (cleared precedent). Voice biomarker pathway aligned with FDA guidance.
Stage 04
AF first. Then PD voice biomarker. Then MCI / AD detection, riding Nature 2026 cross-language evidence.
Stage 05
Clinician dashboards (with patient consent only). Aging-care vertical. Reimbursement pathways.
62+
Peer-reviewed citations
5
Stage validation roadmap
FDA
Voice biomarker guidance · cleared PPG precedent
