SILENT-DS™ Protocol.
Testing Diagnostic Superiority at Hospital Scale.
A prospective, multi-center, observer-blinded cohort study evaluating the diagnostic accuracy of the DeepSensi™ Cognitive Medical Operating System against blinded adjudication and against the physician's own admission note, operating in silent-mode across adult internal-medicine admissions.
SCIENTIFIC METHODOLOGY // NON-INTERVENTIONAL SILENT-MODE HARNESS
Rigorous Observer-Blinded Study Architecture
Diagnostic AI evaluations frequently suffer from benchmark contamination, unblinded retrospective cherry-picking, and clinician confirmation bias. SILENT-DS establishes the first large-scale prospective, frozen-architecture confirmatory study in acute medicine.
Zero Care Interruption
Inside the hospital, on a node with no inbound connection from the internet, DeepSensi reads the EHR record (FHIR R4 / HL7) as standard-of-care clinical notes, vital signs, laboratory panels, and imaging reports are entered; identifiable data never leave the hospital. Treating physicians do not receive or see DeepSensi outputs during active care, ensuring absolute non-intervention and zero observer effect.
Independent Blinded CEAC
Final diagnostic truth is adjudicated by an Independent Blinded Clinical Endpoint Adjudication Committee (CEAC) comprising senior board-certified specialists. The CEAC reviews complete 30-day post-discharge outcomes, biopsy pathology, specialist consults, and autopsy records to determine the authoritative reference diagnosis.
Zero Cloud Egress
All computations execute on sponsor-provided, on-premises sovereign hospital nodes. Zero Protected Health Information (PHI) leaves the hospital intranet.
PRIMARY & SECONDARY OUTCOMES // STATISTICAL ANALYSIS PLAN
Confirmatory Clinical Endpoints (STARD-AI Aligned)
| Endpoint | Scientific Measurement & Metric | Prespecified Analysis |
|---|---|---|
| Top-3 Diagnostic Concordance | Proportion of admissions where the adjudicated CEAC reference diagnosis appears in DeepSensi's top-3 differential diagnosis outputs. | Lower 95% bound above 85%; then non-inferiority (5 percentage points) and superiority against the working diagnoses of the admission note, tested in a fixed sequence. |
| Missed-Critical Safety Event Rate | Proportion of time-sensitive, life-threatening critical emergencies omitted by the system. | Reported with its upper 95% bound (rule of three when no event is observed) and as a paired difference against the admission note. |
| Top-1 Diagnostic Concordance | Proportion of admissions where DeepSensi's highest-ranked diagnosis matches the adjudicated reference diagnosis. | Reported with a 95% interval and as a paired difference against the first-listed diagnosis of the admission note. |
| Structured Refusal (LIMBO Protocol) | Proportion of admissions for which DeepSensi formally declines to rank a diagnosis and lists the investigations needed. | Reported descriptively; no target is set in advance. |
HOSPITAL ONBOARDING // ACADEMIC MEDICAL CENTER PARTICIPATION
Participate as a Study Site Hospital
DeepSensi PBC provides fully funded sovereign compute nodes, EHR integration adapters, and clinical research coordination for qualifying tertiary academic medical centers and health systems. Operating strictly in non-disruptive, silent mode with zero clicks, zero alerts, and zero bedside changes.
SILENT-DS: Hospital Site Participation & Expression of Interest Dossier
Target Audience: Chief Medical Officers (CMO), Chairs of Internal Medicine, Directors of Hospital Medicine, CIOs, and Institutional Review Board (IRB) Leadership.
Zero Clinical Disruption
Operates entirely in the background via read-only EHR feeds. Attending physicians and residents require zero training, receive zero pop-up alerts, and make zero workflow changes. Predictions are cryptographically sealed in write-once storage and evaluated only post-discharge, eliminating bedside malpractice exposure.
Exempt & Pre-Approved
Exempt from Investigational Device Exemption (IDE) under 21 CFR § 812.2(c)(3). Pre-approved for Waiver of Informed Consent under 21 CFR § 50.22 and 45 CFR § 46.116(f), and HIPAA Authorization Waiver under 45 CFR § 164.512(i). Full central IRB reliance (WCG / Advarra) or turnkey local IRB packet provided.
Edge Gateway & Coded Data
Deployed as a Plug-and-Play DeepSensi Edge Gateway (lightweight VM or appliance) within the hospital data center to handle local data anonymization. Strips all direct identifiers on-premise; only Coded Clinical Data leaves the hospital firewall via mTLS 1.3 to the DeepSensi Supercluster for passive inference. Zero inbound ports, zero EHR database modification.
Premier Co-Authorship
Site Principal Investigators and hospital co-investigators receive prominent co-authorship on high-impact multi-center publications reporting diagnostic accuracy against standard hospital care, adhering strictly to STARD 2015 / STARD-AI reporting guidelines. Fully sponsored infrastructure provided at zero cost.
Three-Phase Hospital Onboarding Roadmap
Phase 1: Governance & IRB (Weeks 1 to 4)
Execution of standard institutional Data Use Agreement (DUA) and Study Site Agreement; Central IRB reliance agreement execution (WCG / Advarra) or local hospital IRB approval; designation of Site Principal Investigator.
Phase 2: IT Provisioning & Feed Validation (Weeks 5 to 6)
Provisioning of on-premise locked Edge Node appliance or secure VM; activation of read-only HL7 v2 / FHIR feeds; automated validation across 200 consecutive test admissions (95% data completeness benchmark).
Phase 3: Silent Run & Adjudication (Months 2 to 14)
Consecutive prospective accrual (typically 1,500 to 2,500 adult internal medicine admissions per site); post-discharge clinical chart review by designated site clinician adjudicators; cryptographic matching and aggregate statistical reporting.
Submit Institutional Expression of Interest:
DeepSensi PBC · Division of Clinical Investigations & Health System Partnerships
Email: [email protected] | [email protected]
Direct Partner Line: +1 646 210 1916 · Dover, Delaware, United States
DeepSensi DCT
