Taar1qsub

Executive Summary & Platform Governance

The TAAR-ADAPT Treatment Response Engine (TRE) is a Software as a Medical Device (SaMD) clinical decision support system designed to predict clinical responsiveness to Trace Amine-Associated Receptor 1 (TAAR1) agonist therapy in adult patients with HIV-associated cognitive dysfunction. By fusing genomic, electrophysiological, neuroimmune, and clinical modalities, the system generates probabilistic predictions and dosing tier recommendations.

Device Class

Class II SaMD

510(k) Clearance Pathway

Software Lifecycle

IEC 62304 Class C

Major Level of Concern

Target Discrimination

AUC-ROC ≥ 0.80

Retrospective & Pivotal Validation

Change Protocol

FDA PCCP

FD&C Act Section 515C

Authority Identifier Traceability Matrix

Traceability across Legal Framework (LF), Statutory/Governance (SG), Data Analytics (DA), Procedural (CP), and Scientific (SR) domains.

Authority ID	System Component / Domain	Strategic Function & Technical Scope	Regulatory / Legal Alignment
CP-REG-001	SaMD Classification Engine	Class II 510(k) pathway for moderate-risk decision support	21 CFR 892.2055 / FD&C Act 510(k)
CP-REG-002	Substantial Equivalence Module	Split predicate comparison vs. Watson Oncology, IDx-DR, Neurosteer	21 CFR 807.87 Predicates
CP-REG-003	Intended Use Specification	Decision support for TAAR1 agonist selection in HIV+ cognitive dysfunction	FDA SaMD Guidance
CP-REG-005	Software Lifecycle Pipeline	Major Level of Concern software design, lifecycle, and cybersecurity	IEC 62304 Class C / NIST CSF
CP-REG-006	Clinical Validation Engine	Retrospective (n=200), Prospective (n=30), Pivotal (n=300) clinical protocols	ISO 14155 / GCP Standards
CP-REG-009	Change Control Architecture	Predetermined Change Control Plan (PCCP) bounded retraining gates	FD&C Act Sec 515C / FDA PCCP Guidance
DA-TRE-013	Continuous Learning Infrastructure	Feedback orchestrator (-A), Experiment branch (-B), Performance dashboard (-C)	21 CFR 820.30 Quality System

FDA 510(k) Substantial Equivalence Strategy

Because no single predicate combines genomic, electrophysiological, and neuroimmune data fusion for neuropsychiatric drug selection, TAAR-ADAPT TRE employs a split-predicate regulatory strategy.

Parameter	Candidate: TAAR-ADAPT TRE	Predicate 1: IBM Watson Oncology (K161409)	Predicate 2: IDx-DR (K173671)	Predicate 3: Neurosteer EEG (DEN200034)
Regulation	21 CFR 892.2055 (SaMD)	21 CFR 892.2055	21 CFR 886.1100	21 CFR 882.1400
Intended Use	TAAR1 therapy selection in HIV+ cognitive dysfunction	Oncology treatment selection support	Autonomous diabetic retinopathy detection	Objective cognitive assessment via EEG
Input Modalities	Genomics, EEG P300, Cytokines (IL-6, NfL)	Textual EHR data, pathology reports	Fundus retinal photos	Single/multi-channel EEG signals
Algorithm Engine	Cross-Modal Attention Transformer	Rules Engine + Supervised ML	Deep Convolutional Neural Network	Spectral Processing ML
Lifecycle Level	Major Level of Concern (Class C)	Moderate Level (Class B)	Major Level (Class C)	Moderate Level (Class B)

Predetermined Change Control Plan (PCCP) Compliance

FD&C Act § 515C Approved Architecture

Per FDA Final Guidance on Artificial Intelligence-Enabled Device Software Functions (Dec 2024 / Aug 2025), the PCCP allows pre-specified algorithm retraining without requiring secondary 510(k) filings.

1. Description of Modifications

Feature weight re-calibration on new clinical outcomes
Cross-modal attention fusion matrix updates
Classification threshold adjustments
Prohibited: Changing intended use or adding non-TAAR1 drugs

2. Modification Protocol

≥ 20% out-of-distribution validation samples
Re-validation against locked holdout set
AUC-ROC ≥ 0.80 discrimination floor
Demographic subgroup AUC ≥ 0.75 parity check

3. Impact Assessment

Risk-benefit mitigation under ISO 14971
Safety monitoring for elevated AE rates
Canary rollout (5% traffic) before champion deployment
Automated rollback scripts upon failure

FDA Pre-Submission (Q-Sub) Regulatory Queries

Q-Sub Question 1 (Split-Predicate Approval):

Does CDRH agree that leveraging a split-predicate strategy (Watson Oncology K161409 + IDx-DR K173671 + Neurosteer DEN200034) is acceptable for a 510(k) pathway?

Q-Sub Question 2 (PCCP Re-training Boundaries):

Does CDRH concur that continuous weight optimization via shadow logging and canary rollouts (5% traffic) falls within an approvable PCCP scope without secondary 510(k) filings?

Q-Sub Question 3 (Clinical Trial Endpoints):

Is the 3-phase clinical protocol (n=200 retrospective chart review, n=30 prospective pilot, n=300 pivotal RCT) sufficient to demonstrate clinical validity?

EHR Integration Layer & HL7 FHIR Architecture

The TAAR-ADAPT TRE interfaces with health system EHRs via HL7 FHIR R4/R5 backport guides and CDS Hooks v1.0 event-driven triggers.

1. Clinician Action EHR opens patient record → Triggers CDS Hook patient-view

2. FHIR Data Fetch Adapter pulls Genomics, EEG, Cytokines & MoCA resources

3. ML Inference Cloud enclave runs Cross-Modal Attention Transformer

4. CDS Card Returned Displays RiskAssessment & recommended Dose Tier

TRE Input / Output	Target FHIR Resource	Implementation Guide Profile	Code System / Identifier Mapping
Genomic Variants	MolecularSequence / Observation	HL7 FHIR Genomics IG (v3.0.0)	HGNC: COMT (2228), TAAR1 (11519), CYP2D6 (2625)
EEG Features	Observation	HL7 Observation-Genetics / Device Profile	LOINC: 80628-1 (EEG Study), P300 Amplitude (uV)
Cytokine Biomarkers	Observation	FHIR Core Quantitative Observation	LOINC: 26881-3 (IL-6 Serum pg/mL), sCD14, NfL
Inference Output	RiskAssessment & GuidanceResponse	CDS Hooks v1.0 / Clinical Reasoning IG	LOINC: 75325-1, Prediction Probability (0.0 - 1.0)

CDS Hooks Payload Sample (RiskAssessment) DA-TRE-013-FHIR

{
  "resourceType": "RiskAssessment",
  "id": "taar-adapt-risk-001",
  "status": "final",
  "subject": { "reference": "Patient/PAT-0042" },
  "occurrenceDateTime": "2026-11-04T12:00:00Z",
  "basis": [
    { "reference": "Observation/genomics-comt-val158met" },
    { "reference": "Observation/eeg-p300-amplitude" },
    { "reference": "Observation/cytokine-il6" }
  ],
  "prediction": [
    {
      "outcome": { "text": "Positive Clinical Response to TAAR1 Agonist" },
      "probabilityDecimal": 0.7842,
      "qualitativeRisk": {
        "coding": [{ "system": "http://terminology.hl7.org/CodeSystem/risk-probability", "code": "high" }]
      }
    }
  ],
  "mitigation": "Recommended Dose Tier: Standard (50mg qd). CYP2D6 metabolizer status normal."
}

TAAR-ADAPT TRE

Executive Summary & Platform Governance

Authority Identifier Traceability Matrix

FDA 510(k) Substantial Equivalence Strategy

Predetermined Change Control Plan (PCCP) Compliance

FDA Pre-Submission (Q-Sub) Regulatory Queries

Intellectual Property & FTO Analysis

Freedom-to-Operate (FTO) Competitor Landscape

EHR Integration Layer & HL7 FHIR Architecture

Live Probabilistic Inference Simulator & Performance Analytics

Patient Parameter Inputs

TRE Inference Output

Production Monitoring & Drift Threshold Tracking

Integrated Execution Roadmap (Months 1–42)