--- doc_id: ASTRACBC-OCULAR-EXEC title: Executive Summary - Smartphone-Only Ocular-Biomarker SaMD version: v0.1-draft status: draft owner: Tambua Health Engineering audience: Executive, Product, Clinical, Regulatory device_name: AstraCBC updated_at: 2026-02-15 ---

Executive Summary

1) Objective

Define and operationalize a **smartphone-only, no-attachment** ocular-biomarker SaMD that uses guided eye capture and quality-gated analytics to output **early clinical risk signals** with explicit abstention and confirmatory-care routing.

This package is intentionally execution-focused: it converts the existing AstraCBC architecture into a documentation program that can support productization, validation, and regulated release planning.

2) What We Inherit from AstraCBC

AstraCBC already defines the core safety architecture:

• Guided capture
• Quality gates
• Feature extraction
• Ensemble inference
• Rules + ML arbitration
• Abstain/inconclusive states
• Escalation guidance

It also sets important governance constraints:

• No direct measurement overclaims
• Explicit uncertainty and confidence states
• Conservative failure handling
• Drift and rollback posture

3) Critical Gaps to Close

For ocular SaMD readiness, the current source is incomplete in four areas:

1. No fixed dataset schema and acquisition protocol
2. No quantitative performance results
3. No finalized claim-to-evidence matrix for release
4. No trial-level statistical analysis plan attached to intended claims

4) Phone-Only Pipeline (Locked)

The baseline processing contract for all use cases:

`guided capture -> QC -> feature extraction -> ensemble -> rules+ML arbiter -> abstain or output -> confirmatory next step`

QC gate states (required in all routes)

• `pass`
• `reacquire`
• `abstain`

Output contract (minimum)

• predicted signal or class
• confidence band
• quality status
• escalation timer
• confirmatory recommendation

5) Validation Strategy Summary

Data scope

• Ocular regions: conjunctiva, sclera, pupil
• Fundus without attachments: explicitly high-abstain exploratory mode only
• Label window examples:
• Hb within +/-2h
• BP triplicate protocol
• HbA1c paired reference
• clinician adjudication where required

Stratification

• Device family + OS major version
• Camera capability tier
• Demographic subgroups
• Site-level and device-level holdouts

Robustness simulation

• Illumination variance
• Glare and specular contamination
• Motion and blur
• Compression artifacts
• Ambient-subtraction mismatch

Sample size framing (include in SAP)

• Proportion CI planning: `n ~= z^2 p(1-p)/w^2`
• Mean error CI planning: `n ~= (z*sigma/w)^2`

6) Regulatory and QMS Positioning

Documentation set will be mapped to:

• IMDRF SaMD clinical evaluation framework
• ISO 13485 quality system expectations
• ISO 14971 risk management process
• IEC 62304 software lifecycle evidence
• IEC 62366-1 usability engineering
• ISO 15004-1/-2 and IEC 62471 for eye-illumination safety context
• FDA software, cybersecurity, and PCCP guidance
• HIPAA/GDPR privacy and data governance requirements

7) Prohibited Claims (Must Not Appear)

• "Lab-equivalent for all biomarkers"
• "No confirmatory testing needed"
• "Direct blood counting from phone camera"
• "Universal performance across all phones without qualification"

All external claims must map to the claim-evidence table and pass gates.

8) Delivery Outputs in this Package

• 166-page documentation outline with section-level artifact contracts
• AstraCBC extraction and gap traceability
• Claim-evidence template
• Trial protocol template
• Statistical analysis plan template
• ISO 14971 risk register template
• System architecture and program Gantt diagrams

9) Near-Term Go/No-Go Gates

Release planning should stop if any of the following fail:

• QC failure/abstain rates exceed predefined bounds on holdout devices
• Subgroup performance fails fairness thresholds
• Clinical endpoint precision/sensitivity targets are unmet
• Safety governance controls are not fully testable/auditable

10) Decision

Proceed with dossier authoring using this package as the source of truth for structure and acceptance criteria, while treating all clinical performance claims as provisional until studies close.