Product Requirements Document (PRD)
1. Overview
An HF monitoring application that ingests passive/active signals, computes ambient QoL domain scores, and provides clinician dashboards and trial exports. Phase 1 targets KCCQ‑adjunct validation; Phase 2 adds passive signals and trial‑grade pipelines.
2. Users & Jobs
- Patients: Easy weekly check‑in; optional 20–30s speech prompt; background sensing.
- Clinicians: Track trends, receive safety alerts, view domain breakdown; document actions.
- Research/Trials: Export curated endpoints to EDC; lock versions per protocol.
- Ops/Admin: Manage cohorts, permissions, audits, and integrations.
3. Core Features
- Micro Check‑In (Weekly): 6–8 items (Likert + dyspnea/fatigue), 20–30 seconds.
- Speech Capture: Optional short reading or free speech; on‑device feature extraction where possible.
- Passive Sensing (opt‑in): steps, HR/HRV, sleep, home/away, routine regularity, noise/light.
- Ambient QoL Scoring: Five domain scores + composite, with uncertainty bands.
- Alerts: Rule‑based initial policies; tiered (FYI, Review, Urgent); clinician acknowledgement required.
- Clinician Console: Timeline, sparkline trends, last KCCQ vs ambient QoL deltas, alert feed.
- Trial Exports: Subject visits, locked score versions, JSON/CSV + FHIR MeasureReport.
- EHR Integration: SMART‑on‑FHIR launch; read vitals/meds/problems; write PDFs/notes and DiagnosticReport.
- Privacy/Security: Consent, data minimization, redaction; audit log.
4. Non‑Functional
- Availability ≥99.5%; Latency <300ms p95 for API reads.
- Security: Encryption at rest/in transit; secret rotation; RBAC/ABAC; audit.
- Compliance: HIPAA/GDPR principles; DPA/BAA‑ready.
- Model Governance: dataset versioning, drift monitoring, fairness metrics.
5. Metrics
- Patient: weekly completion rate, time‑to‑complete, passives opt‑in rate.
- Clinical: alert acknowledgement time, false‑alert rate, intervention time.
- Outcomes: correlation to KCCQ, hospitalization rates, urgent call reductions.
6. Constraints/Assumptions
- iOS/Android with background tasks; supports major wearables (HealthKit, Google Fit).
- Early versions avoid continuous audio storage; compute features locally when possible.