How to read this source
Start with headings and summary blocks. Use this page to verify citations, claim wording, and chapter-level evidence support before interpreting conclusions.
Internal Research Source
ISO 14971-aligned risk register template covering hazards, harms, controls, verification evidence, and residual risk.
How to read this source
Start with headings and summary blocks. Use this page to verify citations, claim wording, and chapter-level evidence support before interpreting conclusions.
Used by chapters
Share
In plain language: this template is a structured evidence table. Use it to map claims, thresholds, risks, and pass/fail criteria consistently.
| risk_id | hazard | hazardous_situation | harm | severity | probability | initial_risk | control_measures | residual_risk | verification_method | owner | status |
|---|---|---|---|---|---|---|---|---|---|---|---|
| R-001 | Low-light capture misread | User captures in dark scene; model outputs non-abstained result | Delayed clinical follow-up | High | Medium | High | Hard low-light QC gate; mandatory reacquire; abstain fallback | Low | QC unit tests + field simulation + audit log review | Safety Engineering | Open |
| R-002 | Glare-driven feature corruption | Specular reflections on sclera/conjunctiva pass weak QC | False positive risk signal | Medium | Medium | Medium | Glare detector + ROI mask + multi-frame consistency checks | Low | Synthetic glare stress suite + prospective validation subset | Computer Vision | Open |
| R-003 | Device pipeline shift | OS/camera update changes tone mapping | Performance drift | High | Medium | High | Device compatibility matrix; model monitoring; rollback controls | Medium | Post-market drift dashboard + release gate checklist | MLOps | Open |
| R-004 | Demographic bias | Model underperforms in subgroup | Inequitable missed risk | High | Medium | High | Subgroup thresholds; fairness gate in release criteria; abstain tightening | Medium | Subgroup performance report in every model release | Clinical AI | Open |
| R-005 | Unsafe wording | UI text implies definitive diagnosis | Inappropriate self-treatment | High | Low | Medium | Content lint checks; legal-approved claim library; UI copy tests | Low | Pre-release content audit + legal signoff | Product + Legal | Open |
| R-006 | Insufficient incident handling | High-risk failure not escalated quickly | Patient safety impact | High | Low | Medium | Incident response SOP + on-call escalation + rollback trigger | Low | Tabletop incident drills + postmortem CAPA | Operations | Open |