Project-derived · sanitized

Risk Register (sanitized)

Converts diffuse concern into named, graded, owned, and reviewable risks. The design point is that no risk exists without an owner, a mitigation, and a review date.

What this artifact is for

A risk register turns “we are worried about the model” into a structure an organization can act on. Every entry has a named owner, a mitigation, and a review date. Without an owner, nobody manages the risk. Without a mitigation, it is only a complaint. Without a review date, it gets forgotten.

This risk register is the instrument from the governance pack (v0.2) I developed for my patient-facing cardiometabolic digital twin, aligned with the bounded cardiometabolic use case in my Viewpoint. The entries are a post-deployment projection in the twin’s own domain: the cardiovascular and metabolic risks the system would need to manage once deployed. The twin is pre-deployment, so these are projected risks used to exercise the structure, not figures from a live system. No confidential project or patient detail is included.

Fixed metadata

Field Value
Data provenance Wearable streams (heart rate, HRV, sleep, activity), EHR labs (HbA1c, lipids, blood pressure), validated self-reports, model output logs
Model version Cardiometabolic Risk Model — pre-deployment build (version pinned at validation before any pilot)
Approval owner AI Governance Lead + Patient Safety Committee
Effective date Not yet in force (pre-deployment)
Version v0.2
Last updated July 2026
Known limitation Pre-deployment projection: risk grading reflects anticipated risks and will be revised against real monitoring evidence once deployed

Register

ID Risk Category Likelihood Impact Level Risk owner Existing control Mitigation Review
R1 Model sensitivity for flagging accelerating cardiovascular risk fell from 86% to 72%; patients on a deteriorating cardiometabolic trajectory may be missed Performance / patient safety Medium High High Governance Lead + Clinical Lead Monthly performance monitoring Start drift analysis; add clinician review; assess threshold change 2 weeks
R2 High-risk trajectory flags rose from about 80 to about 150 per week and PPV fell to 30%, producing alert fatigue Workflow / efficiency High Medium High Clinical Lead Clinician feedback channel Tune thresholds; tiered alerts; log handling time 2 weeks
R3 Older patients and women show a higher miss rate, risking inequitable harm Fairness / patient safety Medium High High Patient Safety / Equity Committee No systematic subgroup audit yet Age- and sex-stratified performance audit; restrict use if needed 1 week
R4 The twin surfaced elevated stress and disrupted-sleep signals to the patient as a psychological inference, risking stigma Privacy / stigma Medium Medium-high Medium-high Governance Lead + Clinical Lead Inference-display policy Keep psychological signals as internal features; show trajectory only; patient opt-out 1 week
R5 Continued patient-facing operation during degradation may deliver misleading trajectory narratives Patient safety / trust Medium High High Hospital management + Patient Safety Committee Clinician review Temporarily restrict to advisory; require clinician review before patient-facing output 48 hours

The design principle

Each risk must carry three things: a risk owner, a mitigation, and a review date. The register also separates who identifies a risk, who owns a control, who owns the risk, and who has authority to accept residual risk, so that residual risk is accepted by people with the authority and resources to manage it, not defaulted onto whoever understands the model best.

Citation metadata
Field Value
Title Risk Register (sanitized)
Author Xiaoyan Qian
Public page version v0.2
Publication date 2026-08-03
Stable URL https://xiaoyanqian.com/artifacts/risk-register/
Licence All rights reserved
Suggested citation Qian, X. (2026). Risk Register (sanitized) (v0.2). https://xiaoyanqian.com/artifacts/risk-register/