NR 588AI Week 5 AI Oversight Plan Example

Oversight Plan: Independent Review of the Readmission Risk Model by the Patient Safety Officer, With Powers, Limits, Annual Reauthorization and a Named Recusal

Student Name

College of Nursing, Chamberlain University

NR 588AI: Advanced Nursing Leadership in Artificial Intelligence-Integrated Complex Systems

Instructor Name

Month Day, Year

Oversight Plan: Independent Review of the Readmission Risk Model by the Patient Safety Officer, With Powers, Limits, Annual Reauthorization and a Named Recusal

Independence Before Interval

Oversight of the readmission risk model sits with the System Patient Safety Officer, who reports to the Board Quality Committee rather than to any executive with a budget for the tool. The patient safety officer did not propose the tool, does not manage the transitional care program and has no performance goal tied to it. The tool's sponsor in care management receives the oversight reports but does not prepare, approve or edit them.

Powers

The patient safety officer may:

Obtain, without asking anyone, the model's daily scores, the transitional care list, the overrides recorded by nurses and readmission outcomes, through the clinical quality analytics team.

Require the vendor to provide its quarterly subgroup performance report and any change notice, under the contract terms negotiated in November.

Require the attendance of the vice president of care management, the analytics lead or the vendor's account manager at any review meeting.

Suspend use of the model for up to 30 days, returning the list to the previous diagnosis and age criteria, on evidence of harm, a breach of the equity floor or an unannounced model change.

Limits

The patient safety officer may not end the vendor contract, change the model's threshold or the equity floor, or direct care for individual patients. A suspension longer than 30 days requires the governance committee's decision. These limits keep a single role from canceling a system-wide agreement alone, while leaving it able to act quickly.

What Is Watched, and How Often

Monthly, prepared by the analytics team by the 10th working day:

List size as a share of discharges, by hospital.

Capture rate overall and for the seven groups in the equity review, on a rolling three-month basis.

Nurse additions to and removals from the list, with reasons.

Event reports filed under "readmission model."

Quarterly:

The vendor's subgroup report, compared with our own figures.

Calibration by score band.

A summary to the Board Quality Committee.

Annually:

A full repeat of the bias and equity review.

Gaube et al. (2021) found that clinicians can follow algorithmic advice even when it is flawed, so the monthly reading of nurse additions is watched for a fall that might mean nurses have stopped questioning the list.

Continuation Is a Decision

The tool's approval lasts one year. Each December the governance committee must decide to reauthorize it, change it or end it. Without a reauthorization vote, approval lapses and the list reverts to the previous criteria on 1 January. This prevents the tool from continuing simply because nobody decided to stop it.

A Positive Statement Each Cycle

Each quarter, the patient safety officer signs a short statement to the Board Quality Committee: that the model's capture rate for every monitored group is within 8 points of the overall rate, that the vendor has reported as required and that no unannounced change has occurred. If any part cannot be stated, the statement says which and what action was taken. A report that says nothing is wrong is weaker than one that says what was checked and found right.

Name Who Steps Out

The vice president of care management steps out of governance committee votes on reauthorization, suspension and the equity floor, because the decision affects the department's staffing and budget. The committee member who previously consulted for the vendor also steps out of all votes on this tool. Both may present information before the vote.

Between Meetings

The analytics lead sends the monthly figures to the patient safety officer and the vice president of care management on the same day. If any monitored group trails the overall capture rate by over 8 points in two consecutive months, the patient safety officer decides within five working days whether to suspend.

What the Patient Safety Officer Checks Each Month

Four questions, answered in writing each month: Did the list size stay within the team's capacity at every hospital? Did any monitored group's capture fall below the threshold? Did nurse additions rise or fall sharply, and if so, why? Were any event reports filed under the readmission model, and what did they show? Rajkomar et al. (2018) recommend that fairness be monitored after deployment, not only checked before it, since performance by group can drift as populations and practices change. These four questions are how that monitoring happens in practice.

Why One Year

A tool approved indefinitely tends to stay because stopping it requires someone to argue against it. A one-year approval reverses that: the tool stays only if someone argues for it with evidence. The December vote uses the year's monthly figures, the quarterly statements and the annual equity review, so the committee decides on what the year showed rather than on what was promised at purchase.

When the Patient Safety Officer Is Away

A deputy, the system's associate patient safety officer, holds the same powers during any absence longer than five working days. The deputy is named in the governance committee's minutes.

Cost

The analytics team estimates eight hours a month for the reports, and three days a year for the annual review. The patient safety officer's time is about four hours a month. These are absorbed in existing roles; no new position is needed.

How the Board Learns of a Suspension

If the patient safety officer suspends the tool, the Board Quality Committee chair is told within two working days, with the reason and the expected length. The committee does not need to approve a suspension of up to 30 days, but it should never learn of one from another source.

Outside This Plan

This plan leaves out the quality of the transitional care calls themselves, which the care management program measures separately. Other predictive tools in the system each need their own plan. Nor does it replace the vendor's own monitoring, which the contract requires the vendor to report.

Review of the Plan

After the first year, the Board Quality Committee reviews whether the plan worked: whether reports arrived on time, whether the positive statement was ever qualified and what happened as a result.

Sendak et al. (2020) proposed that each clinical model carry a short label of its intended use and limits; the patient safety officer will maintain one for this model and attach it to each quarterly statement.

References

Gaube, S., Suresh, H., Raue, M., Merritt, A., Berkowitz, S. J., Lermer, E., Coughlin, J. F., Guttag, J. V., Colak, E., & Ghassemi, M. (2021). Do as AI say: Susceptibility in deployment of clinical decision-aids. npj Digital Medicine, 4, 31. https://doi.org/10.1038/s41746-021-00385-9

Rajkomar, A., Hardt, M., Howell, M. D., Corrado, G., & Chin, M. H. (2018). Ensuring fairness in machine learning to advance health equity. Annals of Internal Medicine, 169(12), 866-872. https://doi.org/10.7326/M18-1990

Sendak, M. P., Gao, M., Brajer, N., & Balu, S. (2020). Presenting machine learning model information to clinical end users with model facts labels. npj Digital Medicine, 3, 41. https://doi.org/10.1038/s41746-020-0253-3