NR 588AI Week 2 Technology Governance Brief Example

Technology Governance Brief: Every Body With a Say Over the Readmission Risk Model, What Each May Decide, What Each May Not and Who Can Say Stop

Student Name

College of Nursing, Chamberlain University

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

Instructor Name

Month Day, Year

Technology Governance Brief: Every Body With a Say Over the Readmission Risk Model, What Each May Decide, What Each May Not and Who Can Say Stop

The Tool, the Decision and the Patients

The tool is a vendor's readmission risk model that scores every adult inpatient at our four hospitals daily. The decision it takes part in is which patients join the transitional care list: the 20% with the highest scores each day receive a nurse call within 48 hours of discharge and, if needed, a home visit by a nurse practitioner. The patients it touches are all adult inpatients, about 36,000 discharges a year; those it does not select receive standard discharge instructions only.

The Bodies and Their Remits

System Artificial Intelligence Governance Committee. Approves clinical tools that use predictive models, and their uses. May not: run or configure the tool, or approve its own members' proposals without the proposer leaving the room.

Vice President of Care Management (owner). Runs the tool day to day, sets the list size to match staff capacity and trains staff. May not: change which patient groups the tool applies to, or the threshold rule, without the governance committee's approval.

Clinical Quality and Patient Safety Committee (review). Receives performance reports, including results by patient group, and may recommend changes. May not: change the configuration directly.

System Patient Safety Officer (stop). May suspend use of the tool, returning to the previous selection method, for up to 30 days, pending a decision by the governance committee. May not: end the contract or change the model.

Vice President of Supply Chain with Legal Counsel (contract). Negotiate and sign the vendor agreement and enforce its terms. May not: accept a contract without the conditions the governance committee requires.

Somebody Has to Be Able to Say Stop

The structure has three bodies that can approve or recommend and one that can stop. The patient safety officer can suspend the tool alone, on evidence of harm or of a group consistently missed, without waiting for a meeting. Suspension means the transitional care list reverts to the diagnosis-and-age criteria used before, which remain in the record for exactly this purpose. Without a named stop, every body could approve and none could pause, which is the commonest shape in real organizations.

The Update Problem

The vendor retrains the model about twice a year. A retrained model is, for practical purposes, a new tool. The contract will require 60 days' written notice of any change to the model, its inputs or its intended use. On notice, the governance committee decides whether the change needs a new silent run on our data before use. If the vendor changes the model without notice, the patient safety officer suspends it until the governance committee reviews the change.

Finlayson et al. (2021) describe how changes in the data a model sees, or in the model itself, can degrade performance without anyone noticing; an update rule is how the system notices.

Who May Look Without Asking

The clinical quality analytics team may extract the model's scores, the transitional care list and readmission outcomes at any time, without permission from care management.

The patient safety officer may see any of these and may ask the vendor for its own performance reports directly.

Internal audit may review the governance records annually.

The vice president of care management may not restrict any of this access.

Why the Stop Sits With the Patient Safety Officer

Placing the power to pause with a role that already investigates harm means the tool is treated like any other source of risk. The patient safety officer is used to acting on incomplete information, to deciding quickly and to being questioned afterward. A committee that meets monthly could not act in the days after a pattern of missed patients emerges. A department head who owns the tool would face pressure to keep it running. The patient safety officer has neither problem.

The Groups Most Likely to Be Missed

The governance committee has asked for a bias and equity review before go-live. One widely used algorithm scored Black patients as lower need than equally ill white patients, because its target was spending (Obermeyer et al., 2019). Our tool learns from prior visits, which could produce the same effect for patients with less access. The structure gives the patient safety officer the data and the power to act if that happens.

How the Bodies Fit Together in a Year

In a typical year, the owner runs the tool daily, the analytics team prepares monthly figures, the quality committee reviews them quarterly, the governance committee reauthorizes annually and the contract is reviewed before renewal. The patient safety officer can act at any point in that cycle, without waiting for the next scheduled meeting.

How a Concern Reaches the Structure

Any nurse who believes a patient was wrongly left off the list can report it through the event reporting system with the category "readmission model." Each such report reaches both the safety officer and the care management lead simultaneously.

What This Structure Does Not Cover

It does not cover tools that do not decide who receives care, such as documentation aids, which follow the system's general software policy. It does not decide clinical care for individual patients; nurses and providers remain responsible for those decisions, and may add patients to the list at any time.

Why This Structure, Not Another

An alternative would place everything with care management, which knows the tool best. But care management is also the department that asked for the tool and whose staffing depends on it. Char et al. (2018) note that the interests of those who build or buy a clinical model may not match patients' interests, which is why review and the power to stop sit outside the owning department here.

Gaps

The brief covers one tool; others in the system would each need the same structure drawn around them. The structure was drafted with the governance committee's chair and reviewed by legal counsel, but has not yet been tested in a real suspension. The structure has no patient representative. I recommend the governance committee add one from the system's patient and family advisory council within six months. The brief also does not set how often each body meets; that is part of the oversight plan.

References

Char, D. S., Shah, N. H., & Magnus, D. (2018). Implementing machine learning in health care: Addressing ethical challenges. New England Journal of Medicine, 378(11), 981-983. https://doi.org/10.1056/NEJMp1714229

Finlayson, S. G., Subbaswamy, A., Singh, K., Bowers, J., Kupke, A., Zittrain, J., Kohane, I. S., & Saria, S. (2021). The clinician and dataset shift in artificial intelligence. New England Journal of Medicine, 385(3), 283-286. https://doi.org/10.1056/NEJMc2104626

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453. https://doi.org/10.1126/science.aax2342