MHA 516 Week 5 Policies to Implement Industry Trends Example

Governing the Algorithms: An Artificial Intelligence Policy That Lets a Public Hospital District Adopt the Trend Without Inheriting Its Risks

[Student Name]

University of Phoenix

MHA/516: Operating in Structure: Health Sector Policy and Governance

Week 5 Assignment

[Instructor Name]

[Date]

The public hospital district, its tools, committee and policy are composites written for a model paper; the federal rule and research findings come from the sources cited.

In April, the chief medical officer of a composite Washington public hospital district learned from a nurse that the sepsis alert in the electronic record had been changed by the vendor without notice, and that a readmission risk score used to assign care managers had never been reviewed by anyone at the district. At the same time, physicians were piloting an ambient artificial intelligence scribe. No one was responsible for these tools as a group. She asked for an inventory and a policy. This paper describes both.

The Trend

Algorithms and artificial intelligence are spreading through health care: prediction models in the electronic record, scheduling and billing tools, imaging software and generative tools that draft notes and messages. They promise earlier warnings, less administrative work and better use of scarce staff. They also carry risks of bias, error, privacy breaches and overreliance.

What Was Already in Use

The inventory, completed in six weeks, found 23 tools: nine clinical prediction or alert tools built into the electronic record, four imaging tools, three revenue cycle tools, two scheduling tools, the scribe pilot, a patient messaging assistant and three tools individual departments had bought on their own. Only five had any documented local review.

The Law

Under the federal civil rights regulation for health programs, an organization receiving federal funds may not let clinical decision tools treat people unequally because of protected traits such as race, sex, age or disability. The organization must keep looking for tools whose inputs include any of those traits and, for every tool it finds, take reasonable steps to lower the chance of unequal treatment (U.S. Department of Health and Human Services, 2026). The district, as a recipient of federal funds, is covered.

Evidence of Bias

Research shows how bias enters through design choices. One commercial algorithm, used across many health systems to choose patients for intensive care management, forecast spending rather than sickness. Since the health system historically spent less on Black patients with equal needs, Black patients given the same score were in fact much sicker. Correcting the problem would have lifted the Black share of patients selected for extra support from 17.7% to 46.5% of those chosen (Obermeyer et al., 2019). The algorithm did not use race; it used cost, and cost carried the inequity for it.

Evidence of Error

Generative tools raise different concerns. In a randomized trial of two ambient scribes, physicians reported that clinically significant inaccuracies occurred occasionally with both products, and the authors called for ongoing vigilance (Lukac et al., 2025).

Policy Part One: Scope

The policy covers any tool that uses an algorithm, statistical model or artificial intelligence to inform clinical, operational or financial decisions or to generate content, whether built in-house, bought or embedded in a vendor product.

Policy Part Two: Inventory

Every covered tool must be registered before use, with its purpose, vendor, data inputs, outputs, users and a named clinical or operational owner. The inventory is updated quarterly and whenever a vendor changes a tool.

Policy Part Three: Review Committee

An algorithm review committee, chaired by the chief medical officer, includes nursing, pharmacy, information technology, privacy, compliance, equity and a patient representative. It approves tools before use and reviews them annually.

Policy Part Four: Risk Tiers

Tools are placed in three tiers. High risk: tools that influence diagnosis, treatment or access to services, such as sepsis alerts and care management scores. Moderate risk: tools that generate clinical content reviewed by a clinician, such as scribes. Low risk: operational tools without direct patient effect. Review depth matches the tier, so low-risk tools move quickly.

Policy Part Five: Bias and Accuracy Testing

High-risk tools must be tested on the district's own patients before use and annually, comparing performance across race, ethnicity, sex, age, language and disability where data allow, and reviewing whether inputs such as cost or use could carry inequity. Moderate-risk tools require accuracy audits of samples of output.

Policy Part Six: Transparency

Clinicians must be told when a tool informs a recommendation, what it is designed to do and its known limits. Patients are told when a tool records their visit or drafts messages to them, and can decline recording.

Policy Part Seven: Monitoring and Enforcement

Owners monitor performance and report problems through the safety system. Vendors must notify the district before changing tools. Tools bought outside the process are removed until reviewed. Staff who bypass the process are addressed through the normal disciplinary process.

Applying the Policy

The committee's first reviews produced results. The readmission score was found to use prior utilization as a key input; testing showed it flagged fewer high-need patients who had limited past access to care, so the district added clinical criteria to care management referral. The sepsis alert's vendor change was reviewed and accepted after testing.

Procurement Changes

The policy changed how the district buys technology. Purchasing now asks every vendor whether a product uses an algorithm or artificial intelligence, requires documentation of how it was developed and tested, including performance across patient groups, and requires notice before any change to the model. Contracts give the district the right to audit performance and to end the contract if a tool fails local testing.

Training Clinicians

Clinicians received short training on how the district's high-risk tools work, what they do not do and how to report a tool that seems wrong. The training emphasized that a tool's recommendation does not replace clinical judgment and that overriding an alert with a documented reason is expected practice, not a violation.

Privacy and Data Use

Tools that use patient data must meet the district's privacy requirements. Generative tools may not send patient information to services without a business associate agreement, and staff may not paste patient information into public chat tools. The privacy officer reviews every moderate-risk and high-risk tool before approval.

Board Oversight

The board's quality committee receives an annual report on the inventory, reviews, testing results and incidents, consistent with its oversight of patient safety risks.

Balancing Speed and Safety

Physicians worried that the committee would slow useful tools. The policy sets service standards: low-risk tools are decided within two weeks, moderate-risk tools within six weeks and high-risk tools within twelve, with a fast track for urgent safety needs. The committee reports its decision times so that delays can be seen and fixed.

Measures

The district will track the share of tools inventoried and reviewed, testing results by patient group, reported errors and the time from request to decision for new tools.

Conclusion

Artificial intelligence arrived at the district before any policy did. A federal rule on decision support tools, research showing how cost-based algorithms can carry racial bias and evidence of occasional errors in AI scribes defined the risks. A seven-part policy, tiered by risk and overseen by the board, lets the district adopt the trend while meeting its legal duties and protecting patients.

References

Lukac, P. J., Turner, W., Vangala, S., Chin, A. T., Khalili, J., Shih, Y. T., Sarkisian, C., Cheng, E. M., & Mafi, J. N. (2025). Ambient AI scribes in clinical practice: A randomized trial. NEJM AI, 2(12). https://doi.org/10.1056/AIoa2501000

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

U.S. Department of Health and Human Services. (2026). Nondiscrimination in health programs and activities, 45 C.F.R. pt. 92. Electronic Code of Federal Regulations. https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-A/part-92