DHA 733 Week 4 Leading Technology and Artificial Intelligence Adoption Example

Promise, Proof and Bias: A Leader's Framework for Adopting Artificial Intelligence in Rural Hospitals

[Student Name]

University of Phoenix

DHA/733: Contemporary Leadership Issues

Week 4 Assignment

[Instructor Name]

[Date]

The health network, the vendor proposals, pilot results, costs and governance are composites written for a model paper; research findings come from the sources cited.

In the same week, two vendors approached the regional vice president. One offered an ambient artificial intelligence scribe that listens to clinic visits and drafts notes, promising to give physicians back an hour a day. The other offered a sepsis prediction alert for inpatients, promising earlier treatment and fewer deaths. Together they would cost about $650,000 a year. Rural clinicians were enthusiastic about the scribe and skeptical of another alert. The finance director noted that $650,000 would pay for six nurses. This paper develops a framework for deciding.

The Promise for Rural Hospitals

Artificial intelligence holds real promise for small hospitals. Documentation consumes hours of clinicians' days and contributes to burnout, a serious problem where every physician lost is hard to replace. Prediction tools could help small teams notice deterioration earlier, especially at night when one nurse may cover many patients. But rural hospitals also have fewer staff to evaluate, implement and monitor technology, and they cannot afford expensive mistakes. The region has two information technology analysts for four hospitals, no data scientist and a single clinical informatics physician who works part time. Any tool it adopts must be simple to monitor, or it will run unexamined.

Lesson One: Algorithms Can Carry Bias

The first lesson comes from population health. Obermeyer and colleagues studied a commercial algorithm widely used to identify patients for extra care management and showed that a Black patient and a White patient given the same score were not equally ill: the Black patient was, on average, much sicker. The cause lay in what the tool was built to forecast. It predicted future spending, and because the system spends less on Black patients with a given illness burden, it read lower spending as better health. When the team retrained it to predict illness directly, most of the gap disappeared (Obermeyer et al., 2019).

What Bias Means for the Region

About half of the region's patients are Black, and many have faced barriers to care that lower their recorded costs. Any tool that uses past spending or utilization to predict need could direct resources away from the patients with the greatest need. Leaders must ask what each tool predicts and whether that target is fair. The region's own care management program selected patients partly by past emergency visits and admissions, a measure that could miss sick patients who avoid the hospital because of cost or distance. The population health team will review that selection rule using the lesson from Obermeyer's study.

Lesson Two: Vendor Claims May Not Hold

The second lesson concerns performance. Wong and colleagues externally validated a widely implemented proprietary sepsis model among 27,697 patients with 38,455 hospitalizations at an academic health system and reported modest discrimination, an area under the curve of 0.63. Of 2,552 patients who developed sepsis, the model missed 1,709, or 67%, even though it raised alerts on 6,971 hospitalizations, about 18% of the total, a pattern that breeds alert fatigue (Wong et al., 2021). An alert that fires for one patient in five and misses two cases in three teaches clinicians to ignore it.

What This Means for the Sepsis Alert

The sepsis vendor offered performance figures from its own development data. Wong's findings show why independent, local validation is essential before any purchase. A rural hospital with a few sepsis cases a month cannot afford a tool that adds noise without catching cases. Nurses on the region's units already respond to dozens of alarms a shift, from infusion pumps, monitors and bed alarms, and another low-value alert would compete for the same attention. The existing sepsis screening protocol, used by nurses at each assessment, performs reasonably in the region's own quality reviews.

Lesson Three: Ambient Scribes in Practice

The documentation tool has more encouraging early experience from a real deployment. Tierney and colleagues described a large medical group's rapid rollout of ambient AI scribes, reporting that physicians who used the technology spent less time on documentation and found visits more focused on patients, while emphasizing that clinicians must review drafted notes and that the technology required training, monitoring and attention to patient consent (Tierney et al., 2024).

What This Means for the Scribe

The scribe addresses a real burden, the fourth aim of clinician well-being, and early experience is promising. But questions remain: how well it handles Spanish-speaking patients and regional accents, how often notes contain errors, how patients feel about being recorded and whether time saved is real in rural clinics.

A Leader's Framework for AI

The vice president proposed five questions for every AI proposal. First, evidence: is there independent validation in settings like ours? Second, equity: how does the tool perform for different patient groups, and what does it predict? Third, workflow: does it save clinicians time or add burden? Fourth, cost: how does it compare with alternatives, including hiring staff? Fifth, governance: who will monitor it, and what would make us stop? A proposal that cannot answer all five questions is not ready for purchase, however impressive the demonstration.

Governance

A new AI oversight committee, including physicians, nurses, information technology, compliance, an equity lead and a patient representative, will review proposals, approve pilots, review monitoring data and retire tools that do not perform.

Decision on the Sepsis Alert

The region declined to purchase the sepsis alert. It will instead ask the vendor for independent validation data and, if promising, run a silent pilot in which the model scores patients without alerting clinicians, so its accuracy can be measured locally before any alerts fire.

Consent and Privacy

Ambient tools record conversations, so patients must understand and agree. The region will explain the tool in plain language, in English and Spanish, allow patients to decline without affecting their care and ensure recordings are deleted after notes are finalized, as the contract will require.

Decision on the Scribe

The region approved a six-month pilot of the scribe with twenty clinicians, half physicians and half advanced practice clinicians, with a comparison group. The pilot will measure documentation time, after-hours work, note errors found on audit, patient satisfaction and consent rates, and performance for Spanish-speaking patients.

Monitoring After Launch

AI performance can drift as patients, practices and documentation change. Any adopted tool will have quarterly performance reviews, subgroup analyses and a named owner responsible for pausing it if problems arise. Contracts will give the region access to the data needed for these reviews and the right to end the agreement if performance falls below agreed levels, terms the contract standards adopted earlier already require for critical vendors.

Conclusion

Artificial intelligence offers rural hospitals relief from documentation and help noticing illness, but evidence shows that algorithms can carry bias and vendor claims can fail in practice. A framework built on evidence, equity, workflow, cost and governance allowed the region to pilot a promising scribe carefully and decline an unproven alert.

References

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

Tierney, A. A., Gayre, G., Hoberman, B., Mattern, B., Ballesca, M., Kipnis, P., Liu, V., & Lee, K. (2024). Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catalyst Innovations in Care Delivery, 5(3). https://doi.org/10.1056/CAT.23.0404

Wong, A., Otles, E., Donnelly, J. P., Krumm, A., McCullough, J., DeTroyer-Cooley, O., Pestrue, J., Phillips, M., Konye, J., Penoza, C., Ghous, M., & Singh, K. (2021). External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients. JAMA Internal Medicine, 181(8), 1065-1070. https://doi.org/10.1001/jamainternmed.2021.2626