MMHA 6601 Week 10 Chatbot Governance Review Example

On the Administrative Side of the Line: A Governance Review of a Patient-Facing Chatbot

Student Name

Walden University

MMHA 6601: Technology and Innovations in Healthcare

Instructor Name

Month Day, Year

On the Administrative Side of the Line: A Governance Review of a Patient-Facing Chatbot

The Bot and the Question

Ridgeway Health Partners, the imagined primary care network that frames this term's work, plans to add a chatbot to its patient portal and website. As proposed, it would answer questions about hours, locations, insurance accepted and how to prepare for common visits, help patients schedule and reschedule appointments and route refill requests to the right practice. The vendor's product can also answer health questions using a large language model, and some leaders want that feature turned on.

This review asks four governance questions: what the bot may do (scope), how patients are told they are talking to software (disclosure), how a conversation passes to a person (handoff) and who is responsible when something goes wrong (accountability). Its central recommendation is to keep the bot on the administrative side of a clear clinical line, at least for its first year. The patient access team expects the administrative scope alone to handle a large share of the roughly 1,800 calls a week that reach practice front desks about hours, directions and scheduling, an illustrative figure that the first months of logs will test.

What the Evidence Suggests

Conversational agents in health care have been studied for some time. A systematic review found that most studies were small, used varied designs and rarely reported on safety, though a few raised concerns about agents giving responses that could cause harm if acted on (Laranjo et al., 2018). Newer language models write fluent, often empathetic answers; one study found that a panel of health professionals preferred chatbot responses to physicians' responses on a public forum for quality and empathy (Ayers et al., 2023). That finding is striking but was based on public questions answered without a patient record, judged by evaluators rather than by outcomes for patients.

The case for strong oversight of such models in health care has been made by researchers who note that they can produce confident but wrong answers and that their behavior changes with updates (Meskó & Topol, 2023). For a primary care network, an answer that is wrong about symptoms carries a very different risk from an answer that is wrong about parking.

Scope

Keeping clinical questions out of scope does not mean they are ignored. The bot recognizes them and hands them off, which is arguably more useful than a fluent answer of uncertain quality.

Table 1

Proposed Scope for Year One

The line is drawn at any request that requires clinical judgment about a particular patient.

Disclosure

Patients should never be unsure whether they are talking to a person. The bot will introduce itself as an automated assistant at the start of every conversation, label its messages consistently and remind patients that it cannot give medical advice when they ask clinical questions. Disclosure also covers data: a short notice will say what the conversation is stored for, how long and that it is not used to train the vendor's models.

Handoff

A chatbot is only as safe as its exit. The review sets three handoff rules. First, any message containing urgent language, such as chest pain, trouble breathing or thoughts of self-harm, triggers emergency instructions at once, before anything else, and offers a direct connection to the nurse line. Second, any clinical question offers a handoff to a nurse during business hours or to the after-hours nurse line otherwise, with the conversation passed along so the patient does not repeat themselves. Third, a patient can type "person" at any time to reach a staff member. Handoffs will be logged, and a sample reviewed weekly to check that urgent conversations were handled correctly.

Accountability

The vendor-update row matters because the underlying model may change without the network's knowledge. The contract should require advance notice and allow the network to test before a new version goes live.

Table 2

Accountability Assignments

Every responsibility has a named owner, and clinical decisions remain with clinicians.

Why Not Turn On Health Answers Now?

Leaders who want the health-answer feature point to the evidence that responses can be high quality. The review does not dismiss that. It recommends against enabling it in year one for three reasons: the network has no way yet to monitor the quality of clinical answers at scale, responsibility for an incorrect answer is unclear when no clinician reviewed it and the administrative scope alone will show whether patients trust and use the bot. The question can be revisited after a year with data from the logs, including how often patients asked clinical questions and how often they accepted a handoff, ideally with a limited, supervised pilot in which nurses review drafted answers before they are sent.

Conclusion

A patient-facing chatbot can reduce phone volume and help patients with routine tasks, but only if its scope, disclosure, handoffs and accountability are settled before launch. Keeping the bot on the administrative side of the clinical line in its first year limits the harm it can do while the network learns how patients use it. The line can move later, deliberately, with evidence.

References

Ayers, J. W., Poliak, A., Dredze, M., Leas, E. C., Zhu, Z., Kelley, J. B., Faix, D. J., Goodman, A. M., Longhurst, C. A., Hogarth, M., & Smith, D. M. (2023). Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Internal Medicine, 183(6), 589-596. https://doi.org/10.1001/jamainternmed.2023.1838

Laranjo, L., Dunn, A. G., Tong, H. L., Kocaballi, A. B., Chen, J., Bashir, R., Surian, D., Gallego, B., Magrabi, F., Lau, A. Y. S., & Coiera, E. (2018). Conversational agents in healthcare: A systematic review. Journal of the American Medical Informatics Association, 25(9), 1248-1258. https://doi.org/10.1093/jamia/ocy072

Meskó, B., & Topol, E. J. (2023). The imperative for regulatory oversight of large language models (or generative AI) in healthcare. npj Digital Medicine, 6, 120. https://doi.org/10.1038/s41746-023-00873-0