BUSI 505 Week 7 Discussion Board Forum 4 Example

Discussion Board Forum 4: Emerging Technologies and Generative AI in Clinical Communication

BUSI 505 Healthcare Informatics, Discussion Board Forum 4

Thread by [Student Name]

The health system, its pilot and the classmates are composites; the textbooks, Scripture and the peer-reviewed sources cited are real.

After a vendor demonstration last month, our health system's leaders are preparing to turn on a feature that uses a large language model to draft replies to patients' portal messages. The draft appears in the clinician's inbox, the clinician edits it as needed and the clinician sends it. The vendor presented the tool as a way to save each physician an hour a week and reduce burnout from the growing volume of messages. Before we commit, leadership asked our informatics committee to review what the evidence actually shows and what safeguards we need. Several physicians are enthusiastic, a few are skeptical and the patient advisory council has asked whether patients will be told when software wrote part of a message they receive from their doctor.

Braunstein (2014) describes emerging technologies as promising but warns that their value depends on careful evaluation in real clinical settings, and Nelson and Staggers (2014) emphasize that new systems change workflows, roles and risks in ways designers do not always anticipate. Generative AI is a clear example: it writes fluently, which can make its errors harder to spot, and it shifts the clinician's task from writing to reviewing. Reviewing is a different skill, and research on automation in other fields shows that people tend to trust confident-sounding output more than they should, especially when they are busy.

The early evidence is encouraging but different from the vendor's pitch in important ways. Garcia et al. (2024) evaluated a five-week pilot of AI-drafted replies in primary care and gastroenterology at an academic medical center. Among 162 clinicians, drafts were used for about 20% of replies on average. Contrary to expectations, there was no measurable change in the time clinicians spent reading, writing or acting on messages. However, clinicians reported significant reductions in task load and in work exhaustion, a component of burnout. Crowley et al. (2026) studied 80 physicians, nurses, nurse practitioners and medical assistants across seven clinics and found that replies began from an AI draft about 20% of the time and that 40% of those replies were sent with little or no editing. Two thirds of survey respondents found the drafts useful, and nurses were more favorable than physicians.

Together, these studies suggest the tool may ease the burden clinicians feel more than the minutes they spend, and that a substantial share of drafts reach patients largely unchanged. That second finding raises the stakes for accuracy, tone and accountability. If a draft misstates a medication instruction and is sent unchanged, the error carries the clinician's name and the health system's credibility, and the patient has no way to know a machine wrote it.

I would recommend adopting the tool with four conditions. First, every draft must be reviewed and sent by a licensed clinician, who remains accountable for its content. Second, the informatics committee should audit a sample of sent messages each month for accuracy, tone and safety, with a clear path to report problems. Third, the system should adopt a disclosure policy so patients know when AI helped draft a message. Fourth, success should be measured by clinician burden, patient satisfaction and safety events, not only by time saved, and reviewed by the committee after six months. A message drafted by a machine still carries the clinician's name, and with it the clinician's responsibility.

Paul urged that no corrupting talk come from believers' mouths, but only what is good for building up, so that it gives grace to those who hear (English Standard Version Bible, 2001/2016, Ephesians 4:29). Whether a reply is drafted by a person or a program, the standard for what reaches a patient should be the same: accurate, kind and helpful.

Reply to Monica

Monica, you raised the question of whether patients should be told when AI helped draft a message, and whether telling them might undermine their trust. It is one of the most debated questions in health systems adopting these tools, and policies vary widely. Research suggests the answer involves a real trade-off that health systems should resolve in favor of disclosure. Your concern about trust is well placed, since patient trust is the foundation of the portal relationship, but the evidence suggests that disclosure costs very little satisfaction while concealment could cost a great deal of trust if patients discovered it later. Cavalier et al. (2025) surveyed 1,455 members of a health system's patient advisory group using messages that varied in author, disclosure and seriousness. Participants actually rated AI-drafted responses slightly higher for satisfaction, usefulness and feeling cared for, but their satisfaction dipped slightly when they were told AI was involved. Even so, more than three quarters were satisfied regardless of author or disclosure, and the authors concluded that disclosure should be maintained to respect patient autonomy. In other words, the cost of honesty was small, and patients remained broadly satisfied either way. Barrison et al. (2026) interviewed 30 portal users, stratified by age and education, and found that written disclosure statements often failed to communicate clearly, with nearly half unable to identify AI as the draft's writer, and that disclosures raised concerns about errors, authorship and depersonalization. Most participants nonetheless wanted disclosure, citing their right to know how their care is delivered. Those with the strongest concerns wanted to be told in advance, for example through a portal notice or a conversation, rather than learning about AI from a line at the bottom of a message. That preference suggests disclosure is a process, not a single sentence. Nelson and Staggers (2014) describe transparency as part of trustworthy information systems, and Braunstein (2014) emphasizes that patient engagement depends on trust in the technologies patients use. Both points suggest that a short-term gain in satisfaction from silence would be a poor trade for the long-term credibility of the portal. Paul wrote that he refused to practice deception and instead commended himself by the open statement of the truth (English Standard Version Bible, 2001/2016, 2 Corinthians 4:2), and health systems adopting AI should hold to the same openness. Would your organization test several versions of a disclosure statement with patients before adopting one? The research suggests that clear wording matters as much as the decision to disclose. A patient advisory council could review drafts, and the portal could include a short explanation page describing how AI is used, how clinicians review drafts and how patients can ask for a reply written entirely by their care team.

Reply to Gavin

Gavin, you focused on the finding that many AI drafts are sent with little editing and asked whether that means clinicians are trusting the tool too much. It is the right question to ask, because the safety of the whole approach rests on the assumption that a clinician reads and corrects every draft before it reaches a patient. It is a reasonable concern, sometimes called automation bias, in which people accept a computer's output because reviewing it carefully takes effort. The risk grows when clinicians are tired, behind on their inbox or reviewing many similar messages in a row, which describes many clinicians' afternoons. Crowley et al. (2026), studying a voluntary pilot across seven clinics, found that 40% of replies that began from an AI draft were sent with little or no editing. That could mean many drafts were good enough to send as written, or it could mean some errors slipped through; the study could not tell which, because it measured editing rather than the accuracy of what was sent. Garcia et al. (2024) found no change in time spent on messages, which suggests clinicians were still reading and adjusting drafts rather than simply clicking send, but a five-week pilot is too short to know whether vigilance would last. Attention often fades once a new tool becomes routine, which is precisely when careful auditing matters most. Braunstein (2014) cautions that clinical systems require ongoing evaluation after implementation, not only before it, and Nelson and Staggers (2014) emphasize that safety depends on designing systems that support careful human judgment rather than replace it. Several practical safeguards could reduce the risk, including requiring the clinician to make at least one deliberate confirmation before sending, flagging drafts that mention medication changes, test results or urgent symptoms for closer review and auditing a random sample of minimally edited replies each month, with results shared with the clinicians involved. Feedback from audits should go back to clinicians and to the vendor so that both people and the model improve. Leaders should also watch for drift over time, since a tool that performs well in a pilot may behave differently after vendor updates or as message content changes with the seasons, such as during respiratory virus surges. Proverbs says that truthful lips endure forever, while a lying tongue is but for a moment (English Standard Version Bible, 2001/2016, Proverbs 12:19), a reminder that accuracy in what we tell patients has lasting consequences. Would your organization consider a policy that drafts involving medications or new symptoms can never be sent without edits? Rules like that focus attention where errors would matter most, and they send a clear signal that the clinician, not the software, is responsible for what patients receive.

References

Barrison, P. D., Platt, J., Ackerman, M. S., Friedman, C. P., & Vinson, A. H. (2026). Patient perceptions and preferences for the disclosure of artificial intelligence generated draft replies to electronic messages: A qualitative study. International Journal of Medical Informatics, 217, 106507. https://doi.org/10.1016/j.ijmedinf.2026.106507

Braunstein, M. L. (2014). Contemporary health informatics. AHIMA Press.

Cavalier, J. S., Goldstein, B. A., Ravitsky, V., Bélisle-Pipon, J.-C., Bedoya, A., Maddocks, J., Klotman, S., Roman, M., Sperling, J., Xu, C., Poon, E. G., & Chowdhury, A. (2025). Ethics in patient preferences for artificial intelligence-drafted responses to electronic messages. JAMA Network Open, 8(3), e250449. https://doi.org/10.1001/jamanetworkopen.2025.0449

Crowley, A. P., Hanish, A., Lubken, J., Becker, M., Rosen, C., Moon, J., & Regli, S. H. (2026). Usage of and satisfaction with artificial intelligence-generated draft replies to patient portal messages. Applied Clinical Informatics, 17(3), 510-517. https://doi.org/10.1055/a-2893-7363

English Standard Version Bible. (2016). Crossway. (Original work published 2001)

Garcia, P., Ma, S. P., Shah, S., Smith, M., Jeong, Y., Devon-Sand, A., Tai-Seale, M., Takazawa, K., Clutter, D., Vogt, K., Lugtu, C., Rojo, M., Lin, S., Shanafelt, T., Pfeffer, M. A., & Sharp, C. (2024). Artificial intelligence-generated draft replies to patient inbox messages. JAMA Network Open, 7(3), e243201. https://doi.org/10.1001/jamanetworkopen.2024.3201

Nelson, R., & Staggers, N. (2014). Health informatics: An interprofessional approach. Elsevier Mosby.