In June, the Michigan Value Collaborative (MVC) hosted two virtual workgroup presentations. The first was a cardiac rehabilitation (CR) workgroup that doubled as a Michigan Cardiac Rehab network (MiCR) telehealth forum focused on virtual group CR programming with a live action demonstration. The second workgroup, health in action, focused on how artificial intelligence (AI) can be utilized to improve follow-up care for patients. The MVC Coordinating Center hosts workgroup presentations once or twice per month, covering a variety of topics including CR, post-discharge follow-up, sepsis, rural health, preoperative testing, and health in action.
Cardiac Rehab – MiCR Telehealth Forum
MVC and the Blue Cross Blue Shield of Michigan Cardiovascular Consortium (BMC2) co-lead MiCR, an initiative to improve CR utilization across Michigan. In April, MiCR launched a new virtual forum series to help programs implement and sustain telehealth services under recently extended Medicare reimbursement policies. MVC’s June workgroup hosted the second MiCR telehealth forum in the series, this time featuring a demonstration of group virtual CR by Henry Ford Health’s Steven Keteyian, PhD, Director of Cardiac Rehabilitation & Preventive Cardiology, and Robert Berry, MS, ACSM-CEP FAACVPR, Clinical Coordinator for Cardiac Rehabilitation. The demonstration was followed by a discussion led by MiCR’s Co-Directors, Jessica Golbus, MD, MS, and Mike Thompson, PhD, MPH.
Telehealth Group Example
The workgroup began with a live mock CR telehealth session led by Berry and Keteyian. Three actors representing patients (played by MVC’s Jessie Souva, MSN, RN, C-ONQS; Emily Woltmann, PhD, MSW; and Rachel Folk, MHA) were guided through a simulated telehealth group CR session.
After presenter introductions, Berry began the group CR class greeting the patient actors, who each represented a specific patient profile and common cardiac condition (Figure 1).
Figure 1. Patient Profiles for Cardiac Rehabilitation Mock Session
To begin the session, Berry checked in with each patient on the Zoom session to document the type of exercise or equipment they would be using. As Figure 2 shows, the group session looked similar to a Zoom meeting with all patients exercising simultaneously from locations they established with their instructor in advance. Each patient confirmed their exercise modality, weight, and heart rate as they engaged in their chosen activity. Throughout the exercise process, patients were asked questions about how the exercise was feeling and whether it was causing any pain or discomfort. If a patient was having symptoms, Berry would follow up with more specific questions to help pinpoint the problem area and would give advice for either modifying the exercise, utilizing intentional breathing techniques, or taking a break.
Figure 2. Screenshot of Mock Telehealth CR Session
The patients had an opportunity to ask Berry follow-up questions before the end of the session such as:
- How will I know when it’s safe to push harder?
- Should I continue with CR exercises if my energy level is low today?
- When/how do I share my tracked vitals and symptoms with my providers?
- How do I get in touch with you between sessions?
- Can I still participate in virtual sessions if I’m at my home in Florida?
Following the live demonstration, Golbus and Thompson led participants in a question-and-answer discussion that further clarified how virtual and telehealth CR programs can work. Keteyian and Berry shared that patients must be at their home address for certain payers (Medicare/Medicaid), but patients with other insurance such as Blue Cross Blue Shield of Michigan (BCBSM) can be at any location within the state of Michigan.
Some attendees wondered about confidentiality considerations during the group session. Henry Ford Health said their programming has not required participants to sign an additional Health Information Portability Insurance Portability and Accountability Act (HIPAA) release form; however, they do remind patients to be aware of what information they share in the presence of others during the group session. For people who may be uncomfortable sharing personal information such as weight, different wording can be used. For example, “has your weight changed since your last visit?”
For patients who are unsure about participating in a virtual setting, Keteyian pointed out that they often have the patient come into the facility for the first session to participate as if they were at home in a virtual environment. This often alleviates any concerns about fully participating in a telehealth setting in the future. For those patients who do not have exercise equipment at home, Berry shared that patients can usually find some form of exercise equipment to borrow from a family member, church, or neighbor, or will choose to walk during their session.
A poll of participants showed the current state of CR programming for sites represented at the forum (Figure 3-4), with most sites indicating an interest in exploring both group and individual telehealth CR.
Figure 3. Polling Question: What format of virtual CR is your site considering?
Figure 4. Polling Question: What best describes your site's status on starting a virtual CR program?
According to the polls, most workgroup participants are still considering CR program development and starting to have conversations with staff and leadership. The MiCR team encouraged participants to take these poll questions back to their teams and leadership to inspire discussion on developing a hybrid or telehealth CR program.
MiCR Telehealth Virtual Forum: June 9, 2026
Health in Action Workgroup – Michigan Medicine
MVC was joined by the University of Michigan’s Alexander Janke, MD, MHS, MSc, Assistant Professor of Emergency Medicine, and Florian Schmitzberger, MD, MS, Clinical Assistant Professor of Emergency Medicine, for the health in action workgroup. Their presentation focused on how AI can be used for clinical feedback and quality improvement workflows in various healthcare settings.
Janke shared that emergency clinicians often make high-stakes clinical decisions when treating patients but then have limited feedback on how their patients do as they move through their care pathway. The rapid patient traffic through the emergency department (ED) continuously pulls these providers into new patient cases, preventing them from having the time to manually review past patient charts. A feedback loop project was intended to help address practice variation, missed learning opportunities, and potential quality blind spots. In preparation for the launch of the project, Michigan Medicine sought approval from the institutional review board (IRB), utilized a large language model (LLM) that is HIPAA compliant, and completed a health information technology services review. They also gathered clinical intelligence committee input, Epic database integration support, and input from the division of clinical informatics.
Methods
Previous approaches to reviewing patient progress included manual chart reviews (time consuming), self-imposed reminders via Epic’s in-basket messaging system, or by hearing about a patient’s follow-up by word-of-mouth. With the introduction of AI as a feedback tool, clinicians would have access to LLMs that can read charts faster than a human reviewer, create tailored summaries for the emergency medicine context, and provide another means for learning by supporting clinicians within their limited time. The AI feedback process follows this flow:
- Flag the case – A clinician marks a patient of interest during the shift via a health education research (HER)-integrated workflow
- Wait – A specified time later—three to 14 days—the platform queries downstream documentation
- AI summary – The University of Michigan generative pre-trained transformer (GPT) Toolkit (HIPAA-compliant LLM) generates a tailored clinical summary with a structured prompt
- Deliver – An email arrives in the clinician’s institutional inbox and includes a patient identifier for recall, an emergency medicine-focused summary, and answers to any free-text questions
- Reflect – The clinician reviews, learns from the outcome, and adjusts future practice enabling the adaptive learning cycle
The process begins with the provider flagging a patient case they would like educational feedback on in the future. Flagging a case is done in the “Disposition View” of the clinician decision tree in Epic (Figure 5). After the waiting period of the patient progressing through levels of care and finally discharging from the hospital, the HIPAA compliant UM GPT will send the provider a summary directly to their institutional email.
Figure 5. Disposition View in Epic MyChart
To validate that the AI summary was providing impactful and meaningful patient information, Schmitzberger and Janke developed a structured validation procedure for 200 AI-generated case summaries measuring for accuracy, conciseness, helpfulness, and completeness. The initial scoring showed that the AI summaries were quite accurate and helpful (Figure 6). This early positive feedback has prompted continued work on developing the use of this program in other areas of the institution.
Figure 6. Structured Validation of 200 AI-Generated Case Summaries
Results
Currently the program includes 170+ unique users including faculty, residents, and physician assistants, who are averaging approximately 40 – 70 requests each week, and 30+ power users (a clinical or other staff member who maximizes Epic electronic health record (EHR) efficiency using advanced tools and navigation).
The top three case follow up themes for requested summaries included diagnosis/etiology (~40%), disposition/course/ICU (~17%), and test/imaging results (~10%). With strong approval in the emergency department and endorsement from other committees, the goal is to expand this feedback program to multiple areas within the institution.
Next Steps & Discussion
Janke presented two pathways in which they would like to proceed with expanding the AI feedback program. The first would be horizontal scaling across the institution, including integration into the graduate medical education program for internal medicine/hospitalist care. Second, they would like to apply their approach to a quality measure from the Michigan Emergency Department Improvement Collaborative (MEDIC).
The presenters addressed participant questions about sharing learnings between providers, patient record privacy, and potential applications in other areas like pediatric care and fall prevention, with participants suggesting connections with the Michigan Hospital Medicine Safety Consortium (HMS) metrics and exploring CEUs as incentives for provider participation.
