Majid Afshar, MD, MS
Associate Professor
Director, Learning Health Systems
Departments of Medicine and Biostatistics & Medical Informatics
University of Wisconsin-Madison
Mary Ryan Baumann, PhD
Assistant Professor
Departments of Population Health Sciences and Biostatistics & Medical Informatics
University of Wisconsin-Madison
Keywords
Artificial Intelligence; Burnout; Provider Well-Being; Learning Health System
Key Points
Documentation in the Electronic Health Record (EHR) is a driver of clinician burnout; it can reduce their capacity to connect with patients during an appointment and spill over into daily life, i.e. “work outside of work.” The research team sought to assess whether the use of an ambient Artificial Intelligence (AI) scribe could improve clinician well-being by automating the documentation process.
The stepped-wedge, individually randomized trial took place over 24 weeks at UW Health. 66 providers across various specialties were randomized to receive access to ambient AI at Week 1, Week 7, or Week 13. The primary endpoint was physician well-being, measured by the Professional Fulfillment subscale and a burnout composite.
The research team found a 20% reduction in burnout and a trend towards improvement in professional fulfillment. These effects were sustained over the course of the trial. Use of the technology also decreased “work outside of work” by an average of 30 minutes and “time in notes” by 21 minutes.
Discussion Themes
Qualitative interviews found that patients of the clinicians in the trial appreciated the reduced computer interaction and felt that communication was clearer and more open.
The research team worked with the EHR company to develop a rubric for evaluating the AI-generated notes for accuracy, i.e. the presence of AI hallucinations or falsifications.
Following the trial’s success, UW Health scaled the project from 66 to over 600 licenses. An operational dashboard was developed for real-time monitoring of utilization and documentation metrics to ensure sustained performance and coding compliance.
Hypertension affects 55% of Black adults, more than any other demographic in the US. Diet is the most important mediator of excess hypertension risk among Black adults, and the DASH diet – which emphasizes low-sodium, heart-healthy items like fruits, vegetables, whole grains, and lean proteins – has been shown to be especially efficacious (albeit in tightly controlled settings).
The study team sought to test whether 3 months of dietician-assisted, home-delivered, DASH-patterned grocery delivery to Black residents of communities with few grocery stores would improve their blood pressure. The comparator group received 3 $500 stipends, one every 4 weeks, for self-directed grocery shopping.
The research team found that the intervention reduced urine sodium, systolic blood pressure, diastolic blood pressure, and LDL-cholesterol. Longer-term maintenance of these benefits will likely require sustained access to healthy groceries and nutrition counseling.
Discussion Themes
The 3.4 mmHg reduction in systolic blood pressure is more modest than the 7–10 mmHg typically expected from first-line antihypertensive drugs. Dr. Juraschek emphasized that GoFresh was a prevention cohort for adults not yet on medication.
The health benefits largely decayed after the active intervention ended. While providing food works, structural barriers to accessing healthy food remain a primary challenge. Ongoing qualitative interviews are exploring the specific barriers and facilitators that affected whether participants could maintain the DASH diet after the study ended.
There are nearly 6 million people living with Alzheimer’s Disease and related dementias (AD/ADRD) in the United States, a number projected to double by 2060. Despite the feact that the highest rates of AD/ADRD are found in Black and Hispanic populations, diversity in Alzheimer’s Trials is limited; between 2000 and 2019, 90% of completed trials had 75% – 100% non-Hispanic White participants.
The Alzehimer’s Prevention Webstudy (APT Webstudy) is a remote registry in which participants, aged 50 and older, take quarterly memory assessments. The research team sought to assess whether financial incentives could increase diversity in the APT webstudy. 45,000 patients were invited to enroll; the first ~15,000 (Arm 1) were invited via message only, Arm 2 was offered a $25 incentive, and Arm 3 was offered the chance to win $2,500.
The study team found that the $25 guaranteed incentives increased enrollment the most, but messages alone were the most cost-effective. 29% of new enrollees were Black or Hispanic patients, an improvement over the APT Webstudy baseline of 5.4%. However, the study team concluded, more needs to be done to enroll people of diverse backgrounds specifically.
Discussion Themes
Attendees considered the ethical implications of the $25 and whether it qualified as coercive when offered to low-income communities. Dr. Jacobson believed that $25 was not enough to be coercive in the context of a low-risk registry study, but that the ethics might differ in a clinical trial involving medication risks.
While higher amounts might increase participation, Dr. Jacobson suggested that messaging frequency may be a more critical factor than increasing the dollar amount.
Ankeet S. Bhatt, MD, MBA, ScM
Cardiologist, Kaiser Permanente San Francisco Medical Center
Research Scientist, Kaiser Permanente Northern California Division of Research
Assistant Professor, Kaiser Permanente Bernard J. Tyson School of Medicine
Adjunct Professor, Stanford University School of Medicine
Keywords
Vaccination; Learning Health System; Implementation Science; Nudges; Influenza; Cardiovascular
Key Points
Every year, influenza leads to over 500,000 deaths and 3-5 million severe cases globally. It increases the risk of cardiovascular (CV) events like myocardial infarction and heart failure. Though health guidelines strongly recommend annual influenza vaccination, rates remain suboptimal globally and persistent inequities exist. There’s an urgent need for novel, effective, and scalable strategies to improve influenza vaccination rates.
The KP-VACCINATE trial is one of the largest ever conducted, randomizing over 3.6 million patients in under 30 days. It included several pragmatic elements, such as coordination with existing vaccine promotion efforts; randomization performed by operational health system teams; and endpoint capture fully embedded in the electronic health record.
The study team assessed the effect of a cardiovascular-focused nudge communication on influenza vaccination rates and found that there was no effect. Despite the negative results, the trial establishes that pragmatic and rapid randomization of communication strategies is operationally feasible at scale with routine healthcare workflows in the US.
Discussion Themes
The study team targeted a larger-than-usual population in order to 1) demonstrate the feasibility of randomization within a large-scale health system, and 2) be well-powered for subgroup analyses that could help tailor future interventions.
Dr. Bhatt viewed the negative result as an illustration of the importance of design and context for interventions based in behavioral science, rather than an indication that nudges are ineffective.
Future directions may include involvement of the broader care team, with primary care providers and specialty providers playing a potentially critical role in nudging patients towards vaccine uptake.
Lee H. Schwamm, MD
Sr Vice President and Chief Digital Health Officer
Yale New Haven Health System
Associate Dean, Digital Strategy & Transformation,
Professor of Neurology and of Biomedical Informatics and Data Sciences
Yale School of Medicine
Kristine Olson, MD, MSc
Assistant Clinical Professor
Department of Medicine, Yale School of Medicine
Daniella Meeker, PhD
Associate Professor of Biomedical Informatics and Data Science, Yale School of Medicine
Chief Research Information Officer, Yale New Haven Health System
Keywords
Ambient AI scribe, Ambulatory clinics, Burnout, Clinical notes, Electronic health record.
Key Points
Physicians in clinic-based ambulatory care spend more than half of the workday documenting patient care in the electronic health record leading to burnout.
Both human and older technology ambient scribes have been shown to reduce burnout in single center studies.
From February to October 2024, researchers from the Yale University conducted a multi-center study in 6 health systems enrolling 263 medical professionals in ambulatory clinics to determine if an ambient scribe system reduced provider burnout.
After patient selection and obtaining consent, the Abridge ambient AI scribe (Abridge AI, Inc) generated a standard medical note of the patient visit while a secure online portal allowed for later viewing and editing. The transcript was imported into the clinic’s EHR note template, after which audio was deleted.
The researchers worked with Abridge to modify an existing client satisfaction survey, including 1 item from the mini-Z burnout metric and 3 items from the NASA Task Load Index, to capture change in physician burnout. Participants completed the survey before and after the 30-day intervention.
Analysis of the survey found that the ambient AI scribe reduced burnout from 51.9% to
38.8% – a statistically significant reduction of burnout by 70% in 30 days. Significant improvements were also seen in the cognitive task load required to complete patient documentation. Focus on patients also increased.
Discussion Themes
The study team approached this project as an evaluation of an existing deployment of ambient AI to provide a measure of the ambient AI’s impact due to the perceived crisis around the burden of the electronic health record (EHR) and the need to increase patient access to providers by freeing up provider time. A step-wedge approach may be the best design approach for a study of this nature.
Ambient AI improves the quality of notes in several ways. The billable level of the notes improves, and the transcript provides transparent auditability.
Certain specialties are not seeing as much impact from the ambient AI assisted notes, such as pediatricians examining a non-verbal young child or baby. The physician in this case must learn to verbalize information they gather during their exam in order for the ambient AI to capture this information in the notes.
Patients may benefit from ambient AI as well. Providers may have greater focus and attention on the patient. Providers may capture more information and be able to provide better care.
Connor Drake, PhD, MPA
Research Health Scientist
Center of Innovation to Accelerate Discovery and Practice Transformation (ADAPT)
Durham VA Health System HSR&D
Assistant Professor
Department of Population Health Sciences
Duke University School of Medicine
Susan Spratt, MD
Department of Medicine
Division of Endocrinology, Metabolism, and Nutrition
Department of Family Medicine and Community Health
Duke University School of Medicine
Abigail Rader, MS
PhD Candidate
Department of Population Health Sciences
Duke University School of Medicine
Keywords
Food Insecurity; Groceries; Food as Medicine; Diabetes; Cardiovascular Health; Cardiovascular Disease; Cardiometabolic Health
Key Points
In 2022, an estimated 12.8% of American households experienced food insecurity (FI): a lack of consistent access to safe, nutritious, or sufficient food for every person in a household to live an active, healthy life. FI is also associated with increased cardiometabolic health risk. While promising interventions to improve food security (and, by extension, cardiometabolic health) exist, methodological limitations such as a lack of pragmatic designs limit conclusions on their effectiveness.
The Eat Well pragmatic trial sought to better understand the real-world effectiveness of a produce prescription program when it came to improving cardiometabolic health-related outcomes and utilization patterns. They found that Eat Well did not improve outcomes among diabetic patients at risk for food insecurity. However, an affirmative outreach approach supported rapid scaling of the program.
Produce prescription programs may require greater duration, dose, intensity, and attention to household and implementation factors, including a focus on different at-risk groups, to improve health outcomes. Reducing cost barriers to purchasing fruits and vegetables alone may not be sufficient to improve food security – at least, not enough to improve cardiometabolic health outcomes.
Discussion Themes
Based on initial descriptive analyses, the amount spent on the card had no significant clinical effect – even among the most adherent participants.
Patients often face multiple overlapping social needs (housing, transportation, etc.), and addressing food alone may not be enough for those with the highest complexity. While the intervention was kept simple for scalability’s sake, diabetes management likely requires a more integrated, multi-sector approach that looks at factors like food, exercise, medications, education, behavioral health support, and monitoring.
Future research should identify opportunities to improve implementation, test interventions in higher-risk populations, and collect additional details on patient-reported outcomes.
Laura Galuchie, BS
Senior Director, Global Clinical Development
Merck & Co, Inc.
Zachary Smith, MA
Assistant Director, Data Sciences & Analytics
Tufts Center for the Study of Drug Development
Tufts University School of Medicine
Keywords
Data; Optimization; Data Collection; Protocol Design
Key Points
The TransCelerate Initiative – comprising a group of pharmaceutical companies with research and development organizations – seeks to identify key considerations in protocol design to optimize procedures and their frequency, while providing tools and a value-based framework for internal evaluation.
Optimized data collection can improve patient and site experience, reduce complexity, enhance trial execution through better design decisions, and maintain (or potentially improve) quality.
A seasoned approach to data collection is timely, as the volume of data is rising (and increasingly exceeds that which is needed). Additionally, recent ICH and ethics updates emphasize fit-for-purpose data and eliminating unnecessary complexity in clinical trials.
The TransCelerate-Tufts Center for Study of Drug Development (CSDD) partnership was borne of the need for continued tangible, actionable evidence to demonstrate the opportunity to optimize data collection.
In 2024, they workshopped a data collection instrument and 14 companies collected and provided data. Tufts CSDD conducted data quality checks to ensure accuracy, validity, and completeness and conducted a comprehensive quality control process. Data analysis took place in early 2025. Endpoints were defined as “core” and “non-core” based on procedure type.
The study sought to quantify the collection and use of non-core and extraneous core protocol data; gather updated benchmarks on the amount, purpose, and impact of data collected in clinical trials; and identify ways to improve protocol design by reducing complexity and easing the burden on sites and participants.
The research team found that the mean number of datapoints collected has exploded in the last decade, up from 930,000 in 2012 to nearly 6 million in 2025. More than 1/3 of all data collected comes from non-core and non-essential procedures.
Non-core and other non-essential procedures contribute to 25-30% of total participants and site burden. Note that there may be other benefits to some non-essential procedures; for example, making sure patients are heard through site questionnaires.
The analysis provides empirical evidence encouraging protocol design discussion and a shift towards more intentional and fit-for-purpose data collection strategies. Planning frameworks and collection assessment tools can reduce unnecessary burden on patients, sites, regulators, and other stakeholders, as well as help sponsors critically assess what data are collected and why.
Discussion Themes
When looking at the factors that contributed to the overcollection of data, the study team found that no one function or department was responsible for the majority of the data points and procedures; the distribution of contributing factors was diffuse. It’s an equal-opportunity problem.
Factors driving non-core data collection included teams’ fear of being asked for data they hadn’t collected by regulators and a lack of on-site experience amongst functional areas. In the latter case, teams are focused on their objectives and lack perspective on how data collection translates into the patient experience, the site experience, or an impact on another function within the group.
In addition to financial costs, there are time costs associated with data collection. The study team found a direct correlation between the amount of data and complexity of a trial – and as complexity increased, the time it took to conduct the trial also increased.
Mark Siedner, MD, MPH
Professor of Medicine, Harvard Medical School
Faculty, Africa Health Research Institute
Keywords
Hypertension; Blood Pressure; Community-Based; Implementation Science; Global Health
Key Points
Hypertension (HTN) is the leading preventable cause of death globally. Dr. Siedner’s research typically revolves around HIV, but he turned his attention to HTN after publishing a study on the convergence of infectious and non-communicable disease epidemics in rural South Africa. Unlike HIV, he noted, HTN control remains poor.
The overarching goals of the IMPACT-BP study were to determine causes for poor HTN control in rural South Africa; co-develop an intervention with partners and end-users to address those causes; and implement and evaluate a novel model of care to improve blood pressure (BP) and increase disease control rates.
They began by designing and determining the acceptability of and conducting a readiness assessment for a community-based hypertension control program. The decision to pursue a community-based care model was informed by decades of successful HIV care programs and innovative HTN care programs.
The program had 3 main elements: Patients monitored their BP at home; community health workers (CHWs) visited patients to collect data, address challenges, and deliver medicines; and nurses managed care remotely with mobile health tools and decision support. Program goals included enhancing patient efficacy and self-empowerment; decongesting clinics and decreasing wait times; and task-shifting away from overburdened nurses.
Once the program had been designed and assessed, the study team conducted a randomized trial to determine its effectiveness. The primary outcome was the change in systolic BP from enrollment to 6 months.
Participants were randomized to 3 arms: Standard of care; “CHW,” which included self-monitoring of BP, home visits and medicine delivery by CHWs, and remote management of BP by nurses; and “eCHW+,” which differed from the “CHW” arm in that BP readings were automatically sent to nurses and the CHWs were less involved.
Though the “eCHW+” arm was slightly more successful, the study team observed 8 – 10mm HG reductions in systolic BP and roughly 30% improvements in BP control in both intervention arms.
This was a multidimensional intervention that sought to address multiple barriers to care. The team faced many real-world challenges, including a community health worker labor dispute, persistent nationwide power outages, destructive weather, and a carjacking spree.
Next, the study team will estimate the fidelity, sustainability, acceptability, and cost-effectiveness of the program. Future directions may include an expansion to multimorbidity care; expansion of the model to urban settings; and transportability to the U.S.
Discussion Themes
When it comes to translating these lessons and insights for care coordination in a U.S. setting, a focus on convenience for healthcare workers and for patients will continue to be crucial.
Though eCHW+ arm was successful, participant feedback indicated that the human element was central to intervention acceptability. Participants felt they were getting a tremendous amount of support from their community health workers, and some expressed anger at the possibility of the intervention ending.
Dr. Siedner noted that he sees the success of the trial more as proof of principle that there are fundamental steps we can take to improve chronic disease care than the unveiling of a one-size-fits-all model.
With a trusted healthcare system and provider providing the right kind of health education, this study demonstrates that you can get people to engage in treatment of an asymptomatic disease.
Daniel Pach, MD
Charité – Universitätsmedizin Berlin
Stefanie Lysk, MS
Charité – Universitätsmedizin Berlin
Newsenselab GmbH (app manufacturer)
Keywords
Decentralized Trial; App; Hybrid Care Model; Digital Tools
Key Points
The 2019 Digital Healthcare Act enabled 73 million Germans in the statutory health insurance system to receive digital applications (AKA apps) on prescription. Apps may be prescribed by physicians and psychotherapists for specific diagnoses and integrated into a patient’s treatment plan.
As a result, the Digital Health Application (DiGA) framework was established. The DiGA framework sets out several requirements: Apps must have a medical purpose, not aimed at primary prevention; must be certified as a medical device; and must demonstrate a positive healthcare effect. There are strict requirements regarding safety, data protection, quality, etc., and integration with hardware (e.g. wearables) is possible.
From December 2020 to March 2022, EMMA – a remote, hybrid trial – was conducted to assess the effect of an app for migraine self-management on migraine days per month. The app included a headache and trigger diary; analysis; self-management techniques, including relaxation exercises, endurance training, and education; and heavier change techniques.
The control arm used a diary-only app, with no capacity for feedback or analysis. Symptom intake was the same as the intervention group.
Though both the control and intervention arms saw an average decrease of roughly 2 migraine days per month, the study team found that use of the app lent no added benefit over the diary-only control. Randomized clinical trials are essential to validate digital therapeutics, and the app was withdrawn from the DiGA directory.
Learnings from EMMA were applied to a new project: the Menstrual Health and Hybrid Care Model for All Young Females (MeMaF). MeMaF has 2 stages: 1) digital self-care: evidence-based self-care instructions, symptom- and cycle-diary, knowledge content, and a medical device. 2) For eligible participants, 6 months of on-site and digital hybrid care: medical care, physiotherapy, nutritional counseling, health psychology sessions, additional app content, and telemedicine.
The researchers concluded that sustainable care for complex chronic conditions will be hybrid, combining digital tools with active involvement of healthcare providers. Researchers need to understand hybrid care in real life and plan trials accordingly – though this increases complexity and demands on clinical trials.
Discussion Themes
The control app may have been too powerful; continuous data tracking may itself act as a powerful intervention.. The intervention app also had more entry points for data, so participants in the intervention arm may have been more likely to report symptoms.
There is a need for rigorous and well-designed control conditions. To ensure valid and interpretable study outcomes you need to understand your intervention well.
A post-hoc analysis of engagement with the app found that 85% of participants used the application on a daily basis. Specifically, they interacted with the trigger diaries and the analysis of their trigger factors; the self-care and self-management features of the application weren’t used as much, especially after the first 4 weeks. Dr. Pach noted that this is a common trend in digital studies.
Steps like user verification through insurance and a video consultation with a physician prevented fraudulent participation in the EMMA trial.
Peter Margolis, MD, PhD
Adjunct Professor of Pediatrics
Stanford University School of Medicine
Emeritus Professor of Pediatrics
University of Cincinnati School of Medicine
Former Co-Director of the James M. Anderson Center for Health Systems Excellence
Cincinnati Children’s Hospital Medical Center
Sean C. Dowdy, MD, FACS, FACOG
Chief Value Officer
Robert D. and Patricia E. Kern Associate Dean for Practice Transformation
Professor, Division of Gynecologic Oncology
Mayo Clinic
Sarah Greene, MPH
Consultant and Senior Advisor
The National Academy of Medicine
Keywords
Learning Health System; Healthcare; Knowledge
Key Points
In service to the goal of establishing a Learning Health System (LHS), the National Academy of Medicine developed and shared a foundational set of shared commitments. These principles, published in 2024, sought to define a common cause for all healthcare workers.
To build on the concept of the shared commitments, consider the LHS from a systems perspective: not as a sum of its parts, but as the product of their interaction. Discussions around the feasibility of LHSs often center around individual parts, e.g. data and informatics, incentives, and culture; but an LHS succeeds when these pieces are built and interact as part of a whole.
So, what does a coherent system look like? A learning organization is an organization skilled at creating, acquiring, and transferring knowledge between parts and at modifying its behavior to reflect new knowledge and insights.
Turning an LHS from an idea to a lived reality involves intertwining infrastructure with an adaptive cycle, propelled by a defined population or system; methods for system change and learning; and measurement and evaluation.
The LHS is not intended as a single program or one-size-fits-all structure; it’s a system that learns at multiple levels of scale, from the individual level to the population level. Dr. Margolis shared an example of a patient-physician collaboration that resulted in an electronic health record (EHR)-integrated dosing algorithm utilized across the healthcare system.
The Kern Center at Mayo Clinic demonstrates how the shared commitments come to life within an organization over time. Founded 15 years ago, Kern brings together diverse experts who create and evaluate data-driven solutions that transform healthcare for patients, clinicians, and communities. It seeks to generate both clinical and practical knowledge, emphasizing practice impact over research.
Roughly 4 years ago, the Kern Center was experiencing an existential crisis; a suboptimal focus was negatively impacting on their reputation within the institution. They pivoted to deep practice engagement and a focus on defining and pursuing clinical practice priorities, and have become an essential piece of practice transformation. Resources like the Project Dashboard and HealthLocator are facilitating communication and the diffusion of practice impact.
Every organization faces a design challenge. Organizing in traditional ways means imposing resource constraints based on assumptions about who can contribute and how. When organizations prioritize the capacity for information to flow freely, however, their learning capacity expands; they can overcome constraints by leaning on the amount, quality, and diversity of expertise available to a network. To illustrate this, Dr. Margolis shared a few examples of LH network successes.
The approach to the LHS has changed and adapted since its inception. Ms. Greene shared 10 reasons they think it will endure, to serve as a high-level roadmap to bring organizational leaders to the table and to help distinguish it from other approaches for translating knowledge into action.
Discussion Themes
In its first decade, Kern was established as a research arm; while it was intended to be transformative, it wasn’t well-integrated with clinical practice. A hybrid model, rooted in research and practice, didn’t work either. Transitioning to a focus on practice only has allowed them to start doing transformative work.
The speakers discussed a couple of facets of patient engagement with LHSs. First, during startup, a community of patients, clinicians, researchers, and other stakeholders often work together to identify measures of success. Second, patients who enter LHSs are consented. However, there is always more work to be done in terms of engaging patients in a mutual exchange of information.
The pursuit of standardization can come into conflict with a system’s ability to innovate. Over time, Dr. Dowdy noted, he’s started giving more weight to the unique circumstances of each hospital and their pursuant need for freedom to innovate. They’ve started emphasizing standardized expectations, measured via the Mayo Clinic Value index, as opposed to standardized methods.
