The following is a guest article by Dr. Peter Hahn, MD, MBA, Dr. Lance Owens, DO, and Joshua Wilda, MPA, CHCIO
Over the past two years, clinical and technical leaders at the University of Michigan Health-West have been deploying and assessing the impacts of AI-based solutions intended to improve organizational efficiencies. We focused especially on systems that use advanced machine learning techniques to address the serious problem of physician burnout related to heavy workloads required to document patient care in electronic health record systems (EHRs).
Healthcare has highly specialized requirements for AI and other technologies. We believe our experiences using a combination of performance metrics and user surveys to measure the impacts of these new technologies can inform similar initiatives in other, less-regulated professions characterized by cognitively demanding person-to-person interactions.
The Imperative to Address Physician Burnout
Like most U.S. healthcare organizations, the University of Michigan Health-West has faced rising challenges: increased demand for services from the population, and intense financial pressures from higher costs and reduced reimbursements. In addition, staffing shortages anticipated prior to the COVID-19 pandemic have worsened due to burnout. One source of burnout is the heavy workload required to document patient-provider interactions.
EHRs have multiple benefits. They centralize patient records and make data readily accessible for tracking patient health over time. They also integrate related digital tools, such as those that flag potentially dangerous drug interactions. Moreover, they are essential for billing and receiving reimbursement for patient care from insurance providers.
However, EHRs have also produced a generation of doctors, nurses, and others who have had to become both caregiver and data-entry clerk. Inputting the details of every patient interaction today has become a necessary task with little or no time available for it. Clinicians have had to either enter notes on a computer during patient visits, or outside of clinic hours.
EHR workloads have become a significant cause of physician burnout. Earlier this year, a group of experts collaborated with the National Academy of Medicine to issue a Healthcare Workforce Rescue Package, including a list of recommendations to reduce the time clinicians spend in the EHR.
Finding practical solutions was a key focus for University of Michigan Health-West. We found that ambient conversational AI technology, which uses advanced machine learning to understand and document both the content and clinical context of multi-party conversations, was best suited to meet our needs.
Testing and Deployment
Since ambient conversational AI solutions were first introduced in late 2019, a growing number of health systems have deployed them in pilot programs for physicians in specific medical specialties. One such solution, Nuance’s Dragon Ambient Experience (DAX), uses a secure application running on the clinician’s smartphone to capture the conversation between the provider, patient, and other visit participants. The conversation is then digitized and consolidated into a comprehensive clinical note. A trained specialist validates the quality of the note, which also helps the AI learn over time, before the clinician reviews it in the EHR.
We were among the first health systems to begin an institution-wide rollout of DAX following a four-month pilot program in 2021-22. The hospital also was the first to apply the technology in primary care settings and has since expanded deployment to 110 clinicians, or approximately half of our employed providers.
Quantitative and Qualitative Assessment
Implementing the technology significantly reduced the amount of time clinicians spent taking notes during patient visits. We asked doctors to report the time they spent documenting care and used EHR “signal data” to measure time spent entering data. One provider estimated that instead of spending 75 percent of his time with patients typing and clicking on the computer, the AI system had reduced that to about 25 percent.
We determined that providers saved an average of 5.1 minutes per patient encounter. The aggregate time savings is substantial when you multiply that by the average of 20 to 30 patients that primary care physicians (PCPs) see each day. Physicians effectively regained up to 2.5 hours each day to see more patients or spend more time with each one. Providers who used to take after-hours time – often at home during so-called “pajama time” – say they now have more time with their families.
User surveys enabled us to capture qualitative physician and patient feedback for additional insights:
1. Clinicians described feeling less stressed and having more time to focus on each patient during appointments. One PCP said eliminating the distractions of notetaking enabled him to ask better questions during patient visits.
2. Several clinicians said productivity increased. One used his newfound time to “formulate a strategy and plan with the patient, discuss treatment, therapy, and educate.” Another noted that his time between appointments became less rushed, enabling him to fill prescriptions and reply to emails and phone calls. That also helped his office function more smoothly.
3. A survey of 232 patients who saw their providers before and after implementation asked them to rate on a scale of 1-5 their agreement with statements including:
-
“my visit felt more like a personal conversation” (average: 4.7)
-
“the provider seemed to be more focused on me during the visit” (4.6)
-
“the provider spent less time typing on their computer” (4.7).
One of the most valuable benefits may be the restoration of the human aspects of healthcare – the eye contact and the “face time” essential for building physician-patient trust ― that were eclipsed by the demands of documenting.
Implementing the Technology in Related Fields
The effectiveness and acceptance of the technology among our patients and staff suggests that it could deliver analogous benefits in other high-stakes professional settings where employee burnout, impersonal customer experiences, and record-keeping have become institutional burdens.
A detailed summarization of conversations captured in real time can reduce cognitive distractions so meeting participants can focus on key points, access pertinent background and historical data, and produce a more complete and accurate record of the interaction.
Based on our experiences, we recommend that professional organizations considering use of ambient conversational AI focus at least as much on how it impacts user experiences as it does on worker productivity, service quality and financial performance. As we found, obtaining both quantitative and qualitative data can provide a more accurate and comprehensive assessment of the ROI of the technology with better actionable insights.
