Since 2023, Sutter has been addressing the provider burnout epidemic and the burdensome administrative tasks detracting from the time and focus clinicians dedicate to patients. Dr. Veena Jones, Chief Medial Information Officer, Sutter Health explains "by leveraging generative AI, we’re reducing these burdens, allowing clinicians to focus on meaningful, personalized interactions with their patients while enhancing the overall care experience and the clinicians’ joy of work. It’s truly a win-win for both providers and patients.”

Sutter Health, a non-profit integrated delivery health system providing comprehensive care to nearly 3.5 million patients throughout California, is one of the early health systems thoughtfully exploring ways to test AI-driven solutions to support patient and provider needs.  With a focus on responsible implementation, Sutter is committed to leveraging AI in a way that enhances care while maintaining the highest standards of security and trust.

Specifically, Sutter has invested in AI & GenAI for two key initiatives as “tools for our providers tool kits”:

     1) Supporting Patient Messaging within Epic, Augmented Response Technology (ART) – The ART tool analyzes patient messages, gathers relevant information and drafts a response for the clinician to review, edit and approve, before sending to patients.  Integrated within Epic, ART uses a large language model (LLM) to generate a contextually relevant response for clinician review.

       In Q4 2023, Sutter was one of the first handful of health systems that began piloting this tool.“Our early work with Epic and other leading health systems to pilot the ART tool was instrumental in advancing its development,” said Dr. Jones. “Since October 2023, ART has been in production within the Sutter EHR, following a collaborative effort with primary care clinician champions and patient advisors to help validate and refine its capabilities.”

       Sutter started testing ART with about 70 physicians and advanced practice clinicians to gather feedback on messages generated from the LLM and adjust the prompts before the March 2024 launch. Beyond clinician feedback, the team collaborated with patients for feedback on GenAI generated message response.

      "It was vital to include perspectives from patients in our Patient Family Advisory Council (PFAC) program," said Dr. Jones. "By reviewing message responses with patients, we gained invaluable insights—ensuring that the technology supports clear, compassionate, and meaningful communication. Patients are at the heart of our mission and it’s essential to align AI advancements with the human-centered care we strive to deliver.”

       In March 2024, Sutter launched ART to all primary care clinicians in Internal Medicine and Family Medicine, since they were on the receiving end of the most patient inbox messages. Soon after, ART was made available to all pediatricians and then was expanded to OB/Gyn and geriatric clinicians across the organization.

       Sutter recognized the need to adjust the LLM prompt to ensure relevance and benefit with each specialty rollout, working closely to test ART before making it more widely available. By the end of 2024, all Sutter clinicians receiving patient messages in their in basket had ART-generated draft replies available for their use.  

     AI PATIENT MESSAGING SUCCESS:

     “We started the ART project with the goal of saving providers time when responding to patient emails. What we heard overwhelmingly was that providers want to keep using it because it significantly reduces their cognitive burden”, shares Laura Wilt, Sutter’s Chief Digital Officer. “Think about the level of effort required for a doctor to create a patient response email from scratch versus simply reviewing and editing a pre-drafted message.”

·        To date, 3,000+ Sutter clinicians have used ART to assist with responding to medical advice requests from patients.

·        With over 217,000 messages sent, feedback from Sutter clinicians has been invaluable in refining the prompt engineering process to enhance the quality of draft responses.

·        Clinicians are reporting an average time savings of up to 20 percent when using ART to help manage their patient communications, along with a noticeable reduction in cognitive load and mental fatigue.

Qualitative Provider Feedback:

“I like the empathy and the AI component that reviews the chart and gathers information like last labs, or pending labs. That is very helpful.”

“I like that the responses are often more thorough and considerate than I would have time to write.”

“Includes empathy statements and acknowledgement of patient's symptoms. Really good for quick and straightforward questions.”

“Appreciate the ART-generated responses to long patient messages. Reduces mental workload by having a template initially started.”

2) Ambient Listening for Visit & Follow- up with Patient Friendly Summaries. Sutter has been working closely with Abridge which has a generative artificial intelligence (AI) platform for clinical documentation. It records and transcribes a visit and generates components of the clinical note. In addition it creates a patient-friendly summary of the visit which the provider can paste into their After Visit Summary.  

To date, most clinicians using Abridge have seen a reduction in documentation time and cognitive burden associated with their work. The ambient listening technology allows for more personal connection and focus on the patient interaction, rather than on the computer, thus strengthening the care relationship.    

“Another benefit we are seeing with Abridge’s ambient listening technology to personalize care is enabling providers to conduct the visit in the patient’s native language, while the English documentation is created in real-time. We have patients and clinicians who speak a variety of languages, from Spanish to Cantonese. The mental load is high for these clinicians having to translate everything they discuss into English for the note. Abridge takes care of the translation and puts the note directly into the EMR”, explains Wilt.

From April- December, Sutter enabled 1,000+ ambulatory physicians and advanced pratice clincian users with the ambient technology. Sutter learned about the importance of physicians being able to directly access Abridge through Epic’s Haiku app, having it fully integrated to make it easier to have their work all in one place. Wilt adds “we took a thoughtful approach to implementation, aiming for a smooth adoption. Our Digital Academy Team played a key role in supporting the rollout and ensuring physicians had the resources they needed. It went smoothly, with minimal training needed— just a brief eLearning video to watch. From a physician adoption perspective, it took off like a rocket ship… maybe because it takes tasks away and lets providers focus more on connecting with patients.”

AI AMBIENT LISTENING SUCCESS:

“Again, we initially focused on time savings, and for many clinicians it’s a significant time saver. But we’ve found that —it also helps reduce cognitive burden”, Wilt adds.

      Qualitative Provider Feedback: 

“Abridge has been a life-changing experience. It alleviates 50% of the charting work for me,” said Alice Woo, M.D., a plastic surgeon with Sutter West Bay Medical Group. “[Now] I feel like my conversations with patients are much more intimate and therapeutic.”

“Abridge has changed how I feel at the end of the day,“ said Kevin Chen, M.D., a family medicine physician with Palo Alto Foundation Medical Group. “If you have balance in your life, you can do other things. It makes you a better doctor when you’re in the clinic seeing patients—you’re better centered.”

EXPANDING AI USE CASES:

“We see the patient message response/ART in Epic and Abridge Ambient listening platform as examples of how generative AI technology can help Sutter personalize medicine”, Wilt shares. 

Patient Message in EPIC/ART. Sutter is interested in moving beyond the patient generated messages into other message types such as “results” (i.e. normal results) and “medication” (i.e. medication refill requests). Sutter is also planning a pilot with Epic for another feature to help further streamline physician workflows with “Chart Summarization”, which provides a summary of key patient information within their electronic medical record. 

Ambient Listening/Abridge: Sutter is extending use of the ambient listening tool in a pilot with emergency and inpatient physicians across all Sutter hospitals. They see opportunities with other types of clinical staff, such as with nurses and physical therapists, and look forward to exploring these as the technology evolves. 

Sutter is excited about the continued evolution of Abridge which will "go beyond capturing the conversation and generating a note, to queuing up clinical orders that were discussed. Wilt explains, “so if your doctor mentions getting an Xray, the order would be put in for the physician to just accept, instead of having to type in the order.” 

“AI is all about continuous improvement—what was implemented a few months ago has already evolved, and it will continue to improve over time," said Dr. Jones. She encourages Sutter teams and other health systems to approach these tools with curiosity. "When implemented thoughtfully, AI has the potential to not only reduce administrative burdens but also strengthen the connection between providers and patients, enabling more meaningful care experiences."

MY TAKE: AI INNOVATION BEST PRACTICES

Sutter has invested in the right practices and processes to drive success:

1. Collaborated closely and early with vendors (e.g. Epic, Abridge) to define and pilot AI and GenAI capabilities.
2. Co-created with key clinical and patient stakeholders to ensure a strong experience.
3. Integrated with exisiting workflows spanning development, training, care delviery and support. 
4. Focused on both qualitative and quantitative success measures including the voice of the clinician and patient (e.g. feedback) and user trial/usage/traction. 

During the HIMSS AI for Healthcare Forum in Boston, Dr. Eve Cunningham, MD, MBA, GVP, Chief of Virtual Care and Digital Health at Providence Health presented a few AI use cases including an impactful example which empowers PCPs for better care decisions.

Today primary care doctors are pressured to address patient health concerns in an average of 18 minutes with an average of 220,000 data points in the patient’s electronic medical record. With over one in 3 patients referred to a specialist each year, PCPs are tasked with diagnosing the health issue to and making decisions about the need for a specialist.

Barriers to Specialist Care Decision

With vast amounts of patient data that is not clinically organized around context within the patient’s chart, PCPs must spend time pulling together and reviewing relevant information to support the patient’s health problem. Then the PCP must determine if a specialist is needed, and if so which type of specialist.

Given the shortage of primary care doctors and specialists, optimization of the referral process is critical to access to care. However, the referral process today is both inefficient and ineffective causing problems:

• Patient did not need to go to a specialist. (Note: About 20% of patients referred to specialists do not need to see a specialist at all)
• Patient is sent to wrong specialist and/or at inappropriate level of urgency.
• Patient arrives to specialist without any prereferral work-up such as lab, imaging and first-line therapy for a condition.
• Patient has delayed care until there is a specialist appointment

Accelerating Primary Care Referrals

The September issue of NEJM Catalyst’s article on “A Physician-Created Platform to Speed Clinical Decision-Making and Referral Workflow” describes the problem and process that Providence Health has undertaken to conceptualize and build their MedPearl Platform.

“As primary care clinicians struggle to access the expanding base of specialized knowledge and experiences, they attempt to refer patients to specialists, often without the advantage of usable tools at the point of care.”

Developing this needed platform entailed using a “Human- Centered Design approach with a Jobs to Be Done framework, working closely with our clinicians and UX designers”, explains Dr. Cunningham. “We began by asking clinicians how this platform needed to work and look, and got input from 270 clinicians in the design process, which was essential for gaining buy-in”.

MedPearl Platform Built by & for Clinical Care Decisions

Based on clinician’s requirements, Providence built MedPearl as an application within their Epic electronic medical record (EMR) to integrate a digital assistant with clinical knowledge to support  primary and urgent care clinicians. MedPearl has “concise, human-authored algorithmic primary and subspecialty care next best actions on over 700 conditions”.

Clinicians utilize MedPearl’s AI algorithms to pull up and review essential clinical information alongside contextualized patient data to see next best actions with what has/hasn’t been done for the patient, empowering clinicians to optimize the “work up” while reducing redundant diagnostics.

Since it can take weeks to months for the patient to get in to see the specialist, the MedPearl platform enables PCPs and urgent care providers to take care of the diagnostics in the interim to optimize the care referral decision. With MedPearl, clinicians can see the testing stream, make sure tests do not need a referral, find the most appropriate specialists and triage the urgency of the specialist appointment based on patient history and diagnostics. Think about what this means for rural health where accessing a specialist is even more challenging.

Providence has designed MedPearl with icons to reduce the clinicians cognitive reading load and provides links to guidelines and QR codes to scan for evidence -based applications.

Clinician Experience with MedPearl

Providence Health has deployed MedPearl across their health system which is available in Epic’s Marketplace.

Dr Cunningham as MedPearl founder comments “Our clinicians tell us that with MedPearl they are spending less time in the EMR, fewer outside work hours and are more productive. To date, we have 7,000 clinician users, making 30,000 searches per month on the MedPearl platform”.

During the initial pilot with 216 participants conducting 14,000 searches, clinicians reported the clinical content helped with clinical decision support: they did not need to refer the patient 20% of time, improved care plan and work-up 72% of the time and changed referral specialty and/or level of urgency 20% of the time.  

With MedPearl, clinicians are also realizing better resource efficiency. “We had a challenge with some of our more newly onboarding clinicians overwhelming our Medical Directors with clinical questions and curbside phone calls. The call volume   decreased significantly when these clinicians were advised to reference MedPearl first for answers”, shares Dr. Cunningham.

Clinicians often walk through the MedPearl tool with their patients. They review educational content together, discuss the algorithms and use the decision support tools. 

Feedback from Clinicans Using MedPearl

After using the MedPearl tool, clinicians commented about the ease of use, quick access to the most essential patient data and benefits of leveraging this patient data for the physician, practice and patient.

   “MedPearl provides concise and focused clinical evidence -based summaries, making it easier to quickly access the most relevant information. Its streamlined format saves time and effort, ensuring health professionals can find information without being overwhelmed by unnecessary details”. Medical Director Urgent Care 

   “The cognitive offloading and reassurance that knowing the Prov approved standard of care has been helpful for me, and I love being able to share it with newer clinicians”. Urgent Care Physician

   “The Headache algorithm – perfect example of how to reduce variation in our urgent care practice.” Urgent Care Physician

   “With the demand for neurology consultation greatly exceeding the supply of neurologists for the foreseeable future, MedPearl improves access when it has never been more vital.” Chief of Neurology

   “MedPearl is a point of care tool we need to augment our clinical acumen and to help drive uniformity of patient care path the best -practice level. This allows primary care physicians to provide more care value, improving diagnostic testing, treatment, and referrals and in opening more specialist access.”  Medical Group Director

   “Recommendations re: referrals and specific imaging with and without contrast very helpful. Truly has improved physician satisfaction and flow for the day”.  Family Medicine Physician 

   “Contextualization is pretty cool. Makes Pearl useful in multiple ways.. Quick way to see all relevant data about a work up for a particular topic.” Internal Medicine Specialist

   “Love the patient experience, engagement and shared decision-making components of this tool.”  Practice Senior Director of Operations

   “I love the headache algorithm. During a recent appointment, I had a patient point to each symptom that applied to her.”  Physician

Future Direction of MedPearl for Health Systems

Dr. Cunningham describes Providence Health’s approach with “feedback and iteration as a core principle by which we drive our MedPearl features roadmap and design. We also push our end users to think beyond the workflow of today, and envision workflow of tomorrow.”

The MedPearl team continues to collaborate with users to gain insight into future enhancements. Specifically, they are focused on these platform innovations:

1. "Expanding contextualization beyond labs, imaging and procedures to further enrich the context aware data aggregation experience"

2. "Adding documentation support features to expedite documentation workflow"

3. "Further refine home page to create a more dynamic experience to delight end users"

4. “Since we believe all workflows will converge on ambient technology, we are preparing ourselves to integrate into ambient workflows. We don’t plan to be the ambient provider, rather we envision ourselves as a clinical intelligence plug-in to further enrich that experience.”

Dr. Cunningham confirms plans to begin “deploying MedPearl to other health systems using EPIC and then Oracle EMRs. We see ourselves as ‘headless’ clinical decision support system that is EMR agnostic and application agnostic. We want to permeate and spread the goodness of MedPearl in a seamless fashion, and meet clinicians and patients where they are“, Dr. Cunningham concludes.