Skip to main content

How to Use AI to Improve Patient Referrals

Analysis  |  By Christopher Cheney  
   December 27, 2023

A health system's new AI tool is helping clinicians target the right specialist for a referral, prompt appropriate workups before a referral, and eliminate unnecessary referrals.

Providence health system has developed an artificial intelligence tool to help physicians manage patient referrals more effectively and efficiently.

AI technology is becoming increasingly prevalent in the healthcare sector. For clinicians, AI is being used in a range of applications, including clinical decision support, documentation, and radiology imaging.

Providence identified patient referral management as an AI opportunity and developed MedPearl as an AI product within the health system, says Eve Cunningham, MD, MBA, MedPearl founder and chief of virtual care and digital health at Providence. "MedPearl is a product that lives within Providence. It will probably become its own entity eventually. It has been built and incubated within Providence," she says.

MedPearl is designed to address three common scenarios in patient referrals, Cunningham says.

  • First, the patient can be sent to the wrong specialist. For example, if the patient has a chronic cough and they see a primary care physician, there are several specialists that the patient could be sent to such as a pulmonary doctor; an ear, nose, and throat doctor; and an allergy doctor. All of those specialists could potentially be appropriate for that condition, and sometimes the patient gets bounced around from specialist to specialist because they don't get to the right specialist at the beginning.
  • Second, the patient gets sent to a specialist and the specialist says they have to conduct lab tests and workups, then they will ask the patient to come back a second time. A lot of that work could be done on the front-end with the referral. It is common for primary care physicians not to know what workup the specialist would want. If the primary care physician could be given good information to optimize that workup before the patient sees the specialist, access to care could be improved because it can take months to see a specialist.
  • Third, about 20% to 30% of the time when patients see specialists, they do not need to see a specialist at all. There isn't anything for the specialist to do. If those patients could be kept with their primary care physicians, then the patients who really need to see specialists could see the specialists quicker.

All three of these scenarios are knowledge-sharing challenges in the clinician community, Cunningham says. "The reason why we have these missed opportunities is because we do not do a very good job of sharing knowledge with each other. We are not efficient at it. We do not have a great technology capability to share knowledge with each other in an impactful way that fits into clinician workflows and is easy to use."

Over the past three years, Providence has curated a knowledge bank of referral guides and algorithms to create MedPearl. The referral guides and algorithms have been validated with the clinicians who are using the information. MedPearl features information that is needed at the point of care.

"This information fits into clinicians' workflows, helps get through a patient visit, and helps identify all of the rules of engagement for the next best action for a patient you might need to refer to a specialist," she says. "We have about 600 guides and algorithms in this library. It constitutes about 95% of what a primary care physician does, so when they go into the knowledge base, they are getting what they need."

Generating results

Data shows MedPearl is having a positive impact on patient referrals at Providence.

"We have captured thousands of data points. We have captured search terms. We crowdsource new topics based on when providers are searching for the same thing over and over again," Cunningham says.

Providence conducted a pilot of MedPearl last year. More than 200 clinicians were involved in the pilot, and there were about 14,000 searches in the pilot period. The clinicians reported how the application helped them in their decision-making for referrals.

Twenty percent of the time, clinicians said they did not need to refer a patient to a specialist at all because they got the information they needed from MedPearl and they were able to manage the patient on their own. Seventy-two percent of the time, clinicians said that MedPearl reminded them to order a lab or MedPearl reminded them to start a patient on a first-line therapy before a patient was referred to a specialist. And 20% percent of the time, a clinician changed the specialty or level of urgency for the referral, which seized on the opportunity to make sure a patient did not bounce from one specialist to another.

MedPearl launched at scale in January 2023, and there are now 7,000 users of the AI tool at Providence. Search volume surpassed 150,000 searches this year, and the tool has achieved a 95% search success rate. Early data from 2023 indicates that clinicians have improved their productivity and there is a reduction in unnecessary referrals.

"The way I interpret the data is clinicians feel more confident now that they are using MedPearl that they can refer the patient more appropriately and work them up more appropriately," Cunningham says.

AI and chief medical officers

In adopting AI technology, the primary consideration for CMOs is their workforce, she says. "Their biggest concerns about doctors and clinicians are that they are burned out, there is a shortage of them, or it is difficult to recruit them. CMOs want to make sure that they have a supportive environment for clinicians. CMOs want to be innovative and forward-thinking when they are thinking about different ways of bringing in tools and applications that are going to help assist and augment clinician workflows. CMOs want to be strategic about the types of technology they bring in."

CMOs are the frontline advocates for their clinicians, and they often see problems and prioritize problems that executives in the information systems teams are not prioritizing or are not understanding, Cunningham says.

"When we started building MedPearl, I was the chief medical officer of one of the medical groups when we started identifying the referral problems and started building out the application," she says. "I had many conversations with the information systems team, and they did not understand why we were trying to solve referral problems and why it was such a priority for us. It was because we were living in different worlds to some extent, and we had to translate that for each other and come together with a common understanding."

As members of hospital administration, CMOs must consider the effort required to implement an AI solution, Cunningham says. They need to ask, is it going to scale and is it going to be adopted?

"One of the benefits of MedPearl in this respect is there is not a massive lift for the quality department or the information systems team. We have solved a real clinician pain point, with very little burden on a halo of other teams that must help implement the solution," she says. "In addition, there has been organic adoption. Doctors have talked to each other about the tool and it has caught fire. There was not a need for an internal marketing campaign or a push to get adoption. The tool works in existing workflows. It does not create new workflows. These are all things that CMOs must tick off a list that makes it worth engaging in change management."

Christopher Cheney is the CMO editor at HealthLeaders.


MedPearl, a new AI tool built at Providence, is addressing knowledge-sharing challenges in the clinician community at the health system.

Adoption of MedPearl has soared from 200 clinicians in the pilot phase to 7,000 clinicians since the AI tool was launched at scale in January 2023.

The primary concern for AI tool adoption among chief medical officers is how the technology is going to impact the clinician workforce.

Get the latest on healthcare leadership in your inbox.