Developing and Scaling a Clinical Generative AI Product across Providence

Introduction: The Challenge

One of the core challenges faced by busy front-line physicians and advanced practice providers (clinicians) is recognizing the need for specialty referrals, referring to the appropriate specialists), and knowing what laboratory tests, imaging, or other workup to obtain during the transition to specialty care. It has been estimated that between 26 and 48 percent of all

specialty care referrals are inappropriate or incomplete.

To address this issue, Providence set out to develop its own technological solution - a unique platform that can be embedded within any EMR. This platform, MedPearl,

provides front-line providers with succinct clinical guides and tappable algorithms designed to support their referral workflow.

Goals: Providence's goals for MedPearl were to:

1. Create a dynamic platform that improves patient services and outcomes related to specialty referrals.
2. Provide primary care clinicians with a simple tool to become more effective and efficient in treating patients.
3. Increase clinician job satisfaction.
4. Utilize specialist resources more effectively.
5. Minimize waste and errors.

The platform needed to be highly trusted, useful within established workflows, and adopted voluntarily with minimal formal user training

Implementation

MedPearl was founded by Dr. Eve Cunningham, a chief medical officer and practicing obstetrician and gynecologist. In 2021, Dr. Cunningham created a chatbot called AskProv as the initial solution to the

knowledge sharing and referral issues in her medical group. From 2021 to 2022, Dr. Cunningham and her team conducted over 300 meetings with executive leaders and practicing clinicians to refine the product and gain support for the full platform, which eventually became MedPearl.

During the platform's development, a small team of clinicians collaborated with software developers in an

iterative process of human-centered design to optimize usability and minimize cognitive strain. Multiple prototypes were tested in a clinical setting, leading to the creation of MedPearl's unique no-code platform. This platform allows clinicians themselves to write, curate, and govern all guides and algorithms.

Following a successful pilot, MedPearl was scaled across Providence and continues to be improved

through focus groups and clinician feedback.

Specialist physicians have also noted that the platform is easier and faster to use than their usual methods of knowledge management. As of the writing of this article, the platform has achieved widespread use to over 7,000 users and over 180,000 searches have been completed within referral workflows.

Hurdles

The development of an entirely no-code platform was a significant challenge. Timely content updates were difficult unless clinicians themselves could build, maintain, and govern the clinical guides and algorithms, rather than relying on programmers.

Another hurdle was the creation of an ideal library to address clinical questions regarding specialty referrals. The team created a large language model using Providence e-consult data to identify core topic areas related to referrals. As a result, 95% of the e-consults have been mapped to MedPearl content.

"The Human-Centered Design Approach, Which Considers The Real Lives, Struggles, And Workflows Of Users, Is An Essential Element In Healthcare Technology"

The team also faced the challenge of judiciously utilizing advanced artificial intelligence (Al). While Al was used for semantic search, it was not solely relied upon for content creation, which required expert human curation.

Patient data contextualization proved to be another major challenge. Pulling patient data from medical records and displaying it within MedPearl content presented difficulties due to the heterogeneous lab and conventions. The team had to create 545 unique widgets capable of pulling and displaying useful data.

Key Takeaways

1. Providence's journey was driven by an insistence on a no-code, "by clinicians for clinicians" approach.
2. MedPearl aims to improve patient services and outcomes related to specialty referrals.
3. The platform provides primary care clinicians with a simple tool to enhance effectiveness and efficiency.
4. The utilization of specialist resources is optimized while minimizing waste and errors.
5. Human-centered design is crucial in healthcare technology development.



Conclusion

The human-centered design approach, which considers the real lives, struggles, and workflows of users, is an essential element in healthcare technology. In the case of MedPearl, clinicians themselves build and maintain this novel, no-code knowledge base that combines patient data with clinical knowledge at the point of care.

‍