Automating UCSF’s Workflows and Improving the Patient Experience with Artificial Intelligence
Artificial Intelligence (AI) is revolutionizing healthcare systems like UCSF Health and changing the way they operate. Though healthcare generally lags behind other industries in digital transformation and innovation, UCSF is well underway.
With about 1.4 million faxes received a year at UCSF – 300,000 of which are referrals – UCSF Center of Digital Health Innovation’s Referrals Automation (RefAuto) 2.0 is estimated to save more than 25,000 hours of staff time and 5,000 hours of clinician time annually. To advance this patient access solution closer to fruition and launch, last year, CDHI’s Digital Patient Experience (DPE) program team partnered with H2O.ai.
In January, CDHI’s Bob Rogers, Expert in Residence for AI, and Lead Data Scientist Lu Chen, participated in H2O.ai’s “Better Healthcare with AI Automated Workflows” virtual roundtable discussion with founder and chief executive officer, Sri Ambati, data scientist and product manager Mark Landry, and vice president of product, Prashant Natarajan.
Some Key Highlights from the Discussion
-
By automating mundane tasks such as processing faxed referrals, which are handled in a multi-step process by staff, work can be more efficient and staff burnout prevented, allowing the focus and time to be spent on what’s most important – patient care.
-
Automation makes it possible for UCSF to get patients to the right provider more quickly by removing errors in data processing and variance in timelines, thereby improving the patient’s experience from the very beginning.
-
The impossible became possible: medical referral documents are diverse and complicated. They include varied data types and sometimes include hundreds of pages of patient history. AI is able to apply context to specific pieces of information pulled from the documents. For example, AI can learn whether a name is a patient name, a referring provider name, or some other stakeholder, and can generalize this regardless of format. Compared to the previous template-based Optical Character Recognition (OCR) technology, with AI, the information extraction quality is stable and the model performance will improve over time as more human-in-the-loop data is collected and sent back to further train the AI model. This machine learning feedback loop is critically important to refine the solution and improve the process continually.
-
Automating manual and downstream processes improves the healthcare system internally to impact more patients and give better care, and is changing the way UCSF sees AI.
What do the next 12 months look like for UCSF?
-
According to Chen, currently, there are big streams of incoming referral faxes; once they can be automatically categorized and patient information can be extracted more efficiently, the team is looking for significant improvement in how quickly patients can be scheduled.
-
Rogers sees AI as a gamechanger in healthcare delivery quality and impact on patients. He would like to see UCSF have pipelines using RefAuto 2.0 for all the 1.4 million faxes coming into UCSF, and leveraging the automation both on the upstream intake of documents and downstream scheduling and care. With the administrative burden of healthcare organizations growing rapidly, he believes that streamlining the process is going to be revolutionary for patients and the healthcare system.
To watch the replay, click here.