Practicing medicine is complicated and involves many activities. This is because human body is complex and there are many moving parts. On top of it, there are many activities that need to be done around taking care of patients. These include record-keeping, submitting charge codes to get paid, writing letters to justify doing procedures, and many other things. Although many sectors of the economy have enjoyed from automation from digital technologies in the last few decades, healthcare is a laggard in this area.
One of the great disappointments of the last decade has been that most of the promises about implementing digital information systems to document clinical care never materialized. EHRs, electronic reporting systems for radiology and pathology, data warehouses and other information systems couldn’t show that they improved care or made patients’ lives and clinicians’ jobs easier (quite the opposite). Although it’s better to have as much of the patient information in one place as possible, usually in an EHR, so that the care team can look up patient information quickly, the entry of that data has been a major issue. This turns physicians, one of the most highly trained and important groups of professionals, into glorified data-entry clerks. It wastes their valuable working time that could and should be spent with patients.
The design of these EHRs hasn’t made it easy for healthcare workers to easily find the most important information. Given that a lot of the information is in an unstructured format, it’s also difficult to use it for analysis. That means that the promise of point of care decision support, analytic-driven workflows and automation hasn’t come to fruition, which is disappointing to everyone who uses them to provide care. Worse, the combination of not seeing any of these benefits and having to do the tedious work of entering the information has led to disillusionment and burnout. There’s no question that entering this information has led to additional work hours for the care team.
It’s against this backdrop that we view the promise of AI in clinical workflows. The data that gets entered into the EHRs comes from interactions between clinicians and patients, is generated by devices that monitor the patients in the hospital and the clinic and gets drawn from the review of radiology scans and pathology slides, as well as a number of other sources. All of these activities can generate structured data consisting of voice files, images, numbers from devices and digital text that AI can use to create files inside the EHR. AI can listen to physician-patient interactions and generate notes, place orders for tests, send messages to other providers or care members, create referrals and complete the forms, create testing reports, and more. Also, AI can improve how information is presented to the care team based on the context of the encounter to make their jobs easier and more efficient.
All of this is now much more possible thanks to the recent breakthroughs in generative AI. Much of clinician burnout is due to spending time writing notes, placing orders, generating referrals, writing prior authorization letters and creating patient communication. In other words, burnout is caused by physicians having to generate output! With the emergence of large language models that are used to train generative AI solutions, these use cases will be at the frontier of AI’s applications in healthcare. Given that these language models allow for better training of algorithms on unstructured data without having to annotate so much of the data, it seems that documentation and content generation will be low-risk use cases that could gain traction in the short-term. Solutions are already emerging in this space that we will discuss in the next few blog posts.
