It’s easy to feel the immense transformational capacity of Generative AI as a solution. And healthcare has no shortage of problems to solve. The real insight is in figuring out viable application areas and use cases. Things are becoming a bit clearer in that aspect and it’s worthwhile to keep an ongoing list of where Gen AI application makes sense in healthcare. This post is always under construction, so items may range from fragment-of-idea to name-of-startup-already-doing-it.
- Clinical note writing, especially through ambient listening: Most of the medical care delivered today is represented in some way through written notes. With LLMs we can undoubtedly evolve the writing, understanding and interacting with those notes. And if you combine the listening ability (during patient interaction) and generating useful notes from that, it’s even more viable for adoption. Companies like Abridge are already doing this.
- More accurate billing: Understanding the clinical notes, associated data like labs etc to create a better billing experience for the organization (like claims repricing, automated coding) and the patient (like EOBs).
- Generating referrals: Referrals happen all the time in clinical workflows and are often simple templated documents. Gen AI can definitely enhance them to have them be auto-generated with more relevant information
- Better, faster clinical research enrollment: Matching the in/exclusion criteria for a clinical trial, even communicating with the patient to enroll them and answer questions
- Simplifying prior-authorizations: There is a lot of procedural complexity, documentation and back/forth communication in prior authorizations. Co:Helm is focusing on this (but not sure about details of what they are offering as a solution)
- Generating imaging reports from 2D images like X-rays, as well as 3D images like brain CT scans. As pre-trained models become multi-modal, image and video based reasoning can create really great assistant tools for radiologists.
PS: I’m intentionally skipping clinical applications like suggesting diagnosis, recommending treatments or prescriptions, etc because they are on a longer timeline for viability given they are so risky – the legal, scientific, adoption burden can derail or reshape them a lot as they evolve. No point outlining them right away.