Background:
As a physician involved in pre-authorization determinations I often struggle with the backlog of cases to review, and the frustration of human workflows frought with errors and lost information. I often wonder about using AI to make this process more efficient, and this are my thought after experimenting with automation bots, generative AI Agents and deterministic AI agents.
Premise:
My premise is that AI can be taught to scrape medical records and evaluate them using binary processes.
Most Pre-Authorization guidelines are algorhythms which reach an answer by asking a series of questions and if there are enough affirmatives the requested medical service is authorized.
Since the algorhythm can be made binary, it can be transformed from a ladder diagram of binaries to a custom JSON database and the AI Agent can render affirmative or approved decisions faster then a human clinician. This is how AI can speed up approvals.
However, if insufficient afformatives are obtained, the request is then referred to a clinican (nurse, physical therapist or equipment specialist.) And then if not approved, referred to a Physician. An AI can approve, a nurse can approve, but only the Medical Doctor can deny.
Reason:
Approvals should be fast, and friction between insurance company, provider and member is thus minimized. AI can perform approvals very fast with nearly zero harm.
The proposal speeds up processing of insurnace pre-authorization requests.
This speeds up the delivery of medically necessary medical services.
This should improve outcomes.
This situation creates an entirely different experience for the medical doctor or physican extender requesting services for their patients. The insurance would render nearly instantaneous approvals using AI. This should improve insurance company provider and member relationships. Fast Approvals can only create good will.
Issues:
Unfortunately, and paradoxically, these systems sometimes fail to do all these things due erros in design. Some insurance companies or state regulators require Human in the Loop processes, HITL for both Approvals and Denials. This creates backlogs and slow down the process, because in that instance the AI becomes an additional process for all claims and only adds to the time cycle instead of reducing it.
It's important for everyone involved to understand the workflow for Approvals and Denials and what it's a different situation for an AI to Approve versus an AI to Deny.
I think HITL processes should be limited to Denials, because no medical service requested by a medical doctor should be denied by an AI Agent. There are to many variables which make up a denial for an AI to currently do a good job.
However, I feel HITL requirements for denials shld be 1:1, and I would even state that other then data collection, no AI should be used. But HITL for approvals are 100% inefficient and useless. AI Approvals should require only HITL audits, not HITL in the workflow.
That's enough for one article, as that's a lot to digest.
What do you think?
Posted Using INLEO
