Understanding how underserved the healthcare market is concerning software adaptations doesn't take much. There’s a definitive reason for this, which primarily surrounds HIPAA guidelines and regulations. But we live in a world where regulation and innovation are moving in unison, almost seemingly coexisting instead of battling it out. Neither one could survive without the other in many ways, one way I want to tell you about today is the use case for large language model utilization in the healthcare space.
Very quickly- I work as a data automation specialist at a healthcare management consulting company in the Bay Area, which also happens to be the largest Independent Practice Association as well (IPA). I have taken many roads less traveled to address the need for internal automation requirements across literally every department of this vertical at the legacy enterprise corporate level, and today I will express to you the importance of LLMs and how it could shape the landscape for healthcare’s back office forever.
The IPA; And no, we will not be discussing different hops involved with Lagunitas’ best. We’re talking about IPAs in the healthcare world, the back office heroes that make it all happen behind the scenes, these heroes don't wear capes but they definitely will negotiate better terms for your provider physician so they too can live good lives. They negotiate with your insurance, make sure your health plan is up to date, and much more. But you never hear about them in the tech world, why? Well, I was and still consider myself to be in the said tech world, and I don’t identify anyone in the space to be like-minded for the most part, even IT. So what does this mean for innovation? Is the state of the IPA somewhat malaise to change because of its fear of HIPAA fines? I don't blame them, and the answer is yes. But, there’s still 60% of internal workflow that has gone unchanged since the adoption of the PC, with most of the advances in technology being around Microsoft Office products, Ui Path, and other legacy technologies that also walk a fine line around the HIPAA in the room.
Most technology in healthcare will try its best to avoid getting involved with patient medical reports (PMRs) because of the fear surrounding a data leak and what that means for patient data safety. So you see a lot of stuff being made that doesn’t directly address the important issues, but there’s still a lot of important issues that just don't involve HIPAA, let's take a look; Standardized procedures involve policies and protocols developed by healthcare systems to ensure everyone across the entire organization is on the same page at all times.
- Standardization procedure: takes a long list of email back and forths between the initiator (project manager basically), Researchers, Directors, C-Suite, and sometimes even the board. One could build a standardization procedure language model to aid in the process of standardizing a procedure following company policy, feedback, and safety guidelines. This would be a total aid to everyone involved in the process and ultimately would save time away from a 6-month process, boiling it down to about 3 weeks if built correctly.
- LLM for Employee Onboarding: This is a cool one because it’s a very easy concept; every healthcare company has different onboarding procedures for patients, providers, and employees at the corporate level. By training a language model on the onboarding data, one could build an internal co-pilot for new hires to ask any basic questions they may have about the company’s business and workflow. For example, I struggled for months trying to locate the right documents in different folder drives. A well-trained language model (without being trained on the material in the documents) can understand where everything else is and help everyone locate anything the same way a 20+ year employee would when you ask them.
- LLM for Provider Onboarding: Outside of employee onboarding, when a healthcare company (like an IPA) wants to onboard a new provider, they need to manually review their residency documents, and research papers, validate their degrees and research papers, ensure all of their credentials and licenses are up to date, and much more down to the PPO/ACO/CIN plans that providers need to choose rates for to find which healthcare companies they would like to work with. You’re going to hire this physician provider to serve your patient customer, they better be not lying. This manual process takes 2–6 months depending on the company and candidate. Training a language model to learn the required process to onboard would save at least 80% of the time traditionally required for this process to leave enough time for onboarding specialists to review the findings and make the final judgment. This, of course, is a matter of life and death when you consider that there are more specialized providers doing specific procedures than ever before, which is a great thing when they’re operating. It is in a healthcare company’s best interest to get these people involved as soon as possible to treat patients with rare diseases or require cures.
So the use of LLMs in the standardization and onboarding process procedures looks promising for the future, I would love to raise some money soon and innovate in any one of these verticals. The farther away you can be from HIPAA and still make a meaningful impact in a process optimization situation, the further we get as a society. LLMs are here to stay, I didn’t get into the technical details because my write-ups here just aren't technical, I am also sanctioned under an NDA by my employers and partners at the moment so I would have to effectively quit to share any consultation toward technical details on how to approach either one of these effectively.
Either way, I hope you were able to take something away from this as I use Medium as a personal thought process tool to organize my ideas and share them with the world.