Updated: Sep 29
There’s a lot of noise out there about healthcare innovation and even more noise about AI, but here is the truth: Healthcare needs AI. And not in the way most people are thinking.
September 29, 2023
AI in Healthcare: Dispelling the Myths
There’s a lot of chatter about AI-driven diagnosis, AI-driven genomics, AI-driven drug discovery. I think this is partly due to the fact that in the popular imagination, “AI” is some superhuman intelligence and thus should be applied to the most advanced use cases, like replacing doctors’ diagnostic abilities with some all-seeing, all-knowing omniscient software a la IBM Watson (which failed). These are all noble causes, do not get me wrong and I have no doubt the latest Large Language Models show promise in all of these areas.
But Is AI really at that level right now? Does anyone really trust chatGPT (or any other model) to make life or death decisions about one’s health? And more importantly, is AI-driven diagnostics the most pressing thing clinicians and patients need right now? What are the real problems in real clinics that need to be solved?
A Broken Healthcare System
We live in a world where doctors will use LINAC particle accelerators to accelerate electrons to the speed of
light and beam them into people’s bodies to surgically target and destroy cancer; and then that same doctor will walk over to a fax machine and shuffle through a dense packet of paper to understand why their next patient has been sent to them for critical care.
Healthcare is broken.
The Challenge of Interoperability
The problem is that, in any given region, every patient’s record is spread across multiple disparate systems that are not connected or
designed to work in any meaningful way. Neither the patient nor any doctor (not to mention any AI model) has a complete picture of the patient’s actual healthcare record. There is no real continuity of care. Every consumer of healthcare in the US has experienced the maddening frustration of filling out massive paperwork on their medical history in one clinic only to repeat the process again and again at each clinic they go to for care.
Why Hasn't Interoperability Been Solved?
One would think in the age of the internet and the ubiquity of APIs and connectivity that this problem would have already been solved a decade ago, like it has been solved in most other sectors of the economy. But it hasn’t. Why is that?
In short: interoperability is hard, especially in healthcare given the oligopoly of the entrenched EMR companies. The EMR companies control the data, and they have no incentive to share it; it simply does not align with their business model.
The Search for Solutions
Various initiatives have tried to bridge this gap.
Government-backed Health Information Exchanges (HIEs) were proposed as a panacea. However, hindered by data inaccuracies and misaligned incentives, this solution faced obstacles from the outset. Most HIEs never worked, they never gained enough participation to become useful.
Private ventures have also emerged, attempting to establish exclusive HIE networks across regions. The catch? For these networks to be truly effective, near-total adoption by every healthcare provider is essential. But what we witness is a modest 20% solution coverage for incoming clinic patients.
While certain care coordination tools attempt to rectify this, they merely redirect the problem onto patients, resulting in bad experiences and inconsistent processes.
AI and the Promise of Overcoming Challenges
A glimmer of hope has appeared, though. Thanks to the rise of FHIR and to the 21st Century Cures Act's regulations on information blockages, firms like Health Guerilla, Particle Health, and Redox have made impressive progress. They harness now-accessible FHIR APIs to collate patient records, enhancing both interoperability and care continuity. Yet, the decades-old obstacle persists: the ubiquitous fax.
The real problem is that there is no incentive for clinicians to change their behavior.
True healthcare interoperability hinges on technology that can decode faxes just as humans do (or better) and then render this data API-compatible, without requiring a change in behavior. Until then, complete patient records will remain a pipe dream.
The Role of Generative AI
But with the emergence and ongoing Cambrian explosion of generative AI, there's newfound optimism that this long-standing challenge can finally be overcome. We are standing at the precipice of a digital revolution, something akin to the forming and storming of the internet boom in the early 2000s. It will be amazing to watch.