It was not surprising that much of HIMSS23 was about how Artificial Intelligence (AI) is transforming healthcare. The hot topic was of course Generative AI and the use of Large Language Models (LLM), but there was also a lot of emphasis on Machine Learning (ML) and Natural Language Processing (NLP), with more than 120 companies showcasing various kinds of AI capabilities. It was particularly interesting to see some pragmatic implementations that are demonstrating tangible outcomes.
One health system in Florida (The Villages Health), serving a population of elderly people with chronic conditions and comorbidities, applied NLP on patient records to discover unreported conditions. They set a specific goal to look for the top four conditions that, if missed, would be directly related to unrealized revenue.
Their pilot project combined NLP on over half a million clinical notes in PDF together with ML classification on 7 million encounters with structured problems, medications, and observations data and discovered that 15% of all cases had unreported conditions in one of the four categories, with the highest number being in degenerative neurologic conditions.
They used UMLS CUI and other standard codes for each identified concept and then aggregated the NLP output with the structured data in an OMOP data model, which in a way allows them to use the same pipeline in the future for quality management. It was a good example of using NLP and ML as well as the available standard tools around terminology and semantics.
More work to be done with semantic interoperability
Semantic interoperability appeared to me to be another central theme of the conference. The adoption of standards like FHIR and RESTful APIs, as well as standard terminologies such as SNOMED and LOINC, is well underway but is not enough to achieve semantic interoperability.
There were interesting discussions of how U.S. Core Data for Interoperability (USCDI) is a good first step but leaves many areas undefined: no guidance on how codes should be used in exchanges with FHIR, HL7 v2, and C-CDA, and no constraints on subset of terminologies in areas where they overlap.
These challenges, combined with the fact that terminologies like SNOMED allow for expressing the same clinical meaning in several different ways, and the fact that many of the terms that clinicians use are not even synonyms in these terminologies, leave space for new opportunities.
One proposal went as far as defining a formal clinical language called Clinical Elements Model (CEM) to express concepts with all their qualifiers, allowing to record a blood pressure measurement and its units as well as the location, the body position, and whether it was measured during an exercise. This is quite different from the approach we use today, basing ourselves on the most common characteristics describing a concept — I heard it being called Minimal Viable Data (MVD).
This MVD approach in exchanging information between two systems might actually be dangerous in some situations, for example when not transmitting essential context might prevent clinicians of the receiving system from selecting an appropriate patient treatment.
The impact of genomics, proteomics, and other ‘omics’ on interoperability
In this context, I thought that there was not enough emphasis on precision medicine at the conference, apart from a session on how advancements in genomics, proteomics, and other “omics” are slowly but steadily getting into clinical practice and how interoperability is and will be affected.
Genetics is a significant contributor (30%) to the diseases that we are born with or acquire and is also a key factor in how our body metabolizes some drugs. There are over 600 known drugs that are not working on people with certain genomic variants.
The same is also true for oncology therapies, where sometimes a $4,000-per-month agent might not work for a patient. The impact is not only financial, but also the precious time of that patient going through a therapy with inappropriate chemotherapeutic agent.
It was interesting to see several companies in the pharmacogenomics space that will benefit from some of the new developments in the FHIR and HL7 V2 interoperability standards to include genotypic data in a computable format.
Without adding precision medicine data to interoperability, it will be hard for Clinical Decision Support (CDS) companies to plug their tools into EHR workflows. CDS tools are being developed to help differentiate rare from other maladies using genotyping, which seems like a great untapped opportunity. I was personally surprised to learn that this challenge is not just a “corner case” — in fact all rare diseases together are more common than diabetes.
Pilot projects for decentralization and blockchain
One other theme that was particularly interesting to me but might have gone unnoticed, as there were just a couple of sessions related to it, was decentralization. When you think of decentralization, one of the first technologies that comes to mind is blockchain.
There was just one single session on blockchain at the conference, but it was worth attending. Synaptic Health Alliance presented their pilot project of a provider directory, hosted on a private blockchain, that is covering a part of South Texas and is in active use by Southwestern Health Resources.
It is not too difficult to build a provider directory. The challenge is in keeping it up to date over time. The consortium solved the problem by incentivizing network participants to continuously validate the provider records and rewarding this work with stablecoins. This was a notable example of collaborative work, decentralizing data, while at the same time creating a platform for developing web3 apps.
Another example of decentralization was a set of pilot projects for digital identity created collaboratively by the CARIN Alliance — a multi-sector group of stakeholders representing consumers, patients, health systems, insurers, etc.
The key was the proposed Trust Framework that allows for any compliant Credential Service Provider that issues a credential to a person to be accepted and trusted by data providers (such as an EHR or a HIE). A proof of concept is already available with the portal of the U.S. Department of Health and Human Services, which acts as an identity broker for patients willing to retrieve their data from participating providers. Learn more about the initiative, which is called the External User Management System.
Interoperable digital identity and patient matching
It seems that soon, we as patients should be able to choose any application to act as our patient data steward and receive our health record from all providers that host our data. This also means that patient portals of healthcare providers as well HIE portals will be accessible using the same digital identity.
All this is getting packaged in a FHIR Implementation Guide (IG) called “Interoperable Digital Identity and Patient Matching” focused on both patient-facing (B2C) and payer/provider (B2B) interactions. Patient matching will indeed be greatly facilitated by another identifier next to the probabilistic demographic data matching used today.
This also seems to be an important first step in real patient empowerment, because we as patients are still observers in this process. Hopefully in a more distant future, we as patients should be able to annotate, provide feedback, and even correct our own health record as we as patients are best positioned to say whether we have taken a specific medication on our medications list.
Finally, it feels it is all about APIs — allowing various capabilities, from NLP to semantics to genotyping to CDS to digital identities to portals, to be stitched together in composite secure and privacy-preserving healthcare-specific workflows across networks of care participants. Healthcare is a complex space and deserves specialized healthcare API gateways to facilitate these collaborations.
Learn more about how you can steward healthy data in your organization by accessing these resources:
Guide: Be a champion for healthy data
A how-to guide for promoting healthy data throughout healthcare delivery organizations
On-demand webinar: Be a champion for healthy data
Watch an on-demand webinar about how Rhapsody health solutions customers and partners rely on healthy data to drive healthier outcomes.
Guide: Know your people
Learn the risks of inaccurate or duplicate patient records and how healthcare leaders can strengthen their data foundation with a next-generation enterprise master person index.