December 2024 Update: Interoperability Adoption and the Path to Impact, Value-based Care Projections, New Developments in AI for Enterprise Automation
The December monthly installment once again covers contemporary topics and issues in healthcare, life sciences, AI, and technology. First, this post captures estimates on the adoption of clinical data interoperability, how data sharing is outpacing practical data use, and what we can do about it. Next, value-based care is reviewed with statistics on its rate of growth. Finally, new developments in AI are reviewed, which make AI models more accurate, trustworthy, and capable of automating enterprise work.
Joe Bastante
12/15/20244 min read
Healthcare data interoperability, the good, the bad, and what we can do to make it better
According to ONC data published this year, hospitals sending, receiving, and integrating clinical data increased from 46% in 2018 to 70% in 2023. Interoperability adoption is greater for urban over rural providers, system affiliated over independent providers, and large over small institutions. Increased data sharing is good news since it can help improve care coordination, close care gaps, and support more equitable care. For our clinical research friends, data sharing also helps eliminate manual entry in electronic data capture (EDC) systems, improves patient registries, aids in patient recruitment, and provides more complete real-world data. However, there’s a wrinkle. While 71% of providers have access to external clinical data, only 42% claim to use it at the point of care, and the 42% figure doesn’t imply that data is used often or effectively. Sharing data and delivering value from data are clearly different. For example, having led the implementation of multiple CMS interoperability mandates, such as requirement to grant patients access to all their data via APIs, we watched day after day, week after week, month after month as no members/patients used the capability. Unfortunately, more attention is given to technology deployment over adoption and value, as evidenced by the growing interoperability technology and services industry (estimated at $3.4 billion in 2023 projected to grow as high as $16 billion by 2030). To derive value from interoperability, healthcare players must look beyond technology to data fitness and quality, usability of integrated data within EHRs and core systems, change management, sustainable value measurement, and empowered leaders to adjust course if value measures aren’t attained. Let me know if you’d like to hear more about attaining value and I’ll do a special post on this topic.
See this link for full details on the ONC interoperability data brief:
Is value-based care still growing?
Value-based care (VBC) aligns provider payments to care quality and efficiency. VBC models include those where accountable care organizations take responsibility for a set of patients (i.e., population-based) or payments are bundled for an episode of care (i.e., episode-based), such as a hip replacement. Many, but not all VBC programs have demonstrated savings. For example, one of the largest programs, the Medicare Shared Savings Program, achieved $2.1 billion in net savings in 2023. Some have questioned the savings figures, for example, claiming that increased provider overhead for such programs is not adequately factored in. Nevertheless, aligning payment to outcomes makes sense, and VBC programs are here for the foreseeable future. But are they growing? Yes. First, anecdotally, the CMS Innovation Center is targeting 100% of Medicare beneficiaries in accountable relationships by 2030. According to the 2024 Health Care Payment Learning & Action Network (HCPLAN) Measurement Effort, which analyzed data for 92.7% of U.S. insured, 28.5% of all payments flowed through VBC contracts with downside risk, and this is up from 24.5% in 2022. 50%-55% of Medicare payments flowed through VBC contracts with downside risk, but only 25% of commercial insurance and Medicaid flowed through such contracts. Quite telling, 76% of payers surveyed believe that value-based alternative payment models will increase, and no respondents believe they will decrease. Much will be needed to drive success in VBC programs, such as better data sharing and insights, assistance for independent providers, increased focus on equity, education for healthcare administrators and physicians, and better coordination between providers and between providers and insurers. See the resources below for more data.
2024 Health Care Payment Learning & Action Network (HCPLAN) Measurement Effort report:
https://hcp-lan.org/wp-content/uploads/2024/11/2024-HCPLAN-Methodology-Report-11-13.pdf
CMS 2024 release on Medicare Shared Savings Program performance:
A compendium of statistics on value-based care:
New developments in AI—moving from chatbots to enterprise automation
Given the noisy and confusing perspectives on AI, I thought I’d share practical and important developments in AI, which will become highly relevant for enterprises. The first relates to developments that make large language models (LLMs) more accurate and trustworthy. Many have pointed out that the rate of improvement for LLMs is slowing down with each new version. That may be so, but innovation is happening in the software that wraps around LLMs. For example, OpenAI’s GPT-o1 and Alibaba’s Marco-o1 models began using “chain of thought” and other techniques to more carefully reason about responses from the LLM rather than immediately return responses to the user. A very important announcement was made by Amazon/AWS this week regarding their Automated Reasoning service. This service uses techniques complementary to AI based on math and logic to ensure LLM output aligns with policies and rules. As an example, if the service were provided with a study protocol document, Automated Reasoning can extract the rules (e.g., patient inclusion & exclusion criteria), then validate LLM responses to make sure they conform to the rules. This AWS service is only in beta but it’s a sign of things to come where complementary techniques will protect against hallucinations thus making LLMs far more useful to enterprises.
The second major trend involves capabilities allowing AI models to act on behalf of humans. For example, OpenAI introduced Structured Outputs. Rather than returning natural language text, this allows models to return responses in specific technical formats, which can be used to call enterprise systems to take actions on behalf of users. Anthropic (makers of the Claude models) introduced the Model Context Protocol, which offers a relatively simple way to connect their LLMs to any system or source. Anthropic also announced their Computer Use capability, which allows their models to “see” images of a user’s screen and control the mouse and keyboard to fulfill a user’s request. Imagine asking an AI model, “I have a picture of a receipt on my screen, can you create an expense report for me?” Finally, AI models are being used in actual (humanoid) robots, which is the ultimate form of acting on a human’s behalf. The next big AI wave will be moving AI from chatbots to the real world, and it's beginning to unfold now.
You can find more information about AWS Structure Reasoning here:
Anthropic’s announcement on the Model Context Protocol:
Anthropic’s blog article on the new Computer Use capability:
Alibaba’s advanced reasoning capabilities
As always, feedback or questions are welcomed. I’m here to help. Contact me anytime.
Contact us
Whether you have a request, a query, or want to work with us, use the form below to get in touch with our team.
