As artificial intelligence becomes increasingly integrated into health care, questions about interoperability, policy and regulation that will help enhance patient care and improve outcomes remain on the forefront.
At the recent HIMSS 2025 conference, held from March 3-7 in Las Vegas, Medtech Insight sat down with Aneesh Chopra, chief strategy officer at Arcadia, to discuss among others how the federal approach to health data evolved since his tenure as the first US chief technology officer under the Obama administration.
As a current member of the National AI Advisory Committee, established under the National AI Initiative Act signed into law by President Trump in 2020, Chopra still plays a pivotal role advising President Trump and the White House on a range of issues related to AI.
Chopra transitioned to chief strategy officer at health care data platform Arcadia after its acquisition of CareJourney, a provider of health care data and AI-powered analytics, which he co-founded in 2014.
Key Takeaways
- Chopra stressed the need for better systems to share and analyze health data across different platforms to improve patient outcomes and health care delivery.
- Chopra envisions AI-enabling seamless access to longitudinal patient records, but acknowledges challenges like policy decisions and legal interpretations.
- Chopra discusses how the federal approach to health data has evolved to support AI innovation, emphasizing investments in infrastructure and industry consensus standards.
His vision for Arcadia is to build an interoperable data platform that will equip providers and payers with the tools they need to excel in building high-performing networks and in value-based care to improve patient and financial outcomes.
The interview below has been slightly edited for brevity.
Medtech Insight: Since your tenure as the first White House CTO, AI has become increasingly integrated into health care. How has the federal approach to health data evolved to support AI innovation and where do you see the gaps that still need to be addressed?
Aneesh Chopra: First, a couple of disclosures. I’m currently a member of the National AI Advisory Committee, set up by Congress through the National AI Act, to advise the administration on making sure we do more than we can to ensure America thrives in the age of AI.
The foundation of our approach from the Obama administration forward has been to invest in infrastructure for an innovation economy. For AI that involves R&D, human capital, cloud computing and other digital infrastructure in which one can build models.
The second framework is to work together on industry consensus standards for safe and effective use of AI. You saw this in the Biden administration’s 2023 frontier model commitments. The logic behind that was while there’s no Congressional obligation or regulatory framework today for general purpose AI, there is an opportunity for the industry to self-regulate and work with the administration voluntarily on AI safety and responsibility.
And in the last part of the Obama framework, which continues to this day, is that we want to have an ‘all-hands-on-deck approach’ on the big issues: Delivering a more affordable, equitable, and accessible health care system that involves inspiring the private sector to do more in the use of technology and innovation to advance the health care agenda. But it also, where appropriate, may involve the role of government.
The framework we applied in the Obama administration’s strategy for American innovation very much has [stood the test of] time. The Trump administration had an office of American innovation broadly aligned on these themes. What we saw in the Biden administration, and even in the handoff to Trump, I think, is a lot more alignment than divergence.
The current Trump administration repealed Biden’s [AI-focused] executive order, but the frontier model commitments have continued as far as I can tell so that industry collaboration and public-private partnership look like it’s still alive and well. And the industry is still motivated to self-regulate.
I also helped work on HealthcareAICommitments.com, launched in December 2023, where about 40-plus health plans voluntarily collaborate to emphasize the outcomes we wish to achieve and to monitor for responsible AI to improve outcomes.
This is probably the single biggest policy debate and industry debate we’re having. Is it the model itself that needs to be safe and responsible or [is it] about the outcome, workflow and process in which AI is introduced to achieve an objective?
The coalition wants to hold the health systems and health plans accountable for processes rather than arguing that the model they hired is incrementally more biased or hallucinates. An effective business process tolerates hallucination and ensures that it can respond and protect against the harms that may come from hallucination without having to wait for the perfect model that is not hallucinating.
Is it correct that you are “excited about DOGE,” and if so, why?
Chopra: When President Obama introduced my role to work alongside our chief information officer, he also introduced the nation’s first chief performance officer. So effective and efficient government has been, I believe, a strongly bipartisan agenda. And the objective is to apply technology to innovation to accomplish that objective. We cut costs at the Obama administration, but we did not have a mandate to cut $1t out of the budget, which is the primary objective [of DOGE].
We worked on the margin to say we’re going to introduce a new healthcare.gov and we’re going to introduce APIs and open government so we can build products and services in a way that the private sector can do the building, and we can do the supplying and the underlying technology.
DOGE is doing a lot of its work on USAspending.gov. That was President Obama’s legislative bipartisan accomplishment with Senator Coburn. That bill explicitly focused on making every transactional spending receipt publicly available.
Most of the DOGE analysts are simply reading the very openly available data sets that we’ve put out there in a bipartisan way. Now they may be reaching conclusions that a category of spending isn’t appropriate – waste, fraud and abuse is as much a factor of judgment as it is objective.
This team may be reaching a judgment that certain things are not considered good value for the taxpayer. My judgment might be different. But the underlying idea that the president would direct a team to bring the best and the brightest technologists, data scientists and engineers to make our government more effective and efficient is as kind of mom and apple pie as you’re going to get.
Even if it means mass firings of government employees?
Chopra: If you have $1t goal, there are only so many things you can do to get there. As I understand it, they made a policy decision that any probationary employee would be let go, because they’re not legally entitled to the protections of a full-time job.
Would I have done that? I don’t believe that would be a decision I would make. My hope is that they’ll have some rationality, and I hope some of those folks will come back into the government.
[Elon Musk’s] philosophy is an Elon philosophy. It’s not a bipartisan philosophy. He’s certainly earned the right to have a philosophy. He’s gotten great results in almost every facet of his life, and so this president entrusts him to make those judgments. The American people voted, and this is what they asked for.
Is DOGE bad or good? Well, DOGE is inherently the same as it was in the Obama era. So inherently it’s good, but different people use the tools in different ways, and I don’t pass judgment on this team.
I’ve also heard you say that we desperately need productivity reform in health care. Can you elaborate on that?
Chopra: We’ve done a lot of digitization in health care, but we have not yet done the data liquidity piece at scale. That’s the interoperability question.
Maybe with more interoperability and more data flow, we can be in a position to generate more productivity. Now here’s the issue. Productivity is a function of the economic model in which it operates. In a fee-for-service service model, productivity could mean, ‘I add three more visits to my day as a doctor,’ which may, in aggregate, increase health care costs just because we overutilize a number of services.
On the other hand, in value-based care, productivity would mean, ‘I can accomplish the health care needs for our population for less than the expected costs.’ Productivity in that context means better outcomes, lower cost. I want a productivity revolution in health care. I want to align to the incentives of value-based care.
How can AI and data-sharing help accelerate the shift to value-based care?
Chopra: In health care, we have thousands of fragmented components – physician practices, labs, health systems, health plans, all of whom own a portion of my record – in the proverbial sense of ownership. Before we benefit from AI, we need to stitch together the data so that we have a longitudinal record of a patient.
Arcadia’s DNA is assembling a patient’s longitudinal record. We do it when there’s data supply, which happens because of a value-based care contract. But we are within a year or two of a decoupling, where the data liquidity, the interoperability, will now happen whether or not you are actually in a risk contract.
It may be the case that the supply of data, the ability to generate a longitudinal record [for every patient], may be in an arm’s-length reach of employers, government-sponsored plans.
I see an era where phase zero of AI is to basically hit the longitudinal record button, recruit, aggregate the records, clean up the records, and make them available. This is where the technical question comes in. In standards, we’ve been fighting to get everybody to map data in the same common language.
My name will show up with first name in this field, last name in this field, etc. Pulling all those records and assembling them is easier if everybody speaks the same language. But that’s painful, because every health care company has built their own database, and so they are doing the manual work of converting that data into the OpenFHIR (Fast Healthcare Interoperability Resources-Standard) data model.
The provocative message in AI is, ‘What if I skip that step?’ What if every platform simply posted the data in the format they have, and AI was used to reverse-engineer what each of those different systems did? It’s a policy question: Do we want to continue managing everybody towards a common data model and mapping it, or do we focus on access to the data in whatever form and make that available?
It’s a healthy debate. I see value on both sides. Interoperability may be solved by AI taking raw data in, or it may be solved by everybody mapping the data to the common format to make it easier for developers to use. We’re testing both methods, not just at Arcadia, but as a country.
How many years do you believe are we away from interoperability?
Chopra: The delta between one to three years and three to five years is going to rest on a court case.
What rights do patients have regarding how their data is being used?
Chopra: Today we have an opt-out model. If you enroll in one of these risk programs, you can opt out. CMS has put that opt-out provision in the data-sharing. We share data at Arcadia. [We find that] more patients are opting in for a more coordinated care experience. If I were gambling, [I would project that] more patients are going to opt in for coordinated care than opt out.
What’s in it for the patients to opt in into data sharing?
Chopra: The networks we support [like CMS] present this letter to patients and say, ‘You can see any doctor you want. But if you join my practice in this program, I will coordinate your care better. I will have access to information about your care, and I’m going to reach out to you to make sure that you’re coming in to get all the services you’re entitled to. And if you fall ill and end up in the emergency room, I’m going to be there by your side before you leave, so that you can get the kind of appropriate follow-up care.’ care. So people see that and say, ‘That sounds like better health care to me.’
What are some of the biggest challenges in leveraging health data today for better patient outcomes, and how is Arcadia addressing this?
Chopra: Access is more challenging than data format standards. The profile of a typical Arcadia customer is either a health plan that’s trying to communicate with a provider and helping the providers learn about gaps in care. Or we’re helping providers that work with payers to figure out ways in which they can be more successful given the economic model to deliver better outcomes. So, we’re on both sides.
In nearly every market, there’s information in a database about a patient that if made available in a timely basis to the physician would result in better care. But the doctor doesn’t have access to that data either because that doctor that has the information is not on a national network, therefore not able to have their records queried for data sharing purposes, or they’re on an IT system that struggles with actually physically sharing the data.
The access question has been addressed in rule-making. Physicians have the rights to certain data sets, but those rights have not yet been exercised. That is the situation that we’re in right now. That’s the opportunity I see Arcadia tackle – to help our members exercise their rights so we can do our job of aggregating records and organizing them weather it’s for AI applications or basic analytic tools to help guide patients care.
Will the patient be able to access their medical records digitally and in a format that makes sense to them?
Chopra: Arcadia is not a consumer-facing app, but I am desperately motivated to bring consumer-facing apps into the Arcadia ecosystem. We’re doing more and more of that. Just like how institutions talk to each other, there’s a similar change afoot on how consumers can interact with their data. If you FaceID yourself with a Clear or an ID.me, companies that are in NIST (Institute of Standards and Technology Digital Identity Guidelines) IAL2 (biometric check) then you’ve earned the right to knock on all of those 25 databases that have information about you to say, ‘Please give me my own records back.’ That is coming. It’s called the TEFCA (Trusted Exchange Framework and Common Agreement) Individual Access Service (IAS).
We announced this week (3 March) a partnership with (health care experience platform company) League. Part of that partnership is to communicate when League engages consumers in a digital way and they happen to be in a value-based program, we can bring that information into a more coordinated experience.