The California-based SCAN Foundation is a grant maker, investor, partner, and advocate on a mission to support older adults age well in home and community—and one of its key priorities is data equity. With artificial intelligence playing an increasingly significant role in healthcare, Healthcare Innovation spoke to Anika Heavener, vice president of innovation and investments at the foundation, about the need to address the lack of robust data on older adults and especially those in marginalized populations.
Healthcare Innovation: I was looking at your bio and saw that in a previous role you were executive director of enterprise data and digital health at Mass General Brigham. Could talk about your experience there?
Heavener: I had an incredible experience when I was at MGB. I was there for about five years total. It got started in our innovation group, but then we ended up proposing a $500 million project — the enterprise data and digital program that was a collaborative initiative across the Chief Digital Health Officer, the Chief Data Analytics officer, the Office of the CIO, and our individual partner hospitals. This was seven years after our Epic implementation. Due to capital constraints, we fell behind in a lot of the emerging digital health, but also just tech infrastructure. So that initiative was really to catch us up. And thank God we did, because it was right before COVID hit. We leaned heavily on virtual care during COVID and we really had the infrastructure in place that allowed us to transition much better than a lot of other sites of care that didn’t have those resources and capacity.
HCI: For our readers who may not be familiar with it, could you describe how the SCAN Foundation is related to the SCAN Group, and where the foundation’s funding comes from?
Heavener: We are a supporting organization of the overarching SCAN Group. We’re a public charity and completely independent. That being said, as a supporting organization, we’re still connected, and we have a shared mission. So the majority of the work of the SCAN Group and the SCAN Health Plan is, of course, in supporting older adults, specifically the Medicare and Medicare Advantage populations, as well as dual-eligible members. We focus primarily on the future of aging, specifically in home and community. And with regards to our funding, we got our start with a one-time contribution from the health plan and the government of the State of California that we have managed for the past 16 years.
HCI: Someone mentioned to me that you wanted to respond to an article we recently published, which was basically a summary of some comments from the EHR Association to CMS about social determinants of health data capture. What struck you about that?
Heavener: It is not a surprise that there's no consensus from the EHR Association on what SDOH is. SDOH, by its nature, has long lagged in the clinic. It's one more thing for physicians to do, but the hard truth about it is also that there is a lack of reimbursement incentive to drive the prioritization of SDOH. I think creating standards is a great suggestion. But again, what’s the incentive for that? And what's the catalyst for these already stretched-thin health systems to prioritize SDOH standards?
My greatest concern is that calling for standards can't be a strategy to kick the can down the road and to continue to undervalue the importance of SDOH within clinical care. The article mentioned the gradual transition toward more widespread adoption of SDOH screeners. You know, that's not a strong signal for change. Beyond the idea that we can't kick the can down the road on this further, the different perspective I really want to elevate here is that it's time to reimagine and redefine SDOH.
Everybody knows that life happens outside of the clinic. The EHR is but one component of data that is critical to understanding someone's health and their well being. But the SDOH conversation needs to evolve, and the past ways of working are continuing to lag, so I think there's an opportunity for us to reimagine SDOH because of the amount of information that's at our fingertips, and now, with the horsepower of a lot of emerging technologies, whether that's — dare I say it, AI — but also automation and advanced analytics, we have a lot more at our disposal to make better meaning of the diversity of data that better represents our life.
HCI: The SCAN Foundation's website mentions inadequate data on marginalized populations impeding the ability to understand inequities and evaluate interventions. What are some ways that health systems can get more of that data and have a richer data set?
Heavener: It's not going to be just health systems. If you're hearing the conversations coming from Open AI or Anthropic, there is a quest for new data. That includes protected data, like from the EHR, and it also involves brand new data — data that we haven't gathered from particular populations or slices that we haven't gathered. I think it's an incredible opportunity to leverage that to elevate marginalized populations, whether that's older adults, whether that's Tribal communities. This thirst for more data is going to be a new business model that I think can catalyze a lot of the change that is needed around how we look at social determinants of health. I'm seeing the role of the individual and the consumer as really the catalyst for unlocking the magnitude of data that is needed. Again, it's not just going to be in the EHR. Life doesn't happen in the clinic. So how do we better activate our patient populations to help them steward their data and understand what data they want to share vs. what data they want to protect?
HCI: The SCAN Foundation website states that we’ve got to create the conditions for the healthcare players to compete on equity. Does that mean altering payment mechanisms or quality measures?
Heavener: When we say compete on equity, the argument we're trying to elevate there is, rather than pursuing health equity out of the goodness of our hearts or because it's the right thing to do, making the business case and the ROI case that is going to differentiate and create a more competitive product, that is going to better serve a multitude of patient populations. From a reimbursement perspective, I don't think that's actually the lever we're necessarily trying to pull here. We're trying to demonstrate that diversity brings about stronger competition and stronger product development.
HCI: Let’s talk about some of the foundation’s grant-making and investment work. I interviewed Elliot Green from Dandelion Health a few weeks ago. He spoke about working with the SCAN Foundation on advancing algorithmic equity. What will that work involve?
Heavener: OK, first I’ll zoom out for a second to talk about how we landed on our data equity portfolio. You might ask what a foundation that focuses on the future of aging and home and community is doing working on algorithms. Two years ago our mission and vision around the future of care in home and community had us re-examining what are the biggest opportunities but also the greatest threats to that ability for older adults to stay in home and community. And from a data equity perspective, we recognize that the emerging healthcare landscape — AI, automation, advanced analytics and whatever tech rolls out next month — is solely dependent on big data, and the data these tools need to be effective has to be complete. We have to have better representative data that is not just a piece of history over here around my relationship with this hospital and what this payer knows about me. We have to have complete, holistic data sets, and we really lack that for marginalized, older adults, let alone other marginalized communities.
HCI: What’s the danger to individuals of that incomplete data? The algorithms might have bias in them?
Heavener: Exactly. There's no such thing as bad data, but there's incomplete data, and all that those tools and algorithms have at their disposal is the data provided to them. They will spit out insights based on what's fed to them. And the majority of training data for healthcare AI comes from white males aged 25 to 44. Recognizing that this is the foundation of what healthcare AI is being trained upon right now, of course it is going to be doing a disservice to marginalized populations. We see an incredible call to action for better representation within training data of marginalized populations, but also greater transparency around the data that is utilized for these emerging tools that we are still experimenting with.
It's not just about software deployment anymore. These tools are constantly evolving and learning, so we have to have stronger governance and monitoring practices in place to appropriately protect marginalized populations from bias, and that, to me, is part of the real work to be done here. I'm concerned that health systems and healthcare practitioners are getting “shiny ball syndrome" with AI and they're missing the hard work that comes from effective change management and true adoption that leads to the health outcomes that are going to matter most — clinical improvement, and financial ROI. Otherwise, we're going to spend a lot of money on AI and not really seeing much return for that investment.
HCI: UCSF just announced the creation of a governance structure to vet algorithms, not just initially, but on an ongoing basis and from an equity standpoint, too. But does that effort has to take place within each organization, or does it take place at a higher level in some kind of regulatory framework?
Heavener: That’s one of the bigger questions that we're wrestling with right now. UCSF has a process. Mass General Brigham has a process. Stanford has a process. But what about community sites of care? What about safety net hospitals? What about those who may not have the academic research and the capacity within their staff and funding to think about what frameworks for equity and patient protection they would want to put forward? So there is absolutely a need for some level of regulation. Yet we know that this technology is moving much faster than policy and regulation, so you're seeing emerging coalitions coming together, like the Coalition for Health AI or CHAI. I'm tracking roughly 17 different coalitions, and they all have wonderful aims and intentions, but we still haven't coalesced around what is the standard we want to put forth to ensure that we do this equitably and keep it in the service of patients. So good strides are being made, but it's all individual efforts right now.
HCI: Can we go back to the work with Dandelion Health? They work with big data sets, right? What is that effort going to entail?
Heavener: Our partnership with Dandelion is elevating two things. One, as much as we want to bang the drum that there is not representative data when it comes to training algorithms, we also need a set of tools that helps us appropriately evaluate and monitor for algorithmic bias for marginalized populations. Dandelion is one of those tools that can be that assurance lab or that check and balance for an algorithmic audit. Second, what we are partnering with Dandelion on building is stronger SDOH capabilities and factoring that into their algorithmic bias. Prior to our partnership, their audits did not take census data elements into the evaluation, so we are layering in those different demographic differentiators to better understand and monitor for bias.
HCI: I know that the foundation has a bunch of other initiatives in health equity, but are there some other efforts specifically related to data equity?
Heavener: Our efforts around data equity are really leaning into that representation piece that I shared with you earlier, but then also how we identify, eliminate and prevent algorithmic bias moving forward. One of our current projects is with the New York City Department of Public Health around their CERCA program. CERCA stands for the Coalition to End Racism in Clinical Algorithms. This program was launched by their chief medical officer, Dr. Michelle Morse, about three years ago, and across 11 different health systems in the New York City area. What's so important about the CERCA work is that this was one of the first models in public health around managing algorithmic bias and effective change management.
There are going to be some really interesting studies coming out from the Doris Duke Foundation and the Robert Wood Johnson Foundation later this summer on the efficacy of this program. But early data is showing this is really something to keep an eye on, and the role that we are playing is figuring out how we better codify what they did in New York City, and then “lift and shift” and bring it to other public health entities. Because what we are seeing when it comes to the AI healthcare arms race, it's all at academic medical centers. There is not much going on in public health, in community sites of care. We are working with CERCA to codify their approach and then disseminate that.
HCI: Any other projects you would like to mention?
Heavener: I would just mention one other project that we have under way, and this is around the utilization of AI in California community health clinics. Algorithms aren't brand new. AI didn't just come up overnight. There's actually a lot of this that has been utilized in community settings, but very little attention paid to it. So as we see that healthcare organizations are looking for more tools to identify and address inequities within these algorithms, we have to understand the landscape of what utilization is already taking place. We are sponsoring research with healthcare and public health entities in California to assess the state of play of AI in these settings — FQHCs and safety net hospitals — to understand, frankly, the gap between the conversations that are happening at academic medical centers and the conversations that are happening in lower-resource sites of care. There are definitely going to be some interesting insights around the gaps that are emerging in the way we deploy this new technology.
