For years, the Parkland Center for Clinical Innovation (PCCI), an affiliate organization of Parkland Hospital in Dallas that looks to support underserved populations by bringing together the minds of top data scientists with clinicians, has been planting the seeds for a wide range of innovative work across healthcare.
For instance, in the area of children’s health, PCCI’s pediatric asthma program has developed and tested predictive models to identify children at risk for asthma exacerbations. Those models have been combined with a population health framework to help a Dallas-based insurer, Parkland Community Health Plan, experience a 36 percent drop in the cost of asthma care, or about $12 million in savings, as well as a 30 percent reduction in emergency room visits, and a 42 percent reduction in asthma related-related inpatient admissions.
This is just one example of the seeds that have begun to grow into real-world initiatives that are delivering value for PCCI and its industry partners. Steve Miff, Ph.D., president and CEO of PCCI, recently spoke with Healthcare Innovation Managing Editor Rajiv Leventhal about the innovation center, some of the core work it is prioritizing right now, how health IT is playing a big role, and more. Below are excerpts of that discussion.
Can you give a high-level overview of PCCI?
PCCI has recognized that in order to do transformative innovation, you need to have the freedom to be able to take chances and not get bogged down in the day-to-day elements required to run a healthcare system. So in 2012, PCCI was created as a 501(c)(3) organization that remains affiliated to Parkland [Hospital] through our board, with the goal of bringing together a highly specialized set of data scientists and clinicians to take on some of the harder challenges in healthcare.
Our ultimate goal is to leverage data science and social determinants of health (SDOH) to better support underserved populations across the community. We seek external funding to enable more transformational work. What’s also unique is that we focus a lot on being an early stage R&D organization. We take that further and deploy the models that we build and test in real-time within a community to really understand the impact of the work, and how these [initiatives] can be fully integrated and realized in workflows.
How many data scientists and clinicians are employed at PCCI?
We have 38 FTEs. Within the data science and technology team, we have 10 FTEs and we have had to recruit folks from all over the U.S. The person who leads the group for us was at Epic [in Madison, Wisc.] for 15 years where he started and led their artificial intelligence (AI) team. He joined us two years ago. Our head of technology came from Intel. We also have PhDs who are highly-trained individuals in statistics and in mathematical modeling. And we have five full-time clinicians on staff with a variety of backgrounds. It’s the interaction between those two groups that enable us to design and then test these models that we’re trying to launch.
Speaking of these models, can you dive deeper into a few that you’re working on and the importance of them?
At the core of everything we do is leveraging data science and SDOH, and apply that towards how we better understand an individual’s needs, while supporting underserved populations across communities. We come to that with this principle that health begins where we live, learn, work, play, and pray, and we have the opportunity to use data science to bring those elements together to understand those perspectives, and then be able to disseminate it in a way that’s actionable.
We tend to focus on four key, high-impact areas and populations: maternal health and pediatric care; mental health and behavioral health; patient safety and care quality; and then the high-risk complex populations, such as those with chronic diseases.
One example that we just got first-year pilot results from [stemmed from] the pediatric asthma model that’s now in its fourth year of existence. We took the same model that we used for pediatric asthma and transitioned it to apply to another high-risk population—women at risk for preterm delivery. Those risks are even more pronounced within a population that is socially-economically challenged.
So we built a model to first risk-stratify women—as they progress on their pregnancies—on their chances of having an early delivery, and then tested the engagement of those expecting mothers using a texting approach. Why texting? We learned from the pediatric asthma [program] that it’s the most accessible [engagement method] and used widely—more so than Amazon Echo and Alexa devices.
The first version of this model was built solely on claims and SDOH data, which [posed] challenges because you already have a three-month lag in the data, and the goal is to identify the risks of these women as early as possible. So we were able to still identify over 77 percent of them, either in their first or second trimesters, to predict [preterm delivery] four-times better than the current models being used that are oftentimes more manual in nature.
It’s also important to do things at scale, because if something works we need to be able to expand that out. So during this one-year pilot, we risk-stratified over 26,000 pregnancies in total, and we took a subset of those and engaged them via texting as well as with case managers.
Based on this initial 700-cohort pilot, we saw a 24 percent increase in prenatal visits, and one of the key goals was to be able to get the right women to see their provider through outreach and education that took place earlier in their pregnancies. Through that outreach, as well as via other interventions, we saw a 27 percent decrease in preterm births, which resulted in a 54 percent reduction in baby costs per-member per-month. That’s a cost savings of more than $1 million in just this 700-member cohort. And, this was done with minimal engagement from the clinical teams. We felt like we needed strong proof points before we could get buy-in of the OB-GYNs and the clinical teams at our participating sites.
Broadly, healthcare has historically lagged in the innovation and technology departments. What makes you confident that can change as we move forward?
The payment models, which often have been a barrier to fast innovation, are [going to be] rewarding the drive to more population-based health models. This is slowly, but surely changing. There is also the recognition of the social-economic impact and factors that play into health, and that is also changing through mechanics such as CMS’ Accountable Health Communities Model. So, there are changing elements, and these are prime areas where you can apply advanced analytics and machine learning as you are taking the focus to a broader geography and broader group of key stakeholders who you’re trying to engage.
The progress on AI and analytics [generally] has been slow and steady, but I actually think that’s encouraging. I see it being applied more in some core clinical improvement areas, be it sepsis or readmissions models. I have also seen it be applied for supply chain [purposes], an area that can tremendously benefit form machine learning and AI-driven models, as well as some revenue cycle areas. I do believe that we are one of the very few, if not only [organization] that is applying [AI] to the intersection between health and SDOH, primarily because social determinants data have been elusive, and to apply these models you need to have large data sets and do it at scale.
Then there is the movement from predictive analytics to prescriptive-based analytics, which is not providing just a risk score, but also giving the reasons that are driving the risk score so that the clinical teams can take action on that information. We have been focusing on that a lot; pretty much every model we have displays, at the point of care, what the top reasons are for the score that is generated. And we have seen that drive even more adoption of those models.
