NYU Langone Health oversees a clinically integrated network that covers some 375,000 attributed patients. As such, it depends on a strong care coordination effort to help manage the health of its high-risk patients and steer them to the appropriate care venues.

But the health system knew it could do better when it came to tailoring outreach and boosting patient engagement. While much of its care coordination took the form of telephonic interventions, NYU Langone also sought to drive more effective uptake of its Epic MyChart patient portal.

So it did what so many other hospitals and health systems are doing nowadays to guide their improvement efforts – it turned to data science to discover hidden trends and insights that might help point the way toward improvements in its population health management.

At the HIMSS Big Data and Healthcare Analytics Forum in Boston on October 22, two NYU experts – Harry Saag, MD, NYU Langone's medical director for network integration and ambulatory quality, and Simon Jones, MD, professor in the Department of Population Health at NYU School of Medicine – will explain some of the data-driven insights they've discovered. They recently offered a sneak preview to Healthcare IT News.

Answering the big question: How to be better?

"In current state, probably what most care coordination shops are doing, is you have your list of high-risk patients and you're getting that from whatever data sources, or whatever analytics platform your organization is using," said Saag.

"There's a plethora of vendors," he explained. "There's a plethora of analytics algorithms and platforms that can tell you who will be costly, who you should be worried about if you're in a value-based world. The real challenge is going to that next step and saying, 'Who is not only at risk of becoming ill and becoming costly, who will actually respond to a telephonic care coordinator?"

At NYU Langone, "if we call four patients, the math would say that maybe one in four would pick up the phone, and not only pick up the phone but say yes and accept care coordination right off the bat," said Saag. "After that, the patients who do best are the ones who actually engage their care coordinator longitudinally."

Given that those engagements can last as long as eight to 14 weeks, "as you can imagine, not every patient makes it from that initial acceptance to the finish line," he said.

AI algorithms help predict patterns of engagement

The goal was for NYU Langone, using its existing data, to create a predictive model that doesn't just show which patients may become ill and become costly, "but take me to that next step: Out of your pool of patients who are at risk, this is where you should start,” Saag explained. “These are the patients where you can have the biggest impact and the highest likelihood of succeeding using this care coordination intervention."

Jones added that, "We use a really quite good machine learning technique to predict the likelihood of patients actually engaging in our scheme. Some things, such as how recently they'd had contact an inpatient at our hospital, seemed to be a good predictor of a patient's likelihood.

“Also, how far they are from our hospital – unsurprisingly, patients further away are less likely to engage with us,” Jones said. "It turns out we can actually quite accurately predict whether a patient will engage. We're hoping – and we haven't yet proven this – that using this information we can increase our reach rate."

Beyond the telephone: Data patterns to improve portal uptake

"It's 2018: Part of the challenge is we're relying so much on 1920 technology," said Saag. "And NYU Langone has a very robust patient portal. We're an Epic shop, we use MyChart, which is a fantastic technology and an incredible way to communicate with patients. So can we start understanding who is engaging with it? And can we figure out, when we start engaging that modality, who is most likely to engage in a meaningful way?”

In recent months NYU Langone has been running the numbers on data related to its patients' MyChart usage, he said.

"First of all, who's signing up? We know there's consents, there's downloading the app, there's using the app. But who is signing up at baseline? And then the second question: are you actually engaging with it, interacting with it, using the functionality. Who's looking at their test results? Who's using secure messaging to interact with their provider and their care team?"

Saag's care coordination unit enlisted Jones and his team of data scientists to answer a few key questions: "Can our current data tell us anything about the types of patients – telling us what we can know, not only about who's on MyChart, which is only part of the battle, but who can we really reach and engage with? And what variables make a person more or less likely to use this new technology?"

Intriguing results spotlight new populations to incentivize

"There are some that aren't surprising, such as age: People between 20 and 84 are most likely to sign up," said Jones. "But there are some really interesting things too. For instance, if you get someone who's over 84 to sign up, they have a very, very high chance to be a really significant user of MyChart."

Among his other findings, "we didn't see much difference between men and women. We did see that language preference is highly correlated. We have an English and Spanish version, but the Spanish portal is only available as a desktop version at present.

"Still, we found that uptake of the Spanish version was slightly lower, but if they did sign up, they likely were no different in usage rate to English speakers," he added. "So we need to think a little bit harder about how to get people to sign up whose first language isn't English."

NYU Langone can also make progress by paying close attention to demographic data: "We know that people who use it tend to be more educated, and tend to have a higher median income," said Jones. "So we need to think about messaging and how we get people in, who perhaps don't fit those demographics to sign up for MyChart."

Targeting 'ZNA' for better patient engagement

It's not just education and income that's crucially useful, of course. In a diverse and densely-packed city such as New York, location matters a lot.

"We created heatmaps of the New York area and where our usership is, where folks are using various features of the portal," said Saag.

By focusing on what he called their "ZNA" – the ZIP Code someone lives in – NYU Langone had gotten a much better handle on where it can get the most traction with its patient portal.

"The telephone may work for some," he explained. "MyChart may work for some. But there are some patients for whom neither of those modalities are the best way to manage their health. In recognition, NYU Health launched a community health worker program - they are our boots on the ground, to be that care coordination extender."

It's in instances like that, with actual human beings heading out into the field, where smart, data-driven insights about responsiveness to care interventions are especially useful.

"We're using our own data – not only to tell us who to think about engaging with, but also to help us set our daily schedule and outreach list," said Saag. "We can have these nice heat maps and set up a daily agenda that we know makes sense – maximizes the efficiency of our boots on the ground.”

Not just at NYU – you can do it too

While NYU Langone is an academic medical center, with all the resources and brainpower that entails, Saag and Jones said the lessons they've been learning from their own data are translatable for other providers, regardless of how big or small they are.

"For any provider that's moving into value-based care, I think there's no question that any group that's going to take on risk has to have a care management strategy," said Saag. "You've got to find a way to engage patients, keep them activated in their own health, and engage them in their community in the lowest cost venue that can manage their condition – whether that be at home, an ambulatory clinic, an urgent care center, an ED, an acute care hospital or a post-acute setting, you need a care management strategy."

The key, he said, is that maximizing the value of your care management infrastructure depends on tailoring locally to your own setting. "And to really do it well and get the maximum chance at ROI, you've got to pair it with a thoughtful data science and analytics strategy, so you're maximizing both atmospheres simultaneously and getting momentum to really make a difference.

"You have to be thinking about both of them in tandem," he added, "with the ultimate goal of becoming that aspirational learning health system, where all the data you're generating in your patient care, whether in any of those venues or just making phone calls, you need to be capturing that data in a way your data science team can be using to constantly iterate and reevaluate their models to improve performance."

Twitter: @MikeMiliardHITN
Email the writer: mike.miliard@himssmedia.com