Transcription:

Bill Klaproth (Host): This is a special podcast produced for the 26th Annual Healthcare Marketing and Physician Strategy Summit known as HMPS, October 6th through the 8th in Aventura, Florida as we speak with session presenters and keynote speakers. I'm Bill Klaproth. With me is Chris Hemphill, Vice President of Applied AI at Actium Health. He is also the host of the Hello Healthcare podcast. And also with us is Ryan Younger, Vice President of Marketing at Virtua Health. Their session, AI's Impact on Influencing and Engaging Patients.

Chris and Ryan, welcome and thank you so much for your time. So Chris, let me start with you. In healthcare, AI has demonstrated the most success when focused on growth with tight ties to organizational strategy rather than cost savings. So what are successful uses of AI for patient and physician engagement?

Chris Hemphill: Great question. And I like that frame up that you did, tying it to strategy. Because one thing that's really important with these AI growth use cases for conducting outreach to patients is making sure that these efforts are tied to existing strategy. What service lines are we focused on growing? Do we want to focus on growing, for example, the mammograms that we're conducting or the cardiology consults that we're conducting?

By tying the outreach strategy and way AI is applied to that overall strategy, you're starting to build up a use case for success rather than trying to use it as a tool and do something entirely new that the organization wasn't originally focused on. So when thinking through, we're thinking about from the patient engagement perspective and the marketing perspective, what are some of the things that marketers are focused on already?

Well, when it comes to this outreach, there's the overall strategic goals. There's selecting and building audiences. There's content to be developed. There is the concept of finding the right channels for the right people across that content and even optimizing and analyzing performance.

Each of these use cases, there are AI capabilities that can help -- even with the developing content cases, there's AI capabilities that I would say are more or less advanced at being able to do that. But one place to really focus on is making sure that we're getting the message to the right people, selecting that correct audience. If we look at the challenges that people have had in the past with selecting audiences, there's been the pre-2010 definition of selecting an audience was basically mass marketing. Put up a billboard. But once there was a proliferation of more data, meaningful use brought in this brand new concept of EMR, where people could slice and dice audience segments and things like that. That allowed the targeting to get a little bit more segmented and focused, but still presents its own challenges.

One example of that is with your typical cardiology campaign, if you're just making slices and dices in the data, and you want to reach out to people that are most likely to need cardiology services, a common practice is to identify men who are above 45 years old and women who are above 55 years old and send the cardiology communications out that way.

Well, yes, that's using data that's being a little bit more data-driven and a little bit more focused, but it's creating a bias within the system. It's creating a bias that says that anybody below 45 or anybody below 55 across these gender lines isn't going to be receiving outreach for this campaign whereas anybody above those lines are going to be receiving outreach. Well, we want to reach out to people who have the most need, but it becomes infinitely more complex to say, "Well, if somebody is below 45, but their systolic blood pressure is this and their diastolic is that, et cetera," we want to be more nuanced, but the rules become so complex and on and on and on, that it becomes extremely difficult to form those audiences.

So one big way that AI addresses those challenges is to actually identify all of the nuance that's required to make sure that across the gender spectrum, across age lines, that outreach is maximized to as many people as possible who have that specific need. We're not just blindly saying that anybody below 45 years old doesn't need our outreach. And lo and behold, there's been some major results from that. Like if we're looking at audiences that responded to outreach that otherwise wouldn't have been contacted. And one campaign that we measured, at least 27% of the growth of that campaign came from people that were below that 45-year-old age mark. I don't know if it spoken too much, but I thought it was important to kind of frame out, "Hey, here's a challenge that people do, and here's how AI/machine learning approach has helped with that challenge.

Bill Klaproth (Host): Ryan, you want to chime in on that?

Ryan Younger: Yeah. Thanks. I think Chris covered it really well. I guess I would just say that AI has had a big impact on how we influence and engage patients at Virtua. And really that means cutting through the clutter and personalizing our messaging to be very relevant for distinct audiences. And that can be any number of things, but one of the ways we've really used it as use case is in the reactivation strategy for people that had medically distanced during the coronavirus pandemic, which obviously still is peaking in certain places.

But, in Southern New Jersey, at the end of last summer, early fall, that's when we were trying to really bring people back that had medically distanced. And we looked at people that had heart disease, that had not done some of their screenings like mammograms, maybe had been in some awkward workstations and their spine had issues, digestive disease. So there was a lot of different areas. This really helps us cut through the clutter because it helped us target people that had a propensity for certain disease. And then we could deliver up very relevant messaging for them and really focus on our content to make that known to people and really connect with them. And I think that's how we've used it at Virtua within that broader context that Chris just described.

Bill Klaproth (Host): So it sounds like AI can really help you with targeting, identifying, and qualifying an audience. So Ryan, let me ask you this. When it comes to metrics and I know you'll cover this more in your session, when it comes to metrics, can AI be used to understand which initiatives perform best?

Ryan Younger: Yeah, for sure. We've used it very much as an activation and growth strategy. And we're constantly looking to optimize. We look at that as a strategy in its own right, because sometimes there aren't simple benchmarks that we can go to. We sometimes are just constantly improving what we were doing in a campaign, for the next campaign, or even within that campaign with various A/B testing, we're doing of subjects and images and topics and conversions.

So, we've seen a lot of growth efforts. We've seen higher engagements for people clicking through because they have relevant content and then what they do next, taking that next step to make an appointment and to kind of take that next best action.

Chris Hemphill: Yeah, dovetailing on that too, I like that frame up on how the organization's using it. We can go back to that kind of audience selection example. It's not just the step of identifying who has the need for that service, but once you've done that, you start aggregating those numbers and you start looking at those risk levels across populations, across market segments, across payer types, et cetera. Then it starts answering the question on how to better relate to each set of patients within those groups. So by identifying that risk kind of independent of like ZIP codes, locations, and demographic factors, once you're able to do that, then look at how many people are at risk for cardio illnesses or conditions within this particular segment of the market. It opens the door to start getting more granular about the type of outreach and the way that you're going to be promoting relationships within your market. So that's one area that I see terms of matrix.

Bill Klaproth (Host): And Chris, in your first answer, you gave an example of targeting men over 45. And you said, "Hey, that created a bias because what about the men under 45?" So let me ask you this. When it comes to AI, how do we ensure that these types of approaches aren't perpetuating bias in race, gender, and other areas? You guys talked about ZIP codes too. How do we make sure that doesn't happen?

Chris Hemphill: The first way to make sure that doesn't happen I think is to acknowledge the fact that it can happen. A lot of people have phrases like the data never lies or data's unbiased and things like that. And a lot of my work actually is focused on highlighting that that's not true. So before we can start addressing the issue, we have to acknowledge that the issue exists. And, then, of course like from the AI developer's perspective, demand better of the algorithmic approaches.

So with that first step, identifying bias, there's a quote that I always say from this major healthcare and machine learning ethical AI researcher, Dr. Ziad Obermeyer, which he says that AI kind of presents a crossroads. If we're to build out our AI solutions based on existing data and not acknowledge the bias that generated that data, the processes ultimately that said based on people's ability to afford care or based on how often they show up in the EMR. If we're not to acknowledge those biases, then what we're going to end up doing is perpetuating that bias versus if we start looking at how our models perform by gender, by age, by race. When we start looking at that and start making adjustment and adaptation either to the data themselves or the outputs of the models, then we're in a position where we're not perpetuating by bias, but instead fighting that existing bias that came to us through the system.

So, overall, I would say that you first have to acknowledge it. You can't get defensive when it comes to the concept that, "Hey, maybe our business processes, maybe the economics of the area, maybe the affordability of the services we offer, there's a whole slew of where different types of bias can fit into the system." But what AI presents is the chance to quantify that bias, look at the impact, look at how much outreach is going to whatever groups based on those kind of protective factors that I was mentioning earlier and start making adjustments and adaptations to the models and to the overall strategy to counteract that bias.

Ryan Younger: I thought you answered that really well, Chris. I would probably add one point from again just adding an organizational perspective from Virtua is I think not only do we acknowledge that it exists, but going back to your point earlier about strategy, we have an organizational commitment and strategy for how we want to address bias and how we want to ensure greater diversity and inclusion. And so it permeates everything we're doing, not just our growth strategies related to AI, but how we recruit and how we engage our workforce and lots of other elements that we've really focused on over the past couple of years especially.

Bill Klaproth (Host): Yeah, that makes sense. I like what you both had to say. First, acknowledge that there is bias. And then, Ryan, you were saying have a strategy to address bias probably in your mission or vision statement and make sure you're trying to live up to that through your research in AI. Well, this is going to be a great session. It's called AI's impact on influencing and engaging patients. If I could get your final thoughts, each of you, Ryan, let's start with you. Anything you'd like to add?

Ryan Younger: I'm just really excited for the session. We haven't all gotten together in a very long time. And so it's going to be an exciting time. And in regard to our session, when we think about AI, we often talk about technology is a really critical enabler. But it doesn't take the place obviously of the broader trend going on with consumerism and how we connect with people. And so the technology is really a tool to do that, and it's gotten better and better all the time with what we can do and how sophisticated we can be with how we reach out to people. But we always, again, have to remember that it is a tool.

Bill Klaproth (Host): Right. Thank you for that. And Chris, anything you'd like to add?

Chris Hemphill: Ryan, I think you summed it up really well, is that sometimes a lot of these conversations can go a little bit too much on the technology without acknowledging the people that are having to put it in the place, the organizational culture that's going to have to adjust to maybe we're a little bit more metrics-driven or maybe we use these particular KPIs to guide our strategy. So I hope that the session helps not just understand some of the technologies and the later stuff out there, but really it's going to be exciting digging into how Virtua has incorporated this stuff into their existing strategy, not just from a growth perspective, but also, like Ryan said from a DEI perspective that you've mentioned eariler.

Bill Klaproth (Host): Excellent final thoughts. Thank you both for sharing those. And we are all looking forward to your session and thank you for giving us a preview on this podcast. We appreciate it. Chris and Ryan, thank you so much for your time.

Ryan Younger: Thank you.

Chris Hemphill: Appreciate you.

Bill Klaproth (Host): That's Chris Hemphill and Ryan Younger. And for more information, and to register for the 26th Annual Healthcare Marketing and Physician Strategy Summit, please visit healthcarestrategy.com/summit. Thanks for listening.