Healthcare AI expert Peter Borden, managing director at consulting and services firm Sapient Health, helps healthcare organizations apply innovative AI technologies to their ecosystems.
In this Q&A with SearchHealthIT, Borden talks about how such AI in healthcare applications helps with clinical trials, customizing post-discharge instructions using patients' personal characteristics and population health.
What are some of the top use cases and conditions to which advanced AI is most applicable in healthcare?
Peter Borden: The ones that people are getting most excited about are clinical diagnostics, both for outcomes [and] ... cost is associated with that, and I think that's really exciting. When you combine digital pathology and AI, that's sort of mind-blowing.
Ultimately, it's around finding people who need really contextualized engagement and then giving them support so they can achieve their goals. We think about it as dynamic segmentation and enhanced population insights, and precision engagement once you know who to target.
We're excited about a growing market and personal health diagnostics, which is, I think, different from how diagnostics [is] currently being thought of within the health space.
How will new forms of AI in healthcare affect transitional care when patients leave the hospital for other settings?
Borden: AI can be used in part for much more contextualized discharge planning. What kind of support do people need, what literacy level are they at, what kind of help do they have, who are the different people who [are] either within their community or their family or their care team that can engage with them, and in which ways? Do they live in an urban or rural setting?
So AI in healthcare should make use of socioeconomic health factors as well?
Borden: Absolutely. Many times, people get discharged on some sort of plan that doesn't actually fit in with the context of their life. AI ... can help look for ways to identify people to find the circumstances under which we should provide these services. Maybe they need transport services to get to appointments.
Peter BordenSapient Health
But, also, when you engage people, when you give them alerts and reminders, the way in which you do it is pretty important. So it could be channel selection, but it could also be tone selection. So we're thinking a lot about how understanding of behavior and personality will affect how you engage with a certain message. How would you engage an extrovert to use a Fitbit, and how would you engage an introvert?
So if you start blending some of the clinical data sets, some of the access data sets, with some of these behavioral data sets, you can really understand how to personalize and contextualize.
How could emotional intelligence help AI in healthcare applications?
Borden: Imagine if you added voice to [the] analysis of media usage, digital usage and social commerce. We're doing a lot right now in combining those data sets to get to sentiment analysis, intent analysis and behavioral analysis. Our health team is starting to use this for clinical trials, for example, to find people who are good candidates for trials.
Once candidates are in the funnel of consideration, then you can use different types of AI that will help winnow down to [what is] the right target set based on a variety of factors, inclusions and exclusions, and [what is] right for complex clinical trials. This could take months, normally, for teams to get to, but we have some tools [with which] we have been able to get to it in hours.
As opposed to clinical trials or drug testing, can you see AI in healthcare being incorporated into the big EHR and cloud-based population health systems that providers use?
You want to look at your population to find out who are likely candidates for some future event; like, let's say, going from prediabetes to diabetes. Then you want look within that set to see who are the ones with various complications and [who] are going be more expensive to treat. Who are the ones who are likely to respond to interventions, and who are the ones who are not necessarily as likely to respond to intervention? So you can figure out which are the different segments that we should really focus our efforts on, given the likelihood of progression, given the likelihood of response and given the likelihood of impact from a value perspective.
Dealing with interoperability and security in AI in healthcare
Practical applications of AI in health IT
Healthcare AI has a role at HIMSS 2017