As the delta variant continues to drive COVID-19 infections, some hospitals are being overwhelmed with patient surges as they try to balance virus cases with incoming emergencies and other care.

Kristin Molina, business leader for patient engagement and healthcare analytics at Philips, sat down with MobiHealthNews to discuss how predictive analytics and artificial intelligence can help providers manage triage and patient flow, especially during times of crisis. The interview was edited for clarity and length.

MobiHealthNews: Why are patient flow and triage good places to use predictive analytics and AI?

Kristin Molina: What we see is hospitals more and more are needing to do more with less. And especially with what we've seen with the rise of the pandemic and overcrowding, that it's not just about adding more beds or staff or other resources.

But it's really about how to optimize and give the best care to each patient in the right care setting – and being able to anticipate and predict increases in demand, and that you can have the staff and resources available as you anticipate peaks in demand.

If you don't expect to have that demand, you're not having extra resources or staff around when it's not needed. But of course, with COVID and all that, it just seems to be a peak on peak, sadly.

But really, to manage patient flow, it requires that enterprise view across all the parts of the hospital, and the hospital network, and even what's happening outside the four walls of the hospital.

So that's where really the combination of bringing together clinical and operational data across different care settings and different systems, so that these care teams can have that full picture, and the situational awareness of what's going on in their unit, or their department or even at the enterprise level.

And so that's where we see being able to use predictive analytics, kind of driven by machine learning and AI, is really allowing health systems to have actionable insights into what that next best health action should be so that they can optimize the transitions of care and kind of unlock any bottlenecks in patient flow.

MHN: How has COVID-19 impacted your business in this area?

Molina: COVID has impacted patients and families, and staff and clinicians, incredibly. The burnout that we've seen, the lack of data, the overcrowding, the completely full capacity. Hospitals were just not able to scale up their operations quickly enough to handle these surges in patients.

And then, especially as we look at the ICU, you know that they had capacity shortages, having to create ICU beds in new parts of the hospital or convert general ward and other hospital departments into ICU departments so that they could try to manage this influx of patients.

So what we saw is really that technology with AI to really help distill all this vast amount of data into what was most actionable so that they could best treat their patients and triage them and really help there.

One of our E-ICU programs, we were able to help scale the capacity of a number of our health systems or hospital customers there. Because they were able to augment their limited staff that was in the actual ICU department, but using these technologies and telehealth technologies, especially in the teleICU, to be able to support the bedside staff.

And really it's got the cameras to be more eyes and ears there to help manage and really give more support to the staff that was under such constraints. And so we, in COVID, were able to do some things with our business model to really be able to help some customers scale up much more quickly than what they were expecting in the early part of 2020.

MHN: Do you find providers are willing to trust predictive analytics and AI? Or do you think that maybe has evolved over time?

Molina: I think it's evolving. We're getting better and better AI in general. But I think it's really important that we always focus on delivering that innovation that is people-centric, both for patients, but also the clinicians or the administrators that are using this AI.

And then it's all about helping augment, and especially around clinical AI, helping the clinician do better. We're not trying to replace clinical decisions or automate clinical decisions. We're really trying to distill things that weren't even possible, there were just too many data points.

And now, as digitalization expands, and there's just exponential growth in the number of data, we really do see providers more and more embracing that AI, because it's just not possible to keep up without some level of AI and predictive analytics to be able to really understand that data and what it's telling you.

We have some algorithms in Philips' IntelliVue Guardian that help streamline some of the manual processes there. But then it gives this actionable insight to the clinicians and helps them identify if there's deviations in the patient's vital signs.

So this is really having an impact on the ecosystem longer term, as it's helping them to reduce their patient transfers to the ICU by more than 60%. So it's not about just trying to make something to say, "Yeah, we took a lot of data, and here it is." But actually then helping to say, "Okay, here's an action that you can then take."

So I think that that's important. It's not just trying to predict something, but then helping to guide. How can you use that? Predictive analytics is evolving from that predictive to prescriptive analytics over time.

MHN: How does clinician burnout factor in when considering using this type of tech?

Molina: It's really important that it always fits into the workflow of the provider, the clinician, the administrator. We want to really be seamless, and help and really drive that impact. Not, "Oh, we have this great tool, but now you have to go log into an entirely different system to use it or manually enter."

We have to work really seamlessly or take steps out of the workflow to help make better decisions, but more efficient decisions, and not be an on-top-of thing, because then, when time crunches happen, you just you don't have that adoption.

So for us, for Philips, it's very important that these are really centered around how people will use [them]. That the data visualization is very clean and intuitive. And that it really fits within the workflow that they're used to, and that it doesn't come in as an extra step on top of.

MHN: How do you manage ethical concerns around AI?

Molina: I think there's a couple elements to that. Of course security, and really ensuring that all the data is secure and compliant with all the local requirements. And, of course, we believe in, leverage and utilize industry standards, like APIs, but then also HL7 and FHIR integrations.

But then the other important element is the bias in AI. For Philips, it's all about embracing the fairness, as a guiding principle, to really promote the responsible use of AI. So as we are developing and validating all of our data, we really put an emphasis to make sure that it's representative of any target group for the intended use of our proposition, and take steps to avoid bias and discrimination.

I think this is an area that's getting a lot of focus and intention, and especially from Philips, as we are really investing in data science and AI. Ensuring that we've got all the education and training around that in our end-to-end data science, from the development to the support and implementation of these algorithms.

And then working with our customers, and the CIOs and the other healthcare IT leaders to make sure that we're doing this together. Blind spots are something here. This is very new and evolving, and we're all continuously learning and committed to continuous improvement.

So really teaming up with our customers as well to make sure that their staff are also learning from all of this, and that as they implement and use this, that they have sufficient processes in place on their side so that they can monitor how these algorithms are working, what their performance is.

Is the data quality acceptable and what we intend? But I think for all of us, it's really building in that diversity, and Philips is committed to the diversity in our people and the data and in the validation.

MHN: What do you think is the next frontier for using predictive analytics and AI in healthcare?

Molina: It's really going from the predictive to the prescriptive. It's one thing to tell a customer, "We anticipate that this patient may be deteriorating, and there's an intervention." But then how can we help our customers say, "This could be an appropriate intervention," or, "Based on evidence, this is the right intervention or the right care setting for that patient."

And on the flip side, we see a patient. They're showing clinical improvements. They've been stable. So we've identified that this patient can safely and with the same quality of care be transferred to a lower-cost care setting. So, then, identifying that this is the time, these are the steps in helping to line up all the operational steps that also need to go in place to transfer them.

And then, even more, how can we get better AI when the patients are not in the health system, and be more predictive about the patients as they transition from, say, an acute setting into back into their home? Data quality might not be the same level, there may be inconsistencies there, but having the algorithms that are able to adapt for that, and really be personalized around a specific patient.

And then importantly say, "Here's an intervention that needs to happen so that patient avoids a readmission or an acute event." Again, it's really helping to take the step up from not only predicting, but then to help guide what that next best health action is for that patient.