This article is sponsored by PointClickCare. This article is based on a Skilled Nursing News discussion with Cheryl Field, Group Product Leader for Analytics at PointClickCare. The discussion took place on September 1, 2022 during the Skilled Nursing News RETHINK Conference in Chicago. The article below has been edited for length and clarity.
Skilled Nursing News: I’m really excited to kick us off with a discussion on how innovation can improve advocacy and care collaboration. I’m joined by Cheryl Field from PointClickCare, which is a health care tech platform that spans the care continuum with users including over 22,000 long-term care and post-acute providers and 1,600 hospitals. Can you just start by explaining what do you mean by advanced technologies?
Cheryl Field: Advanced technologies are artificial intelligence and machine learning technologies that we’re utilizing, leveraging the data that we have within PointClickCare, to move forward with aiding the decision-making of caregivers. I’m pretty sure many of you are quite familiar with machine learning and advanced technologies. How many of you used an application for traffic this morning?
As you look at your app to decide, it gives you some predictions about traffic and you made a decision about an alternative route. We do the same thing with weather. We look at a weather map or weather application that gives us some predictions about what’s going to likely occur. Then we actually think about the details of that prediction and then we make a decision. With a 10% chance of rain, you probably skip the umbrella. If it says complete monsoon, you’re going to bring an umbrella and a raincoat. Predictions help us to actually make different and better decisions. That’s really exciting in the area of senior care where we can leverage the data that we’ve been collecting to inform future caregivers and use those advanced technologies.
SNN: Can you elaborate a little bit on that? Why is PointClickCare specifically investing in machine learning?
Field: It’s an investment that we make really from a sense of purpose and responsibility. The simple easy answer is for better outcomes. The real more thoughtful answer is that I’m a clinician, I’m a nurse. In the clinical space as a nurse, I have gone to school. I have data in my database and I have all of my experience in that data. Now I’m going to take a look at you and I’m going to go in as a nurse and ask, “How does Tim look today?” “Is he good?” “Do we need to return him to hospital or is he okay?” I’m using my data and my experience, but I’m filtering in real-time information about respiratory rate.
If I’m a nurse and I go into that room and I’m not sure what I’m looking at, or I’m not sure if what I’m seeing is enough stability, I’m going to go get another clinician. That clinician is just a database of knowledge and experience. Today in your clinical setting, nurses gather other nurses and other consultants to help make decisions. A decision is frankly, just a prediction and you’re making a prediction based on the data you have. What’s really cool about machine learning and why we invest in it is that we can make the decision and the prediction based on all of the data.
It’s not just one or two or three databases. It’s all of the data and PointClickCare’s database that’s used to train machine learning models and make predictions about the future of care. That’s really powerful. What excites me is thinking about taking that power and putting it next to a clinician who’s making a decision.
SNN: Can you talk to us about how PointClickCare sees that happening? How do you leverage the advanced technologies to support that clinical decision-making process?
Field: That’s the art and the science of the end product. What we’ve been talking about is very math heavy and again, as a nurse and someone who’s worked in analytics for a long time, I have to first go in and start to understand. I want to understand the math, but then I need to put it into the hands of clinicians and it has to make sense clinically.
We all care about predicting the future, whether you are a payer, a provider, or a patient. For any of us who’ve ever had a change in health status your first question is, what does this mean? What does it look like? What’s that prediction? Is it a short illness or a long illness? What’s the trajectory? Payers and providers feel the same way. We all want to predict that future state in an ideal sense.
We’ve worked on thinking about predictions that are important to leverage around quality of care, collaboration of care, cost containment so that the health care ecosystem can achieve its goals from a triple aim perspective. In that way, we have begun working on a model that predicts the likelihood of return to hospital.
Also, we’ve made that same information available to hospital liaisons and ACO payer partners. When we’re collaborating on care in real-time, we can be looking at the members’ risks that are changing as they’re within skilled nursing, that post-acute experience, and then working together to mitigate and reduce and see those things ahead of time.
SNN: That’s really interesting the implications it has for care across the whole continuum. Do you have an example you can share of how technology is impacting that care coordination?
Field: As I said, when we put the technology in the hands of health care liaisons, care coordination collaborators, whatever that title is, there’s a human being whose job it is to manage and monitor that member within a hospital network system. Then there’s another human being whose job it is within a nursing center to identify and ward off those risks before it’s occurring.
Again, thinking of return to hospital as a specific risk and knowing the ability to predict those risks in the next seven days and organizing the data in a way that’s clinically relevant to the clinicians who are doing that job has helped us to reduce those returning to the hospital.
What’s really interesting with machine learning is the trends that the neural networks can see. Again, I have a database and you have a database and if we put them all together with enough time and coffee, we could identify these same trends. Machine learning can identify trends and trillions of patterns very quickly. You can make a prediction in seconds that would take a human hours or days or months to assimilate and understand all of that data.
When we put these tools in the hands of those care collaborators, it’s fascinating to see how early they can get ahead of a feature like a change in the median of an O2 saturation, a change in something that as humans we don’t really think too much about. I might look at vital signs and make sure that they’re normal, but what is the trend? When we start to see trends that we can act on sooner, we can ward off unnecessary health outcomes. That’s really fascinating to see.
SNN: Can you talk more about the use of human clinicians and their predictions in building the models that bring this value to the long-term and post-acute care market?
Field: In thinking about machine learning, I think the first pushback you hear is to replace a human. In no way are these kinds of technologies replacing humans. As I described, the way clinicians make decisions today is by gathering other clinicians. We come together with our data and our experience. We come to a consensus, we make a decision.
Today’s staff are stretched so thin and they’re turning over. We have young nurses, agency nurses, float nurses and folks who just don’t have time to know the history of that patient with the speed that they would want to know. By putting a tool like machine learning alongside the clinical decision-making of that staff member, it gives them all of our brains, all of the data brains involved in that decision that they’re making, and then they make a better decision together. That’s been really effective in thinking about how that model is going to be imported into the workflow, the job to be done and the value that it can bring into that health care ecosystem.
SNN: It would be great to hear your take on who benefits from this predictive return to hospital model.
Field: We all benefit. It’s the predictive state of the future. You guys want to know the winning lottery tickets, there’s machine learning models working on that too, but that’s not where our purpose and our drive is. Again, as I really think about those folks who benefit, it’s that patient who is in the center whose homeostasis has been disturbed, who’s having a change. Nobody wants to go backwards. We all want a one-way direction home back to homeostasis, back to the place that we call home. Our providers want that for us, our payers want that for their members. Everyone benefits in that when we use and we harvest these newer technologies and we put them in a very responsible way into the workflow.
That’ll just go back to the question you asked me about using humans. When we did this work building the model, we first started by engaging with experts, clinicians, nurse practitioners, medical directors, and asked them, “What data are you using? Do you make these predictions? Could you tell us who’s likely to go back to the hospital?” They made thousands of predictions and we tested those predictions against our models predictions.
It’s really interesting to learn the biases that we have as humans. We want to get it right, so we have a tendency to over-predict and to overstate risk. Sometimes that’ll cloud our judgment, that’ll cloud our staffing. If everyone’s high risk and everyone gets X number of services, we don’t have the staff to stretch that far. With a model that can be helping to identify those folks that are high risk and actually with more accuracy than those humans, and that’s this model that we’re using. It’s predicting more accurately than our human experts, because of the biases, but when you put those two things together, when you just take pattern recognition that can come from the model and you put it in the hands of a clinician, everyone is going to win.
Most importantly, that patient is going to have early identification of concerns and the model’s going to drive the next decision that the clinician makes. Whether they’re going to go gather more data, which is frankly what we all do. We gather data, we make a decision, we iterate, we gather more data. There’s no such thing as making a mistake, you take action. When you gather data, that data might tell you that you don’t want to take that action again and then you iterate and make a decision.
We’re all going to benefit from these kinds of technologies, but I think today where the staff is not only so stretched and so changing, but also they’re so accustomed to having innovation. Every single one of your nurses use a weather app and a traffic app. Why wouldn’t they expect to have the same technology, leveraging the data from senior care to guide them in the decisions that they’re making? At the end of the day, the staff members are satisfied, the patients get better care, better care costs less dollars, everyone is happier in the end.
SNN: Is PointClickCare providing a risk score specifically for return to hospital and what’s the confidence level on that score?
Field: Yes, we do have a prediction of the likelihood of return to hospital in the next seven days. When you talk about the confidence, the accuracy of the model has been vetted against the baseline predictions from our human experts. We had humans making the predictions on the exact same patients that were then tested by the model. 97% of the skilled nursing patients in a short stay have an accurate prediction more accurate than humans today coming from PointClickCare, which is really exciting.
Now nothing’s perfect. You’re not perfect at making a decision, and when we look back at the number of times that a model will predict someone’s high risk and likely to go back and then the staff really works hard to mitigate those risk factors, and the result is they don’t go back, which is a great result. That’s the result we want. That might be a little confusing to a model who thought, “Oh, wait a minute. I thought Tim was going back and then Tim didn’t go back.” We’re always feeding that data back into the model and monitoring that model over time for its accuracy.
Again, never intended to be the deciding factor, but to be alongside a clinical decision maker and help to identify whose high risk and whose risk is changing. Significant change from yesterday to today, even when you’re within the low or medium category can be an early identifier that something’s on the move and getting ahead of those changes and those features, which we put into clinically meaningful packages so clinicians understand those features are delivered to that clinician so that they can see that. That’s how we’ve addressed that, we do have a model. It is accurate and in the early stages of being integrated in clinical practice.
PointClickCare provides fully integrated solutions powered by a mobile-friendly and regulatory compliant EHR and revenue cycle management platform. To learn more visit: https://pointclickcare.com/.