Artificial intelligence and risk stratification – A winning combination that is improving patient care

 

As the focus on value-based care grows, the ability to triage the needs of cancer patients who are most likely to experience challenges in their care is becoming a larger priority for many community oncology practices. That is why Cardinal Health Specialty Solutions and Jvion, a leader in artificial intelligence technology, partnered to create an artificial intelligence (AI) decision support tool, the CORE^TM, that helps oncology specialists identify high-risk patients and assists with managing their care.

In this interview with Northwest Medical Specialties Medical Director, Dr. Sibel Blau and Jvion Chief Medical Officer, Dr. John Frownfelter, we learn how AI can be used for risk stratification to assist clinicians as they work to provide the best care possible to their patients.

Dr. Blau: Our practice had been in the process of implementing alternate payment model and practice transformation. We realized one of our biggest problems was risk stratification or trying to predict patients’ illnesses ahead of time. We wanted data in real-time rather than using data provided to us from years before. We always acted on the historical data, but for real-time, we didn’t have any alert or risk stratification process in place.

We are a member of Quality Cancer Care Alliance network, which is a clinically integrated cancer network for independent oncology practices across the nation. Our practice CEO heard about Jvion from another practice, brought the opportunity to us and asked, “What do you think?” We jumped at it. It was right on target with what we were trying to do.

Dr. Frownfelter: Jvion provides clinical and socio-economic insights into what is going on with the patient. We can identify the risks for individual patients for “harm events.” We wrap that around an outcome that we are targeting, such as who is at-risk for severe pain, risk of mortality, or deterioration in the next six months, etc. These risks are the drivers for that event occurring.

We provide prescriptive analytics, which goes well beyond just risk predictions to guide the best actions or interventions that are related to socio-economic or clinical factors that can change the outcomes for the patient.

Dr. Blau: One of the benefits of the CORE^TM tool is the ability to identify patients who may be ready for palliative care. For our plan, we had a list for 30-day mortality. We were still putting a system in place, still perfecting it. One of the patients on the list had recently had a CT scan and was doing well. We wondered, “Why is she on the list?” We agreed to bring this patient in, she needed to be seen. She was elderly and had to be brought in by a family friend. At the clinic they drew her blood. She didn’t look sick. Nothing was clinically wrong with her. But when she spoke, she voiced that she, “…didn’t want to do this anymore.” By the time she got home, she collapsed. She was taken to the emergency department and they discovered she had urosepsis. Thankfully, she improved, but she had been experiencing failure when she was taken to the emergency department. This woman lived alone, and she was not very proactive. If her friend hadn’t brought her to the emergency department, she would have died.

Dr. Frownfelter: We have more than 30 million patients in our database, so we are connecting thousands of dots about every patient, their story, and their outcomes every day. Our machine understands connections, learning relationships and drivers about features of each patient and what is leading them into that outcome that we are trying to prevent.

Our brains cannot process thousands of data points on patients simultaneously. At best, we can process six things simultaneously, if we’re sharp that day. This AI tool can process thousands of data points simultaneously and draw the connections.

Another trap clinicians fall into is the concept of heuristic decision making. A physician will think what they are seeing is familiar. They’ll think, “I’ve seen this before,” and then they will come to conclusions quickly. This is wrong decision making that leads to a form of bias. The physician will think, “This is similar to what I’ve seen before, therefore it must be related to what I’ve seen before.” This perspective colors decision making. The Jvion machine can clean the slate on inherent biases, especially when it comes to things like social determinants of health and other factors that are creating risk for patients.

Dr. Blau: We realized we had to make some operational decisions for handling the data when we received it. It left us thinking, “Now we have the data, what do we do with it?”

Often, clinicians are intimidated by the analytics, or they are afraid to let go. The Jvion data showed we needed to provide additional support in key areas of the practice. For instance, we already had a palliative care system in place with APPs, but these AI alerts allow us the opportunity to have some of these patients come in earlier, in a timelier manner to assess their need for palliative care. The data wasn’t alerting or scary, but it led us to providing better pain management and better depression management. That part significantly changed for us. We realized we lacked in areas related to end of life or palliative care area in the past. By having this data, we were able to impact our numbers in a positive way and provide better patient care.

The data also helped build the case for a need to add acute care clinics (ACC) during the weekdays. Previously, we only had one clinic on the weekend, and we staffed it with the physicians and the APPs on a rotational basis. After assessing the data, we decided to hire two additional APPs and add them to regular ACCs during the week. We realized through using the Jvion CORE^TM that we had enough patients we needed to bring into normal clinic, outside of their regular schedules.

Doctors are busy, and when these patients show up on the Jvion list, it is not a quick visit. It takes time as we either need to identify a real illness, like sepsis, or perhaps we’re dealing with something that is more mysterious. We need to analyze what is going on with that patient. It may be a false alarm, but we need to pay attention.

We cannot rely only on whether the patient looks good because their labs could be abnormal. And, with palliative care, if we get into those discussions with our patients, it takes time. Additional clinics allow us the extra time we need, as well as the extra support. This has improved the care we provide, and it has improved our patient satisfaction survey results as patients appreciate the doctor wanting to see them. 

Dr. Blau: I see the benefits. I do not think an artificial intelligence tool means, “Oh my gosh, my job is in jeopardy. It’s going to find something I didn’t recognize.” This is a great tool. It enhances my delivery of patient care.

At the end of the day, we all just want good patient care. In oncology, people are so dedicated and protective of taking care of their patients. If patient care is going to be better, why wouldn’t I use this tool? I am still doing my job. I am still the doctor. I still see the patient and I evaluate and make the decision. AI is not diagnosing the patient. It’s alerting me that something is wrong, and it is telling me that I need to pay extra attention to this patient.

Dr. Blau: At our practice, we are not a very large group, but we have several different locations, and we are spread out. Despite this, we still communicate very well. For the adoption of the AI tool, it is very important to have dedicated training with physicians and other clinical staff. We realized we needed dedicated time, one-on-one with staff because everyone’s experience and understanding with technology is different. I didn’t encounter anyone coming into this with big doubt and paranoia, but bringing them into the education sooner is what I would have done differently. We worked to assign our people to the training, and they learned it deeply. But, looking back, I think we could’ve done it even better.  

Dr. Blau: To do this, you really need a physician to serve as a champion for implementing the tool and have them partner with someone in your practice who is administrative, who understands the flow of the entire organization. Then, have them identify staff to create a taskforce to support implementation, training, etc.

As we mapped things out at our practice, we learned that work flows all look very different. With key clinical and administrative leaders in place and their taskforce group, they can develop the overall team so they can put this into action fast. Within 1-2 months, our administrators made sure to also have a dedicated person looking at the AI data every day. It takes some time and work, but I think implementing an EMR system was much more difficult and took more effort than implementing the Jvion tool. It is much easier to implement this when you identify who needs to be on the team.

To learn more about the Jvion CORE^TM artificial intelligence tool and to request a demo, visit here.