Artificial intelligence is getting better at making predictions that can help keep patients out of the hospital. The results are healthier patients and significant savings.
The transition from hospital to home can be a delicate time for many patients. If something prevents them from following their discharge plans, they could end up back in the hospital. The same is true for patients with serious or chronic illnesses. If their conditions aren’t managed, they can get worse and lead to hospital admissions, too.
That’s why case management programs designed to avoid these kinds of outcomes, which often pair patients with dedicated nurses, have become standard offerings for many insurers and hospitals. But Highmark Blue Cross Blue Shield (Highmark) has taken the practice to a new level. The results are fewer hospital admissions or readmissions and hundreds of dollars in savings.
Recognizing patterns and making predictions
Highmark is using machine learning and predictive analytics to try to identify patients who may benefit from help after they leave the hospital or who have serious or complex illnesses. The process starts with millions of data points, including feeds from health care providers as well as member medical claims and demographic data. Then computer systems learn to spot patterns, and from there, make predictions about what might happen compared to patients with similar situations.
Supporting people leaving the hospital
For hospital transitions, the key is using real-time data, such an “Admit, Discharge, Transfer” notification, rather than relying on medical claims, which can take time to process. This way, a nurse can be ready to reach out as soon as a patient leaves the hospital.
“For transitions of care, days or hours matter,” says Dr. Benjamin Edelshain, Highmark’s vice president of clinical engagement and digital innovation. Edelshain says that if a patient is interested in the program, a Highmark Transitions of Care team member can help ensure medications are filled and delivered and follow up doctor visits booked. They can arrange for meal delivery or help address financial concerns. “Hospital nurses are wonderful—and busy. They don’t always have time to cover everything during discharge planning. Our nurses are available to spend a couple of hours on the phone with a patient, and continue following up as long as a patient needs them.”
Learning to spot the signs that someone needs support to stay out of the hospital
When it comes to preventing complications with chronic or serious illnesses, the machine models are usually looking for particular diagnoses or a number of diagnoses. Edelshain says the model also takes into account claims history to get a baseline. “You’re starting to look for patterns in outpatient care that might be indicative of someone going downhill,” he says. “Maybe someone stops refilling prescriptions for maintenance medications, or maybe they’re seeing the doctor more often.” Or perhaps, says Edelshain, a patient may be facing a high-risk pregnancy or complex psychiatric condition. Those factors could prompt a case manager to reach out and offer help.
Rooting out bias
Reaching out for the right reasons is critical. Algorithms, or the instructions computers use to complete tasks, have been known to be biased. “We test the categories that are prone to bias,” says Blunt. “We ask why has the algorithm selected this variable.” The goal is to make predictions about the outcome of a patient’s trajectory with as much accuracy as possible. Highmark can train its models using the experiences of more than seven million members.
The team is constantly reviewing its computer models for accuracy, as well as for bias. For example, the system may learn that a certain pattern of using more medical services, combined with living in an underserved ZIP code, could mean a patient will struggle to get the care they need and end up in the hospital. But, says Ian Blunt, Highmark’s vice president of advanced analytics, maybe that isn’t always the case for patients who live in that particular ZIP code. The algorithms are then refined to help target extra support to patients who need it the most.
Encouraging patients to enroll
It’s always a member’s choice whether to enroll in a case management program. Some may not want to; some may be difficult to reach. Highmark is also using Google’s artificial intelligence tools to develop strategies for reaching out to people at the most convenient times, in the most convenient ways.
Blunt knows a member might question why their insurer is reaching out to them. “We need to work harder to earn the trust of our members,” he says. “Our case management programs, if they keep a member healthy and out of the hospital, could build that trust,” says Blunt, “and encourage a member to reach out to us for help again if they need it.”
Highmark Blue Cross Blue Shield is an independent licensee of the Blue Cross Blue Shield Association, an association of independent, locally operated Blue Cross and Blue Shield companies.