Say you bang your toe badly enough that you need to see a doctor.
Before anyone looks at the bruise, you are asked to answer a bunch of questions about how you feel in general, your living arrangements and other aspects of your life.
“A lot of patients will question, ‘Why am I getting asked about my mood when I’m here for a toenail injury?’” said Dr. Michael Hasselberg, Chief Digital Health Officer at the University of Rochester Medical Center.
A week, a month, a year from now, you may have a more serious injury or illness. Information that once seemed irrelevant suddenly will matter.
“The reason we’re asking you questions … that don’t always seem completely pertinent to why you’re seeing your doctor today is because it will help us help your doctors keep you healthier and could even save your life,” said Dr. Gregg Nicandri, Chief Medical Information Officer and professor of Orthopedic Surgery at URMC.
Using information collected from individuals and populations in order to gauge when a patient may be at risk for an acute illness or worsening of a chronic condition is not a new idea.
Predictive analytics may date back to when Hippocrates had his fourth patient and compared the ailments in that person to those he had seen in his previous three; but tracking trends just by taking notes is cumbersome and inefficient.
Modern predictive analytics uses machine learning – artificial intelligence – to synthesize massive amounts of data into information that guides health care professionals in diagnosing, treating and managing illness.
Medical personnel use predictive analytics to support clinical decision-making by getting “the right information to the right provider in the right context at the right time, in the right format,” said Dr. Balazs Zsenits, Chief Medical Information Officer at Rochester Regional Health.
The algorithms produce results that are akin to “tapping the clinician on the shoulder and saying, “Hey, there’s a better way to look at this,’” said Dr. David Krusch, vice president, IT Applications at RRH.
In an outpatient setting, predictive analytics make sure patients are up-to-date on screenings and vaccinations based on age and risk, and to guide counseling on disease management.
Predictive analytics also help specializations such as radiology, dermatology and ophthalmology, all of which use images in the clinical setting. Predictive analytics among inpatients can help the care team assess an individual’s risks during and after procedures.
URMC built a machine-learning model based on the global characteristics of people who were likely to go to a nursing home for rehabilitation.
“What we found was the biggest kind of predictor of patients going into a nursing home after surgery was their level of social support,” Hasselberg said. “Did they have family connected to them; did they have visiting nurse services set up?”
How a patient answered questions about social, emotional and physical health at previous appointments help doctors assess your risk for a poor post-surgical outcome.
Hasselberg said the information helps surgeons make better decisions about who is ready for surgery and who needs to be delayed so social supports can be put in place.
Predictive models can also synthesize data from monitors to flag patients at risk for sepsis eight hours before clinicians can spot potential problems. “That actually converts into lives saved,” Nicandri said.
Hospitals face penalties if the same patient has to be admitted again in that time for any reason, so they use algorithms to predict who is at risk for hospital readmission within 30 days of discharge.
To show how the model can be proactive, RRH’s Krusch gave an example of a person who has congestive heart failure, a cause of frequent hospitalization.
“That’s because their condition hasn’t been optimized,” he said. “If we can say, ‘You have an issue with fluid retention, we’re going to start your care management on the day of admission and we’re going to put into place a plan where maybe we can send someone out to your home and give you a diuretic to help reduce fluid retention,’ that patient is not only not going to be readmitted, but that patient is going to have a better outcome.”
The model has helped RRH identify patients most at risk and allowed care teams to focus resources on those individuals.
“These allow us to be preventive rather than fixative,” he said.
But algorithms go only so far. If the computer becomes the one that dictates care, doctors are just robots.
“I’ve always said, ‘I wonder when the day is when people pull up to a drive-thru and a computer,’ and you enter your name and … you just select what the recommendations are,” Dr. Jason Feinberg said. “You’re due for a colonoscopy: ‘accept’ or ‘decline.’Ind You press the button and they schedule you or don’t schedule you. But everything really needs to be synthesized and discussed.”
Feinberg said medicine is still an art, and the data can drive “artful discussions.” He says, “The data that gets run through formulas that are predictive are only as good as the discussion at the bedside or in the doctor’s office.”
He gives an analogy of analytics as red, yellow or green lights to counsel patients in making lifestyle changes to manage illness.
“You don’t run through red lights on (Routes) 5 and 20,” he said. “You probably get through some of them, but it’s a high risk.” Using the analytics helps doctors and patients turn red lights to green, “to prove those predictive models wrong. That’s where the art is,” he said.
The personalized approach also is used by insurers to understand the needs of subscribers and their likelihoods of developing certain conditions.
At Excellus BlueCross BlueShield, the insurer analyzes claims data so it can send healthcare information to a subscriber who is newly diagnosed; have a clinical professional contact a subscriber at risk for a chronic condition; support subscribers who have severe illnesses and promote wellness through reminders about preventive care, according to Joy Auch, director of communications.
URMC’s Hasselberg and Nicandri compare how the health data sets create profiles of patients to the way Amazon uses its own algorithms to predict what you want to buy before you know you want it. The model is based on people who share some of your characteristics, but it also relies on your own traits.
But not everyone likes seeing ads pop up for products they have never bought. At the same time, many people willingly allow their cell phones to track their every move. Consumers may be willing to trade privacy for convenience when it comes to calling an Uber, but will they feel the same with their health information?
“Obviously, with medicine, you have to be very careful,” Nicandri said. “Patient privacy and trust have to be absolutely paramount in terms of protecting the data and protecting and establishing that trusted relationship.”
“It will be interesting over the next decade (to see) how this evolves and what society will tolerate,” particularly as big data companies get more into health care, Hasselberg said. He said medical systems are more likely to lobby the federal government to protect data and ensure it is used to further science.
“Technology is opening up amazing opportunities for improving healthcare. When you think about big data and predictive analytics and machine learning, there are like diseases like Alzheimer’s disease, and we actually may actually have the right data to have a cure for Alzheimer’s disease. We just haven’t had the computing power, the data in one place to find that cure. It may be right in front of our faces, and the computer can help us do that.”
Patti Singer is a freelance writer in Rochester. Contact her at [email protected]l