UPMC’s Health Services Division is harnessing machine learning for better day-to-day outcomes.

It’s not unusual for a leading-edge health care research organization like UPMC to invest in high-powered computers, a world-class data center, and a small army of statisticians and data scientists to study patient outcomes. What differentiates UPMC’s approach to patient analytics is its use of data — not just for study, but for action that improves day-to-day patient outcomes, and not off in the future, but right now.

One recent example is an analytics platform and a machine learning algorithm developed by UPMC’s Clinical Analytics Department that gave doctors the information they needed to identify patients who were at the highest risk of being rehospitalized within seven days of discharge. This enabled them to reduce rehospitalizations by about 50 percent in the unit where the algorithm was piloted at UPMC Presbyterian.

A million-case database

The project used health record data pulled from 1 million hospital discharges to design a proprietary algorithm that helps doctors identify patients at highest risk of rehospitalization within seven and 30 days of discharge. Built with advanced machine learning technology, the algorithm correlates patient clinical data with home location, social determinants of health, referring hospital, and other information. Armed with this tool, UPMC doctors have a better chance of spotting high-risk patients and arranging appropriate post-hospitalization follow-up care.
After development and testing, the algorithm was approved for use within the UPMC system, beginning with a pilot program conducted last year in a UPMC Presbyterian cardiology unit. It features a computer “dashboard” that tracks current hospitalized patients, providing doctors a view of the entire patient population by unit, condition, and risk level. The risk level also is included in individual electronic medical records so doctors are aware of their patient’s risk. “We can use our data to figure out which actions are associated with better outcomes,” says project lead Oscar Marroquin, MD, chief clinical analytics officer, UPMC Health Services Division.

Delivering better care … right now.

Following the successful pilot, the program was expanded and is now being used by nearly half the hospitals in the UPMC system — a number that continues to grow as more doctors use the system to drive better outcomes every day. “Every large healthcare system claims to have a lot of data and analytical capabilities,” says Dr. Marroquin. What’s different about us — and perhaps a few other systems — is that unlike the majority who have built infrastructure mostly for research purposes, we have made the investment to have a team dedicated to delivering better care for our patients right now. Our goal is to be a learning health system, where we develop, train, test, and validate our models in the real world.”

Only the beginning

Chris Carmody, senior vice president of UPMC Information Services, says UPMC is just beginning to tap the value of clinical analytics. “The genius of Dr. Marroquin is that his team takes very practical approaches to the data,” he says. “They’re not creating just one algorithm to solve one problem; they’re creating a platform to solve many problems. The insights that he and his team are garnering are simply awesome.”