UPMC cuts hospital readmission rates with ML algorithm
Model identifies patients at highest risk for re-hospitalization within seven and 30 days of discharge.
The University of Pittsburgh Medical Center’s clinical analytics team has leveraged machine learning to develop an algorithm that rates hospital patients for their risks of being readmitted.
Specifically, the ML algorithm identifies patients at highest risk of re-hospitalization within seven and 30 days of discharge. To date, re-hospitalizations have been reduced by about 50 percent at UPMC Presbyterian hospital.
“Right now, the main area of focus is on seven days,” says Oscar Marroquin, MD, chief clinical analytics officer, UPMC Health Services Division. “The models to predict seven and 30 days are almost identical. The only thing that changes is the statistical weights of each one of the co-variants, which are a little bit different. They are statistically evident but not clinically important.”
Electronic health record data, gleaned from UPMC’s Cerner (inpatient), Epic (ambulatory) and other EHR systems, is analyzed and fed to a dashboard integrated into the clinical workflow that breaks down the findings, helping clinicians to visualize the insights and make it actionable.
The dashboard, which tracks currently hospitalized patients so that UPMC can identify those at high risk of re-hospitalization, allows clinicians to look at the entire patient population by hospital, unit, or condition. According to Marroquin, the dashboard's geo-coding plots where patients live so that the health system can arrange the proper post-hospitalization follow up, as well as track which other hospitals are sending patients to UPMC's facilities.
“We can use our data to figure out which actions are associated with better outcomes,” adds Marroquin, who notes that the ML model was piloted last year at UPMC Presbyterian hospital in one of its cardiology units.
The results of the model, based on an analysis of 1 million discharges, are now being used by about 17 of 40 hospitals at UPMC—the nation’s largest non-profit academic health system—to better manage care for those patients deemed at risk by providing interventions that prevent hospital readmissions.
“Our charge is to use the data in our EHRs, analyze it and derive insights,” notes Marroquin. “Every large healthcare system claims to have a lot of data and analytical capabilities. 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 better deliver care for our patients right now.”
Also See: Decision support tool in EHR helps detect acute kidney injury
Marroquin, who is a practicing cardiologist, leads UPMC’s 15-member clinical analytics team, which applies different analytical techniques to large data sets to develop predictive and prescriptive models of healthcare delivery that result in improved health outcomes.
“We’re less concerned with the use of a specific analytic technique—we use different AI-based techniques to try and figure out the best care using patient data,” adds Marroquin. “We take a pragmatic approach and figure out how to get clinicians answers, regardless of whether we use traditional statistical techniques, tree based models or neural networks.”
According to Marroquin, UPMC is reaping the benefits of a five-year, $100 million investment the health system made starting in late 2012 in the area of enterprise analytics with the creation of a best-in-class data warehouse, bringing together clinical, financial, administrative, genomic and other information that was previously difficult to integrate and analyze.
“What differentiates us is that we are an integrated delivery system,” says Ed McCallister, UPMC’s chief information officer. “There’s an arm of UPMC—the insurance services division and health plan—that have been very deep in data warehousing and analytics.”
Currently, UPMC manages 18 petabytes of online data storage, which comes from sources of information across the 40-hospital enterprise, UPMC Health Plan and outside entities, including labs and pharmacies, according to Chris Carmody, senior vice president of the Information Services Division.
“The genius of Oscar (Marroquin) is taking very practical approaches to the data,” says Carmody. “He’s not creating one algorithm to solve one problem. He’s creating a platform to solve many problems. The insights that he and his team are garnering are simply awesome.”
“My clinical analytics team are the data consumption folks,” concludes Marroquin. “Before our team came about in 2015, it was very difficult to answer clinical questions about patient care longitudinally.”
According to McCallister, UPMC has just started to scratch the surface when it comes tapping the value of clinical analytics. “We now have a holistic view because we have the holistic dataset,” he adds.
Specifically, the ML algorithm identifies patients at highest risk of re-hospitalization within seven and 30 days of discharge. To date, re-hospitalizations have been reduced by about 50 percent at UPMC Presbyterian hospital.
“Right now, the main area of focus is on seven days,” says Oscar Marroquin, MD, chief clinical analytics officer, UPMC Health Services Division. “The models to predict seven and 30 days are almost identical. The only thing that changes is the statistical weights of each one of the co-variants, which are a little bit different. They are statistically evident but not clinically important.”
Electronic health record data, gleaned from UPMC’s Cerner (inpatient), Epic (ambulatory) and other EHR systems, is analyzed and fed to a dashboard integrated into the clinical workflow that breaks down the findings, helping clinicians to visualize the insights and make it actionable.
The dashboard, which tracks currently hospitalized patients so that UPMC can identify those at high risk of re-hospitalization, allows clinicians to look at the entire patient population by hospital, unit, or condition. According to Marroquin, the dashboard's geo-coding plots where patients live so that the health system can arrange the proper post-hospitalization follow up, as well as track which other hospitals are sending patients to UPMC's facilities.
“We can use our data to figure out which actions are associated with better outcomes,” adds Marroquin, who notes that the ML model was piloted last year at UPMC Presbyterian hospital in one of its cardiology units.The results of the model, based on an analysis of 1 million discharges, are now being used by about 17 of 40 hospitals at UPMC—the nation’s largest non-profit academic health system—to better manage care for those patients deemed at risk by providing interventions that prevent hospital readmissions.
“Our charge is to use the data in our EHRs, analyze it and derive insights,” notes Marroquin. “Every large healthcare system claims to have a lot of data and analytical capabilities. 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 better deliver care for our patients right now.”
Also See: Decision support tool in EHR helps detect acute kidney injury
Marroquin, who is a practicing cardiologist, leads UPMC’s 15-member clinical analytics team, which applies different analytical techniques to large data sets to develop predictive and prescriptive models of healthcare delivery that result in improved health outcomes.
“We’re less concerned with the use of a specific analytic technique—we use different AI-based techniques to try and figure out the best care using patient data,” adds Marroquin. “We take a pragmatic approach and figure out how to get clinicians answers, regardless of whether we use traditional statistical techniques, tree based models or neural networks.”
According to Marroquin, UPMC is reaping the benefits of a five-year, $100 million investment the health system made starting in late 2012 in the area of enterprise analytics with the creation of a best-in-class data warehouse, bringing together clinical, financial, administrative, genomic and other information that was previously difficult to integrate and analyze.
“What differentiates us is that we are an integrated delivery system,” says Ed McCallister, UPMC’s chief information officer. “There’s an arm of UPMC—the insurance services division and health plan—that have been very deep in data warehousing and analytics.”
Currently, UPMC manages 18 petabytes of online data storage, which comes from sources of information across the 40-hospital enterprise, UPMC Health Plan and outside entities, including labs and pharmacies, according to Chris Carmody, senior vice president of the Information Services Division.
“The genius of Oscar (Marroquin) is taking very practical approaches to the data,” says Carmody. “He’s not creating one algorithm to solve one problem. He’s creating a platform to solve many problems. The insights that he and his team are garnering are simply awesome.”
“My clinical analytics team are the data consumption folks,” concludes Marroquin. “Before our team came about in 2015, it was very difficult to answer clinical questions about patient care longitudinally.”
According to McCallister, UPMC has just started to scratch the surface when it comes tapping the value of clinical analytics. “We now have a holistic view because we have the holistic dataset,” he adds.
More for you
Loading data for hdm_tax_topic #better-outcomes...