U.Va. predictive monitoring systems improve patient care

Data science helps predict patients’ increase in risk of infection, medical complications

hero

Alongside sustained beside monitoring, CoMET combines medical measures of health with analytical methods of predicting possible outcomes in light of real-time data.

Courtesy AMP3D

As part of the Data Science Institute’s Research Lunch and Learn Series, Cardiovascular Medicine Prof. J. Randall Moorman gave a presentation Oct. 12 entitled “Predictive Analytics Monitoring at the Bedside,” in which he described the ways that better data science practices can make a difference in the patients’ lives.

In his talk, Moorman focused on the research of 41 doctors that have analyzed cardiovascular diseases at different stages in life and the ways in which it has led to the development and tuning of the Heart Rate Observation System and Continuous Monitoring of Event Trajectories.

HeRO returns a score that takes into account a baby’s heart rate pattern as an indicator of increased risk of infection and health challenges in the neonatal intensive care unit, or NICU. CoMET uses a similar, yet more comprehensive, approach for older patients — predictive monitoring synthesizes different measurements of patients’ well-being into a score reflective of their relative risk of complications during recovery in the surgical ICU. 

Specifically, he stressed the need for hospitals to incorporate real-time predictive analytics into the routine monitoring of patients’ conditions.

“The practice of medicine is difficult,” Moorman said. “Clinicians like myself miss things, and then patients get sick and deteriorate under our noses … The [predictive monitoring] display is intended to use all of the data available to physicians and to synthesize for physicians the risk of patients getting sick.”

Beginning in 2001, Moorman and his colleagues introduced the HeRO monitor in the NICU, after observing that the heart rate patterns of premature infants could serve as early indicators of an increased risk of sepsis, a bacterial infection that affects roughly 25 percent of NICU patients, with approximately 40 percent of those cases ultimately leading to death.

“There is a high risk for infection in the NICU,” said Brynne Sullivan, assistant professor of pediatrics in the Division of Neonatology. “The signs and symptoms of infection in premature babies overlap with the signs and symptoms of prematurity, so it can be tricky to recognize infections early. That is why predictive monitoring and having another marker, another reason to pay more attention to a baby, has potential to be really useful.”

Since it can often be difficult to discern whether or not infants are ill, Moorman and his colleagues monitored patients’ heart rate and developed algorithms that could predict the increase in risk that an infant may contract sepsis in the next day, or even in the next few hours. The subsequent analysis is displayed in the form of a bedside HeRO score so that clinicians can easily determine which babies to examine for additional biological markers of infection and then act accordingly.

After piloting the new technology at the University’s NICU, the University partnered with other hospitals throughout the country, expanding the project to nine NICUs nationwide with the help of funding from the National Institutes of Health and Medical Predictive Science Corporation. A randomized trial involving 3,000 babies took place to determine the effectiveness of the HeRO score in multiple environments, resulting in a decrease of over 20 percent in mortality rate for monitored infants. 

“We helped save lives and averted illnesses entirely,” Moorman said. “That is a great and very compelling story, one that I have heard from every NICU that uses this kind of monitoring. It is a strong motivator to take this approach using ubiquitously available data.”

According to Pediatrics Prof. Karen Fairchild, the success of monitoring in the trials reinforced the value of the HeRO score, which the NICU continues to use today as a marker for increased risk of infection and inflammation, and are looking to improve it by adding respiratory and oxygen level analysis.

For Moorman, the results led him in 2011 to seek to adapt the technology to other areas within the hospital. 

As a cardiologist, Moorman primarily cares for adults, and the success of the trials in the NICU prompted him to work on incorporating predictive bedside monitoring into places such as the adult surgical ICU. Along with his research team, Moorman began to identify various events that could increase patients’ length of stay in the hospital and further complicate surgical procedures and treatment plans — including emergency intubation, hemorrhage and sepsis. 

In 2014, Moorman brought his research in medicine and predictive monitoring to the private sector, co-founding Adult Medical Predictive Devices, Diagnostics and Displays, Inc., or AMP3D, a startup that specializes in caring for adult patients through predictive bedside monitoring. 

Developed by clinicians for clinicians, AMP3D’s platform enables patented algorithms to compete against each other to determine which one is best at accurately assessing a patient’s condition. Models incorporate data from electronic medical health records, analyzing physiological data, vital signs and lab test results.

Lawyer Kevin Passarello, the president and CEO of AMP3D, helped refine this method of predictive analytics in the hospital, now known as CoMET by working alongside Moorman and other physicians to determine which harmful events to target. Passarello said that their approach has taken eight years to compile the data, but the work is necessary for the algorithm.

“The research uses a specific, unique approach, where we go chart by chart, looking retrospectively at all of the physiological data, lab tests, vital signs and other things that lead up to a certain diagnosis,” Passarello said. 

As a result and as part of CoMET’s purpose, Passarello emphasized the need to have a platform that could predict up to six hours in advance the potential for deleterious setbacks in patients’ recovery and care, such as intubation, hemorrhage and sepsis. Frequently, if such events are identified early enough, physicians can address them proactively to reduce emergencies, mortality rates and length of stay.

“The idea is to change the paradigm for monitoring from an in the moment and reactive paradigm to an analytical and proactive one,” Passarello said. “We want to enable the physicians to see in advance that certain potentially catastrophic events might have an increased chance of occuring, and so they can treat the patient proactively and mitigate or prevent those things from happening.”

In addition, both Passarello and Moorman touched on the importance of CoMET’s accessibility and its comprehensiveness. 

According to Passarello and Moorman, the interface is user-friendly and allows clinicians to readily discern a patient’s level of risk without having to decipher or interpret multiple graphs. Alongside sustained beside monitoring, CoMET combines medical measures of health with analytical methods of predicting possible outcomes in light of real-time data.

Right now, CoMET is localized to the surgical ICU at the University. Moving forward, AMP3D aims to improve upon CoMET and its current predictive abilities in an effort to better serve adults across all clinical contexts, from the ICU to routine care to outpatient monitoring.

“We wish to address the whole hospital, at every point where clinicians might encounter patients,” Moorman said. “What we loved about the NICU was that signs babies could get sick were detected early, and the results from randomized trial that showed we saved lives. Right now, we are working towards results like that for adults.”

related stories