Predictive analytics needs a bedside, rather than scientific, manner

Early detection of a patient's risk to improve health outcomes is not a new idea.

"Meet the disease on its way to attack you," was first penned by early Roman writer Juvenal. It is a mantra so applicable to predictive analytics that expert Dr. Randall Moorman and others with whom he worked trademarked the quote in 1998.

What is new is the use of big data to accurately predict which patients are at risk for their condition to deteriorate to a subacute potentially catastrophic illness, said Moorman in the HIMSS20 Digital presentation "Who's Sick? Predictive Analytics Monitoring at the Bedside."

Patients who go to the Intensive Care Unit have longer hospital stays and a greater risk of mortality, said Moorman, who is a professor of medicine, physiology and biomedical engineering at the University of Virginia, and who is also Chief Medical Officer of advanced medical predictive devices, diagnostics and displays at the University of Virginia Health System.

For a patient requiring intubation, the risk of death increases from 10% to 50%, Moorman said. If a patient on a hospital floor requires transfer to the ICU, the risk of death goes up 40-fold.

Clinicians are challenged to detect patient deterioration based on current monitoring, which is limited, he said.

"Any improvement could have great benefits to the outcomes of our patients," Moorman said.

Moorman and others developed bedside monitoring that detects physiology going wrong that clinicians can't see on their traditional monitors. The continuous cardiorespiratory monitoring detects vital signs between nurses' visits and uses a much larger data set for an analysis of risk based on all the available data.

"We take the point of view, predictive monitoring inputs need to be complete," he said. "Use every single bit of data you can put hands on to predict illnesses."

Deep learning is not as important as big data in the early detection of illness, he said. Big data refers to massive data sets brought on by new technologies, and deep learning uses algorithms to look for complex relationships in the data.

"It's the data more so than the statistical modeling technique that is important," Moorman said.

Using the new monitor, Moorman and team looked at subacute catastrophic illnesses such as sepsis, bleeding and lung failure, leading to an ICU transfer.

In a trial, mortality was reduced by 20% and the rate of septic shock fell by half.

In studying a previous case, they found that an elderly woman who was admitted for a vascular procedure was doing well clinically, but her rising risk factors predicted by their monitor were not detected. Twelve hours later, the patient presented clinically as being short of breath. A chest X-ray showed pneumonia. She was transferred to the ICU with sepsis and entered a palliative care program the day after.

For 12 hours there was a warning, Moorman said.

The goal is to give physicians and nurses the data they need for clinical-decision support, not to give them a scientific study, Moorman said. Clinicians get a visual indicator of respiratory deterioration through the continuous cardiorespiratory monitoring.

"We should," Moorman said, "be approaching predictive analytics monitoring as bedside clinicians rather than data scientists."

Twitter: @SusanJMorse
Email the writer: susan.morse@himssmedia.com