America’s intensive care units (ICUs) are supposed to showcase the lifesaving potential of modern medicine. They are equipped with the latest technology,staffed with the most skilled staff, and centered in the best hospitals. Yet statistics show that ICUs are failing miserably: One out of every five people treated in ICUs is harmed in some way (medical errors, mostly), costing the U.S. health care system an estimated $20 billion, and exacting a notable human toll.
Johns Hopkins researchers want to reverse this troubling trend. Under a $1.5 million grant from the Gordon and Betty Moore Foundation as well as a $300,000 grant from Florida’s All Children’s Hospital Foundation,
Whiting School computer scientist Suchi Saria has launched a research program aimed at making hospitals safer.
Using ICUs as her proving ground, Saria is wielding innovative machine learning tools aimed at making the health care delivery environment more “intelligent.” Her project is one of the first funded at Johns Hopkins
under a nationwide Moore Foundation initiative to improve patient safety with $500 million in grants over the next decade.
An assistant professor at the Whiting School and the Bloomberg School of Public Health, Saria imagines an ICU where computers equipped with sophisticated algorithms help doctors make more efficient and accurate medical decisions, and where unobtrusive sensors alert hospital staff to potentially harmful errors.
“We have an enormous opportunity here to make a significant difference in terms of the quality of care that patients get, and to enhance their safety while they are being cared for, especially in parts of the inpatientenvironment where fast assimilation of a large amount of information is necessary to respond in a timely manner,” she says.
The approach of Saria’s team is twofold. In the first, it will put to work the reams of data that normally are collected at patients’ bedsides—from symptoms and medications to test results and tracings from bedside sensors like the EKG—to tease out patterns that will indicate which patients need more aggressive care.
“We are developing computational models that crunch enormous amounts of data that are collected on the patient, and [putting] that to work to help physicians track changes in health status over time,” Saria explains.
“These data are being routinely collected in the Electronic Medical Record already. By developing computational approaches that can continuously scan this data, we’re providing caregivers with an extra set of ‘eyes’ that detect salient information that the caregivers can then use to helpmake their treatment decisions.”
Saria points out that there is growing evidence that predictive algorithms are useful in the early detection of which patients are at risk, as well as in reducing the time it takes to intervene in those cases. This work has its roots in research that Saria started at Stanford University. There, she and her team successfully predictedhealth risks in premature infants using the abundance of data collected in bedside monitors in neonatal intensive care units.
The other prong in Saria’s current project involves the development of noninvasive 3D sensors that will unobtrusively monitor care activities in the patient’s room to identify when an error might have occurred or when an important care activity was missed.
“These sensors will track a number of important activities, from safety steps— such as hand washing—that medical staff perform to care-related practices that help improve patient outcomes,” she says. “By giving this information to them in near real-time, quality improvement experts can institute processes for rapid cycle improvement. Doctors think this type of high-fidelity, data-driven, more timely feedback would be a huge leap forward for the field of quality improvement.”
Saria’s work is being done in collaboration with clinical researchers at Johns Hopkins Medicine’s Armstrong Institute for Patient Safety and Quality, an organization that has pioneered the field of quality improvement with tools such as the patient checklist.
“The potential of applying this novel sensing technology for quality improvement purposes is simply breathtaking,” says Peter Pronovost, director of the Armstrong Institute and Johns Hopkins Medicine’s vice president for patient safety and quality.
“By integrating sensor data with existing information from the electronic medical record and patient monitors and analyzing these data, this research can show new approaches for creating a learning health care system.”