Toronto hospital detects infection with analytics

FRAMINGHAM, Mass. — A Toronto hospital is implementing a clinical experiment using big data and real-time analytics to help save the lives of its tiniest patients.

When you watch a premature baby lying still in a clear plastic cradle in a Neonatal Intensive Care Unit, it is hard to imagine that something so small would have a real shot at survival. The obstacles these pre-term infants face are huge: immature lung development, undeveloped immune systems, and a very real change for disabilities — that’s just a small sample of the challenges infants born before 37 weeks of gestation face.

Advances in medical science have made these challenges much more beatable, and many pre-term infants have managed to beat the odds and live successful and happy lives.

Unfortunately, the very medical care that these tiny patients receive can be a vector for infections that, with their compromised immune systems, can be very life-threatening. This class of infections (which can be viral, bacterial, or even fungal) is referred to as nosocomial infections, which is basically any infection a patient picks up in a hospital environment.

Nosocomial infections can be nasty for full-grown adults, and unless doctors and nurses act fast, they can be deadly for premature infants. One of the worst of the bunch is an infection known as late onset neonatal sepsis (LONS). Complicating this is the sad fact that by the time a premature infant starts exhibiting symptoms of LONS infection, things have already gotten pretty bad for the infant. Blood tests aren’t even that conclusive, since false-negative results are a real problem when you can’t extract that much blood to test.

Here, then, comes the miracle. A new application of monitoring technology and real-time analysis is being implemented by The Hospital for Sick Children at the University of Toronto that stands a real shot at alerting health-care providers that a child is being infected — before acute symptoms of infections like LONS show up.

The “before” may sound like some sort of huckster psychic’s trick, but there’s some real science going on. According to a 2007 medical study by Drs. M. Pamela Griffin, et. al., up to a full day before LONS-infected patients start showing signs of trouble, their heart rates show very subtle yet fairly consistent heart rate changes: oddly, their heart beat becomes abnormally steady for a while, which rarely happens to humans even at rest.

These heart rate changes are impossible to detect by a human being watching or listening to a heart rate monitor. But for Dr. Carolyn McGregor of the University of Ontario Institute of Technology (UOIT), with the right tools, this is exactly the kind of task neonatal monitors and real-time analytic software can handle.

Catching LONS is a tricky business in a human-only clinician environment, McGregor explained. Typically, she said, it’s often up to the experienced nurses who will use their knowledge and instincts to see a range of subtle clinical signs to determine a baby is “not quite right.”

Treatment for LONS is a full course of antibiotics, which isn’t something you want to mess around with with infants this age, and could (if overused) lead to antibiotic-resistant forms of LONS, which clearly would be devastating.

McGregor is the lead researcher on Project Artemis, a real-time data gathering and analysis framework that takes signals from the babies’ heart monitors at The Hospital for Sick Children and the Women and Infant’s Hospital in Providence, RI, and processes them to seek out signs of imminent LONS infection and then alert health care workers in each facility of the problem as soon as possible.

McGregor, who is the Canada Research Chair in Health Informatics at UOIT, explained that the Artemis project is being conducted as a blind study, with Artemis-monitored patients being diagnosed alongside NICU patients in each hospital, in order to determine if there’s a real improvement in LONS diagnosis using the traditional clinician methods versus the Artemis monitors.

The Artemis system collects the massive amount of data that’s pulled in from all of the existing monitoring systems that a premature infant is usually connected to anyway. For McGregor, it was the presence of the monitoring systems that helped spark the idea for Artemis: a lot of data was being collected, and nothing was being done with it.

“Massive” and “a lot” deserve some qualifiers, so try this number on for size. McGregor said that currently the Artemis system pulls in 1,256 data points per second per patient. This is what’s known in data circles as a fire hose, and this is truly a powerful fire hose.

Yet, for all the information coming through the network, McGregor emphasized, there’s not a big footprint here that will bog a hospital’s network down.

“These are thin footprints, and are less than one percent of network traffic in a hospital,” McGregor said.

The project’s use of IBM’s InfoSphere streaming technology, has also smoothed out a lot of the real-time analysis issues, since this is right up InfoSphere’s alley. Interestingly, this high-level processing capability allows the entire system to be processed not on big iron, but rather three laptops: one for data acquisition, another for online analysis, and the final machine for stream persistence. All of the data from the Toronto hospital is mirrored at UOIT.

The addition of the Rhode Island location for Artemis added some additional challenges, but was done not only to deliver more data for the project’s algorithms to analyze, but also to demonstrate the proof of concept of the Artemis system as a cloud-based service. While the actual data is gathered in the U.S., it’s processed in Canada before being sent back in a more visual form back in Rhode Island.

Since Artemis is still proceeding as a scientific study, it’s too early to report on the success rate of the system. But McGregor was pleased to see one thing come from the implementation of the study: it seems to dispel the notion that clinicians aren’t interested in such systems.

“They recognize that this is a powerful clinical decision support tool,” McGregor said.

McGregor is also excited about what projects like Artemis, both as monitoring and cloud-based tools, might mean for medical diagnostics. Late-stage, at-risk pregnancies could be monitored remotely (albeit as a slower data rate), thus alleviating the need (and cost) of in-hospital monitoring, she outlined.

Leukemia patients, who also are burdened with compromised immune systems, could also be remotely watched. “We could be looking for infections and catching them a lot sooner,” McGregor said.

The potential benefits are high, and McGregor and her team are working hard to work out any technological, procedural, and legal kinks in the Artemis system to help maximize those benefits.

Big data, it seems, may be helping to save a lot of little lives.

At a glance: Project Artemis