When research scientists at Kaiser Permanente began a decade ago to track data on patients from their neonatal intensive care units, they had a rather simple objective: that all the children in their care would come to their annual reunions alive and healthy.
Over the course time, the research scientists expanded their efforts, monitoring a wide array of health issues — including cases of infant sepsis — and came up with a new treatment weapon: big data. After years of pulling together mounds of data, the healthcare consortium today draws information from across its data stores to accurately predict the potential for early-onset of the potentially deadly infection in newborns.
The original vision came from Dr. Gabriel Escobar, a neonatologist and research scientist at the Kaiser Permanente Northern California Division of Research (DOR). His studies found that, at one Kaiser medical facility, approximately 910 of 7,004 newborns, or 13 percent, were evaluated for early-onset sepsis, while about 700, or 11 percent, were treated empirically. But when all the blood tests came back, only 0.04 percent tested positive for early-onset sepsis.
If Escobar’s findings bore out across the country, that could mean that thousands upon thousands of newborns are treated with antibiotics for no reason.
A big part of the problem was the way early-onset sepsis was predicted. Historically, physicians had to use an algorithm from the Centers for Disease Control, said Dr. Michael Kuzniewicz, a neonatologist physician and director of the Perinatal Research Unit at DOR, who has overseen the data collection and analysis since 2010. But that algorithm didn’t take into account multiple risk factors at once, he said. Take fever, a critical risk factor for sepsis. Physicians used to follow the rule that if a mother had a temperature of more than 38 degrees Celsius, she had a fever; if less, she didn’t. “So kids would be put into one bucket or another bucket based on those very broad classifications,” Kuzniewicz said. “We know that if a mom had a fever of 37.9 versus 38.1, those kids’ risks are not very different.”
The new algorithm that Escobar and his team developed helped physicians put babies in three risk categories: treat them empirically, observe and evaluate, or just continue observation. And that provides three major benefits, according to Terhilda Garrido, vice president of health IT transformation and analytics at Kaiser. First, by predicting and diagnosing properly, patients get better care. Second, it saves Kaiser money by not over-treating patients with unnecessary antibiotics or prolonged hospital stays. And third — and perhaps most important in patients’ eyes — it helps cut down on NICU admissions, giving the mother more time with her newborn child.
Big Data, Big Lift
Kaiser was one of the first healthcare organizations to make significant investments — and headway —in the field of big data. It spent approximately $3 billion planning and implementing its electronic medical record system, which it calls KP HealthConnect. The system runs on a customized version of the Epic Care software platform from Epic Systems, one of the largest medical software developers in the U.S. Patient information is stored in Kaiser’s data centers.
The KP HealthConnect program began in 2002; by March 2010, the system was deployed to all of Kaiser’s facilities. Today, KP HealthConnect is the largest civilian electronic health record system in existence, according to Kaiser.
That’s not the only impressive fact about Kaiser’s data. Kaiser currently cares for more than 10 petabytes of data, according to Garrido, far more than the Library of Congress. And it’s growing every day, she said, as Kaiser adds new members, orders new lab tests, and includes new sources of data, such as genomic testing results. Among the information management tools Kaiser uses, Garrido said, are SAS analytics and SAP BusinessObjects business intelligence software.
The two key drivers for its investments are improving care and cutting unnecessary costs.
“Big data plays a very important part in making healthcare more affordable and driving better services and results for our patients,” said Claudio Abreu, senior vice president for regional IT operations at Kaiser. “We have a dramatic amount of data, and we’re trying to use it in the best ways possible.”
In the NICU, big data powers an online calculator that helps its physicians more accurately predict the potential for early-onset sepsis in newborns. The sepsis calculator is embedded into KP HealthConnect. For physicians, it’s relatively easy to use. The doctors simply take information from the electronic health record, such as the results of clinical examinations of the mother and baby, plug it into the device which taps Kaisers databases and analysis — and receive an instant read-out.
All in all, Kuzniewicz said it took about two years for the research scientists, data scientists and statisticians to develop the databases, complete the regressions, and finalize the algorithm. In July, they rolled out what he called the second phase — combining the maternal and neonatal risk factors in the calculator with a more specific scoring of the infants’ physical exam.
Kuzniewicz said Kaiser over the coming months would conduct formal reviews to determine the system’s impact on infant patient admission, antibiotic usage and other key issues. On average, medical professionals in that region deliver and/or treat 35,000 newborns annually.
Overall, the project is very new, but Garrido is confident that medical professionals will adopt the tool and that they are in talks to expand the project into other regions.
And even before they began tackling early-onset sepsis, Kaiser claimed the data sets they had built led to progress in several areas, including reductions in both infant stays in the NICU and blood-stream infections.
Garrido credited Kaiser’s progress in big data initiatives to coordination between a few key groups.
IT is just the start, she said — but without statisticians and data scientists to make sense of the data, they would be nowhere. “We’re sorting through what’s real and what’s hype in big data,” Garrido said. “IT is not a magician. It can enable that river of data, but you need the analysts and data scientists to make sense of it, and you need clinicians and operations leaders to ask the right questions to help identify where the opportunities are.”
Rarely are information technology tools readily embraced. And Kuzniewicz said physicians in the region didn’t accept its new tool right away. They were skeptical, he said, not only that the calculator would provide results, but because they didn’t fully understand how it was developed. That was overcome fairly easily, though, Kuzniewicz said, by walking them through the research and development of the algorithm.
Another key factor in getting buy-in was demonstrating the impact on clinical management — in other words, how the calculator could help physicians cut antibiotic use. Kuzniewicz pointed to the low number of actual sepsis cases versus those treated empirically for sepsis risk. But today, with published studies finding possible associations between early antibiotic exposure and maladies like asthma, autoimmune disease and obesity, healthcare professionals are on guard.
“While [those reports] haven’t shown to be causal yet, there is concern that overuse of antibiotics in infants can cause problems later on,” Kuzniewicz said. “Any way we can reduce that, we will — especially when we can show them that the risk for sepsis was low and the antibiotics weren’t needed.”
Image: Courtesy of iStock.