Despite the widespread adoption of computerized provider order entry systems, medication management remains a huge patient safety challenge. Could looking at data from multiple sources across the hospital help identify previously unknown safety risks? According to Jonathan Bickel, M.D., M.S., senior director of business intelligence and clinical research informatics at Boston Children’s Hospital, the answer is yes.
Bickel presented at HIMSS15 in Chicago last week on an effort that partnered Boston Children’s with a data science team from Mitre Corp. to pull together diverse data from medication orders, alerts raised by those orders, and reported safety events to help identify precursors to medication safety events.
“Our CEO Sandra Fenwick challenged us to take patient safety to the next level,” Bickel said. “We needed data scientists to help us understand patterns in our own data.”
As Mitre Corp. executive Eric Hughes, Ph.D., said, the effort brought some of the techniques learned from aviation safety to the healthcare setting. Mitre, worked on a project called the Aviation Safety Information Analysis and Sharing with the Federal Aviation Administration and airlines to bring big data to air traffic safety. “Using many sources of data about flights, we went from studying what went wrong to what almost went wrong in near misses to what could go wrong,” he said.
Bickel said bringing the same approach to medication management would allow research teams to focus on important problems they know exist but are hard to solve. They used pattern recognition on detailed log data from the CPOE system, CDS, and medication administration to gain new insights in medication management and then worked on types of interventions to consider, such as changes to clinical decision support alerts. The work of figuring out which re-orders and changes represented real patient safety issues required lots of study, review and validation.
Over a three-year period, they looked at 13 million medication order actions. Than meant studying 5.5 million orders or more than 5,000 per day. “We focused on pattern mining to look not just above the water, but at the iceberg below,” Bickel said. They studied cancelled orders and reorders as precursors to safety events. “Can we look at it way upstream as a precursor and alert then?” he asked. “We wanted to focus on the upstream, right when you order the medication.”
Their research found that in cases in which an order was cancelled and then reordered, the initial dose got dispensed 10 percent of the time, although less than 1 percent were actually administered to the patient. “We want to get that number below what we are seeing, to 0001 percent,” Bickel said.
He also noted that most of these instances are not showing up in the patient safety reporting system. For dosing changes, there are about 100 times more cases than reported safety events. They were able to match about 50 percent of reported patient safety events to these type of cases or orders.
In his summary, Bickel said they found that patient identification issues are prevalent in high-turnover units, and dosing issues depend on medication class. The goal is to move beyond chart review and safety reports to get at early indicators of risk, Bickel said. Insights can be used to address safer dosing for medications with complex dosing rules and better patient ID in high-turnover areas. He added that Boston Children’s is considering using patient photos to accompany the ordering system to ensure the order is for the right patient.
Boston Children’s is working with the National Patient Safety Partnership, which includes Cincinnati Children’s Hospital, Mitre and the Children’s National Medical Center. “A broader partnership is going to help solve this,” he said, including EHR vendors.
In response to an audience question, Bickel said clinical decision support alert fatigue is a huge issue. Many institutions turn down or off some alerts. “The story here is about refining the alerts that do come up to make them more meaningful on what and why they are alerting.”