Keeping Watch
In most aspects of medical care, there is no match for the human touch. But in one crucial area - the battle to protect against medical errors - technology just might trump human frailty.
That is what researchers from our Division of Biomedical Informatics are exploring with doctors and other clinical staff from our Neonatal Intensive Care Unit (NICU), where the stakes of medical errors are about as high as they come.
“We already have systems in place to prevent errors,” says Kristin Melton, MD, a neonatologist in the NICU. “Our error rates are not high, but you are relying on a human system, so errors do occur. We are trying to recognize those times when they do occur and prevent them.”
So Melton and Imre Solti, MD, PhD, a researcher in our Division of Biomedical Informatics, are leading a project funded by a two-year, $500,000 grant from the National Institute of Child Health and Human Development. They are working with a team of experts from Biomedical Informatics and the NICU to develop a sort of digital guidebook for the wary - algorithms that will allow computers to continuously scan patient records for potential problems.
“Our goal is to build a database of adverse events and errors that the computer can search for and identify before they occur,” says Solti, who is principal investigator on the study.
Errors with greatest potential for harm
The researchers are focusing on two types of errors with the most serious consequences for babies in the NICU – the ordering and delivery of medications, and “unplanned extubations,” incidents when a baby’s breathing tube is accidentally dislodged.
Research team members spent the past year reviewing two types of data entered into the electronic health record. For medication errors, they examined “structured data” such as vital signs and other information that is entered routinely into the charting system.
For extubation errors, the annotators pored over clinical notes, the more subjective narrative provided by caregivers about the care and status of each patient. Nearly 750 infants were cared for in our NICU this past year, which generated more than 30,000 “patient days” of notes. The annotators’ job was to identify all the ways in which caregivers describe the errors so that the terms can be included in the algorithms.
Doing no harm
While no error is acceptable in patient care, some have more serious consequences than others. In the coming year of the study, Solti says, the researchers will focus on the errors that resulted in harm to patients.
“As we detect errors in the electronic health records, we will also look at whether these errors contributed to harm to the patient,” Solti says. He adds that creating “categories of harm” will help prioritize which mistakes carry the most serious consequences. The researchers will incorporate this information into their algorithms.
“We don’t ever want errors to happen, but we do want to prioritize which errors should get most of our attention, “ adds Melton.“Harm categories help us prioritize which are the most critical problems to focus on.”
On-the-spot warnings
Once the algorithms are developed, the research team hopes the next step will be putting in place real-time identification of high-impact errors that will send immediate alerts to caregivers. So if a nurse begins to give the wrong dose of medication, an alert would warn that it was not the same as the ordered dose.
The key to making such a system work well is cutting through the clutter of the noise in busy hospital units.
“We want to make sure the alerts are for something significant, and that they fire only if there is a reason,” says Solti. “We don’t want them to be another alert that no one pays attention to.”
Automating the search for potential errors in the NICU should be a significant step forward in improving safety and removing one of the many tasks caregivers must perform. But Solti emphasizes that no automated system will ever replace the need for human judgment.
“The capacity to continuously and rapidly scan the entire record is certainly an advantage,” he says. “But you will always need someone to look and see if something is truly an error or adverse event.”
Solti and his team believe the project can become a model for other intensive care settings.
“If it works for the NICU, which is one of the most complex environments, it should work for other areas as well, adjusted and customized for that particular environment,” he says. “We hope to test it in another institution and see how well our algorithms work there, then roll out to other NICUs on a larger scale.”