Finding Needles in a Haystack: How One Lab Identified Random Errors in a Large Dataset

“As much as we like to believe that errors should not occur in the clinical laboratory, they can and they do,” said Clarence Chan, MD, PhD, a clinical chemistry fellow in the department of pathology at the University of Chicago. “It’s important that we have a rational and objective approach to dealing with these kinds of situations.”

During a roundtable session at the upcoming 2023 AACC Annual Scientific Meeting & Clinical Lab Expo, Chan will present a case study about how his team discovered and investigated random errors that were reoccuring over time. He described the errors as hard to catch even though the lab had quality-control measures in place.

“That’s the whole point of quality control, but sometimes the errors are so sporadic that you don’t always see them,” Chan said. “By coincidence, we had a few primary care doctors asking us about [a couple of results].” That kicked off an investigation into what could be causing such odd and seemingly random recurring errors.

Chan said this case is unique because of its scope–the team reviewed more than 11,000 results over a fairly long period of time–and because it required them to identify a very small fraction of spurious results within that large group. While the details of his experience won’t universally apply to other labs, he hopes participants can learn from how they solved the mystery. This process included brainstorming possible issues and systemically investigating each one until they found the root cause.

Another way labs can benefit from Chan’s insights is hearing how the team applied tools like data analysis and data informatics to their investigation.

“The goal is not to give someone an exact prescription or algorithm of how you deal with these scenarios,” he said. “Every situation will be different.” He added that his team didn’t need high-tech tools to unlock their error-causing mystery. In fact, one of the most important pieces of software they used was Microsoft Excel. “We didn’t have to do any hardcore programming, even though we handle large volumes of data. Understanding how to effectively use Excel, while also recognizing its limits and pitfalls, can also help develop an approach for using more conventional programming software such as R and Python” he said.

The roundtable will empower laboratorians with little to no prior experience in data analytics to gain confidence using large datasets in today’s increasingly digitized healthcare system. This skill is becoming even more critical, given trends towards collecting and analyzing more data across healthcare settings, including clinical laboratories. Globally, “there’s been this push for how do we get that information more efficiently and make more out of it. We’re turning out patient results all the time,” he said. “Not only are you ensuring they're accurate and precise from a technical standpoint, but when there are questions about when things go wrong, how do we get the relevant information?”