Expert Q&A: Labs Must Take a Greater Role in Guiding Clinicians About Abnormal Data

Data is embedded within clinical laboratory operations, and lab managers must seek out the opportunity to use that data as a pathway to cooperation with physicians. That’s because clinical decision support models rely on such information to work smoothly, says Lee Fleisher, MD, emeritus professor of anesthesiology and critical care at Perelman School of Medicine and former director for the Center for Clinical Standards and Quality at the U.S. Centers for Medicare & Medicaid Services from 2020 to 2023.

In this interview with G2 Intelligence, Fleisher outlines the modern role of lab data in patient diagnosis and treatment plans. He also discusses artificial intelligence (AI) within the context of the Clinical Laboratory Improvement Amendments of 1988 (CLIA) and his thoughts about laboratory-developed test (LDT) regulation.

This interview has been lightly edited for clarity.

Q: Where do you see clinical laboratory data being most useful?

A: When added to clinical data, lab data frequently adds value. The idea of predictive analytics tools in the lab has been around for a long time. There are numerous risk indices that combine laboratory data with clinical data to predict diseases, such as heart disease, and complications in the hospital and after surgery.

Q: What does that predictive nature mean today for labs?

A: As we develop better clinical decision support, we need to understand when laboratory data is truly abnormal so that clinicians can make better decisions to take care of the patients. It’s incredibly important that laboratorians get involved, for example, in helping clinicians understand what an elevated troponin is and what comorbidities may raise troponin—and when elevated troponin may not indicate a cardiovascular acute event. The information that any algorithm provides for care needs to be interpreted within a clinical decision support context.

Q: So, in some cases, clinicians might not want to step in after an abnormal test result depending on how the lab interprets the result?

A: This approach brings up Bayes’ Theorem¹, which states that the probability that a test is a true positive or true negative is a function of where a patient starts. So, if someone exercises all the time on a treadmill, and a test showed changes in that person’s EKG but they are otherwise perfectly normal, the probability that the result is a false positive is incredibly high. The lab should be more engaged in helping clinicians to understand what the numbers mean prior to any intervention.

Q: Given the shortage in laboratory staff these days, what you’ve mentioned strikes me as a way for lab leaders to watch for bench scientists who might be adept at interpreting numbers.

A: Yes. I’m an anesthesiologist; I was the chair of the department. But not everyone wants to do everything. Many anesthesiologists just want to provide intraoperative anesthesia. However, there are some anesthesiologists who want to be more involved in quality improvement. It’s the same in laboratories, and that’s great.

Q: Are there any lessons from your career in anesthesiology that you think a lab professional would benefit from?

A: We used to talk about “RAP”: radiologists, anesthesiologists, and pathologists. We’re the hidden people, meaning patients don’t come to the hospital for us. But patients need us. To patients, that unsung group is incredibly important. The lab is integral to the patient’s care, and giving results back accurately and expeditiously is important.

Q: Let’s switch topics. We heard back in late 2022 that CLIA might eventually start treating clinical data—perhaps a genomic sequence or a digital pathology image—as part of a patient specimen. It was data derived from human health conditions, and it had implications for how CLIA looked at AI. Has this debate ever bubbled up in your circles?

A: The question is: What’s the lever to protect those specimens? I wrote a piece in June in JAMA Health Forum about my thoughts on AI, point of care, clinical decision support, and how it should be regulated.² A lot of people think that artificial intelligence and associated algorithms should be regulated in a CLIA-like setting, but not by CLIA itself.

Q: Since we’re talking about CLIA, any thoughts on the debate of LDTs regulated under CLIA versus under the Food and Drug Administration?

A: CLIA certification is meant to determine, “Is the lab performing the test correctly and giving patients back the correct results?” CLIA is not about asking, “Is the test scientifically correct in identifying the organism?” It’s my belief that whether people agree or disagree with the VALID Act or the LDT final rule, CLIA is not the answer. CLIA can’t determine that a test really detects COVID, for example. Instead, CLIA determines whether a lab is performing a COVID test correctly that the FDA or CDC approved of.

Q: A lot of what you’ve talked about centers on the need for labs to move quickly and accurately. How does that look on the patient’s end?

A: For medicine in general, this is the time that we need people’s expertise in interpreting results clearer than in the past. Patients are going into their portals and seeing test results before their clinicians see them. Not that we should prevent it—in fact, we shouldn’t—but labs should think about whether the way results are presented in the portal is the way you as a patient would want to see that information. Otherwise, patients are going to go to Google and/or ChatGPT for help, which could give them wrong answers.

References:

1. Webb, M.P.K. and D. Sidebotham. “Bayes’ formula: a powerful but counterintuitive tool for medical decision-making,” https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7808025/
2. Fleisher, Lee A. and Nicoleta J. Economou-Zavlanos. “Artificial Intelligence Can Be Regulated Using Current Patient Safety Procedures and Infrastructure in Hospitals,” https://jamanetwork.com/journals/jama-health-forum/fullarticle/2820406