Combining education and machine learning to detect contaminated CBC in IV fluids by Carly Maucione, MD

Improved detection of fluid contamination in complete blood counts (CBCs) could prevent some unnecessary downstream transfusions through clinician education and machine learning, according to a new study.¹

Results from a multicenter machine learning study identified the alarming frequency of contaminated blood specimens and highlighted the clinical impact of acting on erroneous test results.¹

“I wish more people would read about this issue and better understand what IV fluid contamination looks like in patient samples,” study researcher Carly Maucione, MD, a resident at Washington University in St. Louis, said in an interview. HCP Live. “When they’re working on the floor and see something suspicious, they might say, ‘I remember hearing about this. I think this might be a contaminated specimen.'” Then they double-check before ordering blood products or take the time to make sure they’re not acting on erroneous results from a basal metabolic panel. ”

Particularly in hospitalized patients with multiple lines and continuous IV fluids, IV fluid contamination can be difficult to distinguish from true clinical changes, increasing the risk of downstream clinical intervention. Blood transfusions are usually initiated in response to abnormal laboratory values, so even small amounts of undetected contamination can have clinical significance.²

Maucione et al. identified machine learning as a potential method to standardize CBC assessment and assist pathologists more accurately. In laboratory practice, techniques for detecting contamination of intravenous fluids are limited, and commonly used approaches such as delta checks are not designed to be sensitive to this problem.^{1, 2}

“At least in the laboratory, there is a lot of potential for machine learning in areas such as flow cytometry,” Maucione said. “A lot of the preprocessing and gating is actually done manually. That’s pattern recognition, and that’s something that machine learning can do very well. I think that’s a bottleneck in our workflow, and anything that can do it faster is where we should focus our efforts.”

In this retrospective, multicenter machine learning diagnostic validation study, we used real-world inpatient data from two institutions to develop and test two machine learning models with area under the receiver operating characteristic curve of 0.972 and 0.957 and area under the precision-recall curve of 0.723 and 0.619 to identify IV fluid contamination in CBC results.¹

Model results revealed that approximately 2% of CBCs resulted in contaminated IV fluids, and the researchers estimated that 6% to 9% of subsequent blood transfusions may have been unnecessary.¹

After identifying an unexpectedly high frequency of contaminated specimens and avoidable blood transfusions, Maucione and colleagues aimed to address that gap. However, she acknowledged that there is no gold standard for confirming IV fluid contamination and further validation is needed to ensure the findings are not due to other confounding factors. Improving the model’s positive predictive value is necessary before it can be used to delay the release of blood products in cases of suspected contamination, but this remains a long-term goal.¹

While real-time implementation remains a long-term goal, education is a more immediate and achievable intervention, Maucione emphasized. Increasing the awareness of pathologists, laboratory technicians, phlebotomists, nurses, and other clinicians who draw blood may help prevent misinterpretation of contaminated specimens and reduce unnecessary blood transfusions.¹

Ultimately, this study highlights both the potential and limitations of machine learning in laboratory medicine. While artificial intelligence may help fill gaps in IV fluid contamination detection, Maucione emphasized that thoughtful integration, rigorous validation, and clinician education remain essential to ensuring patient safety and improving care.¹

Editor’s note: Maucione has not reported any relevant disclosures.

References

1. Morcione C, McCrum N, Zeidman MA, Pearson LN, Metcalfe RA, Spies NC. Using artificial intelligence to identify IV fluid contamination during complete blood counts and subsequent unnecessary red blood cell transfusions. transfusion. Published online August 2026: 10.1111/trf.70072. Doi:https://doi.org/10.1111/trf.70072

2. Spies, NC, Farnsworth, CW. Impact and frequency of IV fluid contamination on basal metabolic panel results using quality indicators. Clinical Laboratory Medicine Journal. 2023;48(1):29-36. Doi:https://doi.org/10.1515/labmed-2023-0098