Uncovering hidden IV fluid contamination through machine learning with Dr. Carly Maucione

Machine learning may help fill gaps in identifying venous (IV) fluid contamination in complete blood counts (CBCs) and prevent unnecessary blood transfusions, according to a new study.¹

Results from a multicenter study suggest that CBC IV fluid contamination may be more common than previously recognized, with machine learning models identifying possible contamination in approximately 2% of samples.¹

“It should shock some people to find out that there is an unmet need like this, but I think it’s okay to be shocked,” Dr. Carly Maucione, a resident and study investigator at Washington University in St. Louis, said in an interview. HCP Live. “The results were higher than we expected. However, there is growing evidence that contamination of IV fluids is a serious problem, and one that can be addressed in the laboratory.”

Contamination of specimens from intravenous fluids has long been recognized as a challenge in clinical laboratories, as it can lead to erroneous CBC results and impact diagnostic interpretation and clinical decision-making.²

As Maucione explained, in the inpatient setting, clinicians may encounter an unexplained drop in hemoglobin that may later normalize or may not have a significant impact on patient management. However, in a more concerning scenario, falsely low hemoglobin values ​​may lead to unnecessary blood transfusion decisions.^{1, 2}

Still, it remains difficult to identify contamination in IV fluids after the sample reaches the laboratory. Detection requires the expertise of laboratory technicians, but standardized methods for training pathologists are lacking. Recognition therefore relies heavily on the technician’s ability to identify anomalous result patterns and deduce possible causes. This approach is inherently variable and difficult to scale across high-volume laboratory workflows.²

Researchers recognized machine learning as a potential solution to standardize care and accurately and accurately identify contamination in IV fluids.¹

To address this gap, researchers conducted a retrospective, multicenter machine learning diagnostic validation study using real-world inpatient data from two institutions. The team developed and tested two machine learning models designed to retrospectively identify IV fluid contamination in CBC results. Since there is no gold standard for confirming contamination, model output was validated against expert chart review.¹

The model was trained using a simulated infusion contamination scenario. These incorporated previous, current, and subsequent hemoglobin concentrations, platelet counts, and white blood cell counts. We then evaluated performance using 1 year of inpatient CBC data from each facility to assess real-world applicability. Transfusions were classified as potentially unnecessary only if the posttransfusion hemoglobin value was unexpectedly higher than the pretransfusion value and exceeded 8 g/dL, consistent with commonly used transfusion thresholds.¹

The model showed strong discriminative performance with areas under the receiver operating characteristic curves of 0.972 and 0.957, and areas under the precision-recall curves of 0.723 and 0.619, respectively. As previously mentioned, approximately 2% of inpatient CBC trios were classified as potentially contaminated across both facilities.¹

Importantly, the researchers assessed the clinical implications associated with the contaminated results. Of the inpatient transfusions for which CBC Trio was available, researchers considered 6% to 9% potentially unnecessary based on a conservative, rules-based definition validated through expert chart review.¹

“We’re seeing that there are clinical outcomes, and those clinical outcomes are something that can be tracked using an algorithm like this,” Maucione said. “We think both of these are worth addressing and moving forward in developing similar algorithms to detect this prospectively and in using these rates as a measure to track.”

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. Newbigging A, Landry N, Brun M, et al. A new solution to an old problem: A practical approach to identifying samples contaminated with intravenous fluids in clinical laboratories. clinical biochemistry. 2024;127-128:110763. doi:https://doi.org/10.1016/j.clinbiochem.2024.110763

‌