Intravenous fluid-induced specimen contamination and detection strategies in clinical laboratories: a narrative review

Introduction

Background

Errors in laboratory results can have serious downstream consequences for patient care (1-4). The majority of these errors occur before the specimen enters the laboratory, in the “preanalytical” phase (5-7). Sources of pre-analytic errors include issues in patient preparation, specimen collection, handling, transportation, and storage before analysis. One source of preanalytical error is contamination of blood samples by intravenous (IV) fluids, which occurs when blood is drawn from a catheter through which exogenous fluids or medications are being infused. It is estimated that IV fluid contamination may occur in as many as 1% of samples. Such contamination can dilute, enrich or interfere with analyte measurement, leading to spurious results that may misguide or delay treatment. These spurious results can be extreme, and are often fleeting in nature, leading to an “anomaly-with-resolution” pattern when observing results in triplicate (8). This triplicate includes a prior, properly-drawn set of results reflecting the patient’s physiologic baseline, a contaminated result displaying significant anomalies in some or all measured analytes, and a subsequent, post result showing resolution of the anomalies if collected with proper technique.

Rationale and knowledge gap

However, there is no consensus method for detecting this anomaly-with-resolution pattern due to IV fluid contamination, nor defining a suspicious result as worthy of cancellation due to contamination (6). The precise impact that a contamination event will exert on measured laboratory results depends on a variety of factors: the composition of the contaminating fluid, its relative proportion (mixture ratio) within the sample, the underlying physiologic values inherent to the specimen, the analytes being measured, and the technology being used to measure them (8). These “degrees of freedom” make for a challenging problem in defining contamination events as clinically significant, and therefore developing a robust mechanism for dealing with them within the laboratory.

Many laboratories rely on a combination of feasibility limits, delta checks (9), and other rule sets or heuristics to flag results that require manual intervention by a technologist. These methods are often insufficient to reliably detect the depth and breadth of contamination events observed in routine laboratory practice (10,11).

Objective

To better evaluate the issue, this review aims to examine the published literature describing the effects of contamination on clinical pathology results (clinical chemistry, hematology, and coagulation parameters), the reported frequency and extent of such contamination in different settings, and methods to detect contaminated specimens—ranging from simple rules to advanced statistical and machine learning (ML) models. We present this article in accordance with the Narrative Review reporting checklist (available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-29/rc).

Methods

The published literature was searched for English language articles published before April 2025. Databases included Embase, Ovid Medline, and Scopus. Keywords used included (pre-analytical OR error OR erroneous OR anomaly OR spurious OR contamination OR dilution OR interference) AND (laboratory OR chemistry test OR metabolic panel OR BMP OR CMP OR complete blood count OR CBC OR assay OR coagulation OR hemostasis OR quality control OR serum OR plasma OR blood). Inclusion criteria included articles published in English before April 2025, covering human clinical blood samples (serum, plasma, whole blood) and studies focusing on reporting effects of IV fluid contamination, dilution, and interference in laboratory chemistry, hematology, or coagulation test. Exclusion criteria included non-English publications, animal studies, articles lacking a clear description of error type or laboratory context, studies involving contamination by substances other than IV fluids, and laboratory tests outside of the chemistry, hematology, and coagulation laboratories. The publications identified were screened for inclusion by at least one reviewer and the full text of articles selected were examined for study characteristics such as author, year, tests examined, error type, detection method and others. The search strategy is summarized in Table 1.

Common characteristics of contaminated laboratory results

When a specimen is contaminated with IV fluid, measured lab values can deviate significantly from the patient’s “true” result. As discussed above, the exact impact of these events depends on a variety of factors. In general, contamination causes a proportional reduction in concentration of analytes not present in the infused fluid, and an increase of any analyte present in the IV fluid. Compounds present in an infusate that interfere with testing can lead to more complex, non-linear patterns of anomalous results, such as heparin interfering with coagulation testing (6). Figure 1 demonstrates these patterns in a schematic overview.

Figure 1 Example effects of IV fluid contamination across three common analytes and contaminating fluid compositions. aPTT, activated partial thromboplastin time; IV, intravenous.

One study examining line draws versus venipuncture found that when best practices were followed (pausing infusions and discarding a waste volume), there were no significant differences in chemistry, complete blood count (CBC), and coagulation study results between samples drawn from a peripheral IV catheter and a direct venipuncture (12). This suggests that discrepancies only arise when contamination with fluid occurs, not inherently from drawing through a catheter. Therefore, emphasis on proper collection technique may aid in the reduction of this issue (12).

Clinical chemistry and metabolic panels

Metabolic panels are commonly ordered in clinical practice, and are used to investigate a wide array of conditions and disorders. The basic metabolic panel (BMP) often contains eight analytes; sodium, chloride, potassium, bicarbonate, blood urea nitrogen (BUN), creatinine, calcium, and glucose. Additional analytes, alanine transaminase (ALT), aspartate aminotransferase (AST), total protein, albumin, alkaline phosphatase (ALP), and total bilirubin, can be added to form the complete (or comprehensive) metabolic panel (CMP). The BMP and CMP aid in evaluating kidney function, fluid status, acid-base and electrolyte homeostasis, and more, making accurate measurements crucial for patient care.

The measurement errors observed when metabolic panels are contaminated by exogenous fluids are a function of the patient’s underlying true concentration, the composition of the contaminating fluid (Table 2), and the relative proportion of fluid in the sample. For example, normal saline (NS) (0.9% NaCl) contains 154 mmol/L (154 mEq/L) of sodium and chloride mixed in sterile water. Therefore, contamination with NS will dilute most measurements down towards zero, but will cause measured sodium and chloride concentrations to converge towards 154 mmol/L (154 mEq/L) (13). This will often result in large decreases in analytes such as calcium or potassium, notable increases in chloride, and only mild increases in sodium, due to the relative differences between normal ranges for those analytes and the composition present in NS. Similarly, Lactated Ringer’s (LR) solution, will dilute most analyte concentrations, but its more physiologic composition makes these changes more subtle. LR’s physiological potassium concentration (4 mmol/L, 4 mEq/L) and small amount of calcium (~1.3 mmol/L, 5.4 mg/dL) make for more subtle changes in these analytes when contamination is present, distinguishing it from NS contamination, and making it much more difficult to detect. Additionally, while not a part of the metabolic panels, LR contamination may also cause moderate increases in lactate (14), but rapid lactate metabolism in vitro and variance in dwell times between collection and measurement make this effect unpredictable and unreliable. LR infusion or bolus may itself also lead to modest increases in measured lactate, despite proper specimen collection technique (15).

Among the most extreme effects seen in contamination by common fluids is that of glucose when contaminated by a dextrose-containing fluid. Standard 5% dextrose in water (D5W) contains ~278 mmol/L glucose (~5,000 mg/dL). Even small amounts of dextrose containing fluids contaminating a sample can lead to spuriously high glucose values, with potentially fatal consequences if acted upon by administering insulin, causing true and extreme hypoglycemia (16).

While glucose and the electrolytes on metabolic panels are present in several fluid types, renal function markers, such as BUN and creatinine, are not. As such, contamination events, regardless of fluid type, will cause proportional decreases in these and other analytes not present in the contaminating fluid (liver enzymes, total protein, bilirubin, etc.).

Finally, acid-base and osmolality measurements can be skewed by contamination as well. Infusion fluids lacking CO₂ will lower measured bicarbonate, causing pH shifts that depend on fluid composition. For example, lactate in LR is metabolized to bicarbonate in the body, but in vitro contamination can elevate lactate and appear acidic. The effect of IV fluid contamination on osmolality is quite predictable (8). Isotonic fluids should exhibit minor or insignificant changes, while hypertonic compositions greatly increase measured osmolality.

Figure 2 summarizes these expected changes by highlighting the distributions of measured concentrations of each analyte in the BMP that would be observed over a range of contamination severities (mixture ratios) from common fluid types. These values are drawn from a publicly available dataset (17) into which contamination by each fluid was simulated. For each analyte and fluid type, a common pattern is observed where the higher the mixture ratio, the more each distribution of results converges upon the concentration for that analyte in the contaminating fluid.

Figure 2 Contamination-induced distribution shifts by fluid type in a basic metabolic panel. Medians (solid lines), interquartile ranges (dashed lines), and middle 95^th percentiles (dotted lines) are calculated for each mixture ratio. Normal saline (NS, blue) is used as a reference. If a particular fluid is not explicitly highlighted, it mimics the effects of NS. BUN, blood urea nitrogen; CO₂, bicarbonate; D5, 5% dextrose; halfNS, 0.45% saline; hyperNS, 3% saline; KCl, 20 mEq/L of potassium chloride; LR, Lactated Ringer’s; NS, 0.9% normal saline.

Quantifying effects and estimating a threshold for “clinical significance”

While the direction of the measurement errors introduced by a contamination event may be largely predictable, the clinical context in which they occur, and their risk to patient care, are much less predictable. Clinical laboratories must balance the quality of reported results with volume and speed considerations. Sample volumes, particularly for common chemistry panels, far exceed what can be feasibly verified by technologists in routine clinical workflows. As such, laboratories rely on “auto-verification” rules to achieve this balance. Defining appropriate boundaries for autoverification, however, requires that we first define the extent to which a contamination event might be deemed “clinically significant”.

In one simulation-based study (8), the effect of contamination by a variety of fluids was quantified and contextualized using common quality or reference metrics, such as total allowable error (TEa), the reference change value (RCV) (18), and the proportion of results that fall within normal, abnormal, and critical ranges. To summarize, the authors concluded that, because a mixture ratio of 10% was sufficient to exceed TEa and RCV, and alter the abnormal or critical flags across several analytes and fluid types, it is acceptable to use 10% as a general, albeit unrefined, threshold for approximating clinically significant contamination in BMP results. Unfortunately, beyond the BMP, there is no published literature estimating clinically significant contamination thresholds for most other chemistry analytes. However, in the absence of literature guidance, it may be reasonable to use the analyte’s TEa as a proxy until more comprehensive evaluation can be performed.

Hematology and the CBC

Unlike chemistry results, the effect of IV fluid contamination on hematology results is somewhat more straightforward, since the analytes measured in hematology testing are absent from administered IV fluids and medications. Therefore, contamination would be expected to cause a uniform, proportional drop in cell counts and concentrations across the CBC. Clinical observations have been used to confirm these dilutional effects in vivo: in a controlled study of adult patients, it was shown that a 1 liter 0.9% saline bolus can significantly decrease hemoglobin, hematocrit, and white blood cell (WBC) count when measured post-infusion (19).

For in vitro sample contamination, the expected effects should be immediate: a sample that is diluted with IV fluid by 10% should result with a ~10% decrease in hemoglobin, hematocrit, platelet, and WBC concentration compared to baseline results. The clinical impact of these dilutional effects may involve a significant change in patient management if results fall below the thresholds for blood product administration. For example, dilution that causes a patient’s hemoglobin to decrease below the red blood cell (RBC) transfusion threshold may subsequently receive blood products based on falsely low results. Pediatric patients, who generally have smaller sample volumes, are especially vulnerable to contamination events, since even small volumes of contaminating fluids such as catheter flushes may lead to significant dilutional effects.

This typical pattern of reduction in all cellular elements (RBC, WBC, platelets) with concurrent increase in analytes present in the infusate has been previously reported (13). A contaminated CBC might be recognized, for example, by a spuriously low hematocrit, alongside a CMP with unexpectedly low total protein and albumin and increased sodium and chloride—a pattern suggestive of sample dilution rather than a physiologic cause of anemia, such as acute blood loss. Clinicians should therefore be aware that a sudden drop in hemoglobin or platelet count without clinical explanation could indicate IV fluid contamination, especially if drawn from an indwelling line.

Coagulation studies effects

Coagulation assays are uniquely sensitive to contamination with heparin or citrate—anticoagulants often present in IV lines or flush solutions. Contamination with citrate (sometimes used as a locking solution in dialysis catheters) can chelate calcium, leading to extremely prolonged clotting times as well as spurious hypocalcemia on chemistry testing. Central lines are often maintained with heparinized saline flushes to prevent clotting. If a sample intended for clot-based coagulation testing of prothrombin time (PT) or activated partial thromboplastin time (aPTT) is drawn from a line containing residual heparin, even a small amount of contamination can falsely prolong clotting times (20). Heparin contamination at levels as low as 1–2 U/mL can markedly elevate PT, with even lower concentrations significantly affecting aPTT. In that same study, the authors estimated the prevalence of heparin contamination at 1.4%, using a drop in aPTT after heparin neutralization as the definition for contamination. Notably, no patient in this study was prescribed anticoagulation therapy at the time, and contamination was attributed to either improper collection from a heparinized line or improper collection tube order (20). In practice, if an aPTT sample is drawn from any line, a common recommendation is to discard 5–10 mL of blood or flush volume first, to avoid this issue (21). One comparative study confirmed that when proper flushing and discard recommendations were performed, coagulation results from arterial lines correlated well with venipuncture, but improper collection technique led to spurious prolongation (22,23).

Heparin interferes with most clot-based assays, producing a pattern of prolonged clotting that corrects with subsequent heparin neutralization. Thus, labs encountering an unexpectedly high aPTT will often perform a heparin neutralization. If the clotting time corrects to normal after adding a neutralizer, it can be interpreted as a confirmation of heparin contamination. Other coagulation tests such as the thrombin time (TT) and reptilase time can be used to detect heparin contamination. Heparin inhibits thrombin through its action on antithrombin III (24), prolonging the TT, but reptilase time is not affected by heparin and will not be prolonged. A pattern suggestive of heparin contamination will therefore show a prolonged aPTT, prolonged TT and a normal reptilase time (25). Another approach is to measure anti-factor Xa activity; any anti-Xa activity in a patient not on heparin or other Xa inhibitor therapy indicates heparin contamination (24).

Undetected IV fluid contamination of coagulation testing can have significant clinical consequences. For example, a falsely prolonged aPTT might prompt unnecessary reversal of anticoagulation or delay needed procedures. Falsely prolonged aPTT or PT/international normalized ratio results may also lead to unnecessary diagnostic testing aimed at ruling out clinically significant causes, which can prolong patient hospitalizations and incur unnecessary expenses. Clinicians should therefore interpret unexpected coagulation results with caution if the sample was drawn through an indwelling line.

Frequency and impact of IV fluid contamination

Despite safeguards, IV fluid contamination of specimens continues to occur in both adult and pediatric settings, though reported frequencies vary by institution and detection method. Because contamination often goes unrecognized, its true incidence is hard to quantify; however, studies that have implemented systematic detection algorithms provide some estimates.

Subjectively, some experts have estimated the prevalence of contamination to be around 1% (26). This estimate aligns well with the 1.4% prevalence of heparin contamination observed in aPTT results (20), as well as the detection rate from manual review of BMP results in studies using machine-learning based detection (11). However, several others have estimated much lower prevalences, including the International Federation of Clinical Chemistry working group’s aggregation of laboratory-reported prevalences (6), which estimated the mean contamination rate for non-microbiology samples at 0.01%. In a 2024 multi-hospital study by Newbigging et al. (27), an automated algorithm flagged approximately 0.03–0.07% of all chemistry samples in large hospitals as likely contaminated, and ~0.003% in a community laboratory setting. These low percentages (on the order of a few in 10,000 samples) reflect only the detectable contamination events meeting the algorithm’s criteria, typically the more severe cases. Another study analyzed >928,000 BMPs over 5.5 years and found that routine manual review only caught contamination when it was fairly extreme—corresponding to roughly 18–24% dilution on average in flagged samples (8). In simulations, that degree of contamination (>25%) was needed to produce obvious aberrations that human reviewers would notice; while contamination at ~10% was rarely identified as such (28).

Differences in patient populations and practices influence contamination frequency. Patient locations with many inpatient draws from indwelling lines (intensive care units, pediatric wards with central lines, etc.) tend to have more opportunities for error compared to outpatient clinics, where venipuncture is standard and IV lines are rare. For example, the Newbigging algorithm’s flag rate was an order of magnitude higher in hospitals (0.03–0.07%) than in a community lab (0.003%), consistent with infrequent line draws in the latter, while a ML-based estimation showed higher prevalence in critical care, emergency, and obstetric patient floors (11). Pediatric hospitals also have unique challenges: infants often have IV fluids and limited blood volume, and drawing from lines is sometimes preferred to avoid repeated needle sticks. However, pediatric-specific contamination rates are not widely published.

The degree of contamination when an event occurs is also important. Studies that have quantified the mixture ratio found that it can vary widely. As noted, flagged cases often involve substantial dilution, with median mixture ratios detected by standard protocols nearing ~20% (8). Given that many detection protocols rely on univariate delta checks, many subtle contamination events may be going unnoticed due to autoverification. Choucair et al. (29) performed in vitro mixing and examined contamination levels from 1% to 50%. They observed that even 5–10% contamination is enough to significantly alter several analytes beyond biologic variability. However, minor contamination (<5%) might produce changes within analytic and biologic noise, thus evading detection entirely. The mixing study results were used to propose a series of more robust, multi-analyte delta check rules specific to NS, LR, and D5W. The goal of newer detection methods is partly to uncover these lower-level contamination events that still could be clinically relevant. For example, a 5% dilution might cause a measured hemoglobin to drop from 7.2 to ~6.8 g/dL (72–68 g/L), possibly prompting an unnecessary transfusion.

Though rare, IV fluid contamination rates are not insignificant: likely occurring on the order of ~1% of samples in environments where line draws are routine. Unfortunately, the majority of incidents likely go unrecognized by current workflows, leading to practice-based estimates much lower than better-equipped retrospective analyses. These discrepancies, and the harms they might cause, highlight the need for improved detection and reinforce the importance of preventive measures (proper flushing, pausing infusions, discarding initial volume) to reduce the occurrence of contamination.

These considerations, while pertinent to many clinical and laboratory environments, can also be subject to context-specific nuances. For example, the “cost” of obtaining a specimen through proper venipuncture is not the same in all patients. Pediatric patients and those populations for which frequent phlebotomy would be necessary (e.g., critical illness, oncology, dialysis, etc.) may be willing to sacrifice the reduced test accuracy in exchange for the added comfort of a line draw. Providers may also be able to appropriately adjust for possible contamination in a specimen they collect, if they know the type of fluid that is present in the collecting line. No literature in our search addressed these important considerations for patient-centric, holistic medical care.

Methods for detecting contaminated samples

The detection of these rare, often complex patterns of measurement error has been the focus of several research groups over the past two decades. Below, we outline the major approaches in the literature and their performance, summarized in Table S1.

Rules-based and heuristic detection methods

Some of the simplest and earliest automated detection methods are still commonly in use today. The most basic rules rely on detecting physiologically implausible patterns of results. Examples of physiologically implausible results include a negative anion gap, hyperkalemia incompatible with life, or unexplained extreme hyperglycemia (30,31). A strength of these simple rules-based systems is their transparency, because each alert is backed by human-intelligible criteria [e.g., “sample flagged because Cl >110 mmol/L (110 mEq/L) and anion gap <0”]. However, these single-analyte, rules-based approaches often lack sufficient sensitivity to detect all clinically relevant contamination events.

Beyond result-based feasibility limits, many laboratories implement “delta checks”, or automated comparison of current results to the patient’s historical results (9). The most basic delta check is a single-analyte delta check, where each analyte is compared individually. For example, a large drop in hemoglobin from prior values could indicate contamination. However, delta checks alone are inadequate to detect all contamination events, as not all patients have recent baseline labs and delta check thresholds may be set too wide to detect clinically significant contamination events. Additionally, if baseline labs are sufficiently abnormal, even a contaminated result might not trigger a delta flag. Single-analyte delta checks are often neither sensitive nor specific enough to reliably detect IV fluid contamination and other common pre-analytic errors (27,32).

To improve upon single-analyte rules, researchers have developed multi-analyte heuristic algorithms that look at combinations of results. Choucair et al. used a series of in vitro mixing studies to develop a set of three-analyte delta checks for NS, LR, and D5W contamination, described above (29). Similarly, Patel et al. (33) evaluated a series of multi-analyte rules for detecting NS contamination, and identified proportional change in chloride and calcium as the most effective rule.

Another approach is to exploit the dilution effect on analytes not present in common IV fluids. Rios Campillo et al. designed a heuristic algorithm using 11 biochemical markers usually absent from IV infusions and normally stable within individuals (34). For each analyte, a “dilution score” was calculated, representing the percentage of analytes showing a significant drop from the previous result. Using retrospectively confirmed contamination cases, they defined contamination as >20% drop in >60% of the measured analytes. This rule achieved a specificity of ~95% and a sensitivity around 64%, compared to expert review as the gold standard. With this approach and threshold, roughly two-thirds of confirmed contaminated samples were detected and only ~5% of clean samples were incorrectly flagged. Notably, since this “dilution effect” approach simply looks for a concerted decline in multiple unrelated analytes, detection of dilution was independent of IV fluid composition. While theoretically sound, in a high-volume clinical lab, a 5% false alarm rate runs the risk of rapidly overwhelming both laboratory staff and clinicians as well as potentially masking truly elevated results as “contaminated”. Additionally, this approach is tailored to the relatively severe mixture ratio of 0.20, making it suboptimal for the more mild, but still potentially significant, contamination events.

Another heuristic-based method was proposed by Newbigging et al., who developed a rule set to capture known contamination “fingerprints” specific to different IV fluids (27). For example, NS contamination was detected with the combination of high chloride, low anion gap, and a proportional drop in total protein, while dextrose contamination was detected with the combination of extremely high glucose with a concomitant drop in sodium and other analytes. When prospectively implemented in a large tertiary hospital lab, it demonstrated 92% positive predictive value and 91% negative predictive value, with 92% agreement with expert adjudication. In 6 months of use, the overall flag rate remained low (0.03–0.07%) and false positive flags were rare, suggesting that a multi-analyte rules-based approach can robustly catch the most severe true contamination events while minimizing disrupting clinicians and laboratory staff with false alerts. However, similar to the Rios Campillo approach (34), this heuristic-based approach will likely fail to capture more mild contamination events.

In summary, heuristic detection combines pattern-recognition rules with explainable thresholds. When well-tuned, these algorithms have achieved high positive predictive value in practice. But while false positive rates are low, they are unlikely to be as optimized as statistical, ML, or “big data” approaches.

Statistical and ML approaches

As computing capabilities in lab medicine continue to grow, so do the opportunities to explore more data-driven or complex techniques (35), ranging from simple statistically guided decision trees to complex ensemble approaches with multiple ML models. A pioneering approach was described by Baron et al. (36). The authors collected a training data set of BMPs with critically elevated glucose concentrations (>500 mg/dL, >27 mmol/L) curated by expert chart review as being physiologic or spurious due to dextrose contamination. They then used the R package {Rpart} to build and optimize simple decision trees to predict these expert-curated labels using BMP analytes and engineered features, such as the patient’s mean glucose concentration over the prior 30 days. The optimized decision trees achieved ~90% sensitivity and specificity on cross-validation, but given practical concerns regarding implementation into their laboratory information system, they also evaluated a simplified model using only glucose and anion gap. Their implementation of the simplified model [glucose >800 mg/dL (>44.4 mmol/L) and anion gap <15] demonstrated substantial increases in sensitivity, labeling 74% of results with extreme elevations in glucose as contaminated, compared to 4% prior to implementation. While limited to dextrose-containing fluid contamination, this approach highlights the potential for data-driven approaches to improve performance, even if the final implemented model is a simplification of a more complex, ML-based approach.

More recent efforts in ML-based detection have included unsupervised, supervised, and even large language model-based approaches. While the exploration of GPT-4’s capabilities in this regard were underwhelming at best, what was more striking was the similarly poor performance from expert reviewers in identifying simulations of NS and dextrose contamination (28).

Given the difficulty in procuring a large, high-fidelity set of training labels, unsupervised ML approaches are highly attractive. To this end, Spies et al. (37) aggregated over 2.5 million BMPs to generate a Uniform Manifold Approximation and Projection (UMAP) model that collapsed the 8 analytes in the BMP into a two-dimensional embedding. Then, using simulations of contamination, mapped each region of this embedding to a corresponding likelihood that a point in that region is physiologic or contaminated. In a test set of 100 results flagged by the UMAP model, 78 were true positives by expert review, with 58 having been missed by the lab’s current workflows, 56 of which contained at least 1 spuriously critical reported result. However, even this model was insufficiently sensitive in detecting contamination at a mixture ratio of 0.10, the threshold deemed potentially clinically significant (8). Of the simulated NS contamination at a mixture of 10%, only 3% were flagged positive by the UMAP approach.

To achieve acceptable sensitivity to mild, but significant, contamination events, the authors next attempted to combine fluid-specific XGBoost (38) and multilayer perceptron models together using simulated contamination as training data, in a technique known as “ensembling” (11). The ensemble approach achieved the desired sensitivity in detecting simulated and real-world contamination events at a 10% mixture ratio, but also flagged a significant proportion of results as contaminated that were not identified as such by the expert reviewers. Whether these discrepancies represent false positives, or mild contamination events the reviewers missed, remains unclear, as no gold-standard confirmation method is available for defining contamination. Altogether, in one year of prospective predictions, the ensemble approach flagged 2.5% of BMPs as potentially contaminated, of which 1.5% were predicted to be significantly mixed (>10% mixture ratio). In that same period, the lab’s current workflow of delta checks and manual review identified 0.1% of BMPs as contaminated, reinforcing the relative insensitivity of the standard-of-care within the field.

It is important to note that while contamination in BMP results is being actively explored via ML techniques, there remains a significant research gap in how well these approaches can generalize to the CBC, coagulation testing, or other common testing panels. Given the success of ML approaches for other error types in these panels, there is reason for optimism that they may be similarly effective in contamination detection (39-44). However, while the ML approaches may be more effective than the simpler rules-based approaches that can be built directly into the laboratory information system, they also come with a substantially increased implementation cost due to significant ML deployment barriers within clinical laboratories (45,46).

Limitations

This review is subject to several limitations. It depends on the quality of included studies and is therefore affected by publication and selection bias. Additionally, despite efforts to be comprehensive, relevant studies may have been missed due to ambiguity in how the errors were described. For example, “dilution” and “contamination” are common descriptors for these errors, but may not be universal. Any laboratory using idiosyncratic language to describe these events may have published relevant work that was not included in our literature search. Finally, it is likely that relevant literature has been published in languages beyond English. However, these works were excluded by study design.

Conclusions

Contamination of blood specimens by IV fluids is a preventable source of laboratory error that can profoundly impact patient management if unrecognized. The studies reviewed demonstrate that even small amounts of IV fluid can cause potentially significant measurement error in chemistry panels, blood counts, and coagulation assays, but the current standard-of-care methods for detecting these errors are insufficiently insensitive. Clinicians and laboratorians should be aware of the subtle, yet recognizable, error patterns: simultaneous dilution of multiple lab values, anomalous electrolyte ratios, or unexpected extreme results. When these error patterns are detected, re-sampling with proper phlebotomy technique is likely indicated before any significant clinical intervention is taken.

On the laboratory side, the last decade has seen improvement in detection algorithms that complement traditional delta checks. Multi-analyte rules and statistically guided checks can be implemented within existing laboratory information systems to respectable effect. More recently, ML models have shown impressive ability to identify subtle contamination events with high reliability, but come with a substantially higher implementation cost.

Future efforts should continue to refine these tools for integration into laboratory information systems, while aiming to establish best practice guidelines for defining and detecting contamination. With these in place, institutions can more robustly evaluate the extent of the issue in their own laboratory, begin refining their detection workflows, and most importantly, educate their specimen collectors and clinical providers in how to reduce the prevalence and impact of these errors. Continued research and interdisciplinary collaboration will further reduce these errors, moving us toward more reliable laboratory diagnostics in both adult and pediatric care.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-29/rc

Peer Review File: Available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-29/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jlpm.amegroups.com/article/view/10.21037/jlpm-25-29/coif). N.C.S. serves as an unpaid editorial board member of Journal of Laboratory and Precision Medicine from April 2025 to March 2027. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Institute of Medicine (US) Committee on Quality of Health Care in America. To Err is Human: Building a Safer Health System. Kohn LT, Corrigan JM, Donaldson MS, editors. Washington (DC): National Academies Press (US); 2000.
2. Bonini P, Plebani M, Ceriotti F, et al. Errors in laboratory medicine. Clin Chem 2002;48:691-8.
3. Plebani M. Errors in laboratory medicine and patient safety: the road ahead. Clin Chem Lab Med 2007;45:700-7. [Crossref] [PubMed]
4. Plebani M. Laboratory-associated and diagnostic errors: a neglected link. Diagnosis (Berl) 2014;1:89-94. [Crossref] [PubMed]
5. Carraro P, Plebani M. Errors in a stat laboratory: types and frequencies 10 years later. Clin Chem 2007;53:1338-42. [Crossref] [PubMed]
6. Sciacovelli L, Lippi G, Sumarac Z, et al. Pre-analytical quality indicators in laboratory medicine: Performance of laboratories participating in the IFCC working group "Laboratory Errors and Patient Safety" project. Clin Chim Acta 2019;497:35-40. [Crossref] [PubMed]
7. Lin Y, Spies NC, Zohner K, et al. Pre-analytical phase errors constitute the vast majority of errors in clinical laboratory testing. Clin Chem Lab Med 2025;63:1709-15. [Crossref] [PubMed]
8. Spies NC, Farnsworth CW. Impact and frequency of IV fluid contamination on basic metabolic panel results using quality metrics. Journal of Laboratory Medicine. 2024;48:29-36.
9. Ladenson JH. Patients as their own controls: use of the computer to identify "laboratory error". Clin Chem 1975;21:1648-53.
10. Iqbal MS, Tabassum A, Arbaeen AF, et al. Preanalytical Errors in a Hematology Laboratory: An Experience from a Tertiary Care Center. Diagnostics (Basel) 2023;13:591. [Crossref] [PubMed]
11. Spies NC, Militello L, Farnsworth CW, et al. Prospective and External Validation of an Ensemble Learning Approach to Sensitively Detect Intravenous Fluid Contamination in Basic Metabolic Panels. Clin Chem 2025;71:296-306. [Crossref] [PubMed]
12. Chen K, Kong W, Liao C, et al. Comparison of laboratory results between central venous access devices and venipuncture: A systematic review and meta-analysis. J Vasc Access 2024;25:1376-88.
13. Truong P, Sargsyan Z. Typical Pattern of Spurious Laboratory Results from Saline Dilution. Am J Med 2023;136:e204-5. [Crossref] [PubMed]
14. Schwartz AL, Basilico MF, Friedman JL. Lactates and Local Knowledge - A Parable of Teamwork. N Engl J Med 2019;380:1392-3. [Crossref] [PubMed]
15. Zitek T, Skaggs ZD, Rahbar A, et al. Does Intravenous Lactated Ringer's Solution Raise Serum Lactate? J Emerg Med 2018;55:313-8. [Crossref] [PubMed]
16. Sinha S, Jayaram R, Hargreaves CG. Fatal neuroglycopaenia after accidental use of a glucose 5% solution in a peripheral arterial cannula flush system. Anaesthesia 2007;62:615-20. [Crossref] [PubMed]
17. Spies N. 2,500,000 anonymized BMP results for the manuscript “Automating the Detection of IV Fluid Contamination Using Unsupervised Machine Learning” figshare; 2023 [cited 2023 Aug 14]. Available online: https://figshare.com/articles/dataset/2_500_000_anonymized_BMP_results_for_the_manuscript_Automating_the_Detection_of_IV_Fluid_Contamination_Using_Unsupervised_Machine_Learning_/23805456/1
18. Fraser CG. Reference change values. Clin Chem Lab Med 2011;50:807-12. [Crossref] [PubMed]
19. Gokcen E, Cevik Y, Pamukcu Gunaydin G, et al. Change in Complete Blood Count Parameters and Vital Signs After Intravenous Infusion of 1 Liter of Normal Saline. Eurasian J Emerg Med 2018;17:71-4.
20. Tientadakul P, Kongkan C, Chinswangwatanakul W. Use of an automated coagulation analyzer to perform heparin neutralization with polybrene in blood samples for routine coagulation testing: practical, rapid, and inexpensive. Arch Pathol Lab Med 2013;137:1641-7. [Crossref] [PubMed]
21. CLSI. Collection, Transport, and Processing of Blood Specimens for Testing Plasma-Based Coagulation Assays. 2024. Available online: https://clsi.org/shop/standards/h21/
22. Lehavi A, Borissovski V, Zisser A, et al. Effect of Heparin Flush in Blood Drawn from Arterial Line on Activated Clotting Time and Thromboelastogram. J Anesth Clin Res 2016;7:610.
23. Lew JK, Hutchinson R, Lin ES. Intra-arterial blood sampling for clotting studies. Effects of heparin contamination. Anaesthesia 1991;46:719-21. [Crossref] [PubMed]
24. Hirsh J, Anand SS, Halperin JL, et al. Mechanism of action and pharmacology of unfractionated heparin. Arterioscler Thromb Vasc Biol 2001;21:1094-6. [Crossref] [PubMed]
25. Newman RS, Fagin AR. Heparin contamination in coagulation testing and a protocol to avoid it and the risk of inappropriate FFP transfusion. Am J Clin Pathol 1995;104:447-9. [Crossref] [PubMed]
26. Marilyn Hamilton, MD, PhD. What percentage of your blood specimens are contaminated with IV Fluid? Available online: https://www.aacc.org/science-and-research/scientific-shorts/2016/what-percentage-of-your-blood-specimens-are-contaminated-with-iv-fluid
27. Newbigging A, Landry N, Brun M, et al. New solutions to old problems: A practical approach to identify samples with intravenous fluid contamination in clinical laboratories. Clin Biochem 2024;127-128:110763. [Crossref] [PubMed]
28. Spies NC, Hubler Z, Roper SM, et al. GPT-4 Underperforms Experts in Detecting IV Fluid Contamination. J Appl Lab Med 2023;8:1092-100. [Crossref] [PubMed]
29. Choucair I, Lee ES, Vera MA, et al. Contamination of clinical blood samples with crystalloid solutions: An experimental approach to derive multianalyte delta checks. Clin Chim Acta 2023;538:22-8. [Crossref] [PubMed]
30. Lippi G, Avanzini P, Sandei F, et al. Blood sample contamination by glucose-containing solutions: effects and identification. Br J Biomed Sci 2013;70:176-9. [Crossref] [PubMed]
31. Cornes MP. Exogenous sample contamination. Sources and interference. Clin Biochem 2016;49:1340-5. [Crossref] [PubMed]
32. Strathmann FG, Baird GS, Hoffman NG. Simulations of delta check rule performance to detect specimen mislabeling using historical laboratory data. Clin Chim Acta 2011;412:1973-7. [Crossref] [PubMed]
33. Patel DK, Naik RD, Boyer RB, et al. Methods to identify saline-contaminated electrolyte profiles. Clin Chem Lab Med 2015;53:1585-91. [Crossref] [PubMed]
34. Rios Campillo C, Sanz de Pedro M, Iturzaeta JM, et al. Design of an algorithm for the detection of intravenous fluid contamination in clinical laboratory samples. Clin Chem Lab Med 2023;61:2002-9. [Crossref] [PubMed]
35. Spies NC, Farnsworth CW, Jackups R. Data-Driven Anomaly Detection in Laboratory Medicine: Past, Present, and Future. J Appl Lab Med 2023;8:162-79. [Crossref] [PubMed]
36. Baron JM, Mermel CH, Lewandrowski KB, et al. Detection of preanalytic laboratory testing errors using a statistically guided protocol. Am J Clin Pathol 2012;138:406-13. [Crossref] [PubMed]
37. Spies NC, Hubler Z, Azimi V, et al. Automating the Detection of IV Fluid Contamination Using Unsupervised Machine Learning. Clin Chem 2024;70:444-52. [Crossref] [PubMed]
38. Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco, CA, USA: ACM; 2016:785-94.
39. Doctor JN, Strylewicz G. Detecting 'wrong blood in tube' errors: Evaluation of a Bayesian network approach. Artif Intell Med 2010;50:75-82. [Crossref] [PubMed]
40. Le QA, Strylewicz G, Doctor JN. Detecting blood laboratory errors using a Bayesian network: an evaluation on liver enzyme tests. Med Decis Making 2011;31:325-37. [Crossref] [PubMed]
41. Farrell CL. Decision support or autonomous artificial intelligence? The case of wrong blood in tube errors. Clin Chem Lab Med 2022;60:1993-7.
42. Farrell CJ, Makuni C, Keenan A, et al. A Machine Learning Model for the Routine Detection of "Wrong Blood in Complete Blood Count Tube" Errors. Clin Chem 2023;69:1031-7. [Crossref] [PubMed]
43. Benirschke RC, Gniadek TJ. Detection of Falsely Elevated Point-of-Care Potassium Results Due to Hemolysis Using Predictive Analytics. Am J Clin Pathol 2020;154:242-7. [Crossref] [PubMed]
44. Graham BV, Master SR, Obstfeld AE, et al. A Multianalyte Machine Learning Model to Detect Wrong Blood in Complete Blood Count Tube Errors in a Pediatric Setting. Clin Chem 2025;71:418-27. [Crossref] [PubMed]
45. Bunch DR, Durant TJ, Rudolf JW. Artificial Intelligence Applications in Clinical Chemistry. Clin Lab Med 2023;43:47-69. [Crossref] [PubMed]
46. Spies NC, Farnsworth CW, Wheeler S, et al. Validating, Implementing, and Monitoring Machine Learning Solutions in the Clinical Laboratory Safely and Effectively. Clin Chem 2024;70:1334-43. [Crossref] [PubMed]