Impact and frequency of IV fluid contamination on basic metabolic panel results using quality metrics

Data aggregation and processing

This study was approved by the Washington University Institutional Review Board (IRB #202202030). All BMP results performed at Barnes-Jewish Hospital between January 1, 2017, and July 1, 2022, were extracted from the laboratory information system (Cerner, Kansas City, USA). Only BMPs for which all eight components were reported into the medical record were included (n=928,742).

Results reported as above or below a threshold value were replaced with that threshold (e.g. a sodium reported as ‘>180 mEq/L’ was replaced with 180 mEq/L). Anion gap was calculated as the sodium concentration minus the sum of chloride and total carbon dioxide concentrations. Only results reported and verified in the clinician-facing electronic medical record (EMR) that may have guided medical decision-making were included in the analysis.

Simulating IV fluid contamination

Motivated by the results of wet-bench mixing experiments [5, 6, 10], IV fluid contamination was simulated as a mixture between the results of real patient samples and the concentrations in common IV fluids. Mixture ratios ranged from 0 to 1, incrementing by 0.01 – where a ratio of 0.50 indicates equal parts patient sample and IV fluid. Supplementary Figure 1 presents a schematic overview of the approach. Simulated errors were validated with an in vitro mixing study with a subset of the fluids described, and specimens were analyzed on a Roche cobas 8,000 chemistry analyzer consistent with previously published studies [10]. Contamination by each of the ten most common IV fluid solutions was simulated, the compositions of which are described in Supplementary Figure 2.

Assessing the analytical impact of contamination via TEa and RCV

TEa thresholds were adopted from the CLIA ‘88 requirements for analytical quality [24]. For analytes without CLIA-defined, absolute TEa thresholds (chloride and CO₂), the absolute thresholds from the Royal College of Pathologists of Australasia (RCPA) [25] were used. Reference change values (RCV) were calculated using the European Federation of Clinical Chemistry and Laboratory Medicine Biological Variation Database [26], which applies the asymmetric approach [27]. As such, contamination that caused an increase in the analyte-of-interest was compared to the appropriate RCV_Increase, while those that caused a decrease were compared with the RCV_Decrease. The mean of the high and low-level analytical coefficients of variation (CV) from our hospital laboratory’s most recent 30 days of quality control analysis were used as the CV_a for the calculation. Median RCV value calculations were used for the final analysis. The TEa and RCV thresholds used for comparison are presented in Supplementary Table 1.

Each of these thresholds were applied to the simulated contamination set, and the mixture ratios at which a majority of contaminated specimens exceeded these thresholds were calculated to mirror the approach used when defining a qualitative cut-off from quantitative measurement data.

Assessing impact on reference intervals and critical flags

To assess the clinical impact of contamination, we applied the abnormal and critical thresholds for each analyte to the simulated contamination set and calculated the proportion of results that would be incorrectly flagged at each mixture ratio. Supplementary Table 2 demonstrates the list of high and low abnormal thresholds and critical values used.

Comparison to the current approach

The procedure used to assess for the presence of contamination of specimens include feasibility flags, delta checks, and manual technologist review. A list of delta check rules and feasibility flags can be found in Supplementary Table 2. Interpretation of potential contamination was at the discretion of each technologist using routine clinical workflows. When a potentially contaminated specimen was identified, the technologists contact the clinical provider and indicate a concern for contamination. If the provider agreed that contamination was the most likely explanation, the results were suppressed and an interpretive comment applied. The absence of the interpretive comment for contamination was used as an objective assessment that a reported result was not identified as contaminated by the current procedure.

Retrospectively flagged results were stratified into those containing and not containing 5 % dextrose using a glucose measurement of 300 mg/dL as a cut-off. The mixture ratio for dextrose contamination was approximated as the increase in glucose divided by 5,000. For example, a glucose result that increased from 100 mg/dL to 600 mg/dL would be assigned a mixture ratio of 0.10 ([600 − 100]/5,000). Non-dextrose contamination was approximated using the decrease in calcium such that a decrease from 9.0 mg/dL to 8.1 mg/dL would be assigned a mixture ratio of 0.10.

Performing the analysis

All statistical analyses were performed in R 4.2.3 using the {tidyverse} [28] and {targets} [29] frameworks. All code written to perform these analyses is available publicly on GitHub at https://github.com/nspies13/contamination_simulation with the anonymized input data available on FigShare at https://doi.org/10.6084/m9.figshare.23805456.v1 [30].

Defining analytically relevant contamination thresholds

Figure 1 summarizes the ratios at which half of the simulated mixtures exceeded TEa for each analyte measured on the BMP for three common fluids; 0.9 % normal saline (NS), lactated Ringer’s (LR), and 5 % dextrose in water (D5W). Supplementary Figure 3 demonstrates the same data for D5NS, D5LR, D5 in 0.45 % NS with and without potassium chloride, 3 % normal saline, and water. A mixture ratio of 0.01 was sufficient to exceed the TEa of 6 mg/dL for glucose in IV fluids containing D5, while mixtures at 0.06 exceeded TEa for non-D5 containing IV fluids. Total CO₂ was also markedly sensitive to contamination across all fluids, with a majority of contaminated results exceeding the absolute threshold of 2 mmol/L defined by the Royal College of Pathologists of Australasia (RPCA) at a mixture ratio of 0.10 for all IV fluids. Chloride, BUN, calcium and potassium exceeded TEa thresholds in a majority of specimens contaminated by NS at ratios of 0.07, 0.12, 0.12, and 0.14, respectively. However, higher mixture ratios were required for a majority of specimens to exceed TEa for LR solutions. Sodium was the analyte least susceptible to saline contamination, though mixtures of 0.04 and 0.08 were sufficient to exceed the TEa of 4 mmol/L for IV fluids that contain water and 0.45 % sodium chloride solutions (Supplementary Figure 3).

Figure 1:

The mixture ratios at which half of the simulated contamination results exceeded the quality metric thresholds for total allowable error (TEa). Red and pink highlight fluid-analyte combinations for which a majority of results exceed TEa at mixture ratios of less than 0.1 and 0.25, respectively.

Incorporating intra-individual variation via the reference change value

Figure 2 displays the mixture ratio from simulated contamination at which the difference between contaminated results and the accompanying non-contaminated result would exceed the RCV in 50 % of cases. More fluid types are displayed in Supplementary Figure 4. D5 at mixture ratios of 0.01 caused a majority of results to exceed the RCV_Increase of 15 % for glucose. Calcium and potassium also demonstrated relatively low mixture ratios at which relevant RCV thresholds were exceeded, requiring a ratio of 0.06 and 0.10 for NS contamination. A majority of LR-contaminated results exceeded the RCV_Decrease of 6 % for calcium at a mixture ratio of 0.16. In contrast to TEa, creatinine results were more sensitive to the RCV_Decrease of 13 %, requiring a mixture ratio of 0.14 for a majority to exceed the threshold across all fluids, while CO₂ and BUN were less sensitive to the RCV_Decrease threshold than to TEa, requiring a mixture of 0.24 and 0.30, respectively, across all fluids.

Figure 2:

The mixture ratios at which half of the simulated contamination results exceeded the relevant reference change value (RCV) by increase or decrease relative to the ground truth result. Red and pink highlight fluid-analyte combinations for which a majority of results exceeded RCV at mixture ratios of less than 0.1 and 0.25, respectively.

Defining clinically relevant thresholds through incorrect normality and critical flags

Figure 3 displays the estimated proportion of results for which the interpretation of the normality of the result would be incorrect due to contamination. The y-axis represents the proportion of results for which the true normality flag does not match that of the simulated contamination. For NS and D5W, a mixture ratio of 0.10 was sufficient to cause a majority of calcium results to be incorrectly flagged, and a ratio of 0.22 caused a majority to display an incorrect critical flag. Similar results were observed for LR contamination at mixture ratios of 0.21 and 0.62 respectively for calcium. For all fluids, a mixture ratio of 0.18 was sufficient to cause a majority of CO₂ results to be incorrectly flagged, while a mixture of 0.58 would cause a majority to display an incorrect critical flag. For glucose, a mixture ratio of 0.01 was sufficient to cause a majority of results to display an incorrect abnormal flag, while a ratio of 0.08 was sufficient for critical flags. Similarly, in D5W contamination, a mixture ratio of 0.03 was sufficient to cause a majority of sodium results to be incorrectly reported, and a ratio of 0.12 caused a majority to display an incorrect critical flag. LR contaminated fluids required the highest proportion of contamination across all analytes to cause an erroneously flagged abnormal or critical result relative to NS and D5W.

Figure 3:

The impact of simulated contamination on the proportion of results reported with incorrect abnormal or critical flags across a subset of clinically relevant analytes and fluids.

Contamination at a mixture ratio of 0.10 is sufficient to cause incorrect abnormal and critical flags

Figure 4 displays the proportion of results that change from normal to abnormal or critical at a mixture ratio of 0.10, chosen due to its impact on the metrics described above. For clarity, only a subset of particularly clinically relevant analytes are shown. A total of 51.39 % of calcium results and 21.17 % of potassium results would be incorrectly reported with an abnormal/critical flag with 10 % normal saline (NS) contamination. Of these, 47.04 % of calcium results that were within the normal range would be reported as low and 0.06 % reported as critical. Of the low calcium results, 2.34 % would be erroneously reported as critical. 13.09 % of truly normal potassium results would be erroneously reported as low, while 0.64 % reported as critical. 6.7 % of high potassium results would be erroneously reported as normal. Similar patterns were observed for the results contaminated by lactated Ringer’s (LR), though to a lesser extent. Contamination by D5W at a mixture ratio of 0.10 resulted in all results flagged as critical for glucose, regardless of their true value. Similarly, a contamination ratio of 0.10 resulted in 99.74 % of sodium results to be incorrectly flagged.

Figure 4:

The impact of contamination at a mixture ratio of 0.10 on the proportion of results displaying each flag. On-diagonal tiles represent correctly flagged results, while off-diagonal results are incorrectly flagged. Grey tiles are those in which the result flag was correctly reported, light red tiles are off by one level (e.g. abnormal to normal), dark red are incorrect by at least two levels (e.g. critically high to normal, or abnormally high to abnormally low).

Estimating the severity of contaminated results by comparison to re-draws

Retrospective BMP results flagged as contaminated by a laboratory technologist (for which a redrawn specimen was measured within 4 h) were analyzed to assess the severity of contamination. Figure 5 displays the distribution of mixture ratios for these results. For results contaminated by a dextrose-containing fluid (blue), the median mixture ratio was 0.18, and the interquartile range was 0.09–0.27. For results contaminated by fluids without dextrose (orange), the median contamination ratio was 0.24 and interquartile range was 0.16–0.38. The bootstrapped 95 % confidence interval for the difference between the medians between D5 and non-D5 contamination was 0.03–0.11.

Figure 5:

Estimated mixture ratios by IV fluids using calcium or glucose as a surrogate measure in specimens identified as contaminated by a laboratory technologist during routine laboratory testing.