Verification of Assayed Blood Gas Quality Control Ranges

Discussion

An analysis of an electronic error reporting system from 30 health care organizations showed that laboratory errors accounted for 14.1% of all quality events.³ Laboratory errors may occur during the pre-analytical, analytical, and post-analytical phases. Whereas pre-analytical errors account for most laboratory events,³ a malfunctioning analyzer is an obvious source of error and is the primary focus of a QC program.

QC is an essential and mandatory part of a laboratory’s quality plan. The acceptable QC range should be 2 SD from the mean value. As mentioned earlier, manufacturers may provide “assayed” expected mean values and ranges, but these must be verified as accurate reflections of the analyzer’s performance. To verify assayed QC mean values and ranges, a minimum of 20 QC measurements (same level from the same lot) should be collected from multiple successive days. Any obvious and infrequent outliers or “random errors” (eg, expected P_O2 value ∼ 50 mm Hg, measured value 100 mm Hg) should be excluded from analysis. If random errors are occurring frequently, the analyzer should be inspected for malfunctions. Using < 20 samples to calculate the mean and SD may not provide an accurate assessment of the analyzer’s performance, and using a very large sample (eg, > 100 values) unnecessarily adds time to the process. Computer software can quickly and accurately calculate mean and SD values; however, laboratory personnel should still understand how these values are derived and what they represent.

The mean or average (denoted by the symbol or μ) is derived by adding the sum of all recorded values and dividing that value by the number of measurements as shown in the equation below:

Where

= mean

Σ^x = sum of recorded values

n = number of values.

SD is a measure of the variability around the mean. Two blood gas analyzers could produce identical mean values for a QC measure (eg, level 1 P_O2); however, the analyzer with the smaller SD has less variability and more precision. Less variability and more precision impact the confidence one should have in patient data because repeating analysis on a blood specimen will not yield exactly the same results. It stands to reason that the data variability present in QC testing may also apply to results reported from patient sampling. For example, consider two identical blood gas analyzers, both report similar mean values for level 1 P_O2, 60 mm Hg; however, the observed range of measured values for analyzer #1 is 55–65 mm Hg, and the observed range of measured values is 45–75 mm Hg for analyzer #2. Clinicians should have less confidence that a patient’s reported P_O2 of 65 mm Hg is truly > 60 mm Hg if analyzer #2 was used as opposed to analyzer #1. The precision of a blood gas analyzer can impact clinical decision-making especially when small differences in blood gas values change clinical classifications such as qualifying for long-term O₂ therapy⁴ or determining ARDS severity.⁵

The lower-case Greek letter σ is used to denote SD. SD calculation starts by summing the difference between each measured value and the mean. For example, if the QC mean is 60, a value of 55 has a difference score of −5, and a value of 65 has a difference score of +5. Because the mean is by definition at the center of all values, the sum of all difference scores will always be zero. To calculate SD, the difference scores will need to be summed; and because some are negative and some are positive, they are squared to transform all difference scores to positive numbers. Now the SD can be calculated by the following equation: where

σ = SD

Σ = sum

x_i = individual values

= mean of all values

n = number of values.

The SD equation subtracts 1 from the number of measurements (one degree of freedom) to calculate a slightly larger SD. Using a larger SD makes the sample variance more generalizable to the entire population. For example, if you were calculating the mean and SD of resting S_pO2 from a sample of 1,000 subjects in a city with a population 100,000, using the number of samples without subtracting 1 might indicate less variance around the mean (ie, smaller SD) than if you sampled all 100,000 residents. This process (n-1) is intended to produce a less biased estimate of variance. Because the SD of blood gas QC is calculated from a sample (eg, 20 QC measurements) and not the total number of measurements projected for the QC lot (eg, 365 measurements for a lot that expires in 1 year), a slightly higher SD is desirable to avoid underestimating the true variance that might be higher if calculated from all 365 measurements in a 1-y period. Using a SD smaller than the true variance for blood gas QC will result in more false “out of control” conditions.

A 2 SD range is used for blood gas QC because the empirical rule states that if the blood gas analyzer is operating normally, and the QC material is unadulterated, 95% of recorded QC values should be within 2 SD of the mean.⁶ Therefore, a QC value 2 SD from the mean should only occur once in 20 measurements (5%). If a recorded QC measurement is between 2–3 SD from the mean, a “warning” condition exists because this is an unusual value (1/20) based on past performance. The empirical rule also states that 99.7% of QC results should be within 3 SD of the mean, so a QC value > 3 SD from the mean should never occur if the analyzer is operating normally and the QC material is unadulterated. Figure 2 shows the distribution or “bell” curve for level 1 P_O2 data. The mean value is at the center of the distribution curve; 68%, 95%, and 99.7% of QC measurements should fall between 1 SD, 2 SD, and 3 SD, respectively. A Levey-Jennings plot charts QC data against the mean and SD as a function of time; whereas the distribution curve is not displayed, it is essentially flipped 90°. Figure 3 shows a cartoon drawing of the differences in the size of the distribution curves for measured versus assayed ranges on a Levey-Jennings plot for level 1 P_O2.

The software of modern blood gas analyzers will alert the laboratory personnel if a QC measurement is out of range, and many devices will lock out the analyzer from performing clinical testing until the QC passes. It is, therefore, important that QC ranges are an accurate reflection of the blood gas analyzer’s true performance.

Our analysis shows the importance of verifying manufacturer-provided QC ranges. Whereas the manufacturer-provided means were essentially identical to our measured mean values, with the exception of level 1 low P_CO2 and level 2 high P_CO2, the manufacturer-provided ranges were markedly outside of 2 SD. This level of discordance could result in undetected analyzer malfunction and potentially impact clinical decision-making.