Inter–laboratory precision and relative accuracy of mechanised and nonmechanised systems for blood gas analysis

Introduction In recent years, quality control has been extended to an everincreasing number of biochemical parameters. Its application to blood-gas analysis is complicated by two main factors: the need for reliable, reproducible quality control materials and the variability of calibration standards among the different clinical departments using blood gas apparatus ]. Moreover, automatic analysers are becoming increasingly sophisticated and complex, which may simplify handling but inevitably increases the risk of error. The purpose of the present study was to compare automatic and non automatic systems for precision and accuracy on the strength of the results of an inter-laboratory quality control survey.


Introduction
In recent years, quality control has been extended to an everincreasing number of biochemical parameters. Its application to blood-gas analysis is complicated by two main factors: the need for reliable, reproducible quality control materials and the variability of calibration standards among the different clinical departments using blood gas apparatus ]. Moreover, automatic analysers are becoming increasingly sophisticated and complex, which may simplify handling but inevitably increases the risk of error.
The purpose of the present study was to compare automatic and non automatic systems for precision and accuracy on the strength of the results of an inter-laboratory quality control survey.

Materials and methods
Twenty-two hospital laboratories from all over France participated in the study, which involved seven mechanised and 15 manual analysers. The two systems were considered separately in order to investigate the possible effects on precision of differences in calibration and operation. All the mechanised instruments were of the same type (ABL I, ABL II from Radiometer). The manual instruments comprised eight Radiometer BMS, two Coming 165 and five Instrumentation Laboratory 413 analysers. All the instruments were in routine use in specialised or non-specialised hospital clinical chemistry laboratories.
A quality control product at three levels (acidosis, normal, alkalosis) was studied. The Preceb6o stated values for these three levels are shown in Table and were calculated from multiple measurements with diffe.rent blood-gas analysers. A single batch of the product was used for all the laboratories participating in the survey. The differences between the three levels must be great enough to enable measurement errors to be characteristic. The quality control programme was conducted over an 18-month period with each laboratory performing ten measurements a month for each level, and for the three parameters measured (pH, pC02 and p02), so giving a .total of 90 analyses per month for each laboratory. At the end of each month, the data from these 22 laboratories were submitted to quality control assessment, with emphasis on two criteria (for each of the three parameters and three levels): that is precision and accuracy.

Accuracy
Accuracy was determined by comparing the monthly mean for a laboratory with the overall mean, having excluded outliers of more than two S.D.S. The overall mean was taken as the target value, in accordance with statistical computations used in most quality control surveys. When the monthly mean for a laboratory did not differ significantly from the overall mean as defined above, the result was considered 'accurate'. The results were judged according to the 't' test for standard deviation of the random error exceeding obvious wrong results and at the 5% level. Conversely, if this condition was not.met, the result was considered 'inaccurate'.

Results
For each parameter, the results from laboratories using mechanised equipment were considered separately from those of laboratories using manual equipment. The percentages of accurate and precise results were compared for each of these two groups. These results are shownTable 2 and Figure 1.
pH For both acidosis and normal levels, the mechanised analysers gave a significantly higher percentage of accurate results than did the manual apparatus. However, for alkalosis results, the difference disappeared. For both types of apparatus precision was similar (Tables 2 and 3).

pC02
With the exception of the acidosis levels, accuracy was much greater for the manual equipment. Precision, however, was better with the mechanised apparatus for acidosis and alkalosis.
In general, the distribution of the results was closer for the mechanised than for the manual apparatus. The results are confirmed by reference to Tables 2 and 3.   p02 The manual apparatus showed a greater percentage of accurate results than did the mechanised apparatus for the three levels, from the lowest to the highest values. The same was true of precision, which was greater for the manual systems, as shown by the distribution of results with Tables 2 and 3.

Discussion
Sources of error revealed during surveys of this type are very varied. Apart from poor sample handling during the analysis, some are common to all analytes studied and some are specific to one only. The lack of a mechanised system for introducing solutions into the different analysers is a possible source of contamination. The calibrating gases must be carefully measured and humidified to ensure correct calibration. The electrodes themselves or the electrode membranes may be faulty. Errors resulting from these defects can cause results for one of the laboratories to beoutside the allowable limits of error for one or more of the analytes. The manual apparatus was less accurate than the mechanised apparatus for pH measurement. The degree of precision calculated by variance analysis is about the same, whatever the type of apparatus used. This is shown on Table 2 and illustrated on Figure 1. This suggests that the poorer accuracy shown by the manual equipment is not related to poor replicability of the measurements but rather to the manual method of calibration. The single-point calibration used by many manual operators is inadequate. Regular two-point calibration as used in the mechanised apparatus, with a correction for drift, may account for the greater accuracy found with the systems. By contrast, pC02 measurements were less accurate with the mechanised systems. Whereas the degree of replicability is about the same for both types of equipment (Table 2) with no loss of accuracy whichever the type of apparatus is used when blood pC02 is low (Figure 1). Results obtained with the mechanised systems are inaccurate, especially for normal and low pC02 values. For the type of apparatus used in this survey preanalysed mixtures of gases were not used. Carbon dioxide and room air mixed in a gas-mixing apparatus are used. With the manual systems, gas mixtures precalibrated with carbon dioxide at two partial pressures are always used. This difference in calibration procedure may account for the lower degree of accuracy found with the mechanised as compared with the manual systems.
p02 measurement was also less accurate with the mechanised appnratus. The difference could be attributed to the fact that the mechanised systems use an electrical zero wheras the manual apparatus is calibrated for p02 with two gas mixtures; nitrogen-carbon dioxide for zero, and oxygen-carbon dioxide for p0 values of about 13.30 kPa. Nevertheless, the data in Table 2 (Table 3) reflects the influence of outlying results. No correlation was found for p0, for pC0 or even for pH measurements, with the percentage of precise results (v 0.53; not significant). So much so that apart from calibration-related inaccuracies, the mechanised systems appear to involve a certain degree of contamination which does not produce an increase in variance and a consequent fall in accuracy. One possible source of contamination could be related to the way in which samples are introduced into the analysers. Inadequately gas-tight measuring chambers might also introduce contamination.
Clearly, the ideal material for quality control of blood gas analysis should be composition and properties identical to those of blood samples being tested. A buffer solution, even if very complex in composition, differs from whole blood and is less satisfactory as a control material. Thus the response of the glass electrode and the residual liquid junction, potential are different for buffers and for whole blood, but since no sufficiently stable commercial whole blood preparation is available, aqueous solutions provide one simple means of assessing th performance of laboratories not specialising in blood gas analysis.

Conclusion
The comparative data obtained in this inter-laboratory survey of mechanised and non-mechanised blood-gas measurement systems point to the need for quality control at three levels, and highlight the fact that inaccuracies often occur with extreme values (at low pC0: and high p0). The findings also suggest that fully mechanised systems are neither more accurate nor more precise under all circumstances. Mechanised equipment lacked replicability in p0: measurements and its use seemed to be associated with a certain degree of contamination.
An important conclusion of this study is that two-point calibration is to be preferred for all measurements, and that equipment involving single-point calibration (non-mechanised) for pH measurement, mechanised for measurements of pC02 and p0z) is inadequate. Undoubtedly, tonometry is still the best method of controlling blood gas analysis, but it is beyond the reach of many laboratories. For this reason, the results of the present survey have important fractional implications.
On page 276 in the second paragraph under the heading "Accuracy" reference is twice made to " Figures 4 and 7". In both cases this should read "Figures 4 to 7". In Figures 11 and 12 on page 280, the line and reference 14] should have been deleted.
In Table 10 the methotrexate concentration should have been given in ktmol.
Some additional information has been provided for Table 9 and for completeness, we have reproduced the entire table. Table 9 Refound values of glucose and urea in pool sera to which various components were added. In the absence of exogenie compounds mean values (x) and standard deviations (s) were calculated from 15 determinations. Results of analysis which are outside the range of confidence (x -+ 3s), are marked with an asterisk. The range of confidence is for glucose 11.14-14.08 mmol/1 and for urea 6.23 6.68 mmol/1.