Analytical performance of the selective, automatic multianalyser Olympus AU 5031

The analytical performance of the selective, automatic multianalyser Olympus AU 5031 was evaluated over four months and assessed for practicability for another eight months. The evaluation followed the ECCLS guidelines. Twenty routine parameters were measured. In addition, sodium and potassium were determined on an attached flame photometric unit. Both the agreement between the eights photometers per unit and the temperature behaviour in the cuvettes was satisfactory. The imprecisions were very good. The within-run imprecision was below 1.5% for the majority of the parameters. The imprecision between days was below 5%, with the exception of creatine phosphokinase (7.4%). Glutamate dehydrogenase gave an imprecision of between 4.0% and 15.9%, which, however, is more likely due to the low activities measured rather than the fault of analyser. The recovery of the assigned values in 12 control sera was between 95% and 105% for 14 tests. Three of the remaining eight tests yielded recoveries with deviations between 10% and 18% (alanine aminotransferase, aspartate aminotransferase and bilirubin). No drift effects were observed and neither a sample carry-over nor a reagent carry-over were detected. Most tests were linear over a very wide range. Only afew tests (mainly lipase and glutamate dehydrogenase) required measurement repetitions with diluted samples. The correlation with routine instruments and tests was close. However, corrections were necessary for 14 of the 22 tests. This was not due to the performance of the analyser but, rather, to the different methodologies of compared tests, or different working temperatures on the comparison instruments, or a lack of accuracy for some of the AU tests.


Introduction
The Olympus AU is an analyser for medium and large laboratories. This paper reports on the performance of the AU 5031. The evaluation lasted six months and the instrument has been in routine use for a further eight months. Although the authors accept that only a multicentre evaluation, as suggested by ECCLS guidelines [1], allows a truly representative assessment for an analyser, multicentre evaluations of analysers of this size require considerable time. The evaluation data are reported here because there is a growing interest in the performance characteristics of the Olympus 5000.
The protocol of this evaluation followed the ECCLS guidelines; changes were made only if they seemed appropriate to the specific features of the AU analyser.
Recently, two evaluations of this analyser have been published [1 and 2]. They, too, followed the ECCLS guidelines but some aspects reported here were not studied or were approached differently. Therefore, the present data and that in the previous reports may serve as a preliminary multicentre evaluation.
General description of the AU 5000 analyser The AU 5000 Analyser is constructed according to a modular concept. Any analyser consists of up to eight units. Each unit is basically a complete analyser and is equipped with either four, eight or 12 reagent lines. The specifications of such a unit are given in table 1. By combining several units the performance of the resulting analyser can be tailored to the requirements of the user in respect to sample throughput, number of tests or both. By choosing the maximum number of 12 reagent lines, the user gains a higher number of available tests at the cost of sample throughput. Possible combinations of available test numbers and throughput capacities are displayed in table 2. However, once a certain configuration is installed it must remain permanent.
In cases of malfunction of a unit, it can be exchanged in less than an hour. Consequently, any AU analyser which contains more than one unit, can, to a certain degree, supply its own back-up system. However, some parts of the analyser have a common function, and, in the event of failure, will cause the analyser to stop. The most important of these is the sample rack transport system and the data-processing unit.
The AU configuration which was evaluated, an AU 5031, consists of three units with eight reagent lines per unit. It processes 150 samples per hour and with an additional (optional) flame photometer carries out 26 tests per sample. Twenty-four cuvettes are assigned to each test and complete one cycle of the cuvette wheel in 8 min, and 24 s. If a test is not required for a given sample, the cuvette passes empty and is washed. Consequently, the processing speed remains constant regardless of the number of tests required for a given sample.
Most of the reagents kits for the AU 5000 were from the Merck Company in Darmstadt, FR Germany. However, there are other suppliers who offer tests which are tailored to this analyser. In addition, since this system is entirely open, many commercial tests can be adapted to the analyser.

Methods
Agreement between photometer and temperature behaviour in cuvettes Two hardware conditions were investigated: the linearity in all eight photometers in a unit and the temperature behaviour in a cuvette after addition of cooled (8 C) reagent 1.
For the linearity experiment, two solutions were prepared with NADH and para-nitrophenol, respectively, which covered in seven dilution steps the extinction range between 0"5 and 3"0 for each solution. The extinctions of each solution were measured in triplicate and were plotted in eight parallel extinction curves for visual inspection of linearity and parallelity.
The temperature in the cuvette was measured with a micro temperature sensor in a cuvette to which 10 [al of sample and, 12 s later, 500 btl of reagent was added. In order to ensure that cooled reagent (8 C) reached the cuvette, the reagent was dispensed repeatedly immediately before the experiment. The rationale of this experiment was to check whether the temperature in the cuvette reached 37 C before reading by photometer 1, 168 s later. Example: An Olympus analyser of the 5030 series consists of three units each ofwhich will be equipped with either eight or four reagent lines. This will, for the total unit, yield 12 or 24 available tests, respectively, allowing a throughput of 150 or 300 samples/h, respectively.  phosphate, uric acid) or 30 C (bilirubin). All enzyme activities which were measured on the AU 5031 at 37 C were converted to corresponding activities at 25 C.

Imprecision
The within-run imprecision was determined from 20 measurements of three commercial control sera. Care was taken to include a control serum containing normal analyte levels (Monitrol I). The series were repeated three times and the medians of the three CV values were taken as the final results.
The same sera were used for the determination of between-day imprecision, which was determined over 21 days (18 working days). Carry-over Carry-over effects are limited due to the construction principle that each cuvette is used for one test only.
However, carry-over might occur as sample carry-over due to insufficient cuvette washing (specimen-related carry-over), through inadequate cleaning of the mixer blades (specimen-independent carry-over).
Specimen-related carry-over Sample carry-over was tested by determining the activity of alkaline phosphatase during two complete cuvette wheel cycles (N 48 samples). All samples were taken from a serum pool with a low activity of 63 U/l, except sample numbers 6 to 10 and 16 to 20 which contained the very high activity of 5200 U/1. Thus, during the second cuvette wheel cycle (sample numbers 25 to 48) samples containing the low activity were measured in cuvettes which were used for the determination of a very high activity of the.analyte in the previous cuvette wheel cycle. If the cuvettes are washed, residue of the sample with high alkaline phosphatase activity should cause elevated results in cuvette numbers 6 to 10 and 16 to 20, respectively.
Drift effects were studied using three commercial and control sera covering different levels of the analytes measured. The sera were dissolved in the morning and split into nine aliquots which were sealed and kept at 4 C until use. Measurements of aliquots were carried out every 60 min for 8 h. For determination of alkaline phosphase and bilirubin the sera were freshly reconstituted every 2 h.

Range limits
Linear ranges were investigated for all tests either by diluting very high concentrations or activities, respectively, of the analyte, or by spiking human serum with the pure analyte. The measured concentrations were plotted against dilutions and the resulting graphs were inspected visually for linearity.
Specimen-independent carry-over In order to investigate the potential reagent carry-over by mixer blades, the combination of lipase and triglyceride determinations was chosen. The first reagent line in a single unit was used for lipase assay and the fifth reagent line for the triglyceride test. Since samples were processed in groups of four, the first of the four mixer blades stirred first the lipase reaction mixture in cuvette and subsequently (after the mixer blade washing procedure) the reaction mixture for the triglyceride determination in cuvette 5.

Imprecision
The within-run imprecision and the between-day impre- It is apparent from this data that 15 of the 22 evaluated tests gave a within-run imprecision of below 1.5%. Six of the seven remaining tests gave values below 2"9% -only glutamate dehydrogenase gave a high imprecision which, however, is not due to a failure of the test or of the analyser, but is a consequence of the low activities which are measured by this test. This performance can be rated as very good.
The between-day imprecision was below 4.5% for at least two out of three control sera. Exceptions were creatine Specimen carry-over The samples below the bars were measured in cuvettes in which an alkaline phosphatase activity of 5,200 U/I had been determined a cuvette wheel cycle before.  to 10 and 16 to 20, respectively, during the preceeding wheel cycle. kinase (7"6, 7"2 and 2"9%) and, again, glutamate dehydrogenase.
The between-day imprecisions of the comparison instrument are also shown in figure 2 (for Monitrol I and II only) the AU 5031 matches this performance.
Close imprecisions were reported by other evaluators who found values surpassing the comparison instrument [3].

Drift
The results of investigations on drift over 8 h were inspected visually after blotting the percentage deviations from the starting value (100%) over a time scale. Figure 3 displays three representative plots. A continuous trend either increasing or decreasing could not be detected for any of the 22 parameters. Deviations were below 5% for all tests. This result is in contrast to Rohac et al. who reported drifts for cholesterol, triglycerides and uric acid, but did not discuss this finding further [3]. Hodgin et al. omitted this aspect from their evaluation [2].

Linearities
Both the linearities found, and the linearity ranges declared by the manufacturer, are given in table 7. The linearities found were better than the manufacturer's for all tests with the exceptions of creatinine kinase and lipase. In fact, a limited linearity was found only for gamma-GT, glutamate dehydrogenase and lipase. For the remaining tests the limitation of linearity is beyond the highest value measured. The findings are in good agreement with the results of other evaluators [2 and 3].
Specimen-related carry-over A potential carry-over of sample residue due to an insufficient washing of the cuvettes was investigated by measuring alkaline phosphatase in cuvettes which had been used before for the determination of the same analyte in a very high activity (5200 U/l). The results were compared with those which were obtained from the same sample in cuvettes, which, a cuvette wheel before, had been used for the determination of this analyte in a low activity (62 U/l). The results are displayed in figure 4 which shows the measured values of this sample in 24 cuvettes (one cycle of the cuvette wheel). It is apparent from the figure that no difference could be detected. The Wilcoxon test did not show a significance difference between both sets of values (p 0"84). Thus, the specimen carry-over was smaller than 0"04%. Carry-over effects between 0"1% and 5"33% were reported for different parameters by other evaluators [1 and 2]. However, neither investigator took into consideration the sequence of cuvettes which were assigned to the respective tests.
Specimen-independent carry-over A carry-over caused by transfer of reagent due to an insufficient washing of the mixer blades was tested in two reagent lines which were served by the same mixer blade (reagent line and reagent line 5). The triglycerides values (means of quadruplicate measurement in reagent line 5) in 10 serum samples did not differ in the Wilcoxon test (p 0"76), regardless of whether concomitant lipase determination had been carried out in reagent line 2 or not. Rohac et al. carried out a similar experiment and found a considerable carry-over leading to threefold lipase measurements in combination with triglycerides. Their results, however, are difficult to discuss because ofa lack of experimental detail [3].

Recovery of assigned values in quality-control sera
The recovered values of all analytes in 10 commercial control sera were expressed as percentage of the assigned values. The medians and the 10th and 90th percentiles of the respective values are displayed in figure 5. Fourteen of 22 tests yielded recoveries within the 5% range. Five tests were in the range between 5% and 10% (y-glutamyl transferase, lactate dehydrogenase, bilirubin, phosphate and total protein) and three tests were in the range between 10 and 18% (alanine aminotrans-ferase, aspartate aminotransferase and creatinine).

Comparison with other analytical routine procedures
The results obtained by determination of all parameters in 100 routine sera on the Olympus AU 5031 and on routine instruments (see tables 3 and 4) were correlated using the method of Bablock and Passing [4]. The results of this statistical evaluation are given in table 8 and figure 6.
The correlations were close for all tests. A correction was unnecessary for eight out of 22 tests (y-glutamyl transferase, sodium, blood urea nitrogen, cholesterol, creatinine, iron, triglycerides). The remaining 14 tests differed either in slope (bilirubin, potassium), or in intercept (alanine aminotransferase, creatinine phosphokinase, glutamate dehydrogenase, lipase, inorganic phosphate, uric acid and calcium), or both (alkaline phosphatase, amylase, aspartate aminotransferase, magnesium and total protein). Statistical data are given in table 8, which displays slope, intercept and the correlation coefficient after. regression analysis according to the equation:y a * x + b where y the AU 5031 and x the comparison instrument. The reasons for the discrepancies were various: different physiochemical methods on the comparison instruments (see tables 3 and 4), different analysis temperatures (see the section on Reagents and comparison instruments) and a lack of accuracy for some methods on the AU 5031 (see figure 5). However, it is unlikely that these discrepancies can be attributed to the performance of the analyser but arise because of the methodological discrepancies summarized above.
The results of these comparisons were discussed with the manufacturers of the respective tests. In consequence, factors of alkaline phosphatase and lactate dehydrogenase have been corrected in accordance with data from the laboratories evaluating the AU analyses.

Assessment of practicability
Despite the size and capacity of the analyser, the Olympus AU 5031. is easy to run. It is easily accessible from all sides and all mechanical movements can. be seen. The data-processing software is sophisticated but can be mastered by a technician after a training period of a few days. The software contains an extensive diagnostic program, which allows a specific detection of technical or electronic problems. Together with a complete system of LED controls of all mechanical movements problems can rapidly be tracked to their origin thus allowing immediate remedy or better communication with the Olympus service.
The preanalytical preparation of the analyser consists of four steps, which require the time periods given in parentheses: warming up to 37 C (60 min); washing of cuvettes and filling of reagent lines with reagent (20 min); reagent blanking plus running controls (20 min). In order to circumvent the long warm-up phase the authors use an electronic clock which starts the warming-up automatically before routine work begins. Thus, the preanalytical phase is reduced to 40 min and this time is required for such routine activities as checking and replacing the reagents.
During the evaluation period and the subsequent routine work no total shut down of the analyser occurred.
However, during .the routine period a single unit broke down twice: in both cases the reason was a faulty adjustment of the sensors which control the movement of the washing units. After correction in all three units this problem did not occur again. The most practical and most rapid solution for this emergency was toconcentrate 'vital' tests on the remaining units. It was easier to transfer reagents and programs from the faulty unit to the other units, than to exchange the total unit as mentioned above ('Description of the AU 5031 analyser'). This transfer of tests was completed in less than 30 min, including washing of the .reagent lines. The units were repaired .by the Olympus service technicians before the next morning. No other breakdown causing a severe delay occurred thereafter.
Two modes of sample identification are possible: via terminal or via bar-code identification. The manual prescribes a correct positioning of the bar code onto the sample cup; the vertical slant should be lower than 5 However, angles up to 15 are read correctly. In addition, the vertical position may vary up to 10 mm.
The following points would improve the performance of the analyser.
(1) Changing from bar-code reading to sample identification via the terminal required a complete termination of bar-code analyses. This causes inevitable delays of 20 min for each change.
(2) The software which produces the print-out is inflexible. For example, all available tests and references ranges are printed, even if only one test has been required for a given sample. In addition, flagging of pathologic values ('high' or 'low') cannot be defined differently for both sexes. As a result, this flag is set only according to one reference range, either female or male. Finally, the matrix printer which is delivered NORCN PHOSPHATE 5MA AU 5031 with the analyser is slow and noisy. The authors have connected a personal computer which runs a versatile custom-made report software and serves a quiet, and very rapid, ink-jet printer (the Epson SQ 2500). (3) The tubing system by which the serum sample is transported to the flame photometer consists of three parts: two narrow tubes made of stainless steel and silicone tubing. Frequent blocks in this system were encountered, which were caused by tiny clots which either pre-existed in the sample or resulted from delayed clotting which could have been triggered by contact with the materials described above. Since an obstruction in the tubing system can result in time losses of up to 45 min this system should be replaced by a continuous Teflon tubing.

Conclusion
The Olympus AU 5031 is a powerful analyser which can be tailored according to the needs of medium and large laboratories. During the evaluation and the subsequent routine period the analyser proved to be very reliable. Both the photometer linearity in a unit and rapid temperature equilibration upon addition of cooled reagent were found satisfactory. Imprecision values very good, while the accuracy of some tests required corrections. Correlations with other routine analysers was close. Neither drift nor a carry-over were detected. Finally, due to the wide ranges of most tests, only a few samples needed to be re-run.