Multicentre evaluation of the Monarch (IL) clinical chemistry analyser

A multicentre evaluation of the Monarch centrifugal analyser is reported. Precision, linearity and accuracy were assessed by comparison with routine methods. Calibration stability, photometric and dispensing accuracy, and carry-over related to samples and reagents were also evaluated. The overall performance of the instrument was good, showing an excellent photometric and dispensing accuracy, absence of sample-dependent carry-over, and almost negligible reagent carry-over. Good precision, linearity and correlation with routine methods were found for the parameters tested. The instrument is reliable and is now used as the routine clinical chemistry analyser in two of the three laboratories taking part in the evaluation.


Introduction
The Monarch is an automatic random access centrifugal analyser for clinical chemistry determinations. Random analysis of samples is possible using a robotic system to sample or control positions and six calibrator positions. The instrument is programmed via a keyboard and a video display unit with a user-friendly menu.
The theoretical throughput of the instrument is about 400 results/h when performing photometric analyses, and about 600 results/h when the ISE module is included. The Monarch automatically organizes the analytical cycle to maximize throughput, which depends upon the number of samples, the number of analyses per sample and whether one or two reagents are being used.
The evaluation reported here took photometric analyses (biochromatic equilibrium, fixed time, kinetic) and the ISE module into account. Additionally, general features, such as photometric performance and dispensing and diluting accuracy using a bichromate solution, were considered.
The work was done on three different instruments installed in three laboratories. For practical and organizational reasons it was not possible to perform every experiment in each laboratory, as defined in the ECCLS multicentre evaluation protocol ]; therefore each participating laboratory performed a different part of the evaluation, but some critical tests were repeated in more than one laboratory. The within-and between-batch imprecision was evaluated using the following materials:

Reagents
(1) Pool of normal sera, subdivided in aliquots and stored at -20 C.    Quality control materials were reconstituted at the beginning of each working day.

Experimental design
Imprecision" materials at three different levels of concentration were analysed five times per day for 10 days.
Calibration stability: the absorbances of the calibrators were recorded during a period of 30 to 71 days, and a regression analysis was performed to evaluate the possible presence of any significant trends.
Method linearity: samples containing high levels of analyte were diluted, in varying proportions, with sera with low levels of analyte. Each dilution was then measured in duplicate.
Method comparison: 60-120 patient samples were analysed on the Monarch, in five-10 runs, over three weeks for each analyte. Results were compared with those obtained by methods and instruments in routine use in the evaluators' laboratories (see tables and 8).

Imprecision
The different components of the imprecision were calculated according to the analysis of variance [2]; the results are shown in table 2. The within-run precision was acceptable in the majority of cases. The overall precision of the electrolyte determinations was excellent (CVs always lower than 1% in the case of sodium, and lower than 1"25 for potassium and chloride) and was good for every other analyte tested, with the exception of cholesterol and urea at low concentrations.  (see table 5 second half, and table 6), the experimental design does not allow the operator to distinguish between imprecision and inaccuracy due to the photometric system or to the dispensing device. The absorbances obtained were therefore compared with the theoretical value for a similar dilution of the solution. So it is possible to evaluate the overall accuracy of the system, but the components of any imprecision cannot be identified. The linearity of response was also calculated using the formulas proposed by Burnett and Martin [3,4]. In each of the three cases, the curvilinear probability was not significant (pre-load mode p 0"5110 r 0"9996, bichromate as sample p 0" 11418 r 2 0"9996, bichromate as reagent p 0"3371 r 0"9998.

Linearity
As shown in table 7, the linearity of the IL methods was very good and almost always higher than that claimed by the manufacturer.

Carry-ouer
The specimen-dependent carry-over was found to be negligible for all the methods studied (table 9). A significant method-to-method carry-over was found only when total protein determination was followed by urate measurement. In this case, the urate value was reduced. This is caused by a falsely elevated reagent blank. Should such a combination occur during calibration, an important increase of all the urate values would be observed. The absorbance increase of an urate reagent blank after a total protein assay (0"046 Abs) was similar to that found by adding one part of biuret reagent to 500 parts of urate reagent and reading the absorbance after an incubation of 5 min at 37 C (0"052 Abs). Therefore, a reagent carryover of about 1/500 can be assumed; this is evident only when particular reagent combinations occur. No carryover was found when an ALT was followed by an LDH assay (a combination that is highly critical in other random access analysers [5]), nor was there any carryover vith any other combination of the analytes tested.

Method comparison
Data and correlation parameters are presented in table 3. The regression line was computed with the nonparametric linear regression model of Passing and Bablock [6]. The correlation coefficient (r) was g6od, or very good, in every case (only for sodium was it somewhat lower, but the range tested was very narrow). The slope of the regression line shows that, in the cases of triglyceride, AST and ALP, there was an important negative proportional bias. This bias could be explained for triglyceride because ofthe differnt kinds ofmethod and calibration material used and for AST and ALP by the different optimization of the methods, and the use (by the Hitachi 737) of a calibrator for enzymatic analyses.

Conclusions
The overall performance of the apparatus seemed to be satisfactory. The instrument is extremely flexible and suitable for development and application of new reagents and research methods, and, moreover, showed a good reliability. It is now used as the routine clinical chemistry analyser in two of the three laboratories who took part in the evaluation.