Multicentre evaluation of the Boehringer Mannheim/Hitachi 911 Analysis System

The analytical performance and practicability of the Boehringer Mannheim (BM)/Hitachi 911 analysis system have been assessed in a multicentre evaluation, which involved six laboratories from European countries. Analytes commonly used in classical clinical chemistry were tested in a core programme, which mainly followed the ECCLS guidelines. In addition, a satellite programme covered other analytes, such as proteins, drugs and urine analytes. In total, the study comprised more than 100 000 data items collected over a three-month period. The evaluation was supported with ‘Computer Aided Evaluation’ (CAEv) and telecommunications. Acceptance criteria for the results were established at the beginning of the study. Nearly all of the analytes met the imprecision limits: within-run imprecision (as CVs) was 2% for enzyme and substrate assays, 1% for ISE methods and 5% for immunoassays; between-day imprecision was 3l% for enzyme and substrate assays, 2% for ISE methods and 10% for immunoassays. No relevant drift effects (systematic deviation ≥ 3%) were observed over eight hours. The methods were linear over a wide range. Sample-related and reagent-dependent carry-over can be reduced to a negligible amount by integration of a softwarecontrolled wash-step. Endogenous interferences were found for creatinine (Jaffé method) and uric acid assays (caused by bilirubin), for creatine kinase, creatine kinase MB isoform and γ-glutamyltransferase (caused by haemoglobin), and for immunoglobulin A (caused by lipaemia) Accuracy was checked by an interlaboratory survey, recovery studies in control materials and method comparison studies. The survey showed that, with the exception of cholesterol and iron in two laboratories, the recovery of analytes did not deviate by more than 5%. Sixty-six of the 77 method comparisons performed met the acceptance criteria. The deviations of the remaining 11 results could be explained by differences in either calibration, application or by the use of different methods. Practicability was assessed using a questionnaire which covered all of the important aspects of an analysis system in the clinical laboratory. Twelve groups of attributes out of 14 were rater higher for the BM/Hitachi 911 than for the present situation in the laboratories concerned. Especially high scores were given for the versatility group. The acceptance criteria for the analytical performance of the BM/Hitachi 911 analysis system were fulfilled in all laboratory segments with few exceptions. The practicability exceeded the requirements in most of the attributes. The results of the study confirmed the usefulness of the system as a consolidated workstation in small- to medium-sized clinical laboratories and in STAT laboratories, or as an instrument for special analytes like proteins and drugs, or for urinalysis in large laboratories.


Introduction
The Boehringer Mannheim (BM)/Hitachi 911 analysis system is the most recent medium-sized analysis system to be introduced to the market by Boehringer Mannheim GmbH. In addition to the well-accepted analytical performance and reliability of previous BM/Hitachi analysis systems, the new instrument has features making it attractive to different sections of clinical laboratories.
These are in routine and emergency analysis (STAT), homogeneous immunoassays for the determination of proteins and drugs and urinalysis. Therefore the BM/Hitachi 911 has to be extremely flexible. This is achieved by incorporation of features such as automatic recognition of four different barcodes for sample identification, use of up to four reagents per test, variable reaction time, random analysis of serum and urine specimens using the same calibration curve, fully automated predilution of specimens and automatic calibration.
The versatility of the new instrument required the design of a comprehensive evaluation. Six European laboratories participated in the multicentre evaluation. Analytes of classical clinical chemistry were tested in a core programme, which mainly tbllowed the ECCLS guidelines [1]. In addition, the behaviour of the system in different laboratory sections was tested in a satellite programme.
The evaluators ran a much less extensive satellite evaluation in order to maintain an acceptable cost/benefit ratio. In total, the study included more than 100 000 data. Processing and analysis of the large data volumes was managed with a program system called 'Computer Aided Evaluation' (CAEv) and telecommunications.
CAEv integrates the definition of study protocols, performance of experiments in the laboratory, online/ 0142-0453/93 $i0.00 @) 1993 Taylor & F is Ltd. Z. Zaman et al. Multicentre evaluation of the Boehringer Mannheim/Hitachi 911 Analysis System offiine data transmission and the immediate assessment and evaluation of the results [2]. The program runs on a standard PC under MS-DOS. It had previously been successfully applied to the multicentre evaluation of the BM/Hitachi 747 analysis system [3].
Telecommunications were used for the first time in an international multicentre evaluation. Installation of the necessary facilities in the participating laboratories allowed rapid and simple transfer of data to the centre coordinating the study. The assessment of practicability was a further goal of the evaluation of the BM/Hitachi 911. The evaluators answered 200 questions each about the system. All results of the multicentre evaluation are presented in this paper.

Description of the instrument
The BM/Hitachi 911 is a medium-sized selective access analyser with a capacity for 35 different tests, including three ion selective electrode (ISE) methods for sodium, potassium and chloride. The instrument specifications are listed in table 1. The throughput is 360 photometric tests/h; this is reduced by automatic sample predilution or additional wash steps which may be needed to eliminate reagent carry-over. The pipetting cycle for photometric tests, for the automatic predilution and for the wash steps, is 10 and for the ISE methods 20 s. Various measurements and calibration procedures that can be applied are: (1) Endpoint measurements with/without sample blank within 16 min. (2) Kinetic determinations with sample or substrate start (49 measuring points within 16 min).
--.Reference electrode" liquid membrane ---Dilution ratio" 1" 31 --Incubation temperature: 37 0"IC The manufacturer provides an application disk to load and store the application settings for the barcoded system reagents. These can be inserted in any free position of the reagent disk. For the use of non-barcoded reagents, the operator must define the appropriate application settings.

Calibration
During the familiarization period, a fixed-factor was determined for the enzyme tests in three independent calibration runs per day on three consecutive days.
The same lot of the calibrator for automated systems (Boehringer Mannheim GmbH) was used for this purpose.
The mean of the nine calibration runs was then taken as the factor, provided that the range of all results did not exceed 3 of the mean.
During experiments for the between-day imprecision and method comparison (21 working days) autocalibration of the analytical system was tested. The autocalibration is triggered by an analyte-dependent calibration interval. In order to avoid any additional effects on the results, a 'start-up' calibration was activated before the remaining experiments. Detailed information about the calibrators employed is shown in the appendix (table 10).

Evaluation protocol
The versatility of the BM/Hitachi 911 analysis system was tested in a core programme and a satellite programme. The core program had been used for evaluations of the BM/Hitachi 704, 717 and 747 analysis systems [-5].
The 13 analytes tested in the core programme were pancreatic 0c-amylase, aspartate aminotransferase, creatine kinase, 7-glutamyltransferase, calcium, cholesterol, creatinine, iron, total protein, uric acid, sodium, potassium, and chloride. This programme included a familiarization period, an initial trial and a main trial. The protocol of the main trial is shown in table 3. The main trial was split between two groups, each consisting of three laboratories, which tested the same set of analytes. For the studies of linearity, drift and sample-related carryover, the different methods were divided between the six evaluators.
The satellite programme covered other analytes such as proteins (C-reactive protein, ferritin, immunoglobulin A, transferrin), drugs (phenobarbital, phenytoin, theophylline, digoxin) and urinalysis (albumin, /-N-a.cetylglucosaminidase, creatinine, sodium, potassium, chloride). In addition, creatinine kinase MB isoform and fruc- On three different days, each day one run with 21 aliquots --:Fhree control materials with different concentrations of the analyte.
--.One human serum pool at the decision level. Between-day Three control materials with different concent'rations of the analyte over 21 days (evaluation and comparison instrument).

Drift
Two control sera and the calibrator were analysed every 30 min over 8 h to test eight methods (aspartate aminotransferase, creatine kinase, 7-glutamyltransferase, calcium, iron, creatinine, total protein and uric acid).
-At zero hour, triplicate measurements were performed and the median was taken as the base value. The percentage recovery of the base value was taken for assessing drift effects.
Analytical range limits [7] --A high level sample was diluted with a low level sample to obtain a series of eleven concentrations with two being the baseline samples and nine intermediate concentrations.
--Triplicate measurements on the 11 concentration levels were performed and the median for each level was calculated.
--The regression line (Passing/Bablok regression) was calculated using a range covering five concentration levels which was assumed to be linear.
--The target values for all concentration levels were calculated from the regression lines.

Carry-over
Sample-related Model of Broughton [ 11]: ---Measurements of five aliquots of a high-concentration sample (hi"" "hs) are followed by --Measurements of five aliquots of a low-concentration sample (11" "15). The experiment is repeated 10 times. If a carry-over effect exists, 11 is the most influenced, 15 the least influenced aliquot.
Reagent-dependent: One control material was used.
Assay A influences assay B. * For details, see reference [3].
(continued) --.Carry-over caused by the cuvettes In a first step, reagents for assay A were requested 21 times. Just after the dispensing of the reagents for the 21st aliquot, the analyser was stopped. In a second step, reagents for assay B were requested 42 times. The first 21 determinations were performed in the cuvettes which previously had contained the reagents for assay A. These determinations would show carry-over effects whereas the last 21 determinations would be uninfluenced. The difference of the medians of both series was the carry-over.
--.Carry-over caused by reagent probes and stirrers Assay B was carried out 21 times. In the second step test A and B were alternately performed 21 times. The carry-over was the difference between the medians of both series.

Interference
Protocol of Glick [8] A specimen with concentrations at the decision level was spiked with the interfering substance and 10 serial dilutions were prepared with the same baseline specimen. The different analytes were measured in triplicate. The percentage recovery of the baseline value for each concentration level was calculated.

Accuracy Calibration
The calibrators of BM/Hitachi 911 and of the comparison instrument were both run on each instrument.
Quality control in three control materials --Median, calculated from the second of duplicate measurements over 21 days.
Interlaboratory survey --One control material with concentrations not known to the evaluators.
--Median, calculated from the second of duplicate measurements over 10 days.
Method comparison in fresh human specimens --10 to 15 specimens were analysed each day for 10 days on the BM/Hitachi 911 and on the comparison instruments. The total number of specimens covered the entire analytical range.
--Comparison of the methods by calculation of the Passing/Bablok regression line [10-].
In the satellite programme within-run imprecision was determined using one or two control materials and one human pool. Between-day imprecision was studied over 10 days with one or two control materials. Drift, linearity, carry-over and interference studies were carried out only for selected analytes.
tosamine assays were run in this part of the evaluation study. The results of these analytes are presented together with those of the core programme.
The protocol included quality specifications which were agreed at the evaorators' first meeting: these are described later in this paper.

Assessmenl of praclicabilily
Practicability was assessed with the aid of a recently published questionnaire [6] comprising about 200 questions which covered all important aspects of an analysis system in the clinical laboratory. The questions were summarized into 14 groups, as shown in table 4. They were related to the installation of the analyser, organization of work, quality assurance and miscellaneous characteristics. A first version of the questionnaire had already been used for the assessment of practicability of BM/Hitachi 747 [3].
Grading was in comparison with the evaluators' present laboratory situation. The assessment was based on a scale from 0 to 10: a score of 0 meant unimportant, useless or poor, and a score of 10 absolutely necessary or excellent. A score of 5 could be interpreted as being acceptable or comparable with the present laboratory situation. The grading was divided into three classes. Scores of up to 3"3 meant 'did not meet the requirements', scores from 3"4 to 6"7 'meets the requirements', and scores from 6"8 to l0 'exceeded the requirements'.

Imprecision
Acceptance criteria for imprecision were based on statistical error propagation [3]. For within-run imprecision in participating laboratories the median of the CVs should not exceed 2 for classical clinical chemistry analytes (enzymes and substrates) in serum (only in the tables is a differentiation made between serum and plasma) and urine at all concentrations tested. The accepted CV was reduced to 1 for the determination of electrolytes by ISE. Taking into consideration the problems associated with immunoassays, such as nonlinear calibration or analytical sensitivity, it was agreed that CVs of less than 5 would be acceptable for these assays.
Within-run imprecision data based on results from all the control sera from all laboratories are shown in figures 1-3, and data for individual control sera are given in the appendix (table 11). The medians of all analytes met the acceptance criteria, except for phenytoin and digoxin in the low level control. In individual control sera, results exceeding the acceptance limits were obtained for  Figure 3. Within-run imprecision for analytes in urine, based on control material data from all laboratories.
cholesterol, iron, sodium, potassium, chloride and for albumin in urine.
All analytes which had not met the quality specifications in control materials showed acceptable CVs in human materials (table 5). Only the CV of creatine kinase MB isotbrm exceeded the acceptance criteria.
Between-day imprecision was defined as acceptable if median CVs for electrolytes were _< 2, for homogeneous immunoassays _< 10 and for the remaining analytes investigated in serum and urine _< 3.  [7]. In the upper range, a method was defined to be linear if this difference was less than 5. The lower range was judged on the basis of such pragmatic considerations as absolute differences between measured and target values, and the diagnostic relevance of these differences. In the case of multi-point calibration, a method could be called linear if a change in the target concentration led to a proportional change in the measured concentration.  (table 6). The lower limit tested for creatinine and calcium was found to be 30 gmol/1, and 0"3 mmol/1, respectively. In urine, the required high linearity ranges were reached for all analytes. Linearity in the low range was analysed for albumin and sodium only. The albumin method was found to be nonlinear below 25 mg/1 and the sodium determination below 10 mmol/1.  Carry-over Carry-over effects were assessed on the basis of the observed change in recovery of an analyte. Instead of adapting an individual deviation for each analyte it was decided to use the within-run imprecision system performance. This was defined as a change of less than twice the standard deviation being acceptable. Sample-related carry-over was only tested for analytes with a large physiological range and between urine and serum specimens. For the potassium assay, a slight carry-over effect (0"17 mmol/1) was observed from urine to serum with a concentration ratio of77:1 (table 7). This was rated as being of no clinical relevance.
Previous experience with BM/Hitachi systems had revealed reagent dependent carry-over caused by the cuvettes for the combination triglycerides/lipase. This effect was avoided by activating a special wash solution step within the normal cleaning procedure for the cuvettes on BM/Hitachi 91 1. Carry-over caused by the reagent probes and the stirrers was tested for the combinations triglycerides/lipase and aspartate aminotransferase/lactate dehydrogenase. This resulted in an elevation of lipase activities by about 400 U/1 and of lactate dehydrogenase activities by about 30 U/1. Activation of a software-controlled wash-step for these combinations reduced the carry-over effects to a negligible amount < 10 U/1).

Interferences
According to Glick et al. [8], a method is resistant to interferences if the deviation between the baseline value and the measured value is less than 10. Only methods not fulfilling this acceptance criterion are shown in the interferograms ofbilirubinaemia, lipaemia and haemolysis (figures 7-9).
Haemolysis caused interference with four of the methods tested ( figure 8). Activities ofcreatine kinase and creatine kinase MB isoform were increased while that of 7glutamyltransferase decreased. The protein assay gave positive bias with haemolysis at high haemoglobin concentrations > 4 g/l). Two of the 16 methods evaluated were susceptible to interference by lipaemia ( figure 9). While immunoglobulin A concentration was increased, an opposite effect was observed on the activities of aspartate aminotransferase. A slight increase was also observed in concentrations of C-reactive protein (+ 12) at high concentrations of triglycerides > 1500 mmol/1).  (see table 11, appendix). In addition, the recovery of pancreatic amylase in one control (106"6%) exceeded the predefined limit.
In the quality control study, recovery experiments with the CEDIA > assays tbr therapeutic drug monitoring, and with urine as sample material, were also included. The recovery of phenobarbital and phenytoin was within the acceptance criterion of 10 at the target value, whereas that of theophylline was 88 in one control serum with a concentration below the therapeutic range, in one of three laboratories on both BM/Hitachi 911 and on the comparison instrument. This effect was not confirmed in a satellite study in which a recovery of 98"4 was obtained using the same control material.
In control urines, all analytes met the acceptance criteria, except creatinine, for which, in one control urine, a recovery of 106"9o was obtained on BM/Hitachi 911 and ot" 110"4 on the comparison instrument. In the other two control urines, the recovery on BM/Hitachi 911 was between 90 to 95.

Method comparison
In total, 74 method comparison studies were performed using fresh human sera or urines; the resulting regression equations are shown in the appendix (table 12). All method comparisons were presented graphically and assessed by the evaluators.
For the determination of enzymes and substrates in serum and urine, agreement with the comparison method was rated as being sufficient if the slope of the regression equation did not deviate more than 53/o from unity and the intercept was less than 5 of the diagnostically important decision levels. The acceptance criteria were increased to -+-10 slope deviation tbr the homogeneous immunoassays. Depending on the analyte, either the detection limit or a deviation of 10 from the relrence value were tolerated as intercept for these assays.
Assessment of electrolyte concentrations in serum is especially critical. Due to the narrow physiological range of these analytes, a relatively large confidence interval for the regression line was obtained. Therefore, method comparisons were judged by the concentration range in which the difference between the methods was less than 5%. The comparisons were accepted if less than 5% deviations were obtained within the following concentration ranges: (1) Sodium 120-170 mmol/1 (2) Potassium 2-10 mmol/1 (3) Chloride 80-130 mmol/1.
As examples, six method comparisons comprising three constituents of serum (creatine kinase, iron, potassium), one of urine (sodium) and two homogeneous immunoassays (one turbidimetric protein assay and one CEDIA " assay) are shown in figure 11.
Method comparisons not fialfilling the acceptance criteria are listed in table 8. For classical clinical chemistry analytes in serum, seven comparisons out of 33 did not meet the acceptance criteria. Two out of 12 method comparisons in urine were unacceptable. Comparisons of homogeneous immunoassays included seven protein assays and 13 drug assays. All method comparisons fulfilled the acceptance criteria except one tbr digoxin.
One of the 12 method comparisons for electrolytes in serum (sodium) did not meet the acceptance criteria. The range in which the results obtained on BM/Hitachi 911 and on the comparison instrument deviated less than 5 was 127-194 mmol/1 instead of the required concentration range of 120-170 retool/. Reliability Reliability during the evaluation phase was rated with the aid of a logbook in which any breakdown, deiict, malfunction or incident of the analysis system was recorded. Some technical problems occurred during the evaluation.
The ISE unit at most evaluation sites had problems, ti0r example noise alarm, air bubbles and crystallization of potassium chloride at the sipper probe of the dilution vessel. These resulted in an increased imprecision. A major modification of the ISE unit at the beginning of the main trial, which caused some delay in several laboratories, improved the performance.
No other systematic malfunctions were observed. Single malfunction episodes occurred with the sample probe and with the stirrer resulting in both situations in a wrong home position and a stoppage of the analyser. Software malfunctions reported during the multicentre evaluation were corrected in two new system software releases.

Assessment of practicability
The practicability of BM/Hitachi 911 was judged in comparison with the present situation in the individual laboratories. The median of all laboratories was calculated from the mean of all scores obtained for each group of attributes. These results are shown in figure 12. The following regression equations were obtained: CK (N 150), y l'OOx + 0"0; FE (N 150)y l'OOx O'l& K (N 150) y l'OOx 0"07; NA in urine (N 100)y 0"98x 1"35; Fr (v oo)y 0"93x + 0"33; .DIG (N 100)y 1.03x 0.14.
2oo More detailed information on the distribution of scores in relation to the main topics is shown in figure 13. Higher scores (8 to 10) were given more frequently for BM/Hitachi 911 than for the existing laboratory situation. For BM/Hitachi 911, out of 194 attributes 12 were rated with a score of 0 to (each by only one evaluation site) and 93 attributes with a score of 9 or 10 (each by one or several laboratories).

Discussion
The versatility of the BM/Hitachi 911 analysis system was tested in the multicentre study in different laboratory sections--routine, STAT, homogeneous immunoassays and urinalysis. The results were assessed on the basis of quality specifications tbr the various performance characteristics previously agreed upon by the group of evaluators. A good pertbrmance was found for most of the analytes in all laboratory sections using different sample materials of serum, plasma and urine. Although some analytes did not fulfill one or the other acceptance criterion, none could be rated unacceptable. The results outside the acceptance criteria are discussed below with the exception of the enzymatic methods for determination of sodium, potassium and chloride. The performance of these methods was unacceptable on the BM/Hitachi 911. New standardization procedures for these tests are under development.

Within-run imprecision
The results for control and human materials (see table 5) show. that overall the instrument had given a good performance. The individual CVs of a few analytes were slightly higher than the acceptance limits. These were found mainly in one laboratory and in one control material. Variable results were obtained for urinary albumin in a control material with a concentration below the decision level (CV vMues of 2"1I, 2"2: and 9"2% at the maximum CVs ot" 3"1 and 3"3 may still be seen as borderfline cases. The maximum CVs of 3"73/0 for sodium and 2"3 for chloride reflected the condition of the ISE unit which had to be modified during the evaluation. In spite of this modification, an improvement was not observed in all laboratories. In one laboratory, the CEDIA :"> Phenytoin and Digoxin assays, and the urinary albumin assay in the low level control, also showed CV values slightly higher than 10.

Quality speccalions
Comparing the between-day imprecision results of the BM/Hitachi 911 with the proposed quality specifications tbr the imprecision of analytical systems tbr clinical chemistry [12,13], it was concluded that the BM/Hitachi 911 analysis system achieved these specifications for nearly all analytes evaluated in the study (table 9). The proposed quality specifications were not met with the calcium assay (interim specification 1"5, BM/Uitachi 911 1"6%), creatinine assay (specification 2"2, BM/Hitachi 911 2"3%), sodium assay (interim specification 0"7%, BM/ Hitachi 911 1"4%) and the chloride assay (interim specification 1"0%, BM/Hitachi 911 1"7%). However, since the imprecision of the calcium and the creatinine assay were only 0"1 higher than the specifications, the results obtained can be judged acceptable. Owing to the low biological variation of sodium and chloride, the imprecision specifications were set very low: 0"3 for sodium and 0"7% for chloride. Since no available technology is capable of producing such a low imprecision at a reasonable cost, interim specifications with 0"7 and 1"0 were proposed. Even these values are unlikely to be easily achieved with the existing technology.

Analylical range limits
The acceptance criteria tbr linearity of the measuring range were fulfilled for all analytes, except for urinary albumin at concentrations below 25 mg/1. This nonlinearity resulted in an underestimation of about 3 mg/1 albumin at the decision level of 20 mg/1. This is within the range found during the multicentre evaluation tbr the albumin test [14]. In view of the large biological variation of this analyte, this deviation was considered acceptable. During the evaluation, single-point calibration had been used for the C-reactive protein assay. This mode of Increased total protein values were measured at haemoglobin concentrations above 4 g/1. This can be readily explained by the ttct that haemoglobin is a protein. The activity of/-glutamyltransferase was decreased in haemolytic samples. As a consequence, the 7-glutamyltransferase test will be modified by the manufacturer in order to reduce this interference. Increased activities of creatine kinase, and especially creatine kinase MB isoform, were found in the presence of haemoglobin. This is caused by the release of adenylate kinase from erythrocytes. Again, the manufacturer intends to modify the application for the creatine kinase MB isoform assay, with the aim of reducing the susceptibility of the method to interference by haemoglobin.
Increased immunoglobulin A concentrations had been measured in lipaemic samples. Theretbre, a modified immunoglobulin A assay is presently being developed to overcome this problem. The effect of turbidity on aspartate aminotransferase activities has previously been observed during other BM/Hitachi system evaluations I-3-1. It can be explained by the fact that the fresh reagents tbr this assay have high initial absorbance and that the addition of a turbid specimen would cause the total absorbance to exceed the photometric measurement range of 3"3 absorbance units.

Avcuracy
During the interlaboratory survey and the quality control study a recovery ofless than 95 was found for cholesterol.
The assigned value of this analyte was confirmed by the reference method of isotope dilution/mass spectrometry [16]. Therefore, the effect could not be attributed to a wrongly assigned value. In two of the three method comparison studies a good agreement was found, whereas in the third increased cholesterol values were obtained on BM/Hitachi 911. These contradictory results could not be explained. The effect was not observed in a satellite study performed in one of the evaluation centres.
Iron values below the acceptance criteria were found in three of the four control sera used in the interlaboratory survey and the quality control study. The effect was confirmed in the method comparison study in two of the three laboratories (table 8). In the third laboratory, in which the comparison instrument was a centrifugal analyser, a slope of 1-00 and an intercept of-0"15 gmol/1 iron were obtained in the regression analysis. The deviant results were probably caused by wrongly low values being assigned to the calibrators of the two comparison instruments.
If the results of the accuracy experiments are compared to the quality specifications proposed by Fraser el al. [12,13], all assays performed on BM/Hitachi 911 met the requirements with the exception of cholesterol and transferrin.
Eleven out of 74 method comparisons gave slope intervals and intercepts that were outside the acceptance limits 2O3 (table 8). Differences in the 7-glutamyltransferase assay were caused by an incorrect temperature conversion factor on the comparison instrument in one laboratory. The method comparison tbr the calcium assay yielded a slope of 1"06. However, the differences between the values of individual samples (from -0"13 to 0"18 mmol/1) were acceptable. In one of three laboratories a slope of 1"11 was obtained in the method comparison for creatinine. In spite of this, agreement was present at the decision level. However, with this high slope, results of samples tiom dialysis patients are not correctly interpreted. In a satellite study in one of the laboratories, the deviation was confirmed. The manufacturer will change the application in order to correct the discrepancy. The lack of agreement in the tiustosamine assay could be explained by the fact that the comparison method was performed with a longer incubation time. The evaluators accepted this deviation.
Albumin in urine showed discrepant results in two of the three laboratories where the assay was compared with nonturbidimetric methods. One out of the four digoxin method comparisons was outside the accepted range. The experiment had been pertbrmed in a laboratory where many samples were derived from haemodialysis patients. These samples might contain digoxin-like immunoreactive thctors which are known to interfere digoxin assays [ 17].
The results ot'the C-reactive protein method comparisons are not reported because single-point calibration used in the experiments had much narrower linear range than was originally thought. ThereIbre, the application was later changed to multi-point calibration.

Praclicabiliy
The outstanding tiature of the BM/Hitachi 911 analysis system is that it can be used as a consolidated workstation tbr assays from diverse segments of a clinical laboratory (clinical chemistry, proteins, therapeutic drug monitoring). All tests necessary tbr a complete patient profile can be pertbrmed using one primary tube without sample splitting and additional manual workload. This simplifies the distribution ofsamples and thus the organization of the laboratory. Hazards ot'sample mix-up as well as cost for sample splitting are avoided. Moreover, analyses requiring a predilution step (tbr example, immunoglobulins) can also be handled easily without the need for high sample volume. The analyser pertbrms the predilution itself.
Urine samples can be run randomly in series with serum or plasma specimens on BM/Hitachi 911. No relevant carry-over effects between the different sample materials were observed during the multicentre evaluation. An excellent analytical pertbrmance was obtained for urine samples, showing that the instrument is well suited for urinalysis.
Suitability of BM/Hitachi 911 for routine analyses depends, of course, on the size of the laboratory. One of the evaluators considered the instruments well suited the routine requirements of his laboratory. The STAT facilities were rated positively in all evaluation centres.