5.2 Transferability of Clinical Laboratory Data

Transferability of clinical laboratory measurement results is an important problem, both within and between laboratories and hospitals (2,4,5,6). For instance, it cannot be taken for granted that measurement results can be transferred over time in longitudinal statistical studies, in the use of reference values, and in the monitoring of patients. Welldesigned statistical control procedures are required which can assure the specified analytical quality. Careful documentation is required of changes performed in measurement procedures and which affect the analytical results. A very common problem of transferability within a laboratory is related to the reporting of measurement results from a group of instruments measuring the same components. Even with instruments of the same type and from the same manufacturer, and with identical reference materials for calibration it may well be inter-instrument differences that jeopardize the analytical quality goals. However, if the instruments show stable performance within defined limits it would be possible to "tune" the instruments on the basis of correction functions estimated from simultaneous measurements on reference material (eg. selected patient samples) covering the whole measurement interval of interest ('analytical bias assessment programme'). Transferability problems are more commonly realised in connection with communication of laboratory results between laboratories, hospitals and health care centres. In the first place in the care of individual patients referred from another hospital, but also in connection with multi-centre collection of reference values, and in using reference limits and decision limits originating from outside the own laboratory. More recently, the transferability problem has also been recognised in the application of decision rules and computer-based interpretative programmes in places not directly involved in the development of the knowledge-bases.


INTRODUCTION
Transferability of clinical laboratory measurement results is an important problem, both within and between laboratories and hospitals (2,4,5,6). For instance, it cannot be taken for granted that measurement results can be transferred over time in longitudinal statistical studies, in the use of reference values, and in the monitoring of patients. Welldesigned statistical control procedures are required which can assure the specified analytical quality. Careful documentation is required of changes performed in measurement procedures and which affect the analytical results.
A very common problem of transferability within a laboratory is related to the reporting of measurement results from a group of instruments measuring the same components.
Even with instruments of the same type and from the same manufacturer, and with identical reference materials for calibration it may well be inter-instrument differences that jeopardize the analytical quality goals.
However, if the instruments show stable performance within defined limits it would be possible to "tune" the instruments on the basis of correction functions estimated from simultaneous measurements on reference material (eg. selected patient samples) covering the whole measurement interval of interest ('analytical bias assessment programme').
Transferability problems are more commonly realised in connection with communication of laboratory results between laboratories, hospitals and health care centres. In the first place in the care of individual patients referred from another hospital, but also in connection with multi-centre collection of reference values, and in using reference limits and decision limits originating from outside the own laboratory. More recently, the transferability problem has also been recognised in the application of decision rules and computer-based interpretative programmes in places not directly involved in the development of the knowledge-bases.
In order to decrease the analytical inter-laboratory variation, improvement is generally required in various ways (9, e.g. with regard to ---Analyticat tnethodology , to get rid of unspecific methods; Calibration procedures, to reduce analytical between-run variation and drift; Internal analytica! qua& control procedures, to achieve stability in analytical performance within defined limits of allowable analytical errors; Ewtemal quati& assessment (EQA) programmes, to obtain a reliable characterization of the individual laboratory with regard to analytical bias ("trueness problems") (for terminology see Chapter 11); Numerical correction of known systematic deviations from conventional true values (analytical bias).
--All these activities also presuppose further development and use of definitive or reference method technology, and well designed reference tnaterials for calibration, analytical quality assessment and control. With other words, the traceability must increase.
Correction of measurement results has been "taboo" in the clinical laboratory field for a long time, but during recent years there has been an increasing awareness of some basic principles of metrology (5), e.g. that u measured value should be corrected for known systenzatic errors before it is stated.
Since metrology is (1) the field of knowledge, which "includes all aspects both theoretical and practical with reference to measurements, whatever their uncertainty, and in whatever fields of science and technology they occur", clinical chemistry and other clinical laboratory disciplines should be no exceptions.
As mentioned in Chapter 2 (7), one of the main subprojects of the present NORDKEM project on 'Medical need for quality specifications in laboratory medicine' is focused on the pragmatic application of metrological principles in order to improve the transferability of clinical laboratory results. The project has been performed in two steps, an introductory phase (ll), followed by a main phase.
The work in the main phase has been supported by the EU-AIM Openlabs project (A) which is gratefully acknowledged.

HOW TO EXPRESS LABORATORY QUALITY SPECIFICATIONS
The Total Error (TE) of an analytical procedure cp(14) can be expressed as a sum of different types of errors, whichin order to be meaningfulought to be possible to estimate. a well-determinedand thus correctablepart of the total systematic error. Usually referred to as 'analytical bias. Can have a positive or negative value. a non-determined, and thus non-correctable part of the total systematic error. May be due to low specificity of the measurement procedure, e. g. matrix effects. Should be estimated as a maximum error. a temporary increase of the systematic error, not detected/eliminated by the quality control system of the laboratory. This increase in error is expressed as a multiple of the inherent random error of the measurement procedure (sAo). Can have a positive or negative value and be of the order (0; ASEdetect >* lower detection limit of QC procedure. a temporary increase in random error not detected/eliminated by the quality control system of the laboratory. The change in random error is expressed as multiples of the inherent random error (sA0), and is in the order ( 1;AREdetect). lower detection limit of QC procedure. a multiplier related to the portion of the distribution exceeding the quality requirement, often set as 1.65 to fix the maximum defect rate to 5 %.
A laboratory 'Analytical Quality Specification' (AQSpec) should be expressed as a 'Total Allowable Error' (TE,) at a defined concentration. The total error TE should always be kept smaller than TE,.
In the process of formulating AQSpecs (7) it is essential to compare the clinical goals with the characteristics of the analytical measurement procedures and the performance of the quality control procedures, and judge what is realistic for the laboratory to set up as its AQSpecs. The quality specifications ought to have probability limitse. g. that 95 % of the produced results should fall within these limits. In a 'Quality policy' document (8), it is also insistent to declare the 'traceability' of the values of the laboratory (for definition see Chapter 10).

OF THE STUDY
In the introductory phase of the project linear regression techniques were used for assessment of analytical stability and bias over a wider concentration range for selected analytical methods. The results of analyses of S--Creatinine and S--Urate methods in 17 laboratories in the Swedish Uppsala-Orebro health care region have been published (11).
The analytical procedures of the various laboratories were described by their  Using S--Calcium as an example, Fig. 3 illustrates typical experiences from this extended study. The results from the laboratories can be well described by ordinary linear least: squares regression analysis, considering the assigned x-values as error-free compared to the measured y-values (c $ Tables 1 and 3 Table 2. The experiences are similar in this 'main phase study': some laboratories demonstrate 'good trueness' and 'analytical stability', some have bias problems and a few have both bias problems and less good analytical stability.   procedures of the external quality assessment (EQA) schemes should be reviewed to better accommodate aspects of transferability and descriptions of laboratories' performance". We propose that "multi-level analytical bias assessment programmes" of the design described in the present paper should complernent/replace the conventional type of passive EQA activities. The proposed statistical approach and the graphical presentation provides a means to assess "analytical stability" (definition see Chapter 11) and "analytical bias" of individual laboratories in relation to specified analytical quality requirements (AQSpecs).
In addition we are also advocating for a more common use of numerical correction procedures in clinical chemistry in order to increase the transferability of measurement results.
An often heard argument against such corrections is that matrix effects may lead to wrong conclusions and that mathematical corrections may introduce ambiguities (cf. 3,12). As emphasized in the beginning of this paper, this type of correction procedure should only be applied to correct for well-known and well-determined analytical bias ("stable bias"); this presupposes that matrix effects on the measurements on "bias assessment material" can be disregarded. The non-determined part of the total systematic error of patient results, e.g. due to matrix effects should not be corrected for, but estimated and reported as a maximum error limit. With a relatively large maximum error estimate of the latter type of systematic error component, it may seem less important to correct for "stable bias".
It goes without saying that non-specific analytical methods should be replaced by more specific methods, if available and affordable. But even with highly specific methods there may be a need for estimation (and correction) for analytical bias due to other factors; this is according to basic metrological principles.
It is also evident that a correction may be more effective for methods with a small ratio of analytical within-laboratory to between-laboratory variation.
A n "Analytical Quality Assurance" (AQA) programme (see Chapter 1 l), combining analytical bias assessment and active correction, can be implemented in different ways, e.g.
1) as an external programme providing the applicable correction functions when there is a need to transfer laboratory results "over time and space"; or 2) as an external assessment followed by internal correction of measurement result before reporting.
It should be noted that "analytical bias assessment and correction programmes" should be run strictly separate from EQA and Proficiency Testing programmes organized by various regulatory bodies.
Supervised correction procedures couldas indicated earlier in this paperbe of great value especially in the following situations: