Methods and approaches to determine "optimal" quality for clinical chemical data

Clinically based quality goals; a NORDKEM project

The three main aspects of analytical quality are 'goals for analytical quality', 'creation of analytical quality', and 'control of analytical quality'. In the NORDKEM-project 'medical need for quality specifications within laboratory medicine' the aspects of analytical quality are combined. The aims is to make an appraisal of the different approaches to goal setting, and to develop a practical procedure for assessing analytical quality requirements. The goals are used to define which demands should be met by the process of establishing and maintaining the quality, and for design of internal as well as external quality assurance procedures. The project is an umbrella project with three main subprojects and several satellite projects. In this presentation a model for evaluation of influence of analytical bias and imprecision on the outcome from a diagnostic classification based on bimodal distribution is described.

Series analyses and quality specifications required for monitoring over time

General specifications of analytical goals for biochemical monitoring have been proposed based on biological within-subject variation. As a complement to this strategy for general global quality requirements there is also a need for other methods and approaches to set more optimal requirements in specific monitoring situations, i.e. application of systems and sensitivity analysis using (i) rules for propagation of errors (uncertainty) in simple algebraic/statistical transformations of laboratory results; (ii) biochemical/pathophysiological simulation models; and (iii) formalized descriptions of clinical classification and decision support processes. The possible gain in medical outcome by improving analytical and preanalytical quality should then be related to the often more important aspects of selection and combination of "tests", period and frequency of sampling/measurement, and the use of correct conceptual and computational models for transformation of data.

The influence of analytical bias on diagnostic misclassifications

Quality specifications for analytical imprecision and bias based on the state of the art; 'biology' and 'analysis of clinical situations' have been proposed by several scientists. Most interesting is the assessment of 'diagnostic misclassifications' based on direct evaluation of the consequences of analytical bias on the percentage of false positives and false negatives from a clinical decision situation, or based on the percentage of healthy individuals outside each reference limit when common reference intervals are used. With use of graphical or computer simulations assuming increasing (positive or negative) analytical bias, the expected percentage of misclassifications can be estimated- and, for the error for which the outcome (the fraction of misclassifications) is considered unacceptable, the maximum allowable analytical bias can be defined. An overview is given of previous proposals for specification of allowable analytical bias, and new examples are presented: (i) for S-transferrin. an analytical bias of +10% will increase the percentage of healthy individuals with measured concentration values above the upper reference limit from 2.5 to 10% (ii) the percentage of healthy men with concentration values for S-cholesterol above 6.2 mmol/l (240 mg/dl) will vary between 25 and 85% for analytical bias from - 1.0 to +1.0 mmol/l (+/- 16%): (iii) for glycated haemoglobin, two examples are given which illustrate the effect of analytical bias on the risk of retinopathy and so-called 'microalbuminuria' for measured values identical to the target 7.5% and 10.1% glycated haemoglobin, respectively. It is concluded that analytical bias may have significant impact on diagnostic performance, better standardization is needed, and quality specifications for allowable analytical bias should be based on medical usefulness criteria or, if such data are not available, on biological criteria.

Analytical quality goals and assessment to ensure transferability of laboratory results

About 3 years ago NORDKEM started a project entitled 'Medical Need for Quality Specifications in Clinical Laboratories'. This communication describes the general approach of this project and a subproject oriented towards problems of transferability of clinical laboratory data. Attempts are made to estimate clinical goals for analytical measurements--either used alone or in combinations with other measurements or observations--in defined clinical situations. Various methodologies are used for the assessment of the clinical goals, starting, e.g., from clinical situations or biological reference data. The analytical quality specifications are expressed as total allowable errors. It is necessary to complement specifications of 'clinical goals' with data describing characteristics of the measurement procedure, preanalytical errors and internal and external quality assurance procedures in order to establish the laboratory quality specification guaranteed by the laboratory. Problems of transferability of clinical laboratory results occur both within a laboratory/hospital and between laboratories/hospitals/health care units, problems that sometimes make communication of laboratory results difficult or even meaningless. In order to study the transferability of clinical laboratory data within a health care region, so called 'transformation functions' could be used to characterize each laboratory in a strict quality assessment/proficiency testing procedure on the basis of analytical quality specifications. Under certain conditions, e.g. for analytical procedures with high specificity and documented stable analytical performance, numerical correction of analytical bias could be performed to decrease interlaboratory variation to a level specified by medical needs.

Transferability of clinical laboratory data within a health care region

Analytical data for S-Creatinine and S-Urate are presented from seventeen laboratories in the Swedish Uppsala-Orebro regional quality assessment program. The bias and imprecision as well as the instability of the measurement procedures in the participating laboratories were estimated over three 14-week periods. Bias was estimated by a linear least squares fit of the difference between measured and assigned values vs. assigned values, and expressed in absolute and relative terms. Instability of the measurement procedures was estimated by comparing slope and intercept of regression lines of measured vs. assigned values from three fourteen week periods. According to our experiences we recommend regression analysis to describe the performance of the analytical methods of a laboratory over time. The results show that most laboratories fell within the limits of +/- 15% bias for S-Creatinine above 100 mumol l-1 and +/- 17% for S-Urate at concentrations above 250 mumol l-1. Various steps to reduce the inter-laboratory variability are suggested, including numerical correction of individual laboratory results using correction functions. In a few laboratories, instability was too high to allow for numerical corrections of analytical results.

Pancreatic iso-amylase in serum as a diagnostic test in different clinical situations. A simulation study

The potential use of a pancreatic iso-amylase test has been studied by computer simulation. This simulation was performed to assess the quality requirements on the test in different clinical situations. It was shown that, in most situations studied, an optimal discriminating level and an optimal analytical quality (imprecision and bias) could be established. In one situation studied in more detail, it was found that the influence of pre-analytical variation, duplicate determinations, reclassification of borderline cases and size of reference sample groups were of minor importance. The weighting ratio of false negatives: false positives was found to be critical for the results. It is concluded that a simulation study of this type can be recommended prior to entering the phases of analytical refinement and clinical testing on patients.

Studies on the required analytical quality of TSH measurements in screening for congenital hypothyroidism

In the present study analytical reliability has been evaluated from the point of view of medical practice. The data were obtained from a screening program for congenital hypothyroidism in newborns using measurements of TSH. The aim of the investigation has been to evaluate the optimal analytical quality of the TSH determinations in the classification of children into two groups, with and without hypothyroidism. Evaluation of analytical quality was attempted at four different levels: (i) the numbers of misclassifications at a certain fixed value of the cutoff point, (ii) the numbers of false positives when the percentage of false negatives is fixed, (iii) the total costs of the screening program, and (iv) the social costs. When the evaluations are performed at level (i) and (ii) the results indicate a need for a high analytical quality. At level (iii) a lower analytical quality is acceptable from consideration of costs only. When social costs (iv) are considered, the acceptable analytical quality will depend on the weights given to the different factors. Finally the results point to a close relationship between the imprecision and the cutoff point chosen.

Diagnosis, size estimation and prediction of acute myocardial infarction from S-myoglobin observations. A system analysis to assess the influence of various sources of variability

Three different ways of using S-myoglobin observations for early diagnosis of acute myocardial infarction have been investigated by computer simulation techniques, viz. classification based on (i) single determination in relation to a decision limit, (ii) the peak serum concentration value, (iii) estimated or predicted value of infarct size from serial serum concentration determinations. The results of the in numero experiments indicate that it is possible to define optimal conditions with regard to time, period and frequency of observation, as well as to assess requirements on pre-analytical/analytical variation. The optimal time for a single observation should be about 10-12 h after onset of symptoms. The influence of pre-analytical/analytical variation is not very critical in this connection and the quality requirements are achievable in the clinical chemical laboratory today (total CV about 0.10). The peak serum value has good diagnostic power, but does not provide a good index of infarct size, no matter how good the analytical quality may be. It should be possible to predict infarct size from early serial S-myoglobin observations. A coefficient of pre-analytical/analytical variation below 0.05 is then required in addition to frequent blood specimen collection from admission up to peak time of the serum concentration curve.