Upsala J Med Sci 99: 231-236, 1994 4. Analytical Quality Specifications Elizabeth M.S. Gowansl, Per Hyltoft Petersen2, Ole Blaabjerg2 1 . Clinical Biochemistry, Telford Hospital, Telford TF6 6TF, Shropshire, England. 2. Department of Clinical Chemistry, Odense University Hospital, DK-5000 Odense C, Denmark 4.1 Overall Goal for Analytical Quality Specifications It has become dogma that reference intervals for biochemical quantities (components) are method dependent and, therefore, are individual to laboratories. This may have arisen largely because a plethora of factors, including laboratory methodology, do affect reference values. The dogma has been further stimulated by external quality assessment schemes and proficiency testing schemes in which considerable differences between analytical methods and commercial kits have not only been tolerated but also stimulated by grouping into so called method dependent or peer groups. There are a few exceptions from this general approach which include the National Cholesterol Education program in USA and German control schemes (Ringversuche and Indstand). However, more correct approaches seem to be developing in US, and through working groups under the auspices of European External Quality Assessment Organizers. In consequence the future may show a general and more informed attitude to the question of transferability of data and, thereby, also to the development and use of common reference intervals. It may be that ethnic differences exist for some quantities, as illustrated by Harris et al. (2) as well as age- and sex-dependent differences but this does not provide a rationale for method dependent reference intervals. These biological differences, however, should be described in detail so that the information could be shared by all laboratories. The purpose therefore is to evaluate the analytical quality specifications required for the performance needed for using common reference intervals in laboratories where the populations are homogeneous for a quantity - irrespective of whether there is one single interval or several are required according to known differences in the populations. 23 I Assumptions for the Models A number of assumptions have to be fulfilled for establishing common reference inter- vals: 1. 2. 3. 4. 5 . 6. 7. 8. The population - or well described subsets - must be homogeneous for the quantity. The inclusion and exclusion criteria for the reference sample group must be clear. The preparation of reference individuals before sampling must be well defined. The sampling technique must be standardized. Handling, preparation, and storage of samples must not influence the quantity, neither the structure nor the concentration or activity. The model for statistical calculations must be in accordance with the actual distribution of data. The number of reference individuals must be sufficient. The assays must be performed with an analytical quality which is better than the quality specifications for sharing common reference intervals - as given below. This infers standardization with traceability of concentration values and specific measurement procedures. The Models For homogeneous healthy groups many quantities are distributed symmetrically or with a positive skewness, allowing for application of one of two statistical models, namely Gaussian and log-Gaussian. The evaluation of the Gaussian model is simpler and it is, therefore, used for the principal evaluation. IFCC has given recommendations for estimation of reference intervals where a major point is that at least 120 individuals should be used for the calculation of reference intervals in order to keep the uncertainty about these limits low (5). For a Gaussian distribution, this corresponds to a 0.90 confidence interval of each limit of 0.24 times the biological standard deviation, or, that the fraction of individuals outside each limit - due to the uncertainty of the sample variation - with 0.90 certainty is between 0.013 and 0.044 - instead of t h e 0.025 which is expected from an interval calculated as mean k l.96*sbiOlogi~. 232 The basis of the model for evaluation of analytical quality specifications is to estimate the reference intervals based on a greater number of individuals (e.g. 800), so as to make the sample uncertainty negligible - and then allow for analytical error instead of sampling error - giving the same maximum uncertainty as allowed by the IFCC. The Quality Specifications Based on this concept, the maximum allowable analytical error - combined bias and imprecision - must not decrease or increase the fraction outside each reference limit more than 0.013 to 0.044. A graphical evaluation (1) reveals a functional relationship between maximum allowable analytical bias and imprecision as shown in fig. 4.1. Analytical bias in fraction of biological standard deviation 0.20 0.10 0.00 Maximum allowable combination of analytical I " " l " " 1 Maximum allowable combination of analytical I \bias and imprecision 0.00 0.25 0.50 0.75 Imprecision in fraction of biological standard deviation Fig. 4.1. Relationship between maximum allowable combination of analytical bias and imprecision according to the described concept. The units on the axes are fractions of the biological standard deviation, Sbiologicap From Hyltoft Petersen and Hgrder (4) with permission. The figure shows that the maximum allowable bias is l B ~ l < 0.24*sbiologi~ when analytical imprecision, SA, is negligible, and that the maximum allowable imprecision is SA < 0.55*sbiOlogid when BA is negligible. For all other combinations, the values have to be interpolated from the curve. These are the analytical quality specifications for sharing common reference intervals when the distribution is Gaussian. 233 For log-Gaussian distributions, the procedure is t h e same for log-values and the analytical quality specifications are the same on t h e log-level. The computations, however, are somewhat more difficult. Therefore, a graph may help in making the estimates by reading off from a curve. This functional relationship is given in fig. 4.2. The most important fact from log-Gaussian distributions is that the analytical quality specifications are dependent on the ratio between the upper and lower reference limits, whereby, the specifications are given as a net of curves. Moreover, the specifications are given in terms of percentage bias and percentage coefficient of variation. BIAS: 6% 0 5 10 15 20 2 5 30 35 IMPRECISION: CV% Fig. 4.2. Approximated maximum allowable combination of analytical bias (%) and coefficient of variation (%) for log-Gaussian distributions. The specifications depend on the ratio between the upper and lower reference limits. From Hyltoft Petersen et al. (3) with permission. When analytical quality specifications are interpolated from the curves (fig. 4.2.), then t h e analytical imprecision should have been subtracted from the dwtribution in order t o get the isolated distribution (see below). Based on the concepts of sharing common reference intervals and of allowing for analytical bias and imprecision instead of uncertainty from samples size it is possible to estimate analytical quality specifications for all endogenous quantities. 234 Possible Modifications Even with the best analytical methods, there will be some uncertainty during measurements of the reference values and, in consequence, in the estimate of reference intervals. Furthermore, laboratories using common reference intervals will dwclose - a t least - some inherent imprecision, which must be included in the stated reference interval. It might, therefore, be considered whether a reasonable low imprecision should be included in the estimation of reference intervals. If so, then the analytical quality specifications are slightly changed, and for laboratories with stable perfomance close t o the imprecision obtained during measurements of reference samples, the specification for analytical bias will be of major importance. For quantities with high ratios between the upper and lower reference limits the effect will be negligible as long as CVA is below 5% but, for quantities with low ratios, it may be necessary to use the combined analytical and biological CV for the estimation of reference intervals. In chapter 7 this pragmatic approach is applied to S-Albumin. References 1. 2. 3. 4. 5 . Gowans EMS, Hyltoft Petersen P, Blaabjerg 0, H ~ r d e r M. Analytical Goals for the Acceptance of Common Reference Intervals for Laboratories Throughout a Geographical Area. Scand J Clin L n b Invest 1988;48:757-64. Harris E q Wong ET, Shaw ST. Statistical Criteris for Seperate Reference Intervals: Race and Gender Groups in Creatine Kinase. Clin Chem 1991;37:1580-2. Hyltoft Petersen P, Gowans EMS, Blaabjerg 0, H ~ r d e r M. Analytical Goals for Estimation of non- Gaussian Reference Intervals. Scand J Clin Lab Invest 1989;49:727-37. Hyltoft Petersen P, H ~ r d e r M. Influence of Analytical Quality on Test Results. In Magid E (ed.). Some Concepts and Principles of Clinical Test Evaluation. NORDKEM, Nordic Clinical Chemistry Project, Helsinki Finland 1992:65-87. Solberg HE. Approved Recommendation (1987) on the Theory of Reference Values. Part 5. Statistical Treatment of Collected Reference Values: Determination of Reference Limits. J Clin Chem Clin Biochem 1987;25:645-56. 235