Drug Information Journal - March 2009 - (Page 146) 146 MEDICAL INFORMATION Dijkman, Fraser, Treasure, Kapke FIGURE 3 Covance lower limit versus empirical lower limit. Derivation of empirical lower reference limits for hemoglobin (g/L). Relative Frequency (b) Global Cutoff Fraction F = 16.9% below Central Lab LRL of 116.9 Reference Healthy Population Clinical Study Population Clinical Study Population for Lab i LRL (Central Laboratory) Empirical LRL (Local Lab) (c) 16.9% of Population for Lab i gives Λ of 120 (a) 2.5% of Ref Healthy Population gives Central Lab LRL of 116.5 80 90 100 110 120 130 140 Test Result 150 160 170 180 190 200 local population, then Li should be low when Λi is low, and vice versa: a positive correlation. If the local laboratory LRLs were chosen arbitrarily, however, there would be no correlation. This is based on Λi being a measure of the characteristics of the local population. It can be seen for laboratory 533 that the results are lower than seen in the whole database, yet the actual laboratory LRLs are higher. The two scatter plots (Figures 4 and 5) presented show visually the relationship between Λi and Li. For ease of interpretation, the ln-transformation has been removed from the axes so that the data points are relative to a value of 100 for the central laboratory LRL. Working Data Extracts. In order to calculate an empirical LRL or URL for a particular laboratory, that laboratory must have enough observations to give a reasonable estimate of that limit. Therefore, the data set actually used for the assessment of, for example, the empirical LRL for leukocyte count (the working data extract) would consist of the leukocyte count results from the data set from only those local laboratories with sufficient observations. “Sufficient” was defined by this rule: a laboratory had sufficient observations if both (a) the number of observations was at least 20 and (b) cutting off a fraction F from this laboratory’s data would cut off at least two observations. Fraction F is the proportion of results in the whole data set cut off by the central laboratory LRL (see the theory section). The rule for sufficient observations was motivated by the need to use as many observations as possible moderated by the requirement to make realistic estimates of a quantile of the distributions. It was thought that 20 observations was the minimum that could be safely used. This number is also discussed by Barry and Westgard in one verification procedure for transfer of reference intervals between laboratories (12). The hemoglobin LRL analysis was repeated using a basis of 50 and of 100 observations. Although the number of laboratories used was reduced
For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.