Utility of a panel of sera for the alignment of test results in the worldwide multicenter study on reference values

Abstract

Background: In a planned International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) worldwide study on reference intervals (RIs), a common panel of serum samples is to be measured by laboratories from different countries, and test results are to be compared through conversion using linear regression analysis. This report presents a validation study that was conducted in collaboration with four laboratories.

Methods: A panel composed of 80 sera was prepared from healthy individuals, and 45 commonly tested analytes (general chemistry, tumor markers, and hormones) were measured on two occasions 1 week apart in each laboratory. Reduced major-axis linear regression was used to convert reference limits (LL and UL). Precision was expressed as a ratio of the standard error of converted LL or UL to the standard deviation (SD) comprising RI (approx. 1/4 of the RI width corresponding to between-individual SD). The allowable and optimal levels of error for the SD ratio (SDR) were set as ≤0.250 and ≤0.125, respectively, in analogy to the common method of setting limits for analytical bias based on between-individual SD.

Results: The values for the calculated SDRs depended upon the distribution patterns of test results: skewness toward higher values makes SDRLL lower and SDRUL higher. However, the CV of the regression line slope, CV(b), is less affected by skewness. The average of SDRLL and SDRUL (aveSDR) correlates closely with CV(b) (r=0.995). The aveSDRs of ≤0.25 and ≤0.125 corresponds approximately to CV(b) values of ≤11% and ≤5.5%, respectively. For all results (i.e., n=80), conversion was allowable (optimal) in 98% (89%) of the analytes, as judged by CV(b). Resampling studies using random subsets of data with a data size (n) of 70 to 20 revealed that SDRs and CV(b) gradually increase with reduction of n, especially with n ≤30.

Conclusions: CV(b) is a robust estimator for judging the convertibility of reference values among laboratories, even with a skewed distribution. Assuming 40 sera to be a practical size for the panel, reference values of 89% (80%) of analytes examined were made comparable by regression analysis with the allowable (optimal) level of precision.