Background: Carryover experiments are widely used for clinical chemistry and immunochemistry analysers to evaluate and validate carryover effects. The experimental design is well described. However, there is no guideline on the statistical approach on data analysis, especially if absence of carryover has to be shown. The only reporting of carryover in ppm is not helpful because its uncertainty is not taken into account. Furthermore, the most commonly used method fails to demonstrate the absence of carryover. We propose a step-by-step guidance applying a new statistical design for analysis of carryover studies based on equivalence testing, and provide a sample based tutorial.
Methods: For statistical analysis of carryover effects an one-sided version of equivalence testing by comparing the difference with a predefined limit (i.e., a test of non-superiority) is used. The methodology is demonstrated by measuring total ßhCG in human serum samples with a UniCel DxI 880 analyser.
Results: A new statistical approach based on equivalence testing has been developed for analysis of data resulting from a typical experimental protocol for carryover studies. Experiments using 8 (11) cycles of high and low concentration samples are appropriate to validate the absence of carryover with 80% (90%) power and an α-level of 0,05 if no carryover is expected. We propose to predefine an acceptance criterion based on the imprecision (here: expressed as one standard deviation) observed for those replicates of the low concentration samples expected to be unaffected by carryover. In the demonstration, the absence of carry-over was concluded with a significance of p < 0.05.
Conclusions: Appropriate statistical methods should be applied when the target of a method-validation experiment is (i) absence of any effect, (ii) non-inferiority / non-superiority or (iii) equivalence. Using the example of carryover studies, we show that one-sided equivalence testing is the proper model, and propose a guidance for analysis of these experiments. The example of carryover illustrates a methodology which is also applicable for analysis of a wide range of experimental approaches, including method comparison, commutability and robustness.