Statistical Test for Equivalence in Analysis of Commutability Experiments

DISCUSSION We applied equivalence testing on step (1) of analysis of commutability experiments. Statistical inference in terms of equivalence is commonly used in guidelines of pharmaceutical industry if similarity of two or more processes within predefined limits has to be proven (e.g. (FDA 2001)). In clinical chemistry literature equivalence testing is mentioned (Lung, Gorko et al. 2003) only seldom, but it is time to introduce these methods. The approach proposed here is advantageous because: (i) The new method is in accordance with definition of commutability. (ii) The statistical procedure is simplified. (iii) Limits of commutability are defined by experts and not by statisticians. In addition, the quality of CS measurements does not influence decision about commutability. Especially usage of prediction interval of Deming regression with variable α-level (!) as proposed in guideline C53-P leads to a dependency of limits on quality of CS measurements on the one hand and an influence of statistician’s decision making on the other hand. – Moreover, same limits are used for all comparisons with the new approach. (iv) Introduction of equivalence testing is connected with a fundamental change of proof of hazard to proof of safety: In conclusion, false positive conclusion of commutability in case of low quality of CS measurements is avoided. On the other hand, false negative non concluding commutability in case of a extremely precise and undistorted measured CS is not possible. Regarding step (2), situations might occur where several clusters of methods with commutable RM are identified. This issue remains to be solved. Fig. 1: Example Method E vs. Method F. Predefined Limits: CV=5% (dashed lines) Assumptions for RM: CV=2%, N = 6 Legend: coloured symbols: RM with 95%-CR. black: CS. All RM with exception of RM D (red symbol) are shown to be commutable for ME/MF. METHODS Data In principle, the analysis of commutability experiments is performed within two steps.