Abstract
Based on the ICH Q14 definition, design space for analytical methods can be described as a subset of investigated experimental conditions for which it can be demonstrated, with sufficient probability, that method fulfills predefined criteria [1]. The response distribution should be simulated and the experimental conditions, for which the probability that the critical quality attributes (CQAs) are within acceptance region is above the desired quality level (p), should be included in the design space. Monte-Carlo simulations, bootstrapping, and Bayesian analysis [2] can be applied, with the last approach preferred as it accounts for both model parameter uncertainty and data correlations [3]. This study aimed to investigate differences between the three approaches to derive the design space for a chromatographic stability-indicating method for citalopram tablets. Retention of impurity A (k) and resolution between citalopram and impurity E (Rs) were selected as CQAs (Y), while acetonitrile content, perchlorate concentration in the mobile phase, and column temperature were critical process parameters (X). The acceptance region for CQAs was set as 𝑆𝑆 = {(𝑘𝑘,𝑅𝑅𝑠𝑠):𝑘𝑘 > 1, 𝑅𝑅𝑠𝑠 > 2} and the design space as DS = {x∈𝜒𝜒|𝑃𝑃(Y∈𝑆𝑆)≥𝜋𝜋} with p = 0.9. Contour plots showing the dependence of the probability on acetonitrile content and column temperature at five different perchlorate concentrations showed no significant differences, which can be attributed to the low uncertainty of the model parameters (0.0015-0.0240) and weak correlation of regression residuals (-0.26). A wider range of parameter uncertainties, residual correlations, etc. need to be investigated to determine areas of practical equivalence of the examined approaches.

This publication has 2 references indexed in Scilit: