Stability studies are an important part of drug development. The aim of a stability study performed according to ICH Q1A(R2) is to determine the shelf life and storage conditions of a drug. Stability studies are designed as recommended in regulatory guidelines for a specific country and include a large number of samples and analysis to be performed on those samples. They are based on the analysis of critical parameters that are believed to be the subject of change during the storage of the drug and thus affect its quality. The stability of a drug can be affected by a number of factors such as temperature, humidity, light, storage time and packaging material. Thus, stability studies can be seen as a complete multi-factor and multi-level mixed experimental design, in which factor values change according to well-defined rules, which is also the main feature of experimental designs. This doctoral thesis presents a comprehensive approach of evaluation and presentation of stability data using Design of Experiments (DoE) and multivariable statistical methods. The results are presented in three separate chapters. Each chapter presents an analysis of a different drug stability study. For the analysis we used multivariable statistical methods, the Partial least squares (PLS) method and the Multiple linear regression (MLR) method, which allow us to analyze the effects of several factors simultaneously, their interactions, and predict the responses of new multi-factor experiments. For the first drug with an unstable active pharmaceutical ingredient (API) Hydrochlorothiazide (HKTZ), we developed a PLS model from all the stability study data obtained. We have shown the influences of individual factors and two-way factor interactions on the content of the major degradation product DSA. We found that the interaction between factors »packaging« and »time« has the biggest impact on 4-amino-6-chlorobenzene-1,3-disulfonamide (DSA) formation. With the use of a contour diagram, we determined that a polyvinylidene chloride (PVDC) blister offers the least protection from degradation of HKTZ. With the model we were also able to predict the drug stability at three different testing conditions (long-term, intermediate and accelerated condition). Furthermore, with a more than half-reduced stability design, we showed that the PLS model predicts similar DSA levels as the full stability design for a drug packaged in a PVDC blister. For the drug with the chemically unstable API saxagliptin (SAXA), we developed a PLS model of up to the sixth month stability data and predicted that the levels of impurity DP-2 were lower by at least 0.2% when the drug was protected from oxygen right after manufacturing. We found that the lower strength of the drug was at least twice less stable with respect to the formation of impurity DP-1. Additionally, we predicted the shelf life of the drug for climate zone II, i.e. 24 months, with high reliability. The predictions for the impurity DP-1 and the Total impurities were more accurate with the PLS model than those obtained with standard linear regression, taking into account that a much larger data set was used for the PLS analysis. Third chapter provides insight into an investigation of possible causes for the acceleration of drug dissolution of a modified-release drug product at stability testing. With an MLR model of up to six months stability results, we found that a decrease of antioxidant Butylhydroxytoluene in the tablet during stability may result in an accelerated dissolution. In addition, a faster dissolution profile can also be influenced by other factors in the starting materials and manufacturing processes of the drug. We further demonstrated the usefulness of the methodology to support the use of the Lean stability approach on a long-term stability study. With the acquired knowledge on drug stability, we designed two optimized long-term stability studies. We have shown that with a reduced long-term design, we can still confirm the 24-month shelf life, given the values of the most critical stability parameter. In the doctoral thesis, we presented the applicability of the DoE methodology and multivariable statistical methods for the evaluation of drug stability. Such an approach for the analysis of all stability data obtained according to the ICH guidelines has not yet been described in the literature. We have shown that the methodology allows us to have a more comprehensive approach to the analysis of a large number of data obtained, compared to classical simple linear regression (LR). Current regulatory guidelines focus mostly on LR and Analysis of Covariance, which include only small amount of data and focus on the stability of the most critical parameter. With multivariable methods we can use all the data available and evaluate influences of the factors and their interactions on the drug stability. The interactions between factors cannot be evaluated with LR. The presented approach can be used to predict drug stability for non-tested factors, to predict results outside the acceptance criteria, and to assess whether stability studies could be reduced without losing the key information on drug stability, that is shelf life determination.
|
