KinetDS: An Open Source Software for Dissolution Test Data Analysis

Because drug quality is the focus for pharmaceutical industry and regulatory agencies, the in vitro dissolution test becomes a standard tool for characterization of manufactured products. However, results of the dissolution test must be expressed in mathematical terms; this is realized by fitting various models to the cumulative dissolution curves. The models might be either mechanistic or empirical. The fitting process requires software (e.g., KinetDS) for automation and determination of possible release mechanisms of drug substances from the dosage forms. The software is FOSS (Free Open Source Software) and is available at http://sourceforge.net/projects/kinetds/. INTRODUCTION Bioequivalence (BE) is the basic criterion of similarity between two pharmaceutical products, namely generic and innovator drugs. Due to the probabilistic character of BE definition, pharmaceutical equivalence (PE) is mandatory for the pharmaceutical industry. In comparison with bioassays, it is based on the more reproducible in vitro dissolution tests. The latter produce time profiles of the drug amount released from the dosage forms, which are expressed in their cumulative form. An expression of dissolution profile similarity is based on two main approaches: (1) the direct comparison of the curves (1) (i.e., by similarity factor f2), and (2) use of mathematical modeling to parameterize curves and compare resulting parameters of the models. The latter method is also recognized as an element of a new development strategy of pharmaceutical products (2, 3), process analytical technology (PAT), and implicates possible mechanisms of drug release from the dosage form, which influences bioavailability (4). Curve fitting of drug release data is also commonly used as a step for in vitro–in vivo correlation (IVIVC) development. The fitting process of the model parameters requires software, which was the aim of this work and resulted in the KinetDS project. Its main objectives were intuitiveness of the graphical user interface (GUI), a high level of software automation, and free availability for commercial and noncommercial use. FEATURES AND THEORETICAL BACKGROUND KinetDS is software for curve fitting, designed particularly to describe the cumulative dissolution curve by a simple equation or set of equations. However, other curves, derived from different data sources, might be also analyzed if their dependent-variable range is between 0 and 100. The equations were chosen from the most popular mechanistic and empirical models (Table 1) applied to the drug dissolution curve description (5) such as zeroto third-order kinetic models, Higuchi, Korsmeyer–Peppas, Weibull (2and 3-parameter), Hixson–Crowell, Michaelis–Menten, and Hill equation. Model-independent descriptions of the dissolution curve are also provided (Table 2) as dissolution efficiency, DE (6), and mean dissolution time, MDT (7). DE and MTD values are automatically computed for every available dataset and included at the end of the report file. The best model choice is based on model performance on the available data. Several measures of goodness of fit are implemented (Table 3): (1) coefficient of determination, R, and empirical coefficient of determination, Remp; (2) Akaike Information Criterion, AIC (8); (3) Bayesian Information Criterion or Schwarz criterion, BIC (9); and (4) root-mean-squared error, RMSE. Additional diagnostics of the linear regression function is the standard error of the regression coefficients (10), also expressed in percent of the relevant parameter value (relative standard error of the regression coefficient). Model parameters are established by linear and nonlinear regression (NLR). The latter is represented here by the simplex method. So far, all the models included in KinetDS 3.0 are susceptible to linearization. Therefore, they might be fit by both linear regression and NLR. It is planned in Version 4.0 to introduce models suitable only for nonlinear regression. Regarding mathematical simplicity of the kinetic models from zeroto third-order, they usually do not require NLR. Specifically, because of its *Corresponding author. diss-19-01-02.indd 6 2/24/2012 1:29:41 PM dx.doi.org/10.14227/DT190112P6