Statistical Analysis of Dissolution Data
To describe and compare the dissolution profiles obtained, linear trapezoidal method was used to calculate the area under the curve of each profile over 4 h (AUC0–4h). This allowed the use of one value representative of drug dissolution to compare the different scenarios tested.
One-way analysis of variance (ANOVA) with a post hoc Tukey honest significant difference (HSD) test was conducted to investigate statistically significant differences (p < 0.05 noting significance level) in the AUC0–4h between direct administration of formulation and mixing the formulations with the different vehicles. t test analysis was used to compare AUC0–4h results obtained between drug dissolution after mixing the formulations with vehicles of same subtype or drug dissolution after mixing the formulations with the same vehicle under different testing conditions (i.e. agitation rate or time between preparation and mixing) (p < 0.05 noting significance). The analyses were performed with GraphPad Prism® v.7 software (San Diego, USA).
Partial least squares regression (PLS-R) analysis was used to correlate the AUC0–4h values of the different testing scenarios (response factor) with the physicochemical properties and macronutrient composition of the vehicles (pH, buffer capacity, surface tension, viscosity, osmolality; percentage of fat, sugars and proteins), drug solubility in each vehicle, type of formulation and testing conditions (i.e. preparation time) (XLSTAT Software; an Add-In for Excel, Microsoft®). The physicochemical properties and macronutrient composition of the vehicles and well as drug solubility values in each vehicle were previously presented (11). When analysing both drugs together, drug characteristics (logP (log octanol-water partition coefficient) and ionisation percentage (obtained from ACD/Labs© 2010–2018)) were also considered as variables. The quality of the model was evaluated with the square of the coefficient of determination (R2) and goodness of prediction (Q2), with values close to 1 being indicative of good fit and prediction power, respectively (33). Full cross-validation (leave-one-out procedure) was used to develop and evaluate the regression model. The optimum number of calibration factors for each model was selected based on the optimum predictability of the model and predicted residual error sum of squares (PRESS). The standardised coefficients of the factors indicated the relative effect (positive or negative) of their corresponding variables on the response. The variable importance in projection (VIP) value was used to evaluate the importance of each factor on the model (33). Model variables with VIP values > 1 were evaluated as the most important in explaining the variation in the dependent variable, whilst values between 0.7 and 1 were considered moderately influential for the model. Values < 0.7 were deemed not of significance for the prediction of the dependent variable (33).