Intelligensys Formulating Knowledge - Phil Plumb's thesis abstract

Developing better products faster

Computer Aided Design and Processing of Encapsulated Pellets

Ahmed Mahmoud Abd El Haleem Ali

PhD Thesis, University of Bradford, 2009

Keywords: Pellets, Capsule filling, Computer simulation, Image analysis, Drug content, Artificial intelligence.

Abstract

Dosing errors in pelletised encapsulated pharmaceutical products mainly result from fluctuations in pellet size and shape polydispersity. Other causes include pellet aggregates, variable methods of filling and non-uniformity of active agent content. In this thesis, the hypothesis that advanced computational techniques will provide new knowledge from modelling the filling of pelletised products and optimising formulations is addressed. Pellets containing lactose and microcrystalline cellulose were prepared, characterised using image analysis, and their filling behaviour into hard gelatin capsules was evaluated experimentally. Modelling of filling was carried out using a computer simulation program based on the Monte Carlo principle. Artificial Neural Networks (ANNs) and neurofuzzy logic techniques were also applied to model the effect of formulation and preparation conditions on the characteristics of pellets. The impact on fill weight variability (% CV) was predicted from modelling the filling into four different capsule sizes.

Computer simulation experiments showed good agreement with experimental filling, with similar trends and values of fill weight variability for all pellet fractions filled using unit capsule filling, tablet die cavity and automatic capsule filling machine. Data from the manual capsule filling apparatus, however, showed higher % CV values due to overfilling and inconsistent dimensions of filling parts of the machine. Simulations and experimental filling of pellet populations containing aggregates demonstrated similar trends of fill weight % CV. High quality predictions of % CV for twins and small triplet aggregates were demonstrated up to an aggregate level of 20 % w/w, whilst large triplet and tetrahedral aggregates showed discrepancy in % CV due to segregation. In general, fill weight values obtained from simulation were less than those from experimental filling by approximately 10 %. This difference was attributed to assumptions made on pellet size and shape distributions and difficulty in simulating the volume of pores within the pellets.

Simulation of filling for commercial propranolol HCl pellets showed similar trends and values of % CV to those from experimental filling. The variability in drug content, found to be approximately 3 times that of fill weight, was attributed to different blends of pellets filled in each capsule, incomplete extraction of the drug and non-uniform distribution of the drug between pellets. Modelling of pellet properties using ANN techniques showed similar results to those found in the literature for optimum pellet properties. Neural computing studies were able to model fill weight and packing fraction with high predictability. However, % CV could not be satisfactorily modelled, except for size 0 capsules. This finding was ascribed to limited differences in % CV values between batches.

Overall, successful results of the research using the new advanced computer techniques detailed in this thesis indicate the value of the knowledge gained in supporting and directing the design and optimisation of pharmaceutical pelletised products.