Intelligensys Formulating Knowledge - Pelleting, Spheronization and capsule filling

Developing better products faster

In Vitro In Vivo Correlations for Inhalation

For drugs administered by inhalation, developing in vitro in vivo correlations has the potential to reduce and rationalize the number of in vivo tests undertaken in development. However, the modelling itself is far from straightforward, since complex relationships exist between aerodynamic particle size of the emitted dose, regional lung deposition, and efficacy. Not only product characteristics (such as particle size) will affect deposition in the lung, but demographic and physiological factors are also important. This means that statistical treatments, with their general assumption of linear relationships, have proved of only limited use.

Working at the Institute of Pharmaceutical Innovation at the University of Bradford, UK, Marcel de Matas and his colleagues have examined the use of artificial neural networks in developing in vitro in vivo correlations for both nebulizer delivery and for dry powder inhalers. INForm, from Intelligensys, was the software package they used for their ANN modelling.

In a paper entitled “Evaluation of an In Vitro In Vivo Correlation for Nebulizer Delivery Using Artificial Neural Networks” (J Pharm Sci 96, 3293-3303 (2007)) de Matas, Shao, Silkstone and Chrystyn examined the relationship between urinary salbutamol excretion within 30 minutes after inhalation (an index of relative lung bioavailability of salbutamol) to specific fractions of particle size distribution, the patient’s body surface area, and the method of nebulization. The data came from 5 different nebulizer systems, administered to 9 different volunteers. This resulted in 44 data points that could be evaluated using the neural network. 27 records were used in the training set, with the remaining 17 used for validation. Care was taken to ensure that both training and validation sets gave good coverage of the experimental space.

The model showed that lung deposition of salbutamol was enhanced by increasing the amount of drug available in specific particle size ranges. Age and gender did not have a large impact on the lung deposition model, but body surface area did. The general conclusion was that despite the relatively small training set of just 27 data records, a model could be developed to predict in vivo nebulizer performance for individual subjects.

De Matas and his coworkers also applied ANNs to another form of inhaled delivery – dry powder inhalers. This was discussed in a poster presented at the Pharmaceutical Sciences World Congress in 2007, and published in a paper entitled “Evaluation of in vitro in vivo correlations for dry powder inhaler delivery using artificial neural networks” by de Matas, Shao, Richardson and Chrystyn in the European Journal of Pharmaceutics 33 80-90 (2008). Here, four dry powder inhaler formulations of salbutamol were tested on 12 different volunteers, and the amount of drug and metabolite excreted up to 24 hours post-inhalation was measured, with co-administration of charcoal to exclude gastrointestinal adsorption (index of total relative lung bioavailability). The same inhalers were also tested in 11 mild asthmatic patients with modelling activities focussing on bronchodilator response to salbutamol. The database for the volunteers was divided into a training set (34 data records) and a validation set (13 data records) using a random selection method. The 44 different data records for the mild asthmatic patients were partitioned randomly into a training set (31 records) and a validation set (13 records). Population demographics (age, gender) were relatively unimportant in the modelling the index of total relative lung bioavailability for healthy volunteers, but (as in the nebulizer study discussed above) body surface area (which they say is probably related to lung size) had a positive effect. The amount of drug in the size range 3.3 to 4.7 microns had a positive effect on lung deposition of salbutamol, whereas the amount of drug at size ranges 5.8-9 and 9-10 microns had a negative effect. For mildly asthmatic patients, demographics were somewhat more important than for healthy subjects with older patients demonstrating poorer response to the medication.

In both the reported studies, de Matas and his colleagues demonstrated that neural networks (as embodied in INForm) have the potential to develop models capable of predicting in vivo performance from in vitro measurements and subject demographics. The particle size distributions of the emitted doses were shown to be the most important variables, with demographics having a lesser (but in many cases, discernable) effect.