Abstract
Purpose
Delivery of accurate doses from dry powder inhalers (DPI) involves many process variables which must be adjusted to ensure patient compliance and optimum therapy. Some of the process variables include: speed of inhalation (flow rate), assumed lung volume of patients, number and duration of inhalations.
Method
Data mining technology based on artificial neural networks and genetic algorithms were used to model the in vitro inhalation process, predict and optimize bioavailability from the inhaled doses.
Results
The delivery of terbutaline doses from Bricanyl Turbuhaler® was modeled and optimized using artificial neural network modeling and optimization software. Highly significant models (
p
< 0.00001) with minimum root mean squared error and high predictability:
R
2
> 81 % and 91 % for the in vitro and the in vivo models were developed, respectively. The optimized models demonstrated that an optimum emitted dose (>76 %) could be obtained if the dose was withdrawn as two inhalations with inhalation volume 4 L and flow rate 60 L/min within 4 s. The same independent variables in addition to % terbutaline emitted were modeled and optimized for % drug excreted in urine. The latter model demonstrated that optimum bioavailability (79.50 %) could be obtained from Bricanyl Turbuhaler® emitting 80–87.50 % terbutaline at a flow rate of 58–60 L/min using two inhalations irrespective of subject forced expiratory volume in 1 s (FEV
1
) or the individual lung capacity.
Conclusion
Optimized in vitro/in vivo inhalation processes using data mining models can offer rapid solutions for dose variability problems and maximize the bioavailability of drugs from DPIs.