Modeling and predicting tablet dissolution slowdown using an acceleration factor approach and constrained neural...

Area of Science:

• Pharmaceutical Sciences
• Materials Science
• Chemical Engineering

Background:

• Tablet dissolution slowdown during storage can compromise drug release and bioavailability.
• Predictive stability models using accelerated data are crucial for assessing long-term storage risks.
• Understanding and mitigating dissolution changes are vital for pharmaceutical development.

Purpose of the Study:

• To compare science-based and machine learning models for predicting tablet dissolution slowdown.
• To evaluate the effectiveness of an acceleration factor (AF) model and a constrained neural network.
• To identify strategies for controlling storage conditions to prevent dissolution changes.

Main Methods:

• Applied an empirical acceleration factor (AF) model to open dish stability data.
• Constrained a neural network with the AF model, leveraging Arrhenius relationships and first-order decay.
• Compared prediction performance against an unconstrained neural network and evaluated packaged storage data.

Main Results:

• Both the AF approach and the constrained neural network accurately predicted dissolution profiles in packaged tablets.
• The AF model identified critical humidity boundary conditions to prevent dissolution slowdown.
• Constraining the neural network with the AF model improved prediction performance.

Conclusions:

• Accelerated stability modeling, particularly the AF approach and constrained neural networks, shows promise for predicting dissolution changes.
• These methods can serve as valuable Modeling Approaches to Reimagine Stability (MARS) tools in pharmaceutical development.
• Insights gained aid in formulation, packaging selection, and stability evaluations to mitigate dissolution challenges.