(218f) Advancing Drug Product Stability Prediction Via Integrating Data Science, Decision Tree, and Kinetic Modeling for Continuous Improvements

Pharmaceutical drug product (DP) early development can offer an initial prediction of DP stability with limited data of lab scales for the kinetic modeling exploration. By continuing the DP stability study through further development at pilot scales and manufacturing scales and complementing it with additional data of kinetic modeling analysis, the accuracy and reliability of DP stability prediction can be improved.

The objective of this study is to advance drug product stability prediction beyond just using lab generated data. The modeling process is proposed to include, but not limited to employing a) data science methodology to analyze available stability data, b) decision tree analysis to reveal attributes-performance classification for trend analysis, and c) integrating both of them into kinetic modeling with accessible manufacturing scale data, to develop drug product’s real-life scenarios for innovations and improvements of the DP stability models for modeling and prediction.

Once the stability kinetic modeling results from the early lab stage trends are in alignment with the observed trends at the manufacturing scale, it signifies that the DP stability can be strongly dependent on few critical parameters, as observed at the lab scales. This indicates that there are no additional key attributes from the scale up and tech transfer that have an impact on drug product stability. In other words, the kinetic modeling projection from the early lab scale development can act as a representative for scaled manufacturing system.

In contrast, the misalignment observations will suggest from a system-level that there should be additional attributes impacting DP stability. These are particularly related to the equipment and process scale up to pilot and manufacturing scales. The kinetic modeling from accessible manufacturing scale data can guide additional attributes impacting the DP stability. Accordingly, the integrated methodology of data science, decision tree, and kinetic modeling can enhance the probability of success for stability prediction. This can be achieved with the newly identified additional key attributes, either from scale up or tech transfer to improve the manufacturing process control, and thus to enhance DP stability and modeling prediction.