Evaluation of formulation composition and process settings for a new API
The application of model simulations in drug product development has long been recognized as a promising approach. However, it can sometimes feel abstract and challenging to understand how to practically incorporate tablet press model predictions into a Drug Product Development (DPD) workflow. Using the power of the Elegent data-driven predictions, formulators and process scientists could advance their capabilities in formulation and process development, and speed up the DPD process for new APIs.
To shed light on this matter, we will explore the application of models in various DPD development scenarios, beginning with an example in formulation development. By illustrating real-world cases, we aim to provide a clearer understanding of how models can be effectively employed in the development of drug products.
To demonstrate the practical application of models, let's consider a formulation scenario where the investigation focuses on finding the optimal ratio of two fillers while maintaining a fixed concentration of the active pharmaceutical ingredient (API). The critical quality attribute under evaluation is tablet tensile strength. Based on experience, it is known that the main compression force is the most influential variable affecting this attribute.
In this case, the advantage of utilising a dataset and models becomes evident. They enable the estimation of the critical quality attribute while simultaneously varying important formulation parameters, such as the concentrations or ratio of the two fillers, alongside varying the main compression force. This comprehensive analysis allows for a better understanding of the relationship between these variables and their impact on tablet tensile strength, leading to possibly more informed decision-making in the formulation process.
With the availability of a unique dataset that captures the tabletting performance under varying formulation, API, and tabletting process parameters, a model can be constructed to predict the output for an unseen API. This model exhibits a non-linear nature, allowing it to capture the essential relationships present in the dataset.
The sheer size of the dataset, containing nearly half a million tablet recordings, enables robust prediction of the tablet tensile strength. The hyperparameters are optimized through cross-validation techniques, which ensures that the model is fine-tuned, not overfitted and capable of delivering reliable predictions. This leads to a valuable tool in support of drug product development.
The concentration of the API is set at 30% weight-basis, 1% magnesium stearate is added as lubricant and sodium carboxymethyl cellulose at 5% as disintegrant. The remaining formulation components for enhancing tablet strength and ensuring proper dissolution, are varied between the remaining range of 0 to 64%. The concentration of the two fillers is thus inversely proportional, as the constraint holds that the total concentration must match 100%.
A tablet of 400mg weight is targeted and the other process settings are set as follows: precompression occurs at 10 MPa, the speed of the press is set at 20 rpm and the paddle speed at 10% of the maximum level. The resulting tensile strength is now simulated for varying levels of the two filler concentrations and main compression force.
Evaluating the model output
We can see that the range of tensile strength for the predicted tablets lies between 0.5 to 2.75 MPa, with a clear gradient in function of the concentration of filler 1 and main compression force. Naturally, a low MCF leads to the lowest levels of tablet strength, exhibiting a zone where formulation aspects even barely make an impact. The story becomes different as the MCF is elevated though, the overall tensile strength increases and the formulation aspects clearly start to make more of a difference. At 30% of filler 1 in the tablets, meaning roughly a 50/50 ratio of the two fillers, a plateau is reached of the highest predicted strength levels between 2.5 and 2.75 MPa. Increasing the concentration of filler 1, or increasing the MCF further above 205 MPa, does not affect tablet tensile strength significantly. At lower concentrations of filler 1, the obtained tensile strength shows a clear gradient with its concentration at high MCF. The non-linear patterns of tablet quality in function of formulation and process settings, present in the large tabletting dataset, are clearly visible in this model output.
How can a formulator use this output in practice?
For starters, the model output resulted in a process setting vs. formulation aspect map, that is only obtainable with a multitude of experiments without the dataset and models. Considering that API is not available in large amounts at the start of a product development process, or can be very costly, this is not an achievable feature using only experimental ways. For further use, the most straightforward function is to choose the point, or region, of optimal tablet tensile strength. In this case, the target could be a range of 1.75 - 2.25 MPa, as it is expected that the balance of disintegration and tablet strength is optimal here. It is possible of course that the values of the other CQAs still change in this zone. Thus, this output can already serve as a plausible safe zone to vary these degrees of freedom, without compromising on tensile strength. The optimal operating point can be chosen based on other tablet CQA values, or based on economical drivers, for instance when filler 2 would be much less costly than filler 1 and it is economical to limit the concentration of the latter to a certain maximum. As another input that allows targeting experiments for the most promising input conditions, zones could be noticed where experiments would not be interesting.
Conclusion
The support of the dataset and the simulations allowed the estimation process behaviour for a new API powder. This output can be used for informed decision-making for targeted experiments in various ways, using a plausible mapping of formulation vs. process setting that is not obtainable experimentally in this phase. Using the power of the Elegent data-driven predictions, formulators and process scientists could advance their capabilities in formulation and process development, and speed up the DPD process for new APIs.