Mixer-Independent Soft Sensors to Predict Tablet Critical Quality Attributes in a Continuous Direct Compression Process

Mathematical models that predict critical quality attributes (CQAs) play a significant role in quality by design. Models whose input variables include process parameters (PPs) associated with equipment require reconstruction when equipment configurations change. This study proposes a systematic method for selecting input variables in a continuous direct compression process with the aim of eliminating the need for model reconstruction when mixer configurations change. We conducted 19 experiments by varying nine PPs and measured five CQAs and 24 material attributes (MAs) of intermediate products at two mixing processes. Multiple statistical models were developed by incrementally adding mixer-independent PPs, MAs, and mixer-dependent PPs to input variables. Partial least squares regression provided high prediction accuracy for the disintegration time, dissolution rate, and tablet weight relative standard deviation (RSD), while Gaussian process regression was effective for the hardness and tablet APAP mass fraction RSD. These high-performance predictions were achieved using only MAs and mixer-independent PPs. These results demonstrate the potential for constructing mixer-independent models and thereby avoiding the need for model reconstruction when mixer configurations change. This study contributes to facilitating flexible equipment configuration changes.