Natascia Meneghetti

Knowledge management in secondary pharmaceutical manufacturing by mining of data historians-A proof-of-concept study.

In this proof-of-concept study, a methodology is proposed to systematically analyze large data historians of secondary pharmaceutical manufacturing systems using data mining techniques. The objective is to develop an approach enabling to automatically retrieve operation-relevant information that can assist the management in the periodic review of a manufactory system. The proposed methodology allows one to automatically perform three tasks: the identification of single batches within the entire data-sequence of the historical dataset, the identification of distinct operating phases within each batch, and the characterization of a batch with respect to an assigned multivariate set of operating characteristics. The approach is tested on a six-month dataset of a commercial-scale granulation/drying system, where several millions of data entries are recorded. The quality of results and the generality of the approach indicate that there is a strong potential for extending the method to even larger historical datasets and to different operations, thus making it an advanced PAT tool that can assist the implementation of continual improvement paradigms within a quality-by-design framework.

Uncertainty back-propagation in PLS model inversion for design space determination in pharmaceutical product development

Abstract The inversion of latent-variable models is an effective tool to assist the determination of the design space (DS) of a new pharmaceutical product. A challenging issue in partial least-square (PLS) regression model inversion is to describe how the uncertainty on the model outputs (product quality) relates to the uncertainty on the model inputs (raw material properties and process parameters). In this study, a methodology to relate the uncertainty on the output of a PLS model to the uncertainty on the model inputs is proposed. Two uncertainty back-propagation models are formulated and critically compared. Frequentist confidence regions (CRs) for the solution of the inversion problem are built. These CRs represent a subspace of the historical knowledge space within which the DS of the product to be developed is likely to lie with assigned confidence level. The input combinations that belong to these CRs (and that are consistent with the historical calibration data set) should be primarily investigated when an experimental campaign is to be performed to determine the DS. The proposed methodology is tested on three different case studies, two of which involve experimental data taken from the literature, respectively, on a roller compactor and on a wet granulator. It is shown that both uncertainty back-propagation models are effective in bracketing the DS, with the second model outperforming the first one in terms of shrinkage of the space within which experiments should be carried out to identify the DS.

Supporting the transfer of products between different equipment through latent variable model inversion

Abstract Product transfer is a problem commonly encountered in industry when a production has to be transferred from a source plant to a target plant. In this paper a strategy to assist product transfer is proposed. The strategy is based on latent variable models (LVMs) to relate the data available from one or more source plants with the (usually scarce) data available from the target plant, and on LVM inversion to estimate the target plant operating conditions. The inversion is performed within an optimization framework, which can handle constraints for both product quality variables and input variables. An experimental nanoparticle production process is used as a test bed to illustrate the benefits of the proposed strategy.

Mining Information From Developmental Data: Process Understanding, Design Space Identification, and Product Transfer

Abstract Many pharmaceutical industry environments are characterized by the availability of large amounts of data deriving from developmental or production activities. In this chapter we show how information can be extracted from these data sets using a class of multivariate statistical methods called latent-variable (LV) models. In particular, we refer to three LV modeling approaches (principal component analysis, partial least-squares regression, and joint-Y partial least-squares regression), and show how they can be used: (1) to improve the understanding of a continuous process for the manufacturing of paracetamol tablets; (2) to assist the determination of the design space for a new pharmaceutical product; and (3) to scale-up the manufacturing of a nanoparticle product from a smaller device to a larger one.

Automated Data Review in Secondary Pharmaceutical Manufacturing by Pattern Recognition Techniques

Abstract A methodology is proposed to support the periodic review of manufacturing data in the pharmaceutical industry. Pattern recognition techniques are employed to isolate and analyze operation-relevant data segments to the purpose of automatically extracting the information embedded in large databases of secondary manufacturing systems. The results achieved by testing the proposed methodology on two six-month datasets of a commercial-scale drying unit demonstrate the potential of this approach, which can be easily extended to other manufacturing operations.

First-Principles Model Diagnosis in Batch Systems by Multivariate Statistical Modeling

Abstract Process/model mismatch may arise when a first-principles model is challenged against historical experimental data. In this study, a methodology recently proposed to diagnose the root cause of the mismatch in continuous processes is extended to batch systems, taking a batch drying process as a case study to test the proposed methodology. The likely sources of the mismatch are identified using a multivariate statistical model and analyzing the model residuals as well as the scores shifts. Two simulated examples demonstrate the effectiveness of the proposed methodology.

Diagnosing Process/Model Mismatch in First-Principles Models by Latent Variable Modeling

Abstract A methodology is proposed to diagnose the causes for the process/model mismatch (PMM) that may arise when a process is simulated using a first-principles (FP) model. To this purpose, a latent variable model is used to assess the consistency between the correlation structure of a historical operation dataset and that of a similar dataset generated using the FP model. Inconsistencies between the two correlation structures are analyzed by means of diagnostic indices. Engineering judgment is then used to pinpoint which equations or parameters of the FP model are mostly responsible for the observed PMM. The proposed methodology is tested on two simulated case studies, and it is shown to provide clear indications on where the mismatch originates from.