The Inspiration Behind Discoverant

A major pharmaceutical manufacturer wanted to find a more effective way of minimizing lot failures and increasing process stability. They knew this would have a significant positive impact on their manufacturing profitability. Stand-alone applications such as statistics and graphics packages had previously been used at the facility for this purpose, but proved less than ideal. These partial solutions required that more than 50% of their analysis time be spent on non-analytical tasks such as the gathering, conditioning, formatting, importing and exporting of data between applications. This inefficiency translated into lost opportunities for process stabilization, lot failure avoidance and increased plant capacity through increased yield and quality. In addition, the problem of high failure rates had not been resolved.

Several process measurement systems were already in use at the company’s facility to record process parameters. The raw data gathered and stored in their databases served as a critical strategic resource for manufacturing productivity and quality improvement, as well as for specification failure avoidance. The manufacturing data was crucial to informed decision-making, but the information content wasn’t readily available in an understandable form. However, without an integrated, easy-to-use, enterprise-wide data analysis and visualization software system like Discoverant, there was a high inertial barrier to making process information available for decision-making by manufacturing technicians, scientists and managers.

The Challenge Discoverant Faced

The manufacturer needed to identify all the key process drivers to determine their combined effects on the process outcome, namely the Tablet Dissolution Rate. They also wanted useful methods for reporting key process parameters and outcomes, including process signatures, for making informative comparisons between groups of lots.

Using Discoverant’s analysis and visualization capabilities, this leading manufacturer was able to efficiently analyze data from over 60 product batches from one pharmaceutical manufacturing campaign.

Discoverant Opened Their Eyes

The manufacturing-related data needed for statistical analysis and visualization at this facility was stored in several locations. Supervisory Control and Data Acquisition (SCADA) and Distributed Control Systems (DCS) were being used to monitor instruments, control process parameters at their set points, and to archive these measurements in a process data historian. Associated lot-history and product-quality data was found in several other locations such as their LIMS and SAP systems, as well as on paper records. Historically, it had been very difficult and time-consuming to manually gather and analyze all the data for a particular lot or group of lots. Discoverant solved this problem by providing a single, process-centric view of the combined data with point-and-click access to all the manufacturing-related databases on the corporate network. It also allowed easier input of data from paper records. The absence of a system like Discoverant at the facility had been slowing their ability to get useful infor-mation out of increasingly large manufacturing databases.

Discoverant's Data Analyses and Graphics Worked Beautifully

The following general workflows were used for the analyses reported in this study.

1. Correlated Variables Identified. Discoverant created a customized matrix to look for single pairs of correlated variables. The matrix used all available data. Missing values were replaced with the mean of the remaining values using data-conditioning routines. Discoverant’s ability to do this type of data conditioning can be extremely valuable when compensating for the effects of missing data—a benefit not readily available in most statistics packages.

2. Outcome Modeled. Principal Component Analysis and Stepwise Multiple Linear Regression (PCA/SMLR) were used to condense the correlation information in the raw variables into a set of key factors. These factors were then used to model the outcome variables using SMLR with cross-validation. This multivariate analysis method is called “Principal Component Regression” or PCR.

3. Patterns Identified. Discoverant used static and animated multi-dimensional imaging techniques to show patterns in the data. This allowed examination of the behavior of key process parameters in groups of lots ranked either by lot number (production date) or by the process outcome parameter (Tablet Dissolution Rate). Cusum charts were used to analyze historical process performance. This analysis showed that several changes had occurred in the manufacturing process, perhaps inadvertently, which caused changes in the averages for the outcome variables at specific dates. Once this pattern of process drift was detected, Cusum plots helped target the lot numbers at which possible changes occurred. Standard control charts were already in use to monitor the manufacturing process at the facility, but using Cusum plots to supplement standard SPC charts proved to be much more informative.

4. Key Process Drivers Displayed. Discoverant created sophisticated Visual Process Signatures™ that were easy to understand. These Signatures used a single, animated image to display the relationships of many factors to each other, to the rest of the data, and to the process outcomes. Combining multi-dimensional visual enhancements of the tabular information along with the quantitative results of the PCR analysis, the Visual Process Signatures also provided clear and independent confirmation of the major quantitative findings in a readily understandable form. This was especially beneficial for those whose core technical expertise was in areas other than statistics but who did have a high degree of influence over process outcomes—chemists, engineers, and supervisors as well as the manufacturing operators. By illustrating how and why their process was performing the way it was, without resorting to mind-numbing tables of numbers and statistics, they quickly achieved significant enhancements in process performance.

Discoverant displayed the Process Signatures as dynamic images that were rotated in three dimensions on the screen to show additional information. Historical performance of the manufacturing process was also displayed as rolling average Visual Process Signatures for selected groups of lots, for the complete set of parameters, and then as the selected group of parameters of most interest.

Discoverant Showed All the Key Drivers to Maximize the Outcome

Using PCR, the combination of most critical, controllable process parameters was identified. Discoverant found the smallest combination of process parameters that had the greatest effect on the process outcomes. There were only five such parameters and they were located in widely disparate parts of the process. Specific recommendations for process improvements were then formulated based on the combination of key process indicators that Discoverant had identified.

Improved Outcome With Lower Cost

For all the data used in this study, the process parameters for each batch had been operated within their approved ranges. Therefore, the manufacturer was able to test Discoverant’s findings directly within the manufacturing process without the need for additional small-scale experimentation—a major cost advantage of retrospective data analysis. The process improvement recommendations were derived from the variations that occurred in the manufacturing process when operated under approved conditions. When implemented the recommended changes allowed the manufacturing staff to bias their process toward the best possible outcomes without the need to change their manufacturing technology or get it re-approved by the FDA—a tremendous savings in time and money.