Quality Applications

SAS JMP Statistical Discovery Software Benefits Uses D-optimal algorithm for building designs based on unique problems Augments designs easily Includes statistical quality control, gage R&R and reliability analysis tools www.jmpdiscovery.com

DOE Software Refines Processes at Pharmaceuticals Lab
SAS JMP Statistical Discovery Software

It's not uncommon for established laboratories to begin relying on the same old methods for tests and development proc-esses, which is exactly what chemists at Sepracor Canada Ltd. of Windsor, Nova Scotia, wanted to avoid. Sepracor Canada manufactures bulk active pharmaceutical ingredients for commercial purposes or use in clinical trials. When a new drug compound went into development, the lab adopted SAS Institute's JMP Statistical Discovery Software for design of experiments (DOE) to break away from the usual approaches.

 "DOE opens up the possibility of safe exploration beyond the usual confines while including what you already know and trust," says Mark Bailey, the SAS consultant who helped Sepracor Canada implement the software.

 Sepracor chemists wanted to optimize the purification proc-ess for the drug compound using recrystallization. One of the primary techniques used to purify pharmaceuticals (and chemicals in general), crystallization involves dissolving the solid pharmaceutical in a solvent by heating the mixture. As the solution cools, the pharmaceutical precipitates due to decreased solubility at the lower temperature, and many impurities remain in the solution. To remove impurities, the precipitated solid is then separated from the solution by filtration.

 "Many factors can influence the crystallization," says Craig Sturge, principal chemist at Sepracor Canada. "Without using statistical process design, there's a lot of experimentation, changing one parameter at a time based mainly on the chemist's experience and knowledge. JMP allows us to throw everything into the mix and eliminate the factors that don't have any impact on the purification, yielding better results more quickly with less effort and allowing us to understand the purification process in more detail."

 A recent problem involved nine process factors, one categorical factor (whether to introduce a small "seed" amount of pure crystalline material during the cooling) and three solvent components with simple and linear constraints. Traditionally the mixtures would be studied separately and blocking could be utilized with the process factors but not directly in the mixture design.

 Although the process was integrated, the time required for each recrystallization meant that the experiments had to be divided up over several days. "The complete group of experiments is too large to be conducted together under homogeneous conditions," explains Bailey. "Dividing the experiments into smaller groups is necessary, but it also invites 'special causes' to intrude and add to the variation in unpredictable ways. How we group the runs, or experiments, can either make us more vulnerable to unwanted influences or impervious to them. Blocking is a scheme of grouping our runs to achieve the latter result.

 "The Custom Designer in JMP version 4 was very valuable here because we could define the problem initially and work sequentially toward the final model of the recrystallization with confidence but with minimal extra work," Bailey continues.

 The chemists could have run the experiment with fractional factorial for the process variables and a nine-run extreme vertices simplex design for mixture main effects, which totaled 25 runs, but it wouldn't have tested the process factors and solvent components together. Integrating the two would require replication of the mixture design at each design point, resulting in 108-126 runs plus blocking. But to test the main effects, JMP's Custom Designer suggested only 18 runs.

 According to Sturge, confidence in the process is where the real difference lies in terms of using DOE. "With the DOE approach, there's a high degree of confidence that all of the possible factors affecting the process have been investigated," he says.

 "The potential benefits to process optimization are enormous," Sturge concludes. "The main way that the systematic approach of statistical DOE helps me is allowing me to look at a large number of process variables that otherwise might be unmanageable. It makes it possible to identify with confidence the small number of factors that are important to my processes. In this way I can spend more time optimizing these factors instead of wasting time trying to identify them."