Experimental design and inferential modeling in pharmaceutical crystallization

A fractional factorial experimental design was used to investigate relative effects of operating conditions on the filtration resistance of a slurry p roduced in a pharmaceutical semicontinuous batch crystallizer. The six oper ating variables were seed type, seed amount, temperature, solvent ratio, ad dition time,and agitation intensity. An empirical model constructed between the operating variables and filtration resistance was used to define the f actor operating procedure, which reduced filtration time 3.7-fold. Several chemometric techniques were used to construct inferential models between th e in-process measurement of particle chord-length distribution and filtrati on resistance to help detect operational problems before completing the bat ch and decide when batch crystallization runs should end. Depending on the model quality criterion, the most popular chemometric methods of partial le ast squares and top-down principal-component regression can produce lower q uality models. Another chemometric approach, confidence-interval principal- component regression, predicted 70% more accurately than the best OLS model . The main effects and inferential models serve different but complementary roles in developing and implementing high-performance crystallization proc ess operations. A main-effects model constructed from statistical experimen tal design data determined optimal operating conditions rapidly, while th i nferential model can determine operational problems and batch end times dur ing batch-process operations.