Biomedical Engineering Reference
In-Depth Information
performance. A two-sided confidence level (
a
)was set at 0.05 for this analysis. Thismeans
that when the p value is
0.05, the effect of an operational parameter on the related
performanceparameter is recognizedas statistically significant.However,whenp is
<
0.05,
the impact of operational parameter on the performance parameter is not statistically
significant. For the case study, analysis of product yield, product quality 1, and product
quality 2 data was performed. Prediction profiles based on regression analysis for each
operational parameter were generated and presented in Fig. 5.7. They illustrate how each
operational parameter affects each performance parameter within the characterization
range.Forexample, anincreaseinthetemperatureduringgrowth-phaseresults inadecrease
inproductyieldandanincreaseinbothproductqualityattributes(Fig.5.7a).Figure5.7band
c illustrates similar trends for the induction phase of the production fermentation step.
While the prediction profiles are useful in illustrating the trends, review of the scaled
estimates provides a more quantitative comparison. Figures 5.8-5.10 show the scaled
estimates for the three data sets presented in Fig. 5.7. Figure 5.11 further illustrates
the interaction effects that are observed in the process characterization studies. As
mentioned earlier, the p-value denotes the statistical significance of the effect while the
scaled estimate represents themagnitude of the effect. For process robustness, effectswere
examined that are both significant in magnitude and statistically significant. By defining
>
>
10% change from the mean for the scaled effect estimate to be considered significant in
magnitude, the following conclusions can be made upon review of the data in Figs. 5.8-
5.10. For the growth phase of the production fermentation (Fig. 5.8), effect of temperature
on product yield; effect of temperature, feed-1, and pH on product quality 1; and effect of
temperature and pHon product quality 2 are both statistically significant and of significant
magnitude. Further, first-order interactions between feed-1 and pH show a statistically
significant effect on both product yield and product quality 1 (Fig. 5.11a and b), but only
for product quality 1 is this interaction effect also significant in magnitude. For the
induction phase of the production fermentation (Figs. 5.9 and 5.10), effect of dissolved
oxygen on product quality 1 and pH on product quality 2 are both statistically significant
and of significant magnitude. Further, first-order interactions between feed-2 and pH have
a significant effect on product quality 2 (Fig. 5.11c), in contrast to the interaction effect
between Feed-2 and dissolved oxygen that only is statistically significant.
5.6 WORST CASE STUDIES
Aworst case study was conducted to demonstrate process robustness and to determine
whether process consistency and product quality performance would still be acceptable
when operating each unit at the edge of the normal operating range. On the basis of the
identification of operating parameters affecting product quality and process consistency
(including product yield), a set of best case andworst case cell paste product intermediate
was produced that exhibited either a high product titer and high purity or a low product
titer and low purity (Table 5.4). The worst case product intermediate was further purified
over an anion exchange (AEX) column. It was demonstrated that even under worst case
operating conditions the acceptance criteria for product quality 2 impurity levels after the
AEX purification step was never exceeded, assuring acceptable product quality of the
