SUPPORTING PRODUCTION PROCESS CONTROL AND ROBUSTNESS
When laboratory test results are reported into a data management system used for process monitoring, the corresponding assay
control results can be reported simultaneously to permit an overlay of production process data and test system control data.
This readily facilitates the monitoring of statistical process control data and analytical test system control data. Data
from these overlay charts provide several important process and analytical performance criteria. In brief, process and test
method performance can be monitored individually and immediately compared to each other. Analytical inaccuracy and imprecision
can be a major contributor to the observed process variability. This is particularly true for many bioassays used to test
for biopharmaceutical drug efficacy, purity, and stability. It is therefore extremely useful to monitor and relate the observed
process variability results to assay control results that are generated simultaneously by the same method. As stated earlier,
many outlier result investigations could be shortened and simplified because outcomes often are based on the investigation
results regarding whether a test result was valid or within the expected test system control range.
Figure 4 illustrates the overlay of sample results and the corresponding assay control results.10 The control and each sample were manufactured by the same process and are routinely tested using this quantitative limit
test. In Figure 4, a characterized and quantitated process impurity is directly monitored against the test system assay control.
One immediately observes that both process and analytical testing are not out of statistical control. However, the measured
process performance is not ideal — slightly less than three standard deviations (SD) for the distance of the process mean
(1.10%) to the specification limit of no more than 2.0%. Individual assay control and process data are correctly reported
in this chart to 1/10th of a percentile, based on the established significant digits in the specification limit. One also
readily recognizes that the observed process variability (SD=0.36%) includes a significant contribution from the day-to-day
assay variability (SD=0.25%) reflected in the assay control data.
We took this a step further by using the data from the overlay chart together with our mix-ing/blending study results from
the process development and validation data to estimate the actual process variability. A simplified relationship of main
factors contributing to the overall observed process variability is shown here (V = Variability):
[Vobserved for process]2 = [Vassay]2 + [Vsampling/batch uniformity]2 + [Vactual for process]2
The results are illustrated in Figure 5. One immediately recognizes that the actual or true process variability (SD=0.18%)
is relatively small when compared to the observed process variability. Given this, the easiest way to decrease the observed
process variability is to increase both the number of samples collected and increase the number of assays performed before
averaging all results. This significantly lowers the variability factors, assay variability, and sampling/batch uniformity
to a more tolerable level. We could probably bring our less-than-3-SD process much closer to a desirable (but currently not
measurable) 5-SD process. Although this would likely increase resource demands, mostly for the analytical testing, it would
mean less OOS results, leading to a more robust process and ultimately more product to market. In addition, an improved day-to-day
precision (intermediate precision) in analytical results converts to more reliable test results for measuring deliberate process
Figure 5. Contributing Factors to Process Variability
A good AMM program can indicate weak points within the overall process quality when control charts are combined. With AMM,
one can estimate the time, money, and effort needed to correct imprecise results and achieve more accurate data. Remember
that testing and production process adjustments are costly and can be risky when test results are not reliable.
Stephan O. Krause, Ph.D., is validation manager of QC assay support for Bayer HealthCare LLC, 800 Dwight Way, Berkeley, CA. 94701-1986, 510.705.4191, Fax: 510.705.5143, firstname.lastname@example.org