Specifications versus control limits
Using control limits as specifications will hinder a manufacturer's ability to monitor product and to make process improvements.
The advantage of maintaining a separation between specifications and control limits is that the customer is adequately protected
through properly defined specifications, and the manufacturer is protected from discarding product that is fit for use. This
distinction promotes attention to process shifts or trends without affecting product distribution, and the manufacturer is
encouraged to improve its process to reap the benefits of improved process capability. Improved process capability means fewer
failures, and thus greater capacity to provide safe and effective product to the market.
Furthermore, the vision of Quality by Design and design space cannot be achieved when control limits are used as specifications.
The design space may be viewed as the region of process settings that yields acceptable product (i.e., product that meets
specifications). When control limits are used as specifications, the design space reverts to the control space for the process,
leaving no opportunity for process improvement.
The ideal relationship between limits and process spaces is illustrated in Figure 2. Two response factors (X1 and X2) are studied across the knowledge space, yielding a response surface in the quality attribute (panel 1). The response surface
intersects the lower specification limits (LSL, panel 2) and upper specification limits (USL, panel 3) to yield the design
space (panel 4). The control space represents a normal operating range for the factors, falling well within the design space
(panel 5). Product that is manufactured within this control space will yield measurements falling within the upper and lower
control limits (UCL and LCL in panel 6). As long as a control space falls within the design space, the associated control
limits will fall within the specifications, yielding good process capability and the opportunity for process improvement.
Interpreting data from annual stability studies and process validation
Another conflict exists in the guidance documents regarding establishing shelf life and interpreting individual stability
results. ICH Q1E, Evaluation of Stability Data promotes the use of statistical methods for establishing the shelf life of the product. As mentioned previously, this approach
uses the confidence bound on the mean regression line. However, individual stability measurements are more variable than is
predicted by the confidence bound, and are thus likely to become OOS throughout shelf life. This is illustrated in Figure
3. Following ICH Q1E, a regression analysis is performed on development stability data, yielding a lower 95% confidence bound
(dashed curve) that intersects a minimum requirement for potency (dotted line), predicting 24-month shelf life for the product.
However, measurements from an annual stability lot are not restricted by the confidence bound but instead by a prediction
bound (dotted curve). The probability of experiencing one or more OOS measurements throughout the shelf life of the annual
lot is predicted from this example assuming the annual lot degrades like the development lot, and the assay variability is
the same as that observed from the development study. The proportions of measurements that are predicted to fall below the
minimum requirement at 12, 18, and 24 months are 5%, 9%, and 18% respectively. The probability of experiencing one or more
OOS measurements is calculated as:
=1 – Prob (No OOS)
=1 – Prob (No OOS at 12 mos.)
* Prob (No OOS at 18 mos.)
* Prob (No OOS at 24 mos.)
=1 – 0.95 * 0.91 * 0.82 = 0.29
Thus, in this case, material that was judged to have 24-month shelf life by ICH Q1E has a ~30% probability of yielding an
OOS result over the course of an annual stability study.