3. Take action that is appropriate to the type of attribute and the purpose of different types of limits.
The action that should be taken when a quality attribute exceeds its specifications is not the same as the action that should
be taken when an attribute exceeds its control limits. Exceeding a specification signals that the product may not be fit for
use, while exceeding a control limit signals that the process may not be in adequate control and should be monitored for shifts
Figure 6 compares the actions that might be taken when a quality attribute exceeds a specification or control limit, depending
on whether the attribute is linked to fitness for use or is a consistency attribute used to monitor the product or the process.
Following down the left side of the flow chart, a quality attribute that is linked to fitness for use may have specifications that were defined during development as well
as control limits that have been updated throughout the lifecycle of the product. Failure to meet specifications results in
a formal OOS investigation, which may result in withholding the lot or withdrawing it from the market. A lot that meets specifications
but fails to meet its control limits will result in an OOT investigation. This may result in corrective action when an assignable
cause has been linked to the OOT event.
Figure 6. Flow chart that compares the actions that might be taken with a quality attribute that is linked to fitness for
use with those that might be taken with a consistency attribute that is used to monitor the product or the process.
On the other hand (following down the right side of the flow chart), a quality attribute that is not linked to fitness for
use but is used to monitor product or process consistency may have control limits that are updated throughout the lifecycle
of the product. A failure to meet control limits will result in an OOT investigation. That investigation may result in corrective
action when an assignable cause has been linked to the OOT event.
4. Treat post-marketing stability studies as a control strategy for the process, rather than an assessment of the lot's stability.
Annual post-licensure stability studies should be designed to monitor stability characteristics of the product, such as changes
in a quality attribute, rather than the level of the quality attribute at points throughout shelf life.
The evaluation of a pharmaceutical products does not end with its approval for marketing. After marketing, sponsors monitor
their products to ensure that they conform to their experience during development. The US Food and Drug Administration generally
requires a commitment to monitor the stability of at least one lot of each approved product produced each year. This approach
to ensuring post-licensure stability could potentially be enhanced to provide increased confidence that the manufacturer produces
product that is safe and efficacious until its expiry. This must be balanced, however, against the risk of a program that
would yield false signals of product instability, resulting in unnecessary effort on the part of industry and regulatory agencies
to ascertain the cause.
The industry and regulators may wish to consider novel models that combine strategic design and analysis for monitoring post-licensure
stability to detect meaningful changes in the stability profile of the product. These models require a shift from the historical post-licensure stability paradigm of requiring that individual stability time-point measurements
meet expiry specifications to a new approach of stability monitoring in which common statistical techniques are used to evaluate
the composite data representing product on the market. A standard regression analysis can be used either to obtain slopes
that can be monitored for shifts or trends in the stability characteristics of the product, or to model ongoing lots for shelf-life
characteristics such as predicted potency at expiry.17
Such a paradigm of stability monitoring promotes data collection. When the data are analyzed using regression analysis, the
inclusion of additional time points or lots provides a more precise estimate of product stability, with less risk to the customer
and manufacturer alike. In contrast, when time-point results are held to specifications, there is an increased risk of "failure"
because of repeat testing, and thus a disincentive for data collection.
The following summarizes the potential opportunities that can result from the design, analysis, and interpretation of annual
stability studies that promote data collection to achieve better control of product on the market:
- One or more commercial lots may be enrolled into the post-licensure stability program. A single lot may be used if product
is stable, or if there is an adequate range between product release and predicted value at expiry. More lots might be enrolled
when there is greater risk that commercial lots may fall below their specification limit at expiry.
- Standard stability time points, as described in regulatory guidelines, need not be used, if this approach is balanced against
the total number of lots on stability. Studying more lots with fewer time points per lot provides a more representative profile
of product on the market than studying more time points on a single lot.
- Data from a single lot, or from all lots currently on stability, might be analyzed using common statistical analysis techniques.
The predicted value from such an analysis will yield a better estimate of the true value of a lot over time than an individual
stability time point measurement will, and the combined analysis from lots on stability provide a reliable forecast of the
quality of product on the market.
- The annual stability program might come to be recognized as "quality control on the product," rather than a study of the particular
lots on stability. In this manner, the annual stability program becomes a part of the overall product quality system.
5. Use statistical approaches, where possible, to enhance the design of a study and to obtain reliable product understanding.
Quality by Design (QbD) means that product and process performance characteristics are scientifically designed to meet specific
objectives, not merely empirically derived from performance of test batches. Statistical methods should be used to enhance
the design of manufacturing experiments and to make optimal use of the experimental information.
Multifactor design of experiments (DoE) can be used to identify parameters that affect the process or to define the design
space for the process. Proper analysis of the data provides scientific understanding and empirical confirmation of the process.
Mathematical modeling of the factor responses helps reveal process parameters that affect a critical quality attribute, and
thereby offers a basis for maintaining product quality. Such modeling provides more value than evaluating conformance lots
Likewise, stability studies should be designed to meet the goal of the evaluation. Typically, the goal of the stability study
is to estimate the quality attribute at significant points in time (e.g., at expiry) or the degradation rate at meaningful
temperatures. In these cases, the study should be designed and analyzed to obtain the most reliable (i.e., least variable)
estimate of the parameter of interest. A study to predict the quality attribute at significant points in time will select
intervals that adequately represent the targeted region and will use common statistical methods to forecast response. A study
of the effect of temperature or physical exposure might ideally compare results only at the beginning and end of the exposure
to obtain the most reliable estimate of the effect. These goals, as well as those of validation, are best met using simple
statistical methods rather than comparing results to specifications. Holding stability and validation results to specifications
engenders a disincentive for collecting valuable product and process information.