Compositing Samples and the Risk to Product Quality

Compositing involves taking samples from only one portion of units or containers, combining or blending the samples together to form a composite, and then taking a single sample from the composite and testing it once. The classic example of compositing is the receipt of, for example, 100 containers of powder. Rather than test all 100 containers, companies want to sample a few containers, mix material from several containers together, and perform only one test. Clearly, this type of sampling loses information about variability because compositing is physical averaging.

Lynn D. Torbeck

The usual reason to do a composite sample is to save the cost of sampling, testing, and documentation: not a statement FDA wants to hear. The agency's response would simply be, "That is the cost of compliance." From a regulatory point of view, there is a need to justify compositing versus testing all of the containers on some grounds other than just cost. This testing is probably best done using statistical arguments that the composite test is sufficient in the specific situation, that additional testing would not materially improve results, or decrease patient risk.

Compositing risks

Consideration of compositing raises several related topics. First, compositing is only done on incoming materials and not on in-process materials or on finished products. Second, compositing should not be done on crucial materials, precursors, APIs, anything that comes in direct contact with the API, in-process materials, or the finished product. The risk is too great. The probability may be low, but the consequence could be lethal to the patient if product quality is not met.

The first justification for compositing is that the risk to the patient is not materially increased by compositing. This justification is supported by the physical situation. Incoming materials generally include liquids, viscous liquids, pastes and emulsions, suspensions, powders, granulations, and solids. The nature of the material affects the sources and structure of the variability.

Take for example a tank car full of a nonviscous liquid. Generally, it is reasonable to assume that the liquid is homogeneous and one sample would be all that is needed. Thirty samples would not give more information about the liquid. The variability in the results would be from the test method itself. Now, suppose the tank car of liquid is contaminated with a heavier liquid that settles in the lowest point of the tank, usually the drain opening. If the single sample is taken from the top part of the tank, we would miss the contamination at the bottom. The sampling scheme must take into account the physical structure of the material and sources of variability.

The most common situation for compositing is the receipt of shipment of drums of a powder. A company wants to open as few of the containers as possible to take samples. The receiving company wants to mix those few samples into a blend and test one sample once. Clearly, all information about variability is lost. Further, if one of the samples that is taken and blended is low (or high) and out of specification, the blend may not be low (or high) enough to signal a difference. Again, the risk to product quality could be high.

Given 100 containers of a powder, many sources of variability can be identified within and between containers. Further, there are issues of deliberate fraud and counterfeit materials that add to possible variability. For example, drums may be filled with a cheap material such as lactose, and then only a small layer of API is put on the top. The common practice of only sampling from the top of the drum misses the lactose.

Skip-lot and skip-test sampling

Another practice that companies want to do is to do "skip-lot sampling" and literally skip sampling and testing some lots. Given the necessary assumptions, skip-lot testing could be supported statistically but not from a practical point. Again, the risk to product quality is too high.

In Europe, the regulatory agencies require at least an identity test on every drum or container received. That is good practice. That requirement should be expanded to specify that some samples from the middle and bottom of some containers should be taken, and not just samples from the top.

In the past, the industry has tried to argue for "skip-test sampling", meaing that each incoming lot of materials would be tested for identity but some lots would be skipped for a full battery of tests. Again, this type of sampling increases the risk to product quality and, therefore, the patient.

So, how to address this issue? As noted previously, a case needs to be built for compositing that does not include cost savings. This case would include how well do we know the supplier, what historical data can be used, what validation studies are available, and how is risk to the patient minimized? Statistically, good estimates of variability are needed because compositing averages out valuable variation.

Summary

Composite sampling may be acceptable when the material is known to be homogeneous or the variability structure is well estimated with high confidence. It is a bad approach, however, when used for critical materials, APIs, or in the absence of information about variability of the material.

Composite sampling is not prohibited by FDA, but it is suspect from the get go. It must be supported by data, facts, and documentation.

Lynn D. Torbeck is a statistician at PharmStat Consulting.
tel. 847.424.1314
LDTorbeck@PharmStat.comhttp://www.pharmstat.com