Mastering Method Development: A Comprehensive Guide to Sampling Protocols and Robust CCI Testing

The first step following system characterization is to evaluate baseline sample behavior. Negative control characterization may also include master sample testing. Whether system characterization occurs with or without master samples, the primary objective is to test a sufficient number of negative samples to represent negative control behavior. When considering sampling, the number of samples is often only the first question. Clients must also determine whether their plan accounts for differences between production runs. Are they attempting to set up the same method for two packages that are actually handled differently? Any potential difference warrants increasing the sample size and evaluating samples separately. PTI recommends using at least 30 samples for a basic negative control set. This value is derived from the convergence point between a normal distribution and a Student’s distribution. Classically, 30 samples is considered the point where the Student’s distribution, designed for handling smaller sample sizes, begins to approximate the behavior of a normal distribution. This allows for the use of straightforward Gaussian analysis instead of a more complex model.

The Central Limit Theorem (CLT) provides an alternate basis for evaluating sample size sufficiency. According to the CLT, for a population characterized by a mean (μ) and standard deviation (σ), if sufficiently large random sample sets are taken from this population with replacement, the distribution of the sample set means will approach a normal distribution. However, if the underlying population is not normal and the sample set is too small, even the sample set means will not result in a normal distribution. The typical size recommendation remains 30 samples. This method is only effective if a representative set of test samples is provided.

Once the negative sample set is evaluated, appropriate confidence intervals may be computed using a normal distribution. Confidence intervals estimate a range of values likely to encompass an unknown population parameter. This estimated range is derived from specific sample data. In the context of Container Closure Integrity Testing (CCIT) , this sample data represents test results. Furthermore, the direction of failure is often known. This knowledge makes it possible to assess the probability of a false negative, which occurs when a test fails to detect a specified breach size in container integrity.

Positive control samples are also required to develop a method. Their purpose is to demonstrate that specific defect types are statistically distinguishable from the general population. Before selecting the quantity of positive controls, the types of positive controls must be appropriately selected. Leaks calibrated to a required size and leaks simulating natural defects may be necessary for positive control evaluation. A comprehensive risk assessment may help identify defect profiles and their underlying causes, forming the basis for a robust positive control strategy.

Defects fall into three categories: catastrophic, gross, and micro. Catastrophic defects are apparent upon visual inspection and are often associated with visible container damage. Gross defects, larger than 100 μm, may not be readily visible upon inspection. Micro defects refer to leaks smaller than gross defects. To effectively challenge the method, it is advisable to include four distinct positive control sizes: three sizes of micro leaks and one gross leak profile. This comprehensive strategy ensures that the testing method reliably detects various leak sizes and types, enhancing its robustness and validity of the results. Depending on the production or laboratory situation, testing for catastrophic defects may also be necessary.

Positive controls are evaluated with the assumption that a given defect type will perform consistently according to a normal distribution. Defect type refers to characteristics such as size, location, creation method, or defect channel length. PTI recommends a set of three sizes and 15 samples for each defect type. For mass-flow-based technologies, the assumption of normality aligns with the measurement method. Exceptions, such as alternative fluids or detection methods like visual inspection or electric current-based detection, require increasing the number of positive controls. Regardless of the assumption of normality, baseline signal variation for positive controls must be assumed to match the negative control baseline variation. Without this assumption the number of positive controls could not be reduced below 30 samples per defect type.

Positive control evaluations must also consider harmonization needs, allowable drift, and rejection limits. Recipes are typically established for long-term use or multi-site applications. A well-validated method with reasonable margins is essential. A general rule of thumb is to allow at least 1.5 standard deviations for long-term variability, on top of the classic six-sigma rejection criterion. For normally distributed detection technologies, long term variability should be added to the minimum of 3 to 6 standard deviations that should fall between the positive control sample average and the rejection limit. For non-normal distributions, such as high-voltage detection, PTI recommends using the lowest measured positive control value instead of the average. If additional margin is required, PTI suggests doubling the rejection limit as a target for lowest positive control result.

For processes where speed is critical or the limit of detection (LOD) is low, clients are encouraged to contact PTI for assistance. Methods using a well charactized system where the lowest positive control results range from 15–30 standard deviations above the negative sample average are considered robust. Clients using recipes with results below this range, especially for multi-site considerations, should conduct a careful risk assessment. Where defect type normality is well-characterized, or theoretically justified, positive sample sets may be reduced to 8–15 controls per defect type, compensated for by using a Student’s distribution evaluation.

Establishing a robust method for CCIT requires understanding baseline sample behavior, positive control evaluation, and appropriate statistical techniques. While Gaussian statistics, Student’s distributions, and the Central Limit Theorem provide valuable tools, thorough risk assessment is essential for evaluating defect profiles accurately. A carefully developed method supports long-term success in product line testing. Implementing a positive control strategy that includes various leak sizes enhances testing reliability, ensuring the detection of potential breaches in container integrity. This safeguards product quality and patient safety while improving robustness, transferability, and long-term applicability of recipes.

REFERENCES

1. Online MPH and Teaching Public Health Modules.
2. Valerie J. Easton and John H. McColl’s Statistics Glossary v1.1
3. What Is T-Distribution in Probability? How Do You Use It?