The views presented in this document do not necessarily reflect those of the author’s respective companies or organizations. Furthermore, they should not be interpreted as being representative or endorsed by any regulatory authority.
Do we need blend uniformity – why is the CU testing with appropriate sampling plan not acceptable?
CGMP (21 CFR.211.110) requires an in-process testing of powder blends to demonstrate adequacy of mixing, but it does not state that the blend has to be directly assessed for uniformity. It was on this premise that the original draft stratified sampling guidance document allowed the use of in-process dosage unit data as a surrogate to demonstrate dosage unit uniformity during routine manufacture. Blend uniformity should be assessed during process design (Stage 1 Validation) and process qualification (Stage 2 Validation). Defaulting directly to the testing of in-process dosage units is discouraged, the exception being when blend sampling presents severe risks to the operators taking the samples (which should be discussed and accepted by regulators).
Is this an updated version of the withdrawn stratified sampling draft guidance document?
No. The BUCU’s publication has been put forth as potential modifications to the withdrawn draft stratified sampling guidance document, which FDA may (or may not) consider for future guidance.
What if I have too small a batch for 40 samples?
The sampling plans presented in the proposed approach are only one way to perform blend and content uniformity. As stated in the publication, other sampling plans may be used, if justified, including reduced quantities for smaller batches. However, sampling plans should be representative of the entire blender or batch.
Do I have to do the systematic sampling for the routine manufacturing release testing?
Systematic sampling plans covering the entire compression or filling operation (including beginning and end of batch samples) should be used during routine manufacture, if the results are also to be used to demonstrate blend homogeneity.
What if I cannot afford to purchase SAS to run the sampling routine?
Tables have been posted on the ISPE website for sampling plans discussed in the publications.
What is coverage? Routine manufacturing acceptance criteria (50/95) is looser. Does that mean I’m only 50% confident future samples will pass USP Testing?
In this context, the coverage is the probability of passing the USP test with a future or alternate sample from the batch. Therefore, 50/95 refers to having 50% confidence of having at least 95% probability of passing the USP test on a future/alternate sample from the batch. Therefore, 50% of the time, there will be at least a 95% probability that a future/alternate sample from the batch will pass the USP test and the other 50% of the time there is less than a 95% chance of passing the USP test. However, the probability will still be high (much higher than 50%) when the 50/95 criteria has been met. This can be observed graphically from Figures 10 through 15 in the technical paper, “Assessment of Blend and Content Uniformity. Technical Discussion of Sampling Plans and Application of ASTM E2709/E2810.”
When and for how long would enhanced testing be required during Continued Process Verification (Stage 3A) before reverting to routine testing? Until the results of development work lowers the SD and/or it levels off?
Stage 3 of validation - continued process verification - is often divided in two (3A and 3B), with 3A being heightened sampling and testing until sources of variability is understood, and 3B being routine monitoring of an understood, in-control process. If a product needed tier 2 testing (e.g., triplicate blend or 40x3 dosage units) during Process Qualification, additional samples may need to be analyzed during Continued Process Verification (Stage 3A). The number of dosage units that should be assayed depends on the magnitude of the risks associated with the process and product. If a product has standard deviations for the blend and/or dosage units that are consistently high, the number of dosage units may be comparable to sample sizes used during Process Qualification Stage 2. For products with moderate blend and/or dosage unit standard deviations, fewer dosage units may suffice.
Enhanced testing may be reduced if process changes are implemented that demonstrate an improvement in blend and/or dosage unit uniformity. Testing may also be reduced to after a due diligent effort to improve the process still results in high but consistent (and acceptable) standard deviations for the blend and/or dosage unit data, indicating that is the best uniformity that the process can achieve.
How much variability in the manufacturing process and analytical methods should be tolerated?
It is in a company’s best interest to minimize variability in the process and/or analytical methods, especially for processes that have SD’s > 3%. Analytical methods should be validated in accordance with CGMP criteria (including linearity, selectivity, limit of quantitation, precision, accuracy, range, repeatability, robustness, etc.). In general the underlying precision of analytical methods should be better than the variability target for the product (i.e. 3%).
Can CU have different levels of importance? For example, what impact does therapeutic index considerations have when defining specifications?
In some instances, tighter acceptance criteria may be necessary for narrow therapeutic drugs. Conversely, wider acceptance criteria could ensure adequate therapeutic effect for drugs with broad therapeutic ranges.
Reference: Bergum J., et. al., Current Events in Blend and Content Uniformity, Pharmaceutical Engineering, March/April 2014, Vol. 34, No. 2, p. 1-10.
What is the level of testing (batches, number of samples) expected during:
a. Stage 1 Process Design and Stage 2 Process Qualification?
During the manufacture of development, clinical and stability batches, scientists should seize the opportunity to generate data regarding the process’s ability to produce acceptable blend and dosage unit uniformity. The number of batches manufactured during process development will likely vary from product to product. Factors such as the availability of drug substance, number of batches required for development, clinical supplies and registration stability studies will impact the number of batches manufactured during Stage 1 Process Design. The number of locations sampled in the blend should be sufficient to adequately “map” it, and target problematic areas such as above discharge valves and along the center axis of rotation. Triplicate samples should be taken and assayed for each location to allow VCA to be performed on the data, which can then be used to improve formulation and/or process performance. The number of dosage unit samples to be tested should ensure comprehensive coverage of the batch. Batches made with bench-top or pilot scale equipment will likely require fewer sampling locations. Note that acceptance criteria depend on the sampling plan. Therefore, if data obtained from the sampling plan are evaluated using acceptance criteria, the criteria for blend and content uniformity should be appropriate for the Phase of development. For example, looser criteria may be acceptable during the manufacture of Phase 2 clinical supplies (e.g., tested against USP <905>). At Stage 2, the sampling plan should be chosen so that acceptable batches will meet the associated acceptance criteria. As the process is further developed and additional knowledge/experience gained, the acceptance criteria should approach that to be used for the commercial product.
b. During Stage 3 Continued Process Verification?
Stage 3 Continued Process Verification applies to the lifecycle of the product. A risk based approach should be used to define the level of testing used during Stage 3A. Products that have higher SDs for blend and dosage unit uniformity may require larger sample sizes. Other considerations that can impact Stage 3A sampling plans include the overall control strategy, use of SPC charts, the amount of confidence in the predictive powers of model(s) used to predict content uniformity.
When the acceptance criteria are easily met during Stage 2 Process Qualification or Stage 3A (if reduced testing is required based on Stage 2 results), the team’s framework for CU (and inferred BU) sampling and testing is to progress to Stage 3B process verification. For each production batch, one acceptable sampling plan is to take a sample of 30 dosage units across the batch. 10 of these dosage units, selected across the batch, should be tested and if needed, the additional 20 dosage units tested using Sampling Plan 1 or the company’s chosen acceptance criteria/methodology.
The level of testing required for post-approval changes to the formulation and/or manufacturing process that have the potential to impact blend or dosage unit uniformity should be comparable to sampling plans during Stage 2 Process Verification, rather than those used during routine production (Stage 3B Continued Process Verification).
Does quality refer to the probability of passing the USP or to the probability of falling between 85-115% or something else?
It refers to the uniformity of the product. This can be established by applying the probability of future samples taken from the batch passing USP <905> (ASTM method or tolerance interval approach), or the probability of individual values falling between a potency range such as 85-115% or any other defined range (tolerance interval approach).
How sensitive is the ASTM method to non-normal data?
The ASTM method is sensitive to non-normal data and the will generally catch a uniformity problem if present. If the data is representative of the batch as a whole but non-normal, the ASTM approach will still appropriately catch and reject the batch.
The OC curves to pass the USP UDU test and ASTM E2709/E2810 depend on the distribution. An investigation of non-normality requires assuming a specific non-normal distribution that would represent the dosage unit distribution. There are many different possible non-normal distributions. Some of these distributions may be well known (e.g. log-normal) with parameters that define the distribution (such as the mean and standard deviation that define the normal distribution) while others could be generated using a mixture of distributions. For example, the distribution could be normal for a high percentage of the data and take on another distribution for the remaining data. Once a non-normal distribution is specified, data from that distribution needs to be generated (probably by simulation) and tested against the USP UDU test. An OC curve showing the probability of passing the USP UDU test could then be constructed for that distribution. OC curves depend on assumptions of the distribution. For the normal distribution, there are parameters (mean and standard deviation) that are fixed and then the OC curve is generated. If the non-normal distribution has parameters, then a similar approach could be used. If the non-normal distribution is more complicated, then defining the OC curve becomes more difficult. Once the OC curve is generated for the non-normal distribution, the same non-normal distribution is used to generate data that would be compared to the ASTM E2709/E2810 acceptance limit tables which assumes that the data is normal for the confidence intervals. An OC curve would be generated to determine the probability of passing the ASTM E2709/E2810 acceptance limit table. The USP vs ASTM E2709/E2810 OC curves can then be compared to each other to evaluate the effect of the non-normal distribution. This exercise has not been performed for the ASTM E2709/E2810 methodology. If a non-normal distribution can be provided that makes sense for content uniformity, then this evaluation could be done. Based on anecdotal data, batches that fail the ASTM E2709/E2810 acceptance table have shown issues with the data (e.g., too much variability, mean far away from target, and non-normality due to skewed distribution or outlying observations). It is possible that for larger sample sizes, a single outlying observation could be present in the data but the data passes the acceptance limits. This supports the recommendation that the data be plotted to look for unusual observations/ extremely skewed distributions and/or other anomalies.
How can one draw conclusions for a batch of 1 million tablets from a sample of 10 tablets during routine manufacture?
Systematic sampling plans with a sufficient number of dosage units should be taken from specifically targeted locations in the compression or filing operation, including areas that have the greatest potential to yield extreme high or low values, that can monitor the manufacturing process outputs that are most responsible for causing finished product variability. Correlate systematic sampling of the dosage units with significant events in the blending process. Formulations that have extremely low dose and/or high potency may call for more rigorous sampling to assess blend and/or dosage unit uniformity.
A difference between dosage unit sampling plans presented in the ISPE BUCU papers and that contained in the withdrawn draft stratified sampling guidance document is with regards to the number of sampling locations and replicates taken within each location. The ISPE BUCU believes better information about the uniformity of the batch is achieved by sampling more locations (20x3 à 40x3 in the example) instead of more replicates from the same location (20x3 à 20x7). [Three replicates provide sufficient information for within location error; 40 locations provide superior information than 20 locations for content uniformity across the batch.]
This recommendation is built on the premise that by the time 10 tablets are used as a basis for determining uniformity, that many development and qualification batches have been produced, and that the processes are well understood, well monitored, and are under control. Therefore, aberrant units (e.g. over- or underweight, broken) will be identified by the control systems and removed from the population. In this case, the analysis of 10 tablets can provide sufficient assurance that the uniformity is achieved since the process is sufficiently similar to prior successful batches. Analysis of ten tablets alone to determine uniformity in an unknown process would likely be insufficient. How you take the samples is more important than the number of samples.
In addition to how the samples are taken, the criteria that are applied are important to the conclusions that can be made from the sample. Because the criteria are based on the sample size, an appropriately conservative criterion is built in for that sample size to account for random sample-to-sample variability. (It is assumed that the dosage units are normally distributed. If the sample or population is not normally distributed see Question 12 above for its effect on the conservativeness of the test.)
Can in-process dosage units be used as a surrogate for BUA?
The ISPE BUCU Group recommends the use of in-process dosage units (weight corrected if necessary) as a surrogate for blend uniformity analysis. CFR states that adequacy of mixing must be assessed, but it doesn’t have to be directly measured and hence allows the use of a surrogate test. Proper justification is necessary, which should include the comparison of blend uniformity results with dosage unit uniformity data (with or without weight correction) to support the approach. It should also be noted that the issues leading to the withdrawal of the draft stratified sampling guidance document Powder Blends and Finished Dosage Units — Stratified In-Process Dosage Unit Sampling and Assessment, October 2003) were not so much the use of in-process dosage unit uniformity as a surrogate for blend uniformity, but the acceptance criteria (which were based on USP <905>) that were used to pass/fail the batch.
Can in-process dosage units be used for batch release?
The results from the in-process dosage units can also be used for batch release testing (non-weight corrected). If the in-process dosage units are not the finished dosage form (e.g., tablet core vs. film coated tablet), content uniformity data for the in-process and finished dosage forms should be compared to demonstrate similarity.
Regarding in-process dosage unit sampling, when do you use Sampling Plan 1 and Sampling Plan 2?
During Stage 1 Process Design and Stage 2 Process Qualification there is a need (and FDA expectation) to understand both between location and within location variability. This knowledge will provide insight into both blend and dosage unit uniformity and where opportunities for improving blend/dosage unit uniformity should be focused, if needed. Once this variability is understood, then based on the amount of variability and level of risk to the patient the choice of Sampling Plan 1 or 2 can be made for Stage 3 process verification. For the vast majority of products that pass the ASTM criteria during Stage 2 process qualification, use of Sampling Plan 1 in Stage 3 is justified. However, if the between location variability is shown to be large, then it may be more appropriate to use the number of locations from process qualification and Sampling Plan 2 during Stage 3 Process Verification to assure adequate continued monitoring and control of the process. (Note: In coated tablet cases where Sampling Plan 2 is warranted for Stage 3, samples may need to be taken at compression since location information is typically lost during coating.)
Can the approach be used for liquids?
The validation concerns for solutions are similar to those for other blended products. Upon blending, solutions are homogeneous. For the evaluation of the finished solution dosage forms, weight uniformity may serve as a surrogate for content for single-unit products. Although the ISPE BUCU Group focused on powder blends and solid oral dosage forms, the approach may also have some application for suspensions, powders for reconstitution and semisolid dosage forms.
Did you use HAZOP (Hazard and Operability Analysis) to identify sampling locations?
The BUCU Group did not specify a specific approach to use during any risk assessment process to identify problematic areas to sample. Problematic areas within the process should be identified, and the systematic sampling scheme modified to cover those events. Any extra samples and data collected around critical events should be assessed separately.
If the BU standard deviation is >5.0, is BU acceptable or not acceptable? Do you progress to analysis of the dosage units?
During Stage 2 validation (process qualification), batches with BU SDs >5.0% fail if one cannot attribute the root cause of the high SD to be due to be non-mixing/material related (such as blend sampling bias or an analytical error) and supported by dosage units data. If the standard deviation for the blend is > 3.0, replicate blend samples should be analyzed and an investigation into the cause of the high standard deviation, in particular to see if it is due to a non-blending related cause (such as sampling error or analytical error). Repeatability and reproducibility tests should be conducted to assess variability in the analytical methods. VCA should be performed on the blend and dosage unit data as part of the investigation, to determine within location and between location variance. Situations where within location error is high for the blend, but low for dosage units may imply sampling bias. If the within error component is high for both the blend and dosage units, micro mixing issues exist (possibly poor dispersion, segregation or agglomeration). If between location variances are high, the blend and dosage units have poor uniformity. Regardless of the fate of the batch, scientists companies can use the analysis on blend and dosage unit samples to identify underlying causes for poor content uniformity and make process improvements.
Why can’t we just go directly to testing dosage units?
The ISPE BUCU Group believes that a good faith effort to assess BU should be done, except in instances where exposure to the blend could put operators at risk. Valid BU data (i.e. with minimal sampling errors) is the best way to assess the performance of the blending process. FDA inspectors found too many companies were not putting forth a good faith effort to access BU (including the development of a suitable sampling technique considering sample size, thief type, etc.) and immediately defaulting to IPDUs, which was one of the reasons for withdrawing the draft guidance document.
The Agency currently recommends that all replicate samples taken from various locations in the blender be evaluated to perform a statistically valid analysis. The Group recommends only assaying one sample per location if SD is < 3.0% following the analysis of one sample from each point. Will this be acceptable?
The ISPE BUCU Group acknowledges the FDA Q&A published in August 2013, which recommends that all replicate samples taken from various locations in the blender be evaluated to perform a statistically valid analysis. However, the ISPE BUCU Group feels that this is unnecessary if after the analysis of the first sample from each location, the SD is <3.0. VCA performed on all samples (including the remaining replicates) would only be assessing random variability and noise.
Is the Group’s recommendation the only option to assess BU/CU?
As stated in the paper, the statistical approach, acceptance criteria and sampling plans are but one way to assess blend and dosage unit uniformity. Other statistical approaches and the use of PAT/large n for blend end-point detection should be acceptable.
What are the plans to go global with the approach?
The manuscripts are directed at a common activity that can be applied in a global setting. It should be noted that the PQRI Blend Uniformity Working Group’s paper (2003) that served as the foundation for the FDA draft stratified sampling approach that was subsequently withdrawn was referenced in many regulatory filings and approved in regions outside of the US. Since the technical issues are universal, it would be highly desirable to harmonize blend and content uniformity testing with Japan and the EU. This could be initiated through discussions within ICH. The challenge with the current harmonized <905> procedure is that it uses fixed sample sizes. Greater flexibility is desired to allow the use of manufacturing data to demonstrate a high probability of passing UDU perhaps without destructive analysis of any samples.
How do you identify blend sampling bias?
Publications exist in the pharmaceutical literature that describe ways to identify blend sampling bias. Also see the response in #21 to see how VCA can be used to identify sampling bias.
What is the difference between systematic and stratified sampling?
As stated in the paper, stratified sampling plans partition the batch into “strata” (e.g., first 1/20, second 1/20, …, and final 1/20). The combination of all strata must cover the entire batch. Then random sampling is performed within each stratum. Systematic sampling is performed by taking a sample(s) at equal intervals throughout the batch typically by total number of dosage units or manufacturing time.
For routine manufacturing (3B), can a company pick any nx1 for stage 1 and 2 testing? So rather than 10x1 then 30x1, could a company go to 30x1 then 90x1? What is the rationale then for two stages of testing - why not one and done?
The sampling plans provided in the publications are only meant to be examples for demonstrative purposes. Alternative sampling plans, if justified, should be allowed. When defining sampling plans, it is advised to use a risk based approach that balances consumer and producer’s risk. The two stage testing approach is consistent with USP <905>, and affords a second chance to pass the acceptance criteria if the initial sample fails. Individuals can determine the sampling plan they wish to use based on the expected variability of the blend/dosage units.
What other statistical approaches are available that can be used in place of the ASTM method? How do they fit into the Group’s recommendation?
Other statistical approaches based on tolerance intervals and Bayesian statistics exist and/or are being developed. Alternate statistical approaches, acceptance criteria, and sampling plans can be used by insertion into the appropriate boxes in the framework identified by the ISPE BUCU Group, with justification. Process analytical technology (such as NIR monitoring of blends and/or tablets, with corresponding acceptance criteria) is also an acceptable approach to assess blend and content uniformity.
Relevant Information From: “Questions and Answers on Current Good Manufacturing Practices, Good Guidance Practices, Level 2 Guidance - Production and Process Controls”