Summary of Data-Mining Effort - PQRIpqri.org/wp-content/uploads/2015/08/pdf/bu032802.pdf · from blend sampling requirements, ... 2. 1X to ≤3X, ... Product Quality Research Institute

Results of Statistical Analysis of Blend and Dosage Unit Content Uniformity Data Obtained from the Product

Quality Research Institute Blend Uniformity Working Group Data-Mining Effort

Garth Boehm, Jon Clark, John Dietrick, Thomas Garcia, Muralidhara Gavini, Jean-Marie Geoffroy, Pedro Jimenez, Gerald Mergen, Fernando Muzzio, Jerry Planchard, James Prescott, Jozef Timmermans, Neeru Takiar, David Whiteman I. Introduction In an effort to enhance the assurance of product uniformity and to provide potential relief from blend sampling requirements, the Blend Uniformity Working Group (BUWG) of the Product Quality Research Institute (PQRI) has drafted a recommendation that describes an alternative approach for the assessment of the content uniformity of powder mixtures and the resulting dosage units. Although the sample sizes and acceptance criteria stated in this proposal have been derived through computer simulations, the BUWG called upon pharmaceutical companies to submit content uniformity data for blends, in-process dosage unit samples, and finished dosage forms to challenge the performance of the acceptance criteria stated in the recommendation. The data were also used to test the hypothesis; “blend uniformity testing in routine manufacture is not predictive of the uniformity of dosage units.” In response to concerns expressed by ANDA applicants regarding inconsistencies in review chemists' recommendations to demonstrate adequacy of mix, the FDA issued the draft document Guidance for Industry, ANDAs: Blend Uniformity Analysis (August 3, 1999). Both the generic and brand industries raised a number of concerns following the issuance of this document. As a result of industry feedback on this draft guidance, one of the primary goals of the BUWG was to address the gap between scientific principles and the regulatory policy stated in this document. In September 2000, the BUWG sponsored a workshop on blend uniformity. At the conclusion of the workshop, it was recognized that limitations in current sampling technology and subsequent handling (powder segregation) might limit the effectiveness of using blend sample analysis to ensure adequacy of mix. Alternative solutions were sought to address the shortcomings associated with sampling and analyzing powder blends. The BUWG drafted a recommendation (Appendix 1) which provides an alternative approach that collectively demonstrates the uniformity of both powder blends and their corresponding dosage units. The document addresses key stages in the product lifecycle of solid oral drug products, including process development, validation and routine production. This recommendation does not apply to those drug products where the determination of dosage form uniformity by weight variation is allowed, and may only be applied to products in which the active ingredients are introduced into the powder blend. The approach described in this recommendation is being proposed to satisfy the cGMP requirement for in-process testing to demonstrate adequacy of mix, as well as USP testing

Results of Statistical Analysis of Blend and Dosage Unit Content Uniformity Data Obtained from the Product Quality Research Institute Blend Uniformity Working Group Data-Mining Effort March 28, 2002 Page 1 of 37

of finished dosage forms for content uniformity. Alternatively, traditionally employed methods (such as the direct sampling and analysis of powder blends, in conjunction with content uniformity testing of finished dosage forms) may continue to be used to satisfy cGMP and compendial release requirements. The elements of the BUWG recommendation were established using computer simulations. Data mining was conducted to check and further substantiate the elements of the recommendation. The acquired data was used to assess whether or not blend sampling is predictive of the content uniformity for the dosage forms during routine production, and if so, identify those instances where it provides beneficial information.

II. Submission of Data

A. Process Overview for Data Submission Data was electronically submitted to the PQRI Executive Secretary according to the format described in the template contained in Attachment 2. Companies were asked to electronically submit data in a blinded format by the close of business September 15, 2001. The BUWG asked that products having manufacturing excesses not be submitted as they would have further complicated the analyses. The PQRI Executive Secretary confirmed receipt of the data and verified that it was not submitted by a fraudulent source. Prior to forwarding the data onward for analysis, information regarding the source of the data was removed by the PQRI Executive Secretary. As a result of the above precautions, the identity of the contributing company was blinded to all individuals and organizations performing data analyses.

B. Template for the Submission of Data Appendix 2 contains the template used for the submission of data. A letter designation was used to codify each product (such as “A, B or C). For products with essentially exact manufacturing formulas but slightly different strengths (e.g. a 2 and 4 mg strength of the same compound where the formula differs only in drug quantity or fill weight), the same letter designation was used, but appropriate clarifying information was entered in the spreadsheet (e.g. adjust the final dosage weight or active concentration). Each lot of a given product was labeled consecutively, starting with 1. The amount of active as a percentage of total fill weight was entered. Information regarding the dosage unit weight (mg) and the amount of active (mg) contained in the dosage unit was also requested. Batch size was entered in kilograms. One of four methods of product manufacture was entered: dry blend, wet granulation, dry granulation, or other. Information regarding the type of dosage form (tablet, capsule, or powder fill) and the final dosage weight (mg) was requested. For a powder fill dosage form, the weight (mg) of the entire contents including overages was requested.

C. Notes for In-Process Blend Testing Results The sample locations for each blend sample were numbered. Multiple containers were labeled as independent locations. A multiple bin or drum process where multiple samples Results of Statistical Analysis of Blend and Dosage Unit Content Uniformity Data Obtained from the Product Quality Research Institute Blend Uniformity Working Group Data-Mining Effort March 28, 2002 Page 2 of 37

are retrieved from each container was addressed by labeling the location for each container as its own separate entry. Sample naming locations within a lot were not duplicated unless multiple samples were taken from the same location of the same container. For example, a four-bin process where 10 samples are retrieved from each bin would have sampling locations labeled from 1-40. If location 2 of the 40 samples was sampled 3 times, all three samples would be labeled as coming from location 2. The size of the blend sample was expressed as a function of the final dosage unit weight. The choices were:

1. <1X, 2. 1X to ≤3X, 3. >3X but ≤5X, 4. >5X but ≤10X, and 5. >10X

where X refers to the weight of the final dosage unit (including manufacturing overages). It was requested that blend uniformity results be reported for individual samples only (not as average results), and expressed as a percentage of the target quantity. Products containing manufacturing excesses were not solicited.

D. Notes for In-Process Dosage Unit Results The sample location was expressed as a percentage of the entire compression, encapsulation, or filling operation. For example, a lot that is sampled at 11 regular locations throughout compression would have sample locations of 0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%. Where sampling was performed at the beginning, middle and end of a batch, the locations were labeled as 0%, 50% and 100%, respectively. If multiple dosage unit results were determined from a single location, all were labeled with the same sampling location. For example, when four tablets are tested at the 20% time point, all results were labeled with 20% and individual values reported. Results were requested for individual dosage units only (not average results).

E. Notes for Product Release Results Individual dosage unit content uniformity results (not average results) were reported as a percentage of the label claim for each lot submitted.

III. Summary of Data Submitted Eight pharmaceutical companies responded to the BUWG call for data, and submitted a total of 149 batches. Table 1 summarizes the characteristics of the submitted data. Although the BUWG would have liked to have data meeting all possible combinations of dosage forms, percent active, manufacturing process, and batch size, the group realized that this was unlikely prior to the call for data. Therefore, the intent of the exercise was to check the validity of the sampling plans and acceptance criteria derived through computer simulations. The BUWG believes that the batches provide a representative sample of industry data as a wide range of RSD values were obtained. Results of Statistical Analysis of Blend and Dosage Unit Content Uniformity Data Obtained from the Product Quality Research Institute Blend Uniformity Working Group Data-Mining Effort March 28, 2002 Page 3 of 37

All data submitted were for tablet dosage forms. No capsule or powder fill data were submitted. Data from wet granulation, dry granulation and direct compression processes were received for products with a wide range of drug concentrations in the formulations. Batch sizes for the data submitted ranged from 24-2382 kg with a median of 500 kg. Table 1. Summary of Batch Characteristics for Submitted Data A. Percent Active in Formulation

% Active in Formulation <5% 5-15% 15-25% >25%

Number of Batches

75 21 9 44

B. Manufacturing Process

Method of Manufacture Direct Compression Wet Granulation Dry Granulation

Number of Batches 12 67 70 C. Type of Dosage Form

Product Tablets Capsules Other

Number of Batches 149 0 0 D. Batch Size

Batch Size (kg) ≤100 101-400 >400

Number of Batches 24 32 93 E. Breakdown by Company Company 1 2 3 4 5 6 7 8 Number of Batches Submitted

4

7

9

21

16

60

14

18


IV. Results and Discussion A. Validation of Assumptions Used for Computer Simulations The BUWG has proposed a series of criteria in their stratified sampling recommendation (Attachment 1) to address concerns raised following the issuance of the FDA document Guidance for Industry, ANDAs: Blend Uniformity Analysis, August 3, 1999. In establishing the parameters of the criteria stated in this recommendation (i.e., number of locations sampled, number of assays per location, maximum allowable RSD, etc.), extensive use was made of Monte Carlo simulation. The use of this methodology allowed a balance to be achieved between the need for sufficient assurance of adequacy of mix, and the efficient use of analytical resources. In these evaluations it was assumed that a stratified sample of the blend and dosage units were taken (in the case of validation studies) or a stratified sample of only the dosage units (in the case of routine manufacture). Underlying the computer simulation was the assumption that dosage-form content uniformity data (both between and within location) are normally distributed. Therefore, the first objective of this effort was to evaluate the appropriateness of these assumptions for the simulation of product data by the examination of validation data submitted from industry sources. The Shapiro-Wilk test for normality was used to evaluate the appropriateness of the normality assumptions used for Monte Carlo simulation. This test has been shown to provide a superior omnibus indicator of non-normality judged over the various symmetric, asymmetric, short- and long-tailed alternatives and over all the sample sizes used [Shapiro, Wilk & Chen, "A Comparative Study of Various Tests of Normality," Journal of the American Statistical Association, 1968.] Distance tests (those that make use of the empirical distribution function) are typically inferior in sensitivity against continuous distribution alternatives, with few exceptions. Power comparisons on W-statistic (Shapiro-Wilk), D-statistic (D'Agostino), Lilliefors' test, Anderson-Darling statistic and others have shown that although the D-test is quite powerful for symmetric long tailed alternative hypotheses, its extreme insensitivity to some other hypotheses from other groups indicates that it does not appear to be an omnibus test for normality. The Lilliefors test is an omnibus test but its power is generally inferior to the other tests. The W-test is a powerful omnibus test of normality, i.e., it has reasonable power over a broad range of alternative hypotheses. Power studies indicate that the W-test is more powerful than the other tests. Therefore, based on this information, the Shapiro-Wilk test was the choice when the alternative is not specified.

Normality of Between Location Means

The simulations performed by the BUWG assumed that the location means within a dosage-form batch were normally distributed. The Wilk-Shapiro test for normality was applied to the location means for each batch that possessed at least 10 locations.


Although the test for normality could be conducted with fewer than 10 location means, it was thought that at least 10 were needed to be representative of the process. Table 2 presents the number of batches with at least 10 location means, the number of batches rejecting normality, and the percent of batches rejecting normality. Table 2. Summary of Normality Testing for Between Location Means

Company Number of batches with

at least 10 location means

Number of batches with Shapiro-Wilk OSL ≤

0.01

Percent of batches with Shapiro-Wilk OSL ≤

0.01

C1 1 0 0%

C2 4 1 25%

C3 0 0 -

C4 12 1 8%

C5 16 2 13%

C6 25 2 8%

C7 12 1 8%

C8 18 3 17%

Total 88 10 11% Even though the hypothesis of normality is rejected at α = 0.01, the data have very little variability and the consequences of non-normality on the simulations would be minimal.

About 11% of the batches had at least one location mean that was statistically different from the remaining location means. Of the 9 batches with location means that are distributed in a non-normal fashion, 7 of the batches had either initial (start-up) or final (run-out) means that were dramatically different from the other location means resulting in normality being rejected. The above analysis concludes that using the normal distribution to generate batch location means to perform computer simulations to estimate criteria rejection rates will yield rejection rate estimates that are slightly smaller (more conservative) than criteria rejection rates based on actual data. It should be noted though that the majority of batches have location means that are normally distributed, thereby justifying the simulation assumptions.

Normality of Within Location Data

The simulations performed by the BUWG also assumed that the within-location data for a given dosage form batch were normally distributed. To be able to pool the data across the various locations within a batch, the data from each location within a batch were


translated by subtracting the location mean from each value at that location. The Shapiro-Wilk test for normality was applied to the translated data for each batch that possessed at least 10 values. Since the values at each location were translated by subtracting the estimated location mean, a degree of freedom for each location in the batch should have been subtracted from the reported degrees of freedom. The reported OSL (or P-value) is smaller than the correct OSL due to the reported degrees of freedom being larger than the correct degrees of freedom. Table 3 summarizes the results of this testing for the number of batches with at least 10 location means, the number of batches rejecting normality, and the percent of batches rejecting normality. Table 3. Summary of Normality Testing for Within Location Means

Company Number of batches with at least 10 values

Number of batches with Wilk-Shapiro OSL ≤

0.01

Percent of batches with Wilk-Shapiro OSL ≤

0.01

C1 4 1 25%

C2 7 1 14%

C3 9 0 0%

C4 21 7 33%

C5 16 4 25%

C6 60 6 10%

C7 14 2 14%

C8 18 2 11%

Total 149 23 15%

Overall, about 15% (10%-33% for individual companies) of the batches had at least one transformed value that was statistically different from the remaining batch transformed values. Typically, a single anomalous transformed value occurred at a single location within the batch about 5-6 standard deviations from the location mean. In a few cases, normality was rejected due to the shape of the distribution (skewness and kurtosis). However, in most of the cases, normality was rejected due to the single anomalous value. The anomalous value occurred at no particular batch location. The implications of the above analysis are that using the normal distribution to generate individual values at specific batch locations to perform computer simulations to estimate criteria rejection rates will yield rejection rate estimates that are slightly smaller (more conservative) than criteria rejection rates based on actual data. It should be noted though that the majority of batches have within location data that are normally distributed, thereby justifying the simulation assumptions.


B. Evaluation of the Hypotheses, “Blend Uniformity Testing in Routine Manufacture is Not Predictive of the Uniformity of Dosage Units”

The second objective of the analysis was a multi-part evaluation of the hypothesis that blend-uniformity testing in routine manufacturing is not predictive of the uniformity of dosage units. This hypothesis was evaluated by examining the predictive value of blend data variability in determining dosage unit data variability. It should be expected that in an observational study such as this one that the sample sizes do not match up with the criteria sample size requirements. For example, the USP Content Uniformity test calls for 10 units to be tested first and then augmented with another 20 units if the first stage test does not pass. In this observational study, the number of units tested could be quite large with no regard to first or second stage. The effect on how the various criteria perform depends on the criteria themselves. The multi-part analysis was conducted on the submitted industry data as follows:

a. Evaluation of the correlation between blend and product data for batches having low blend variability (RSD’s < 3.0%), moderate blend variability (RSD’s between 3.0% and 5.0%), and higher blend variability (RSD’s > 5.0%).

b. Determining if the blend data meet either the OGD acceptance criteria (RSD ≤ 5% with the average between 90% and 110%) or the FDA validation criteria (RSD ≤ 5% with each individual value between 90% and 110%) for each batch in the database. Additionally, determining if the individual sample dosage units provide a high degree of assurance that compendium requirements for content uniformity can be consistently met for each batch in the database.

c. Evaluation of each batch to determine if the following criteria are met: BUWG Validation Criteria, both blend and dosage-form1 BUWG Routine Criteria readily pass Stage 1 BUWG Routine Criteria readily pass Stage 2 or marginally pass USP Content Uniformity 2 ICH Content Uniformity 3 Criteria proposed in PDA Technical Report No. 25 4 Office of Generic Drug (OGD) Blend Uniformity Analysis Guidance 5 FDA Blend Validation Guidance 6

1 PQRI Blend Uniformity Working Group’s draft proposal- “THE USE OF STRATIFIED SAMPLING OF BLEND AND DOSAGE UNITS TO DEMONSTRATE ADEQUACY OF MIX FOR POWDER BLENDS” (see Appendix 1)

2 USP 23, <905> Uniformity of Dosage Units, United States Pharmacopeial Convention, pp. 1838-1839. 3 Pharmacopeial Forum, Vol. 27(3), May-June 2001, United States Pharmacopeial Convention, pp 2615-2619.

4 Blend Uniformity Analysis: Validation and In-Process Testing, PDA Technical Report No. 25, Supplement, Vol. 51, No S3, 1997.

5 Guidance for Industry, ANDAs: Blend Uniformity Analysis, (Internet) http://www/fda.gov/cder/guidance/index.htm, FDA CDER Office of Generic Drugs, August 1999.


6 J. Dietrick, presented at a “Special Forum on Blend Analysis,” sponsored by PDA, Rockville, MD, January 1996.

http://www/fda.gov/cder/guidance/index.htm

Eight companies submitted data that were used in the analyses. All dosage-form data were from tablets. Of the 8 companies that originally submitted data, only one (Company 5) submitted weights for use in weight correction. Because of this, no weight correction was performed on the comparative analyses. However, an analysis of Company 5 data for the BUWG Validation Dosage-Form criteria was conducted using the weight-corrected data to evaluate the effect on the analysis. Those results are presented and discussed in the following section, Table 7. Table 4 provides the details of the blend and dosage unit criteria as modified in these analyses due to the nature of the data. Appendix 3 provides the details of the criteria as they were implemented.

Table 4. Criteria as Used in the Analyses Criterion Blend Dosage Unit

BUWG Validation

• Use all data • Stage 1 only • RSD ≤ 5.0% • Each value must be within 10

percentage points of the mean

• Data are not weight corrected • Number of locations ≥ 10 • Each location mean must be 90% - 110% • Each value must be 75% - 125% • RSD ≤ 4% ⇒ Readily Pass • 4 < RSD ≤ 6% ⇒ Marginally Pass

BUWG Routine- Readily Pass Stage 1

• Data are not weight corrected • Use all data • Mean must be 90% - 110% • RSD ≤ 5%

BUWG Routine- Readily Pass Stage 2 & Marginally Pass

• Data are not weight corrected • Use all data • Mean must be 90% - 110% • RSD ≤ 6%

USP Content Uniformity • Use all data • Stage 2 only

ICH Content Uniformity • Use all data • Stage 2 only

PDA Tech Report #25 • Use all data • SDPI ≤ 6% (Stage 1 only)

• Number of locations ≥ 10 • Bergum7 SP28 90/95

Office of Generic Drug (OGD) Blend Uniformity Analysis Guidance

• Use all data • Mean must be 90% - 110% • RSD ≤ 5%

FDA Blend Validation Guidance6

• Use all data • Each value must be 90% - 110% • RSD ≤ 5%

7 Bergum, J. S., Constructing Acceptance Limits for Multiple Stage Tests, Drug Development and Industrial Pharmacy, 16(14), 2153-2166 (1990).

8 A Bergum SP2 sampling plan is similar to a stratified sampling plan except that the strata may be randomly selected. The strata are the locations within the blend or production of product.


Relationship Between Blend & Dosage-Form RSD

The relationship between the blend relative standard deviation (RSD) and the RSD values for the corresponding dosage units was evaluated by plotting the respective values (Figures 1-4). Figure 1 contains all 149 pairs of values plotted with reference lines at 3% and 5% on the blend RSD axis. The plot also includes a 45° line that represents perfect correlation between the two variables. Figure 2 contains only pairs of values whose blend RSD values are less than or equal to 3%, with the 45° line. Figure 3 contains only pairs of values whose blend RSD values are greater than 3% but less than or equal to 5%, with the 45° line. Figure 4 contains only pairs of values whose blend RSD values are greater than 5%, with the 45° line. Of the 149 batches of data plotted: • 120 batches had blend RSD values that were less than or equal to 3%. Of these 120

batches, 114 of the corresponding dosage units had RSD values less than or equal to 3%, and 6 batches had dosage unit values in the range of 3% to 6%.

• 16 batches had blend RSD values that were between 3% and 5%. Of these batches, 3 had dosage unit RSD values that were essentially the same as their corresponding blend RSD values. Twelve batches had dosage unit RSD values less than their corresponding blend RSD values, and 1 of the dosage unit RSD values was greater than its corresponding blend RSD value.

• Of the 13 batches having blend RSD values greater than 5%, all had dosage unit RSD values that were less than 5%.

This data demonstrates that blend uniformity testing is not always predictive of the uniformity of dosage units. For those instances where the blend RSD is < 3%, the data suggests some consistency between the blend and dosage unit uniformity. In some instances in this region of the plot (Figure 2), the dosage unit RSD was larger than the blend values. Two potential causes for the higher dosage unit RSD values could be failure to account for the weight variance in the dosage unit data, or segregation of the blend following the mixing process. The consistency between blend and dosage unit uniformity begins to diminish when blend RSD values fall in the 3-5% range. In many instances, dosage unit RSD values were 1-2% lower than their corresponding blend values. As the blend RSD values exceeded 5%, blend RSD values are not predictive of dosage form uniformity. In certain instances, blend uniformity testing may be valuable if the blend RSD is ≤ 3% and higher RSD values are observed for the dosage units. However, when the blend uniformity RSD is > 5%, the ability to predict final dosage form uniformity with blend uniformity data becomes unreliable.


Figure 1. Correlation of Blend and Dosage Unit RSD Values


Figure 2. Correlation of Blend and Dosage Unit RSD Values (Blend RSD Values ≤ 3%)


Figure 3. Correlation of Blend and Dosage Unit RSD Values (Blend RSD Values 3-5%)


Figure 4. Correlation of Blend and Dosage Unit RSD Values (Blend RSD Values > 5%)


C. Evaluation of Various Blend Criteria Performance

Tables 5 and 6 summarize the results of testing the data set according to the BUWG Validation, OGD Blend Uniformity Draft Guidance, FDA Blend Validation Guidance, and PDA Technical Report #25 criteria. Table 5 demonstrates that all the blend criteria gave similar results (83%-91%) in terms of batches passing. Table 6 shows that all the dosage unit criteria performed equally well (94%-100%) in terms of batches passing, with the exception of the PDA Technical Report #25 criteria passing fewer batches (70%). Table 5. Comparison of Test Criteria for Blend Data

Criteria Percent of Batches Passing

BUWG Validation 131 / 149 (88%) OGD Blend Uniformity Analysis Guidance 136 / 149 (91%)

FDA Blend Validation Guidance 123 / 149 (83%) PDA Technical Report #25 132 / 149 (89%)

Table 6. Comparison of Test Criteria for Dosage Unit Data

Criteria Percent of Batches Passing

BUWG Validation

Readily Pass

Marginally Pass

83 / 88 (94%)

79

4

BUWG Routine-

Readily Pass Stage 1

Readily Pass Stage 2 (Marginally Pass)

88 / 88 (100%)

86

2

USP Content Uniformity 85 / 88 (97%)

ICH Content Uniformity 86 / 88 (98%)

PDA Technical Report 25 62 / 88 (70%) Table 7 summarizes the results following the analysis of the 16 weight corrected batches of data submitted by Company 5, which satisfied the BUWG requirements as given in Table 4. Weight correcting had no effect on passing/failing the individual criteria. However, a shift is noticed in the number of batches readily and marginally passing the BUWG Validation Criteria, and in the number readily passing Stage 1 and Stage 2 of the BUWG Routine Criteria.


Table 7. Evaluation of BUWG Weight-Corrected Dosage Unit Criteria Performance

Criteria Percent of Company 5 Batches Passing Non-

Weight Corrected Criteria

Percent of Company 5 Batches Passing

Weight Corrected Criteria

BUWG Validation

Readily Pass

Marginally Pass

12 / 16 (75%)

9

3

12 / 16 (75%)

11

1

BUWG Routine-

Readily Pass Stage 1

Readily Pass Stage 2 (Marginally Pass)

15 / 16 (94%)

14

1

15 / 16 (94%)

13

2

D. Predictive Relationship of Blend Criteria for Dosage-Form PDA Criteria

Ideally, the predictive relationship between a criterion and the USP Content Uniformity should be determined. However, the use of observation data does not allow for following the sample size specific to the two-stage USP Content Uniformity Test. The Bergum procedure is a method capable of demonstrating the ability of a batch of dosage units to meet the USP Content Uniformity requirements. Since the Bergum procedure is single stage and does not require a specific sample size, it was chosen as a surrogate to determine the predictive power of various tests to comply with the USP Content Uniformity requirements. The predictive relationship of the OGD Guidance Document, FDA Validation Criteria, BUWG recommendation and the PDA Dosage-Form criterion was evaluated using Fisher’s Exact9 right-sided alternative hypothesis test for 2x2 contingency tables. For 2 ×2 tables, Fisher's exact test defines the probability of observing a table that gives at least as much evidence of association as the one actually observed, given that the null hypothesis is true. The null hypothesis for this analysis states that passing the dosage form PDA criterion is not related with a passing value on a blend criterion. The one-sided alternative hypothesis states that passing the dosage form PDA criteria is more likely to be associated with a passing value on a blend criterion. A small right-sided OSL (<0.0500) supports the alternative hypothesis. The results of this analysis are as follows: • Of the 79 batches that passed the OGD Blend Criteria, 54 (68%) passed the Bergum

Dosage Form Criteria. Of the 9 batches that failed the OGD Blend Criteria, 8 (89%) passed the Bergum Dosage Form Criteria. Fisher’s Exact right-sided OSL is 0.9645 indicating that there is no evidence of a predictive relationship.

9 SAS Institute Inc., SAS OnlineDoc®, Version 8, Cary, NC: SAS Institute Inc., 1999. Results of Statistical Analysis of Blend and Dosage Unit Content Uniformity Data Obtained from the Product Quality Research Institute Blend Uniformity Working Group Data-Mining Effort March 28, 2002 Page 16 of 37

• Of the 67 batches that passed the FDA Blend Criteria, 48 (72%) passed the Bergum Dosage Form Criteria. Of the 21 batches that failed the FDA Blend Criteria, 14 (67%) passed the Bergum Dosage Form Criteria. Fisher’s Exact right-sided OSL is 0.4282 indicating that there is no evidence of a predictive relationship.

• Of the 78 batches that passed the PDA Blend Criteria, 53 (68%) passed the Bergum Dosage Form Criteria. Of the 10 batches that failed the PDA Blend Criteria, 9 (90%) passed the Bergum Dosage Form Criteria. Fisher’s Exact right-sided OSL is 0.8087 indicating that there is no evidence of a predictive relationship.

• Of the 75 batches that passed the BUWG Validation Blend Criteria, 52 (69%) passed the Bergum Dosage Form Criteria. Of the 13 batches that failed the BUWG Validation Blend Criteria, 10 (77%) passed the Bergum Dosage Form Criteria. Fisher’s Exact right-sided OSL is 0.9762 indicating that there is no evidence of a predictive relationship.

The dataset analyzed through various analyses in this study demonstrate that there is no predictive value in any of the above blend criteria in regard to meeting the dosage form criterion.

IV. Conclusions A. Use of Computer Simulations to Define Sampling Schemes and Acceptance

Criteria contained in the BUWG Stratified Sampling Recommendation The results of the data mining analysis justified the assumption that the batch location means (both between location and within location) were normally distributed when using Monte Carlo simulations to generate the sampling plans and acceptance criteria contained in the BUWG recommendation. The majority of batches had within location data and location means that were normally distributed. The use of normally distributed computer simulations to generate individual values at specific batch locations to estimate criteria rejection rates will yield rejection rate estimates that are slightly smaller (more conservative) than criteria rejection rates based on actual data. B. Use of Dosage Unit Uniformity Data to Demonstrate Blend Adequacy of Mix The conclusions following the analysis to determine whether or not blend data is useful in the prediction of the uniformity of dosage units varied. In certain instances, blend uniformity testing may be value added, particularly when the blend RSD is <3% and higher RSD values are observed for the dosage units. In these instances, it is possible that the powder mixture may segregate upon transfer from the blender to the compression/ filling equipment, resulting in poorer RSD values. The availability of both blend and dosage unit uniformity data in these situations would be valuable information to assist in the identification of the root cause of the problem.10 However, when the blend uniformity RSD is > 5%, the ability to predict final dosage form uniformity with blend data becomes unreliable. It is possible that this unreliability may be due to difficulties encountered when sampling the blend, which can bias the results.


10 JK Prescott and TP Garcia, Pharmaceutical Technology, 25 (3), March 2001, p. 68-88.

The BUWG believes that blend uniformity testing provides useful information during process development, validation and as part of investigations to identify causes of issues that may arise during commercial production. Blend data is a critical part of process development. Blend analysis is used to assess the impact that formulation and process variables have on blend uniformity and during the optimization of mixing processes. The BUWG supports the use of expanded testing during validation to identify significant events during the manufacturing process. Blend data can demonstrate process control and may be useful when defining stratified sampling plans for dosage unit analysis. The purpose of validation campaigns is to demonstrate control in each unit operation; therefore the uniformity of the blend should be assessed and considered in conjunction with the resulting dosage unit data. The BUWG supports the use of stratified sampling in lieu of blend analysis during process validation for situations where it is not advisable to sample the blend (e.g. breaking process containment to sample a toxic compound). During routine production, if high dosage unit RSD values are suddenly observed, then performing blend analysis on subsequent batches may be a useful diagnostic tool for trouble-shooting the problem. The BUWG does not believe that blend uniformity testing is necessary during routine manufacture of solid dosage forms when stratified sampling and testing of in-process dosage units is used as an alternative. The results of the data mining analysis demonstrated that high blend RSD values are not necessarily indicative of poor content uniformity for the dosage units, and therefore adds little assurance that the product is in conformance. When comparing blend and dosage unit data, four possible combinations are possible: 1) low blend - low dosage unit, 2) high blend - low dosage unit, 3) low blend – high dosage unit, and 4) high blend – high dosage unit. Regardless of which combination you have, testing of the dosage units will always be necessary. If the blend RSD is low, dosage units need to be tested to ensure segregation did not occur during material transfer. If the blend RSD is high, testing of the dosage units is necessary as part of the investigation to determine whether the product truly is not uniform or if sampling error occurred. The analysis of the dataset using the various techniques in this study demonstrated that there is no predictive value in any of the blend criteria in regard to meeting the dosage form criterion. Therefore, at this time the BUWG advocates the testing of stratified samples of dosage units (in lieu of blend testing) to demonstrate adequacy of mix and dosage unit uniformity for routine production batches. The BUWG acknowledges that the stratified sampling approach described in their recommendation is only one means of accessing blend and dosage unit uniformity. CGMP regulations are flexible and suitable alternative approaches are also appropriate to assess blend and dosage unit uniformity. Emerging on-line technologies such as NIR, may be used for the sampling and testing of both blends and dosage units for uniformity.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the contributions of Loren Gelber and John Hoblitzell during the review of this paper.


Appendix 1

PQRI Blend Uniformity Working Group Recommendation for the Use of Stratified Sampling to Demonstrate Blend and Dosage Unit Uniformity


THE USE OF STRATIFIED SAMPLING OF BLEND AND DOSAGE UNITS TO DEMONSTRATE ADEQUACY OF MIX FOR POWDER

BLENDS11 I. Scope

This proposal is meant to address concerns raised following the issuance of the FDA document Guidance for Industry, ANDAs: Blend Uniformity Analysis, (August 3, 1999) as it relates to filing requirements and post-approval commitments. It also applies to ANDA exhibit batches, process validation batches and post-validation commercial batches for solid oral drug products. It does not apply to those drug products where the determination of dosage-form uniformity by weight variation is allowed. This proposal is applicable to active ingredient(s) contained in the blend. The approach described in this document is proposed as a means to satisfy the cGMP requirement for in-process testing to demonstrate adequacy of mix, as well as USP compendial requirements for the content uniformity of finished dosage forms. Alternatively, traditionally employed methods (such as the direct sampling and analysis of powder blends, in conjunction with content uniformity testing of finished dosage forms) may continue to be used to satisfy cGMP and compendial testing requirements.

II. Definitions Stratified sampling is the process of selecting units deliberately from various

locations within a lot or batch or from various phases or periods of a process to obtain a sample.12 Stratified sampling of the blend and dosage units specifically targets locations either in the blender or throughout the compression/filling operation, which have a higher risk of producing failing content uniformity results.

Potency refers to the content of drug substance (also referred to as active

ingredient) present in the tested dosage unit. Alternate methods of analysis for the content of drug substance, such as a quantitative spectrophotometric method, may be used in place of a more elaborate HPLC method.

To weight correct is to adjust the dosage unit potency result to eliminate the unit

weight effect. This method is used to demonstrate blend uniformity using dosage unit results. For example, a tablet with potency of 19.4 mg and weight of 98 mg = 19.4 ÷ 98 = 0.198 mg/mg. Label claim is 20 mg per each

11 The proposals in this document assume that an on-line, in-process measurement system is not currently available for demonstrating blend uniformity (e.g., on-line NIR measurement of in-process blend or dosage units).


12 Glossary and Tables for Statistical Quality Control, ASQC Quality Press, copyright 1983.

100 mg tablet, so the weight corrected result is 0.198 ÷ 0.20 * 100 = 99% of target blend potency.

Unless otherwise specifically stated, all dosage unit potencies are to be weight corrected prior to evaluating the acceptance criteria described in this document. All weight-corrected potencies are to be expressed as a percentage of the blend target concentration. Calculations to satisfy compendial testing requirements and the 75.0 –125.0% criteria for individual dosage units stated in Attachment 1 are not weight corrected. Further, the potencies are expressed as a percentage of the label concentration.

ANDA Exhibit Batches refer to any batch submitted in support of an ANDA.

This includes bioequivalence, test and commercial production batches of a drug product.

Compendial testing mentioned in this document refers to USP <905> Uniformity

of Dosage Units, by Content Uniformity. RSD is relative standard deviation. RSD = [(standard deviation)/(mean)] x 100% III. Background In response to concerns by ANDA applicants regarding inconsistency in review chemists’ recommendations, the FDA Drug Product Technical Committee published a draft guidance in August 1999. The guidance proposes routine blend sample analysis on commercial batches for ANDA products when USP Content Uniformity testing is required on the product. As a result of industry feedback on this draft guidance, a primary goal of the Product Quality Research Institute (PQRI) Blend Uniformity Working Group (BUWG) was to address the gap between scientific principles and the regulatory policy stated in this document. In September 2000, the working group sponsored a workshop on blend uniformity. At the conclusion of the workshop, it was recognized that limitations in current sampling technology and subsequent handling (powder segregation) might limit the effectiveness of using blend sample analysis to ensure adequacy of blending. Alternative solutions were sought to address the shortcomings of sampling and analyzing blends. The PQRI BUWG felt that any solution should possess the following three qualities:

1. The test should be simple to perform, maximizing the use of the data. 2. Acceptance criteria should be easy to evaluate and interpret. 3. Acceptance criteria should demonstrate when lack of homogeneity is

suspected.


In-process dosage unit analysis (of tablet cores, hard gelatin capsules, or other solid dose forms) is proposed as an alternative to routine blend sample analysis. Current GMPs state “control procedures shall include…adequacy of mixing to assure uniformity and homogeneity.” [21CFR 211.110 (a)(3)]. Dosage unit analysis satisfies this in-process control requirement by indirectly measuring the uniformity of the blend by sampling and testing in-process dosage units. Stratified sampling techniques are incorporated to collect in-process dosage units throughout the compression or filling process. In-process dosage unit analysis has many positive aspects:

• It is an accurate and reflective measure of homogeneity of the product. • It eliminates blend sampling error issues related to thief sampling. • It applies resources where they produce reliable, accurate information

about the quality of the product given to the patient. • Weighing errors involving blend samples are eliminated. • It removes the safety issues surrounding blend sampling of toxic or potent

drugs manufactured in isolated environments. • It accounts for segregation after blending .

The following proposal presents strategies for in-process dosage unit analysis and blend sample analysis. The PQRI BUWG advocates the use of the proposed strategy defined in Section V or Attachment 1 during the manufacture of ANDA exhibit batch(es) and during the validation of the commercial manufacturing process. The rationale for each sample size and acceptance criteria for the proposal contained in Attachment 1 are provided in Attachments 2 and 3. If this proposal is used to test the ANDA exhibit batch(es) and three commercial scale validation lots, and the results comply, then it may be sufficient to perform stratified sampling and analysis of in-process dosage units for routine commercial batches in lieu of sampling and analysis as advocated in the current draft of the ANDA Blend Uniformity Guidance Document. The level of testing required to satisfy cGMP requirements would be dependent on the quality of the data generated by testing the batches in accordance with the proposal. For those products that readily comply with the defined acceptance criteria, a modification of the USP Content Uniformity Test may be used to satisfy the cGMP requirement for routine monitoring of production batches for adequacy of mix (see Section VI and Attachment 4). Processes that do not readily comply would require additional testing for routine production batches.


IV. Process Development In general, content uniformity of the final dosage form is dependent on the homogeneity of the powder mixture in the blender. The development of robust blending and transfer processes that will not cause post-blending segregation of the mixture and thus result in manufacture of a product of acceptable content uniformity remains a critical objective during formulation and process development. Blend sample analysis should be conducted on development batches by extensively sampling both the blender and intermediate bulk containers (IBCs), when applicable, to identify an appropriate range of blending times, dead spots in blenders, segregation in IBCs, and the presence of sampling error. Appropriate blend sampling techniques and procedures should be developed for each product, including the consideration of sampling thieves of various designs, and defining the impact of sample size (for example, 1-10X dosage unit range) in an effort to develop a technique capable of measuring the true uniformity of the blend. Sample quantities larger than 3X can be used if they can be scientifically justified. Blend sampling plans should be designed to allow variance component analysis to be performed on the data to quantitate the variability attributed to the uniformity of the blend as well as any sampling error that may be present. If there is high between-location error in the blender, the deficiencies in the blending operation must be addressed. If blend-sampling error is detected, more sophisticated statistical analysis should be applied to assess the situation such as the use of methods described in PDA Technical Report 25.13 In addition to extensively sampling the blend, stratified sampling and testing of the dosage units should also be performed, taking samples at defined intervals and locations throughout the compression or filling process. During development, a minimum of 20 uniformly spaced dosage unit sampling points must be defined. Additional samples may be taken to further assess events such as filling or emptying of hoppers during the compression or filling process. Comparisons between the blend and dosage unit data should be conducted throughout product development. Investigations should be conducted to identify potential causes of any discrepancies observed between the blend and dosage unit uniformity data.14 A basic foundation of this proposal is the strong technical opinion that the proposed approach is likely to reveal formulation and processing problems that might remain hidden if less discriminating techniques are used (e.g. thief sampling; fewer sampling points during dosage form manufacture).

13 PDA Journal of Pharmaceutical Science and Technology, Technical Report No. 25, Blend Uniformity Analysis: Validation and In-Process Testing.


14 JK Prescott and TP Garcia, Pharmaceutical Technology, 25 (3), March 2001, p. 68-88.

V. Use of Stratified Blend and In-Process Dosage Unit Sampling During the Manufacture of ANDA Exhibit and Process Validation Batches to Demonstrate Uniformity of the Blend

During the manufacture of both ANDA exhibit and process validation batches, an assessment of the uniformity of both the powder blend and in-process dosage units should be made. However, for some products sampling errors make it very difficult to validate blending operations using only blend data. As a result, it is proposed to use in-process dosage unit data in conjunction with blend sample data to demonstrate blend uniformity for those instances where sampling error has been shown to exist. Both blend sampling and dosage unit sampling are proposed according to sampling plans defined in Table 1 during the manufacture of ANDA exhibit batch(es) and/or process validation batches. Sampling locations must be identified prior to the manufacture of the ANDA exhibit batch(es) and/or validation batches. Blend uniformity is demonstrated by assaying blend samples and dosage unit samples. If a blending problem exists (for example, significant variability is attributed to between-location error), then the blend is not uniform and further process development exercises should be conducted to address the deficiencies in the blending process. Demonstrating Blend Uniformity, Option 1:

See Attachment 1 for the flowchart of this option. Demonstrating Blend Uniformity, Option 2:

Alternatively, procedures described in the PDA Journal of Pharmaceutical Science and Technology, Technical Report No. 25, “Blend Uniformity Analysis: Validation and In-Process Testing” can be used to obtain assurance that the blend is uniform.


Table 1. Sampling Plans for ANDA Exhibit and Process Validation Batches

Blend Dosage Unit Identify at least 10 locations in the blender to pull blend samples. Locations must be carefully chosen to represent potential areas of poor blending. For example, in tumbling blenders (such as V-blenders, double cones, or drum mixers), samples should be selected from at least 2 depths along the axis of the blender. For convective blenders (such as a ribbon blender), a special effort should be made to implement uniform volumetric sampling, including the corners and discharge area (at least 20 locations are suggested to adequately validate convective blenders). Take at least three replicate samples from each location.

Identify at least 20 locations throughout the compression or filling operation to obtain dosage units. The sampling locations must be carefully chosen to represent significant events (e.g. hopper changeover) during the compression or filling process including samples from the beginning and end of the compression or filling operation.15 Take at least 7 dosage units from each location.

If exhibit batch blend uniformity is adequately demonstrated in an ANDA using the stratified sampling criteria described in this document, the Office of Generic Drugs would consider the blend uniformity requirement of the exhibit batches to be satisfied. The use of routine stratified dosage unit testing as described in this document for commercial production batches (Section VI) would satisfy the cGMP requirements for in-process testing as defined in 21CFR211.110 (a)(3). VI. Proposed Stratified Testing Plan and Acceptance Criteria for

Routine Monitoring of Production Batches The following section proposes a method to satisfy both the cGMP requirement for an in-process test to demonstrate adequacy of mix as defined in 21CFR 211.110 (a)(3) (in lieu of blend testing) and compendial testing, through the analysis of a single set of in-process dosage units (Attachment 4). To utilize this approach, the content uniformity results (not weight corrected) obtained for the in-process stratified dosage unit samples must be demonstrated to provide the same or better control (sensitivity to lack of uniformity) as the content uniformity data generated during compendial testing of the corresponding finished dosage units. This relationship must be established ideally for each of the ANDA exhibit batches and validation batches manufactured. If the stratified sample is representative of the final dosage unit (e.g., if the final dosage unit is an uncoated tablet), or the previous relationship is established, then it would not be necessary to perform compendial testing on finished dosage forms to


15 The beginning and end samples are taken from dosage units that would normally be included in the batch.

demonstrate content uniformity. In this instance, the results from testing the stratified in-process dosage unit samples would be sufficient to demonstrate acceptable content uniformity for the batch. If the relationship between in-process and compendial testing cannot be demonstrated, then both compendial testing of the finished product and in-process testing of stratified dosage unit samples must be performed separately. Products Which Readily Comply with Acceptance Criteria in Section V or Attachment 1 Products with RSD values ≤ 4.0% for in-process dosage units, all mean results within 90.0 – 110.0%, and all individual results between 75.0 – 125.0%, for each exhibit and validation batch, are considered to have readily complied with the Section V or Attachment 1 criteria. These products may use a modification of the USP Content Uniformity Test to satisfy cGMP compliance requirements for routine monitoring of production batches for adequacy of mix (in lieu of blend testing), as well as for compendial testing. Rather than pulling random samples for compendial testing at the conclusion of the manufacture of the product, stratified samples of the dosage units are taken in-process during the compression or filling operation. Identify at least 10 stratified sampling locations throughout the compression or filling operation to obtain dosage units. The sampling locations must be representative of the compression or filling process and include samples from the beginning and the end of the batch.16 Remove at least 3 dosage units from each location. The product meets specifications if the results comply with the acceptance criteria stated in Table 2. Alternatively, if the product complies with the criteria stated in PDA 25, then the procedures outlined in Table 2 may be used.


16 The beginning and end samples are taken from dosage units that would normally be included in the batch.

Table 2. Acceptance Criteria to Demonstrate Blend Adequacy of Mix and Dosage Unit Content Uniformity for Routine Commercial Batches for Products that Readily Comply with the Acceptance Criteria Discussed in Section V or Attachment 1.

cGMP Requirement Stage 1: Sample at least 3 dosage units

from at least 10 locations. Assay one dosage unit from each of at least 10 locations and weight correct the result. If the mean is between 90.0 - 110.0% of target and the RSD is ≤ 5.0%, adequacy of mix is demonstrated. Stage 2: Assay the remaining 2 samples from each location and weight correct the results. If the mean of all individual samples from Stage 1 and Stage 2 testing is between 90.0 – 110.0% of target and the RSD is ≤ 6.0%, adequacy of mix is demonstrated.

Compendial Requirement The same dosage units are tested according to the compendial procedure described in the USP. Values are not weight corrected.

Products Which Marginally Comply with Acceptance Criteria in Section V or Attachment 1 Processes yielding marginal results (e.g. validation and ANDA exhibit batches have one or more RSD values > 4.0% but ≤ 6.0% for in-process dosage units, but comply with the criteria for mean and individual values when tested in accordance with the procedure described in Section V or Attachment 1) require additional testing to satisfy cGMP compliance during routine production batches. Stratified sampling should be performed by taking a minimum of three dosage units from at least 10 locations during the compression or filling operation. At least 30 dosage units are assayed and weight corrected, then compared against the following acceptance criteria: mean of all individual dosage units lies between 90.0 – 110.0% of target, and RSD of all individual dosage units is ≤ 6.0%. If after testing 5 consecutive batches the criteria for the mean is met and the RSD routinely is ≤ 5.0%, testing to satisfy the cGMP requirement may be performed according to the procedure defined in Table 2.


Attachment 1

Demonstration of Adequacy of Mix and Content Uniformity for ANDA Exhibit Batch(es) and Validation Batches [Option 1]

Validating Blend

Option 1

From blend, sample atleast 10 locations, with at

least 3 replicates fromeach location

Blend sample criteria:RSD < 5.0% and allindividuals are within

the mean +/- 10% (absolute)

Meet criteria?

Dosage unit criteria:RSD of all individuals < 6.0%,

Each location mean is within 90.0% -110.0% of target potency, and

all individuals are within 75.0% and125.0% of target potency*

Assay 2nd and 3rdblend samples from

each location

yesno

During filling or compression,sample from at least 20

locations, at least 7 dosageunits each

Investigate originalcriteria "failure"

Mixing problemhas beenidentified?

Go back todevelopment

yesInvestigation pointsto blend sampling

error or some otherattributable cause

noMeet criteria?

Pass blendvalidation

yes

Dosage unit criteria:RSD of all individuals < 6.0%,

Each location mean is within 90.0% -110.0% of target potency, and

all individuals are within 75.0% and125.0% of target potency*

no

Meet criteria?

yesno

Blend is notuniform

Blend is notuniform or post-

blending practicesare causingsegregation

Assay 1 per location

Assay at least 3 dosageunits per each location,

weight correct each result

Assay at least 4 more dosageunits from each location (at

least 7 per location altogether),weight correct each result

Assay at least 7 dosageunits per each location,

weight correct each result

* When comparing individual dosage units to 75.0 - 125.0% of target potency, use the ‘as is’ results (not corrected for weight).


Attachment 2

Rationale for Blend Samples and Acceptance Criteria Sampling Locations

The minimum number of sampling locations suggested is based upon current scientific knowledge of blenders. PQRI BUWG supports the use of 10 - 15 locations to validate a tumbling blender and at least 20 locations to validate a convective blender. Sampling less than 10 locations will not adequately identify lack of blend uniformity.

Acceptance Criteria

In the past the FDA has proposed that in the testing of blends, either as part of a validation exercise or in routine blend testing, the RSD of the samples should not exceed 5.0% when the assays are expressed as a percent of the target concentration. In the current proposal we have retained the use of this limit during the testing of blend powders. We find this standard consistent with the intent of providing sufficient assurance, given the relatively small sample size that the blend is adequately mixed. In addition to the RSD criteria, it is proposed that all individuals fall within “90.0 – 110.0% of the mean result” for validation (allowing for thief bias). This is a reasonable requirement for blends given the adjustment of thief sample quantity to accommodate for thief error (see Section IV, Process Development), and the use of dosage unit data to validate the blending process if the blend data continue to demonstrate thief error.


Attachment 3

Rationale for Dosage Unit Sample Sizes and Acceptance Criteria

The number of locations and sample sizes within each location were chosen along with the acceptance criteria such that the test would: 1. Always be tighter (more stringent) than the USP test for content uniformity. 2. Be harder to pass for a process with significant between-location

variability (indicating blend uniformity issues) across filling or compression, but easier to pass when the process did not demonstrate between-location variability in the dosage units.

The dosage unit criteria have three requirements. For results that have been weight corrected: • The RSD limit defines the uniformity requirements when there is no

between-location variability. • A comparison of each location mean to 90.0% - 110.0% of target identifies

between-location variability For results not weight corrected: • A comparison of each individual to 75.0% and 125.0% of target is used to

pick up super-potent or sub-potent units. A value outside 25.0% of the target potency may indicate inadequate blend uniformity.

Numerous computer simulations were performed to identify a sampling plan and acceptance criteria that would meet the above requirements. At the same time, consideration was also given to requiring a sufficient number of locations to adequately represent all parts of the batch while trying to minimize the excessive use of analytical resources. Two of the simulations will be described below. In these simulations, the batch mean was centered at 100%, while the weight variation was set at 1.5%. Data for 5000 batches were generated for a given level of variability, and the results compared to the acceptance criteria. The percent of batches meeting the criteria was computed for each variability level. This process was continued until the percent passing was established for batches with total variability up to approximately 10% (RSD). In the first simulation, it was assumed that there was no between-location variability (no blend uniformity issues after filling/compression), but only increasing within-location variability. Figure 1 is the plot of these results. The legend indicates the sampling plans being compared. The plot labeled “20x3, 7” represents the sampling plan recommended for validation. “USP” indicates the plot for the USP content uniformity test for tablets. As seen in this figure, the “20x3, 7” plan is equivalent to or tighter than the USP test at all


levels of variability, and shows good discrimination (the curve is very steep) once it breaks away from the horizontal. Batches have a high probability of passing with an RSD of up to 5.5%. As the RSD increases beyond 5.5% the probability of failure rapidly increases. A second simulation assumed that there was increasing between-location variability, while maintaining the % RSD values for both weight variation and assay each at 1.5%. Figure 2 is a plot of these results. The key is the same as in Figure 1. All results obtained using the “20x3, 7” sampling plan are tighter than the USP test. Further, as the between-location variability increases (in other words, as blend uniformity issues arise), batches have a higher probability of failure. At RSD’s greater than 4%, the probability of failure starts to significantly increase. Thus, the “20x3, 7” proposal meets the requirements described above.


Figure 1 - No Between-Location VariabilityPopulation Mean = 100%, Wt. RSD = 1.5%Within-Location RSD Varies from 1 - 10%

Total RSD, %0 1 2 3 4 5 6 7 8 9 10 11 12

Prob

abilit

y of

Mee

ting

Crit

eria

, %

0

20

40

60

80

100

120

20x3, 7USP

Figure 2 - Between Location Variability ExistsPopulation Mean = 100%, Assay RSD = 1.5%, Wt. RSD = 1.5%

Between Location RSD varies from 1 - 10%

Total RSD, %

0 1 2 3 4 5 6 7 8 9 10 11 12

Prob

abili

ty o

f Mee

ting

Crit

eria

, %

0

20

40

60

80

100

120

20x3, 7USP


Attachment 4

Demonstration of Adequacy of Mix and Content Uniformity During Routine Manufacture

Note: Validation and Exhibit batches met dosage unit criteria (Attachment 1) for location means, RSD, and individuals.

Which routinetest?

Remove at least 3 dosageunits at each sampling

location

Identify at least 10 samplinglocations during filling or

compression to represent theentire batch

Assay 1 dosage unit perlocation and weight correct the

results

Assay at least 3dosage units per

locationand weightcorrect the results

Meet criteria?

Stage 1 criteria:mean is within 90.0% to

110.0% of target andRSD is < 5.0%

Criteria:mean is within 90.0% to


Meet criteria?

Adequacy of mix isdemonstrated

Assay remaining 2dosage units per

location and weightcorrect the results

Stage 2 criteria:mean is within 90.0% to


Compute mean and RSDof all samples combinedfrom both Stage 1 and

Stage 2

Adequacy of mix isNOT demonstrated

no

Meet criteria?yes no

yes no

yes

All Validation andExhibit batches had an RSD < 4.0% for the

dosage units

At least 1 Validation orExhibit batch had an

RSD > 4.0%forthe dosage units

Adequacy of mix isNOT demonstrated

Adequacy of mix isdemonstrated

Marginallycomply

Readily comply

If 5consecutive batches

have a mean within 90.0% to110.0% of target and an

RSD < 5.0%

Use Readily complytesting and acceptance

criteria for future batches

Continue usingMarginally comply

testing and acceptancecriteria for future batches

yes no

Note: use Readilycomply test for all future

batches


Results of Statistical AnProduct Quality Research InMarch 28, 20

alysis of Blend and Dosage Unit Content Uniformity Data Obtained from thstitute Blend Uniformity Working Group Data-Mining Effort

02 Page 34 o

Appendix 2

Template Used for the Submission of Data PQRI Datamining Template

Company Name Enter company code name, not the name (e.g., A, B, etc.) Notes for Data Entry Product Specific Information Product Example: Product A, B or C Lot no Validation lot number (e.g. lot 1, 2 or 3) % Active Percent of drug in final dosage form. Include manufacturing excesses. For

products with multiple actives, please enter each active as a separate, individual product, and correcting the sample size so that it reflects the sample size relative to that active only (e.g. 1-3X for one active, and 5-10X for the other active).

Method of Manufacture Four choices only: dry blend, wet granulation, dry granulation or “other” for processes adding active into core.

Final Weight of Dosage Unit In milligrams, the tablet, capsule or powder fill weight, including fill weight overages for powder fill bottles.

Notes for In-Process Results Sample Location Number the sample location. For multiple containers, label as independent

locations (e.g. For a 4 bin process where 10 samples are retrieved, number the sample location from 1-40)

Blend Sample Size Express as <1X, 1X to <=3X, 3X to <=5X, 5X to <=10X, & >10X where X refers to the dosage unit weight or total fill weight of the bottle including the fill overage.

Blend Results Express as percent of theoretical amount, do not include products with manufacturing overages

Notes for In-Process Dosage Form Sample Location Enter a number 1-100%. For multiple samples at the same location, number all the

same. For beginning, middle and end samples, label 0%, 50% and 100%, respectively.

In-Process Dosage Form Results Enter percent of theoretical label amount. (Enter 0% for Beginning samples, 50% for Middle samples and 100% for End samples.)

Actual Dosage Form Weight (optional) Enter the actual weight of the individual dosage form, if available. Notes for Product Release Results Final Dosage Form Results Enter results per individual unit as percent of theoretical amount.

Product Specific Information In-Process Blend Results

Pro-duct Lot No % Active Batch Size (kg)

Method of Manu-facture

Dosage Form

Dosage Weight (mg)

Sample Location

No. (e.g. 1, 2, 3)

Target Blend

Sample Size

Blend Results (% of

Theoret-ical)

Product Specific Information Table (Continued)

In-Process Dosage Form Results Product Release Results

Sample Location No. (% of Run)

Dosage Form

Results (% of

Theoretical)

Actual Dosage Form Weight

(mg) (Optional)

Results (% of Label Claim) Comments

e

f 37

Appendix 3

Acceptance Criteria Used to Compare Various Methods to Determine Blend and/or Dosage Unit

Uniformity


Criterion Blend Dosage-Form

PQRI Validation

Sample at least 10 locations with at least 3 replicates from each location Stage 1 • Assay one from each

location • RSD ≤ 5.0% • Each value must be (mean ±

10%) Stage 2 • Not specified

Sample at least 20 locations with at least 7 dosage units from each location Stage 1 • Assay 3 dosage units from each location

& weight-correct each result • Each location mean must be (90% -

110%) • Each non-weight-corrected result must be

(75%-125%) • RSD ≤ 4% ⇒ Readily Pass • 4 < RSD ≤ 6% ⇒ Marginally Pass Stage 2 • Assay 4 more dosage units from each

location & weight-correct each result • Each location mean must be (90% -

110%) • Each non-weight-corrected result must be

(75%-125%) • RSD ≤ 4% ⇒ Readily Pass • 4 < RSD ≤ 6% ⇒ Marginally Pass

PQRI Routine- Readily Pass Stage 1

Sample at least 10 locations with at least 3 dosage units from each location. Stage 1 • Assay 1 dosage unit from each location &

weight-correct the result • Mean must be 90% - 110% • RSD ≤ 5%

PQRI Routine- Readily Pass Stage 2&Marginally Pass

Stage 2 • Assay 2 additional dosage units from each

location & weight-correct the results • Mean must be 90% - 110% • RSD ≤ 6%

USP Content Uniformity for Tablets. (There are two USP CU criteria; however, since only tablet data were submitted, only the tablet criterion is discussed in this report.)

Sample 30 dosage units. Stage 1 • Assay 10 dosage units. • Each result must be 85.0% - 115.0% • RSD ≤ 6.0% Stage 2 • Assay 20 additional dosage units • Each result must be 75.0% - 125.0% • NMT 1 result outside 85.0% - 115.0% • RSD ≤ 7.8%

ICH Content Uniformity

Sample 30 dosage units. The calculation of M is as follows: • 98.5% ≤ mean ≤ 101.5% ⇒ M = mean • mean < 98.5% ⇒ M = 98.5% • mean > 101.5% ⇒ M = 101.5% Stage 1 • Assay 10 dosage units. • AccValue = | M – batch mean| + 2.4S • AccValue must be ≤ 15%



Stage 2 • Assay 20 additional dosage units • AccValue = | M – batch mean| + 2.0S • AccValue must be ≤ 15% • Each result must be M ± 25%

PDA Tech Report #25

• Sample at least 10 locations with at least 1 replicate from each location

• Assay at least 1 from each location

• Calculate the SDPI (See technical report)

• SDPI ≤ 6%

• Bergum Procedure SP2 90/95 (See technical report)

Office of Generic Drug (OGD) Blend Uniformity Analysis Guidance

• Sample at least 10 locations with 1 replicate from each location

• Mean must be 90% - 110% • RSD ≤ 5%

FDA Blend Validation Guidance

• Sample at least 10 locations with 1 replicate from each location

• Each value must be 90% - 110%

• RSD ≤ 5%

Documents

Summary of Data-Mining Effort - PQRIpqri.org/wp-content/uploads/2015/08/pdf/bu032802.pdf · from blend sampling requirements, ... 2. 1X to ≤3X, ... Product Quality Research Institute