Uniformity of Dosage Units Using Large Sample Sizes

Published on: 
Pharmaceutical Technology, Pharmaceutical Technology-10-02-2012, Volume 36, Issue 10

New European Pharmacopoeia chapter aims to resolve problems with applying the harmonized UDU test to large sample sizes.

Recent development in analytical technology has made possible the fast determination of unit content in a large number of dosage units from a batch using nondestructive analytical methods during production. These measurement techniques are often referred to as process analytical technology (PAT). Using such methodology, a better understanding of the manufacturing process, in line with the quality-by-design (QbD) concept according to the International Conference on Harmonization (ICH) Quality Guidelines Q8–Q11, and a closer control of the drug product can be obtained compared with the use of traditional analytical methods. The increased process control that is achieved by PAT is attractive both from the patient's point of view (improved product quality) and from the industry's point of view (increased production efficacy, less batch rejection).

Ø. Holte

Acceptable batch quality is demonstrated by compliance with the drug product specification. Usually, several of the tests of a specification refer to pharmacopeial test methodologies and acceptance criteria. One such test is the European Pharmacopoeia (Ph.Eur.) General Chapter 2.9.40 on Uniformity of Dosage Units (UDU). To take full advantage of the increased batch control that is gained by PAT in general and large sample size in particular, there has been a demand for a test method that utilizes large sample sizes to demonstrate compliance with UDU. Such a test has recently been adopted by the European Pharmacopoeia Commission, and will be published as Ph.Eur. General Chapter 2.9.47. In this paper, the new test is presented and explained.

M. Horvat


To ensure the consistency of dosage units, each unit in a batch should have an active substance content within a limited range around the label claim (1). Ph.Eur. General Chapter 2.9.40 on UDU addresses the recommended test to demonstrate this critical property in a batch of drug product. The general monograph was introduced in Supplement 5.2 of the Ph.Eur., and is harmonized with the Japanese Pharmacopeia (JP). The test is also included in the US Pharmacopeia (USP), but with a reservation against the possibility to demonstrate UDU by mass variation rather than content uniformity, which is allowed in Ph.Eur. and JP under certain circumstances (2). When justified and authorized, acceptable dose uniformity may be demonstrated by compliance with Ph.Eur. General Chapter 2.9.5 Uniformity of Mass of Single-Dose Preparations (2.9.5) or General Chapter 2.9.6 Uniformity of Content of Single-Dose Preparations (2.9.6) instead of the UDU test (3).

With the harmonized UDU test, acceptable and nonacceptable batches, respectively, are more precisely judged than with the 2.9.5/2.9.6 tests, as the sample size is larger (n = 30, as opposed to n = 20, and n = 10, respectively). The UDU test returns a numerical measure of the dose consistency—that is, the acceptance value (AV). In addition, UDU takes into account sample mean: a stricter standard deviation requirement applies if the sample mean is more than 1.5% off-target. The performance of the old and the new General Chapters has been discussed by Limberg and Savsek (4).

Although it is assumed that the sample is representative for the batch, it is acknowledged that the evaluation of a small sample will only provide an estimate of the batch quality. There is always a risk that a highly variable batch would pass the UDU test and be released. Likewise, there is always a risk that a good quality batch can fail the UDU test and be rejected. Increasing the sample size leads to a more precise estimate of the batch variability.

Concerns have been raised that the UDU requirements discourage the use of modern analytical techniques that are fast and nondestructive (e.g., PAT techniques) (5–10). It was unfortunate that a pharmacopoeial requirement could be regarded as a disincentive to the implementation of such analytical methods.

The main concern with the UDU test when applied to large samples was the requirement that no single result of the test sample is outside ± L2 % of the reference value M (M = "sample average"; L2 = 25.0, unless otherwise specified. For a precise definition of M, refer to Ph.Eur. 2.9.40). Such an unconditional requirement is included in both General Chapters 2.9.5/2.9.6 and the UDU chapter. The requirement was established to disclose batches with largely deviating units, even if the sample mean and the overall sample variance is acceptable. This "safety net" does not assume any distribution in the sample or in the batch (e.g., normality), and it seems reasonable enough not to allow any largely deviating unit in a small sample.

Even in normal distributed batches of good quality, a small number of largely deviating units is expected. As sample size increases, the probability to detect one of these units becomes significant. In the new General Chapter 2.9.47 (2.9.47), a small number of largely deviating units is allowed for large sample sizes. This allowance is not considered an acceptance of largely deviating units as such, but rather recognizes that the large sample has a greater probability to contain such units, even when the batch in total is considered to be of acceptable quality.

A proposal for 2.9.47 was published in Pharmeuropa 23.2 in March 2011, together with a background paper explaining the elaboration of the proposal in detail (9). During the public consultation period, several comments were submitted by industry and regulators. The feedback was fairly uniform, and the European Directorate for the Quality of Medicines (EDQM) PAT working party accordingly elaborated a revised proposal for Chapter 2.9.47. The revised text was adopted by the European Pharmacopoeia Commission in April 2012, and it will be published in Supplement 7.7 of the Ph.Eur. and implemented on Apr. 1, 2013.

Industry comments and applied feedback

The following section summarizes the comments received during the public consultation, and explains how the industry feedback has been taken into account in the revised text. The primary concerns raised during the public consultation were related to four key issues, as outlined below.

"What is the relation between the Ph.Eur. new General Chapter 2.9.47 and the existing chapters (2.9.5, 2.9.6, and 2.9.40)?" Before the adoption of 2.9.47, there were already three general chapters in Ph.Eur. addressing dose variability. The new chapter does not represent a fourth set of acceptance criteria for the determination of dose variability. Rather, as an alternative to demonstrating compliance with 2.9.40 with a traditional sample size n = 30, compliance with the UDU test could be demonstrated by compliance with the criteria of 2.9.47 with a large sample (sample size ranging from n = 100 to n = 10,000). Chapter 2.9.47 should always be applied in conjunction with chapter 2.9.40, where the relevant parameters (e.g., acceptance value, reference value) are defined and explained. In fact, 2.9.47 is meaningless without a reference to 2.9.40. There is no formal link between 2.9.47 and the older dose variability tests described in 2.9.5 and 2.9.6.

General Chapter 2.9.47 presents two alternative sets of acceptance criteria: one parametric and one nonparametric test. It is the user's choice which of the two sets of criteria to apply. For a given sample, the two sets may not give the same result, due to their fundamental difference (parametric versus nonparametric). However, both alternatives are considered equivalent in the demonstration of compliance with 2.9.40. The nonparametric test criteria for largely deviating units (L2/c2-criteria) are identical in the two alternatives.

There is no regulatory expectation that 2.9.47 should be used by a marketing authorization (MA) applicant or a MA holder, in the determination of compliance with 2.9.40. There is also no regulatory expectation that any of the two alternative sets of test criteria should be favoured over the other. The new chapter does not represent a new requirement. It is the user's decision to demonstrate compliance with 2.9.40 by any of the criteria described in the new 2.9.47.


However, it is not acceptable that a batch failing the criteria of 2.9.47 is retested by the traditional criteria of 2.9.40, with the intention to achieve a more fortunate result. It is also not acceptable to retest a batch using the other alternative set of criteria in 2.9.47 if a batch has produced an unsatisfactory result with any of the two alternatives.

"The general acceptance criteria of the new chapter are too wide." The feedback from both industry and regulators was harmonized in that both parties argued that the new proposed acceptance criteria of 2.9.47 were too wide. From Monte–Carlo simulations, it was evident that for certain batch distributions, with unusually high standard deviation, a large sample fulfilling the acceptance criteria of 2.9.47 could easily fail the criteria of 2.9.40 when evaluated on a subset of the sample (n = 30). This concern is illustrated in Figure 1, where the red oval represents batch characteristics where there is a larger probability to pass the test criteria for large samples, than the UDU criteria for a small sample. For the batches with a standard deviation between 6 and 8.8 %, the probability to pass the previously proposed version of the 2.9.47 test is greater than the probability to pass the harmonized UDU test.

Figure 1: Operations characteristic (OC) curves of the initial proposal for 2.9.47: Selected sample sizes (including the obsolete sample size n = 75) are compared with the UDU test (n = 30). The red oval represents the higher probability to pass the 2.9.47 test than the UDU test for certain batch distributions. The simulated batches follow normal distribution with a certain standard deviation (indicated along the X-axis).

Consequently, the revised criteria of the adopted 2.9.47 are such that a very small range of batch characteristics gives a greater possibility to pass the new criteria, than the 2.9.40 criteria for n = 30. These batches already have a high probability (> 90 %) to pass the UDU test (indicated by the red oval in Figure 2):

Figure 2: OC curves of selected sample sizes for the adopted 2.9.47 (Alternative 1 and 2, respectively), compared with the uniformity of dosage unit (UDU) test (n = 30). The red oval represents the higher probability to pass the 2.9.47 test than the UDU test for certain batch distributions. The simulated batches follow normal distribution with a certain standard deviation (indicated along the X-axis).

"The specific acceptance criteria for largely deviating units in the large samples are too strict." In the original proposal for large sample test criteria, the first largely deviating unit (LDU) was allowed at sample size n = 500 (see Table I). A batch that complies with these acceptance criteria for LDU when evaluated on a large sample would have a 90% probability to pass the zero-tolerance criterion for LDU when evaluated on a small sample n = 30 (9).

Table I: Draft proposal (Pharmeuropa 23.2): Number of largely deviating units allowed for a selection of sample sizes.

In practice, current technology typically returns sample sizes of a few hundred, so that if the first largely deviating unit is allowed at n = 500, it was argued by several stakeholders that such acceptance criteria would not fully resolve the main concern—the 2.9.40 zero-tolerance criteria for a largely deviating unit.

On the other hand, a batch that complies with the adopted acceptance criteria for LDU when evaluated on a large sample would have a 75% probability to pass the zero-tolerance criterion for LDU when evaluated on a small sample n = 30 (9). An extract of the revised acceptance criteria that are now integrated in the adopted chapter 2.9.47 is presented in Table II:

Table II: Adopted test (Ph.Eur. supplement 7.7): Number of largely deviating units allowed for a selection of sample sizes.

"Editorial issues." In the adopted Ph. Eur. text, the introduction to the general chapter has been rewritten to further clarify the relationship between the two alternative tests of 2.9.47 and the existing 2.9.40 (as discussed above). The tables of acceptance criteria (k, c1, c2 versus sample size n) have been expanded, and there has been no rounding of the sample sizes performed where a certain set of acceptance criteria apply.

The criteria for a "medium-sized" batch sizes (30 < n < 100) have been removed, as these were found to be less relevant for the problem statement (demonstration of UDU using large sample sizes).

Demonstration of the performance of the adopted 2.9.47 test. In the following, a series of operations characteristic (OC) curves are presented to demonstrate the performance of the new test, compared with the performance of the harmonised UDU test (Note: in the figures, Alternative 1 and 2 are denoted as "Option 1 and 2"; reference is also made to Figure 2).

Figure 3 represents the same situation as shown in Figure 2, except that the simulated batches has an off-target mean at 96 %. The batches are normal distributed around the off-target mean.

Figure 3: OC curves for normal distributed batches with an off-target mean.

In Figure 4, the probability to pass the criteria of Alternative 1 for long-tailed and bimodal batches, respectively, is compared with the OC curve for normal distributed batches (n = 1000). The long-tailed distribution could typically appear in a batch where there is an inadequate blending process, or where demixing occurs. The bimodal distribution could typically appear where several independent pieces of equipment are involved in a crucial stage of the process, and one of the pieces is faulty. Examples include a rotary tablet press and single-dose preparations (e.g., powders) that are filled by several independent filling stations. Obviously, the long-tailed and the bimodal batches always have a smaller probability to pass the test than the normal distributed batches, provided that overall standard deviation is the same. The UDU test is also sensitive to the distribution of the batch, but there is a wide range of standard deviations where the UDU test is indecisive or in some cases even less discriminating for non-uniformly distributed batches. The new alternative tests are more precise as they evaluate a larger sample.

Figure 4: 2.9.47 Alternative 1: OC curves (sample size n = 1,000) for long-tailed and bimodal distributions, as compared to normal distributions with the same standard deviation. The results are compared with the UDU test (dotted curves). An illustration of the batch distributions is presented in Figure 5 below.

In Figure 5, the same simulated batches have been evaluated by the nonparametric test criteria of Alternative 2. Comparing Figures 4 and 5, it is evident that the two alternatives are very similar in their evaluation of the tested batches, in particular for the long-tailed and the bimodal batches. When looking at Figure 6, it is evident that any differences in evaluation based on the general L1 criteria would be compensated for by the L2 criterion, which is identical in the two alternatives.

Figure 5: OC curves (sample size n = 1,000) for long-tailed and bimodal distributions, as compared to normal distributions with the same standard deviation. 2.9.47 Alternative 2.

Figure 6 illustrates whether the different simulated batches are rejected based on the general L1 criteria (AV/ c1), or based on the additional L2 criteria for largely deviating units (c2). The green + red area represents the probability that the batch passes the L1 criteria, and the red area alone represents the probability that a batch that passes the L1 criteria but fails the L2 criteria. Consequently, the green area alone is equivalent to the area under the curve in Figure 4, which represents the probability to pass the 2.9.47 test. It is apparent that the L2 criterion is important to disclose bimodal- and long-tailed distributions, as well as other deviations from normality. For the normal distributed batches, the L2 criterion hardly contributes to the evaluation at all. However, when evaluating the dose uniformity of a batch, and in particular by a third party, it is not practical, nor necessary to make any assumption as to the distribution of the batch or the sample.

Figure 6: Illustration of the relative importance of the L1 and the L2 criteria in the evaluation of acceptable dose uniformity according to 2.9.47 (n =500). Green + red area: Probability to pass the L1 criteria. Red area: Probability that a batch that passes the L1 criteria, fails the L2 criteria. Green area: the probability to pass the 2.9.47 test.


The recently adopted Ph.Eur. General Chapter 2.9.47 should resolve the problems that have been addressed regarding the applicability of the harmonized UDU test (Chapter 2.9.40) when applied to large sample sizes. With the new test criteria, more information from the large sample is taken into account in the evaluation of dose uniformity than is available in a subset of the sample (n = 30). Thus, manufacturing processes where a large sample size is available are more precisely evaluated.

The new test does not represent new regulatory expectations. Chapter 2.9.40 represents the requirements for acceptable dose uniformity, and 2.9.47 is just an alternative means to demonstrate compliance with the 2.9.40 criteria.

The proposed test criteria are at least equally stringent as the requirements of Ph.Eur. 2.9.40, and more discriminating due to the larger sample size. Although the new test originally has been motivated by PAT applications, it is applicable also to traditional sampling and analysis.


The initial draft and adopted chapter were elaborated by the members of the Ph.Eur. PAT Working Party (Chair: Prof. G. Ragnarsson, Medical Products Agency, Sweden). We also acknowledge many helpful proposals and comments from experts from industry, industry associations, and regulatory authorities that participated in an expert hearing on Sept. 29, 2010, and contributed through the public comment process.

Ø. Holte is a scientific officer with the Norwegian Medicines Agency.

M. Horvat is a leading scientist with Lek Pharmaceuticals. Both authors are representing the European Pharmacopoeia (Ph.Eur.) PAT Working Party.


1. Ph.Eur., General Chapter 2.9.40 Uniformity of Dosage Units (European Pharmacopeia, Council of Europe, France).

2. USP, General Chapter <905> Uniformity of Dosage Units (US Pharmacopeial Convention, Rockville, Maryland).

3. See Ph.Eur. Dosage Form Monographs (e.g., <Tablets>).

4. J. Limberg and M. Savsek, Pharmeuropa Scientific Notes 2, 45–48 (2006).

5. D. Sandell et al., Drug Information Jrnl. 40 337–344 (2006).

6. M. Diener et al., Drug Information Jrnl. 43 287–298 (2009).

7. L. Foust et al., Pharm. Technol. 31 (9) 108–115 (2007).

8. J.R. Murphy and K.L. Griffiths,Pharm. Technol. 30 (1) 52–60 92006).

9. J. Bergum and K.E. Vukovinsky, Pharm. Technol. 34 (11) 72–79 (2010).

10. Ø. Holte and M. Horvat, Pharmeuropa 23.2, 286–293 (2011).