A Solid Dosage and Blend Content Uniformity Troubleshooting Diagram

James K. Prescott* and Thomas P. Garcia

T
For both therapeutic and safety reasons, pharmaceutical dosage forms must accurately deliver the proper dose to the patient each time the product is consumed. Content uniformity of the finished product cannot be achieved without the preparation of a uniform blend that does not segregate between the blending and compressionfilling operations. This article examines a number of variables that may contribute to content uniformity problems for both powder blends and final dosage forms. It is meant to serve as a troubleshooting guide to assist pharmaceutical scientists in the identification and resolution of root causes of content uniformity problems.
James K. Prescott is a senior project engineer at Jenike & Johanson, Inc. (Westford, MA), tel. 978.392.0300, e-mail [email protected]. Thomas P. Garcia is a manager in the Solid Dosage Form Technology Transfer Group at Pfizer, Inc., Global Research and Development (Groton, CT).
*To whom all correspondence should be addressed.
JENIKE & JOHANS0N

he need for routine blend uniformity testing, following process validation, for ensuring content uniformity of final product has been a topic of extensive debate over the past several years. Public health concerns with respect to suboptimal blend and content uniformity is expressed in several regulatory policy documents. Recently, FDA issued an abbreviated new drug application (ANDA) draft guidance document that further elaborated on the issue of blend uniformity. Eliminating inadequate potency and content uniformity problems for marketed product is the public health objective of regulatory policies. For the past several years, a lack of adequate potency and/or content uniformity has been the number-one product quality reason for the recall of marketed solid dosage forms. From a regulatory perspective, recalls as a result of potency or content uniformity raise doubt in the adequacy of manufacturing controls and product release testing methods. The lack of blend homogeneity is one of several possible reasons that may contribute to unacceptable potency or content uniformity in marketed products. A clear understanding of the root causes of these problems is desired. The uniformity of a solid dosage form, as opposed to that of a liquid or cream, is of particular concern. Unfortunately, the science of powder blending and powder handling often is not generally well understood by those who formulate products or select processing equipment. Therefore, some processes that are prone to problems with blending, segregation, and flow have been developed and introduced into production, ultimately resulting in varying degrees of content uniformity issues. This problem is further complicated by the difficulties encountered in physically sampling a stationary powder bed using sample thieves, which have been demonstrated to be very prone to sampling error. The result of this error is samples that do not represent the state of the blend from where they were collected, which may lead to erroneous conclusions about the true uniformity of the powder blend. The Product Quality Research Institute (PQRI) a consortium of industry, academia, and regulatory scientists has been formed with the specific goal of addressing gaps between scientific knowledge and regulatory policy. The Blend Uniformity Working Group (BUWG) was formed to examine issues involving blending principles, sampling, and analysis of powder blends. To address these issues, a number of initiatives have
www.phar maportal.com

68

Pharmaceutical Technology

MARCH 2001

% Label claim

% Label claim

115 105 95 85 75 0 2 4 6 8 10 Sample location (time) 12

115 105 95 85 75 0 2 4 6 8 10 Sample location (time) 12

% Label claim

125

125

125 115 105 95 85 75 0 2 4 6 8 10 Sample location (time) 12

Figure 1: Example of satisfactory product data; satisfactory blend data are similar.
(a) 125 % Label claim

Figure 2: Example of high within-location variation product data; high within-location variation blend data are similar.

Figure 4: Example of stray value in product data; stray value in blend data is similar.

been identified, including 115 q enhanced in105 process product 95 testing to indirectly 85 assess blend uni75 0 2 4 6 8 10 12 formity and avoid Sample location (time) (b) bias introduced by 125 powder-sampling 115 devices 105 95 q the use of on85 line techniques 75 such as NIR to A B C D G H I J K L monitor blend uniSample location (spacial) formity Figure 3: (a) Example of high betweenq the evaluation location variation (wandering) product data; and identification (b) example of high between-location of meaningful acvariation blend data. ceptance criteria q identifying key formulation and process variables that impact the quality of blending. However, addressing sampling error and modifying acceptance criteria alone cannot improve upon deficient processes that are not providing adequate uniformity. One cannot expect to manufacture a product of acceptable quality without first preparing an adequately mixed blend that maintains its degree of homogeneity during subsequent transfer and compressionfilling operations. The purpose of this article is to provide formulation and process development scientists an aid that will assist them in troubleshooting content uniformity problems. A number of potential causes for uniformity problems that may be observed are discussed. Areas for further investigation and possible corrective actions are also presented. As with any tool, misuse can result in erroneous conclusions. This article is not intended to serve as the basis for regulatory policy or to serve as acceptance criteria for manufacturing. Instead, this article should be used as a means of qualitative assessment of processes where uniformity could be improved.

RSD

samplestandarddeviation samplemean

100%

[1]

Product and blend data definitions discussed in the troubleshooting diagram

The troubleshooting diagram is intended to link poor blend and/or product uniformity data to possible root causes of the problem. The diagram and supporting information in this article may be applied in principle to any powder-derived dosage form such as tablets, capsules, powder-filled bottles or vials, and sachets. The relative standard deviation (RSD) is calculated as follows:
69

The situations that are described are general behaviors only and are not correlated necessarily to any particular specifications such as the USP content uniformity test. For example, the term high RSD relates only to an RSD that one wishes to improve. Therefore, a failing or out-of-specification result does not need to be obtained to use this diagram. Many of the problematic situations presented can be improved once the root cause of the behavior is understood. The plots in Figures 16 describe the six basic trends commonly observed for product uniformity and blend samples. The data collected to prepare the plots were obtained by stratified nested sampling. Stratified sampling is the process of collecting blend samples deliberately from specific (planned) locations within a blender or by collecting product samples during the entire compressionfilling process. Nested sampling is the simultaneous collection of multiple samples within a location and is required to provide the data necessary to demonstrate the variability inherent in a single location. For the purposes of this article, the term sampling location refers to a physical location in the blender (i.e., for blend data) or a sampling time during the course of the compressionfilling operation. The described trends are based on tendencies of the mean, the between-location variance, and the within-location variance. Note that blend data based on samples taken outside the blender (e.g., bin or drum), though informative, are less applicable to the diagram because segregation may have been induced during discharge of the blender. 1. Satisfactory. Satisfactory data demonstrate that the process produces a product of acceptable content uniformity, which is reproducible for all batches (see Figure 1). For instances in which the data are for the product, the product should pass Bergums criteria; for the cases of blend data, the blend sample should comply with the standard deviation prediction interval. 2.High within-location variability. The variability of individual assay values obtained within each sampling location is wide (see Figure 2). When the data are subjected to component variance analysis, the within-location error term is larger than the between-location term. No clear trend of data is observed within a batch, and a consistent pattern between multiple batches is seldom observed. Potential causes of this problem include poor microblending, insufficient particle distribution, sampling thief error (i.e., for blend data only), or poor powder flow resulting in variable fill weights (i.e., for product data only). 3. High between-location variability. When data contain high
Pharmaceutical Technology
JANUARY 2000

% Label claim

Pharmaceutical Technology

MARCH 2001

69

115 105 95 85 75 0 2 4 6 8 10 Sample location (time) 12

125 115 105 95 85 75

Between-location RSD

% Label claim

% Label claim

(a) 125

16 12 8 4 0 0 2 4 6 8 10 12 14 16 18 Sample interval (time)

B C D G H I J Sample location (spacial)

(b) 125 % Label claim 115 105 95 85 75 A B

Figure 6: Example of blend data with an assay shift; assay shift with product data is similar.

Figure 7: Example of satisfactory temporal blend data.

Between-location RSD

Between-location RSD

16 12 8 4 0 0 2 4 6 8 10 12 14 Sample interval (time) 16 18 Too short Too long

16 12 8 4 0 0 2 4 6 8 10 12 14 16 18 Sample interval (time)

C D G H I J K Sample location (spacial)

Figure 5: (a) Example of trending product; (b) example of hot spot in blend data.

between-location variability, although Figure 8: Example of poor blend times Figure 9: Example of poor macroblending the difference in the mean values for (incomplete blending left, deblending right). (RSD remains high). samples taken from the various locations is large, little variability is observed in the values of individual hot spot are significantly greater than are those for the remainder samples within a sample location (see Figure 3). Component of the batch. Unlike trends, hot spots do not necessarily occur variance analysis demonstrates that the contribution of the at the top or bottom of a blender or at the beginning or end of between-location error to the overall variability observed is the compressionfilling operation. The trouble spot generally much greater than that attributed to the within-location term. occurs in the same location for each batch. Potential causes of A distinct pattern may or may not be apparent, both within a hot spots are dead spots in the blender or, though less likely, bisingle batch and between multiple batches of product. The ased sampling locations. term wandering is used to describe high between-location vari6. Assay shift. An assay shift occurs when the mean assay valability for the product because the mean of the samples seems ues are no longer centered on 100% label claim (see Figure 6). to wander over time. Potential causes of high between-loca- Both between-location and within-location errors are typically tion variability include poor macroblending (i.e., quality of the low, and the abnormality may be repeatable between batches. blend on a large scale), segregation, and poor weight control Potential causes for assay shifts are the loss of one component mechanisms (for product data only). (through adsorption or extraction during processing), analyt4.Stray value. Single or multiple stray values may be observed ical error, factoring or dispensing error, sampling bias (for blend well beyond typical variability (see Figure 4). The problem may data), or improper fill weight (for product data). not be observed in each batch because the probability of finding such samples may be low. Potential causes of stray values Potential root causes of blend or product content are agglomeration of the active, an analytical or sample-han- uniformity problems dling error, or a dead spot in the blender. The magnitude and Although factors not discussed in this article could contribute direction that the value(s) deviate from the mean can assist in to blend and product content uniformity problems, seven comidentifying the problem (e.g., greater than 150200% label claim mon root causes are as follows: may suggest that the sample is super potent as a result of ag1.Non-optimum blending. Non-optimum blending results when glomeration of the active). the blender does not provide the best blend that is theoretically 5.Trending and hot spots. Trending occurs when one observes possible (i.e., a randomized blend of particles). Failure to achieve in the data a distinct direction in the assay values (see Figure adequacy of mix could be the result of poor formulation de5a). The trend may be observed as one progresses from the velopment, inadequate blender operation (e.g., fill level, loadtop to bottom locations when sampling a blend or may be seen ing, number of revolutions), or poor selection of blending equipas the compressionfilling operation progresses over time. ment. Figure 7 demonstrates the trend that should be observed Trending commonly is associated with product made from for acceptable blending processes. The RSD value decreases inithe end of a bin, drum, or batch and generally is repeatable tially. With further blending, the RSD levels off at the point at from batch to batch. Although the locations mean often is which the best possible uniformity for this system is achieved substantially different from that for the remainder of the over- and remains stable regardless of further blending. all batch, within-location error typically is low. Potential causes Figure 8 illustrates a situation whereby a powder blend segof trending are segregation by particle size, which results in regates with further blending. The blend operation will yield an assay or powder density variations, or static charge bias (for optimal product when the blender is operated in the time range blend data). that the RSD is at its lowest values. Poor macroblending is obHot spots generally are the result of incomplete blending in served if the lowest RSD value achieved is unacceptably high (see a specific region (i.e., a dead spot) of a blender (see Figure 5b). Figure 9). This may be caused by a persistent dead spot (perhaps Individual and mean assay values for samples taken from the caused by using an improper volume of material in the blender),
70

Pharmaceutical Technology

MARCH 2001

www.phar maportal.com

Between-location RSD

16 12

8 by errors in loading the blending container, or by create poor dose uniformity of 4 using an improper blend speed. If blend-sampling the finished product, even if the bias is present, a similar plot could result. Figure concentration of the blend re0 0 2 4 6 8 10 12 14 16 18 10 shows an example of poor microblending. In mains uniform across the batch. Sample interval (time) this instance, the blender is incapable of blending Poor weight control could result the materials on a microscopic scale, which could Figure 10: Example of poor microblending if the formulation possesses correspond to the products unit dose. Potential (RSD wanders). poor flow properties or if causes of poor microblending include segregation, equipment weight controls malagglomeration, or large differences in particle size, shape, or den- function during the compressionfilling operation. sity. Note that with a sufficiently high number of blend samples, 5.Wrong mass or loss of component. The wrong quantity of drug poor microblending may manifest itself as a steady, high RSD substance may be added to the batch as the result of dispensidentical to the example of poor macroblending. ing errors, improper factoring of the drug substance (for those In many cases, especially after process development, the blend cases in which the amount of drug added to the product is adRSD is not measured across the blend cycle and instead is mea- justed to account for inherent drug substance potency variasured only at the endpoint. If during routine monitoring the tions), or low assay values of the input drug substance. Loss of final RSD increases significantly over previous runs, one should drug or excipient also can occur during processing for exinvestigate the root cause of this observation. This problem ample, by adsorption of a component onto an equipment surcould be the result of many factors, including a poor choice of face, by becoming trapped in filter socks, or by being physically blenders, poor microblending, overfill or improper filling of removed through powder extraction devices. the blender, sampling problems, or formulation problems. In 6.Analytical error. Analytical error leads to results that are not this case, when the final RSD varies widely from batch to batch, representative of the sample collected for analysis. These errors blend sampling such as complete blend cycle information along could be the result of poor sample splitting (particularly with with nested sampling and enhanced product sampling may be powder blends), dilution errors, improperly prepared standards, useful in further diagnosing the problem. weighing errors, and container tare errors or vialcap mix-ups 2.Thief sampling error. Sampling error results when the sampling (i.e., for blend samples). device does not obtain samples that are representative of the blend. 7. Insufficient particle distribution. If the particle distribution This can be caused by a number of factors such as the design and is not considered during formulation development, the ranoperation of the thief, the sampling technique, static charge, and dom mixture of particles can be incapable of meeting uniforthe physical properties of the formulation being sampled. Sam- mity requirements. This could be the result of the particle-size pling bias, a form of sampling error, results when there is a re- distribution of the drug, improper sizing of granulations, or peatable shift in the mean of the samples because of preferential the agglomeration of a component at any point in the process. flow of one or more components into the sampling cavity. For the particles in the system, this random variation is irreNote that thief sampling error can result in false negatives ducible and not a function of blending or segregation. (i.e., the blend is poor but thief data say otherwise) in addition The appendix discusses how a shift in assay and/or increased to the more common concern of false positives. A thief prone variability is related to root causes. to false negatives is sometimes called a counterfeiter. 3.Segregation after discharge. Segregation occurs when the blend Using the solid dosage and blend uniformity demixes as a result of powder transfer from the blender to the troubleshooting diagram compressionfilling equipment. The blend also can demix in Step 1: Identify the product trend for your product. Plot the prodhoppers during the course of the compressionfilling operation. uct content uniformity data as a function of sample location Three segregation mechanisms common with typical pharma- (i.e., time of production). Identify the product trend behavior ceutical powders are sifting, fluidization, and dusting. from Figures 16 that is most similar to your data. Sifting segregation is a process by which smaller particles move Step 2: Identify the blend sample result of your product. Plot the through a matrix of larger ones. During the filling of a bin or blend content uniformity data as a function of sample location drum, a concentration of fine particles develops under the fill (i.e., position in the blender). Again using Figures 16, identify point while the larger particles roll or slide to the periphery of the blend sample behavior that is most similar to your data. the pile. This often results in fines discharging first, followed by Step 3: Identify a reference number. The combination of the coarse particles at the end of a container. product trends and blend sample results, in addition to those Fluidization segregation results in a top-to-bottom segregation situations in which data are not available for either the prodpattern with fines at the top of a bin or drum. This can be the uct or the blend, yields 48 cases. Because the diagrams cover result of air counterflow, such as discharging from one closed several pages, each of these cases is identified by a reference container to another, or it can be caused by high discharge number for convenience. rates. The reference numbers consist of a composite of the prodDusting, or entrainment in air, results in fines accumulating at uct trend and blend sample result numbers. The reference numthe perimeter of a bin or drum. These fines often discharge at ber is derived in the following format: X.Y, where X is the prodthe end of a container unless design precautions are taken. uct trend number, and Y is the blend sample result number 4. Weight control. Wide variance in product fill weights can separated by a period. For example, the case of trending with

71

Pharmaceutical Technology

MARCH 2001

Pharmaceutical Technology

JANUARY 2000

71

tablets (5) along with satisfactory blend results (1) yields a reference number of 5.1. Once a reference number is identified, stay within that reference numbers row across the next pages to identify root causes (see Step 4). Each combination of product and blend trends (i.e., the reference number) also is given a relative probability of occurring (e.g., high, medium, or low) as a result of a single root cause. If a low-probability case is encountered, multiple root causes of the problem probably are occurring. Step 4: Identify potential root causes. This article identifies seven common root causes of blend sample and content uniformity problems. Each reference number has a relative probability of being caused by any of these seven root causes. For each reference number, a root cause is assigned a qualitative probability based on theoretical grounds and the practical experiences of the authors. These probabilities are presented in matrix form on the diagram as follows: 4 This is a common and highly likely root cause for the problem. Start your investigation here. 3 This is a likely root cause of the problem, but seek supporting data to confirm it. 2 There is a good chance that this root cause is contributing to the problem, but be aware of other possible causes. 1 This is not very likely to be linked to the problem, and other more likely root causes should be ruled out first. Be aware that multiple root causes may be present. 0 It is highly unlikely that this is a contributing factor to the problem. Seek other reasons and be aware that multiple root causes may be causing the problem. Select the highest probable root cause and drop down vertically within that column for the next steps. Note that in some instances several root causes have equally high probabilities of occurring. These cases warrant further investigation for multiple root causes. As described earlier, a thief sampling error occasionally can result in false negatives. Reference numbers 2.1, 3.1, 4.1, and 5.1 all have acceptable blend data but leave open the possibility that the blend was poor even though the blend sample data looked (falsely) acceptable. For this reason, the root causes of nonoptimum blending and thief sampling error are boxed together. Using the troubleshooting diagram is much easier if all of its pages are placed next to each other. This allows the user to read easily across the rows of the diagram. As an example of using the chart, reference number 5.1 (trending with product and satisfactory blend results) has a high probability of a single root cause. Although non-optimum blending and thief sampling error can be a possible root cause (together with a probability of 2), segregation is a much more likely candidate causing the problem with a probability of 4. One should consider a number of additional points such as the entire history of the product and process when interpreting the recommendations from the troubleshooting diagram. Questions one should ask include q Is this a new product or an existing one with a significant body of data? q Has this problem been seen with this product or one similar to it?
72

What is unique or different about this product or process? Have the materials, processes, operators, equipment, or environmental control changed recently? q How do the physical characteristics of the materials used for this batch compare with what was intended? q Is the problem repeatable among multiple batches, or is this an isolated incidence? q Did the operators observe any anomalies during the manufacture of the batch? q Were any equipment malfunctions encountered during batch manufacture? q How do the mean and RSD values for the blend and product compare? q How do the measured RSDs compare with the theoretical RSD of a randomized blend of particles? Addressing each of these questions will further help the scientist identify the cause(s) of the problem and its successful resolution. Steps 5 and 6: Further investigations and possible solutions. Once a likely root cause has been identified, stay within that column and drop down to Steps 5 and 6. Step 5 presents some starting points to initially consider in identifying and confirming the suggested root cause. If further investigation reveals data that do not support the root cause selected, go back to Step 4 and identify other likely root causes. Once the root cause has been confirmed, Step 6 provides suggestions for corrective actions. For example, in the case discussed above, segregation was identified as the most likely root cause. Further information about each root cause can be found within the references that are listed at the end of this article.
q q

Handling multiple product or blend problems

Each of the problems presented in this article has been reduced to the simplest behavior. However, real-world situations seldom follow such neat guidelines. If multiple behaviors such as product trending with a product assay shift are observed, then this problem should be split into two cases and be considered separately. In this example, if the blend data were satisfactory, then reference numbers 5.1 and 6.1 must be evaluated. In cases where multiple problems are present, multiple root causes also are probably responsible. However, one should give special consideration to those root causes that are common to both cases.

Conclusion
Many variables can affect a processs ability to produce a blend and product of acceptable content uniformity. This article discusses a number of potential causes leading to content uniformity problems. However, additional factors also could be contributing to the quality of the blend and product. It should be emphasized that without proper formulation and process development, as well as the selection of appropriate blending and transfer equipment, the chances of obtaining a blend and product of acceptable uniformity are reduced significantly. The authors hope that this article will provide pharmaceutical scientists with a useful reference tool that will allow them to work through future problems that they may encounter during the manufacture of solid dosage forms.
www.phar maportal.com

Pharmaceutical Technology

MARCH 2001

Appendix
One can describe the mean and standard deviation of samples from a given batch as follows:
Population Initial 1. 2. 3. 4.
b p 2 b 2 p

Blend

Thief xthief

Contributing Components* Segregation Weight xwp

Loss xlossb xlossp

Analytical xab
ap s2ab

Random

Sampling xab x

Xinitial Xinitial { { s2b s2b

s2thief} s2seg} s2wp

s2r s2r

sp s2sb

s2ap

s2sp

*With the exception of the sampling term (discussed below), each contributing component above is one of the root causes of the troubleshooting diagram. The initial and loss terms together make up the wrong mass of component root cause.

Equation 1 is the population mean assay of the blend (normalized to the weight of the b sample). Xinitial is the initial content of the blend based on the actual mass of each component used,which could be different from the theoretical because of dispensing errors. xthief is a change in mean sample content caused by preferential fill of one or more components into the thief cavity (this term is called bias). xlossb is the mean loss of content up to the point of collecting samples from the blender (a negative term for loss of active,or positive for a loss of excipient). xab is the mean shift in assay caused by analytical and other lab errors for analysis of the blend samples (e.g.,calibration). xsb is the shift in assay caused by the measured value being based on a finite number of blend samples.As the number of samples increases,this term is reduced.The magnitude of this term also is larger for a larger population variance.The measured assay value is in fact only an estimate of the population mean assay.The actual mean assay of the entire batch of blended powder (population) is equal to the measured value less xsb.This term is not deterministic,but rather is a statistical or probabilistic term. Equation 2 is the populations mean assay of the product. p Xinitial is the initial content of the blend. xwp is a change in mean assay caused by weight changes of the product (i.e.,a shift in mean weight). xlossp is the mean loss of content up to the point of creating the dose. xap is the mean shift in assay caused by analytical and other lab errors for the analysis of the product. xsp is the shift in assay caused by the measured value being based on a finite number of product samples (again,this is a probabilistic term). Note xap xab if the analytical methods are the same for both product and blend. Equation 3 2 is the population variance (standard deviation squared) of the blend. b s2b is the variance of the powder blend caused by non-optimum blending. s2thief is the variance caused by thief sampling error. s2ab is the variance caused by analytical and other lab errors for the analysis of the blend samples. s2r is the variance caused by random distribution of individual particles. s2sb is the shift in variance caused by the measured value being based on a finite

number of blend samples.As the number of samples increase,and as the population variance decreases,this term decreases. Note the terms {s2b s2thief} are bracketed.It is often assumed that thief sampling error is always additive to the variation of the blend.This is not always the case.In some instances,thieves can disturb the powder bed in a way to smear the sample.This in effect improves the uniformity of the blend where the sample is collected,resulting in samples that are more uniform than is the initial state of the blend.Because s2thief cannot be a negative term,the terms {s2b s2thief} must be considered together.This is why the thief error and non-optimum blending are tied together for reference numbers 2.1,3.1,4.1,and 5.1. In a variance components analysis,it is often assumed that the betweenlocation error term is equal to s2b,the within-location error term is equal to s2thief, and the terms s2ab,s2r and s2sb are neglected. Equation 4 2 is the population variance of the product. p s2b is the variance of the powder blend caused by non-optimum blending. s2seg is the increase in variance caused by segregation upon discharge and handling. s2wp is the variance to weight variability of the product. s2ap is the variance to analytical and other lab errors for the analysis of the product. s2r is the variance caused by random distribution of individual particles. s2sp is the shift in variance caused by the measured value being based on a finite number of product samples. Note s2ap s2ab if the analytical methods are the same for both product and blend. Note the terms {s2b s2seg} are bracketed.In actuality,these terms should form a single irreducible term describing the variance of the powder blend caused by the combination of blending and segregation; this mathematically cannot be separated into two components.However,showing this as two terms is illustrative.If the blend segregates and powder uniformity gets worse,then the term s2seg is positive.If additional handling improves the uniformity of the blend,the powder becomes more uniform than the initial state of the blend.In this instance,the term s2seg would be negative.Because this cannot be a negative term,the terms {s2b + s2seg} must be considered together. Finally,note that s2r is formulation specific,and highly depends upon the particle-size distribution of the active.This term will not change between the blend samples and the product,unless the powders particle size changes upon transfer from the blender to the creation of the dose.Of all the terms used here, this is the only term that can be calculated,a priori,on first principles.

73

Pharmaceutical Technology

MARCH 2001

Pharmaceutical Technology

JANUARY 2000

73

Table I: Solid dosage and blend content uniformity troubleshooting diagram

Steps 1 and 2: Describe the product (dose) and blend data 1. First, describe the PRODUCT
(see Product and Blend Data Definitions and Figs. 16)

2. Next, describe the BLEND SAMPLES

(see Product and Blend Data Definitions and Figs. 16)

Step 3 Reference number

Keys to probabilities of possible root causes 4 Highly likely root cause.Start here first. 3 Likely,seek supporting data. 2 Good chance,but keep your eyes open for other possibilities. 1 Not likely,rule out other reasons first; multiple root causes may be present. 0 Very unlikely,seek other reasons; multiple root causes may be present. Some additional considerations: q Is this a new product or an existing one with a significant body of data? q Has this problem been seen with this product or one similar to it? q What is unique or different about this product or process? q Have materials,processes,operators,equipment,or environmental control changed recently? q How do the physical characteristics of materials used for this batch compare to what was intended? q Is the problem repeatable among multiple batches or was this an isolated incidence? q Did the operators observe any anomalies during the manufacture of the batch? q Were any equipment malfunctions encountered? q Compare the mean of product to the mean of blend. q Compare the RSD of product to the RSD of blend.
74

(CONTINUED ON NEXT PAGE)

Poster available A poster-size version of the troubleshooting diagram is available by contacting Jenike & Johanson,Inc.at [email protected].

Pharmaceutical Technology

MARCH 2001

Pharmaceutical Technology

MARCH 2001

74

Table I: Solid dosage and blend content uniformity troubleshooting diagram (continued)
Step 3 Reference Step 4: Correlate the data with possible root causes; continue with Steps 5 and 6 below number Probability is given on a scale of 04 (see Key, pg. 80)
CONTINUED ON NEXT PAGE)

Step 5: With possible root causes identified, continue with further investigation

Step 6: With additional data to support root cause, consider possible solutions

Pharmaceutical Technology

Pharmaceutical Technology

76

MARCH 2001

MARCH 2001

76

Table I: Solid dosage and blend content uniformity troubleshooting diagram (continued)
Step 4: Correlate the data with possible root causes; continue with Steps 5 and 6 below Probability is given on a scale of 04 (see Key, pg. 80)

Step 5: With possible root causes identified, continue with further investigation

Step 6: With additional data to support root cause, consider possible solutions

77

Pharmaceutical Technology

MARCH 2001

Pharmaceutical Technology

MARCH 2001

77

Acknowledgments
This article is part of the efforts of the PQRI BUWG to address the gaps between scientific theory and regulatory compliance. The authors would like to acknowledge those scientists, engineers, and operators who have worked with them over the years to identify and solve various content uniformity problems. The authors would also like to acknowledge input received from the Product Quality Research Institute Blend Uniformity Working Group.

Bibliography
Powder blending N.A. Orr and E. Shotton, The Mixing of Cohesive Powders, The Chemical Engineer (1), 1219 (1973). J. Bridgewater, Fundamental Powder Mixing Mechanisms, Powder Technol. 15, 215236 (1976). Z.T. Chowan, E.E. Linn, and Li-Hua Chi, Mixing of Pharmaceutical Solids II: Evaluation of Multicomponent Mixing of Cohesive Powders in Cylindrical Shear Mixer, J. Pharm. Sci. 70 (3), 243 (1981). Z.T. Chowan and Li-Hua Chi, Mixing of Pharmaceutical Solids III: Multivariate Statistical Analysis of Multicomponent Mixing, J. Pharm. Sci. 70 (3), 247 (1981). J.N. Staniforth, H.A. Ahmed, and P.J. Lockwood, Quality Assurance in Pharmaceutical Powder Processing: Current Developments, Drug Dev. Ind. Pharm. 15 (6,7), 909926 (1989). R.L. Lantz, Jr., and J.B. Schwartz, Mixing, in Pharmaceutical Dosage Forms: Tablets, H.A. Lieberman, L. Lachman, and J.B. Schwartz, Eds. (Marcel Dekker, Inc., New York, 2d ed., 1989). L.T. Fan, Yi-Ming Chen, and F.S. Lai, Recent Developments in Solids Mixing, Powder Technol. 61 (6), 255288 (1990). J.W. Carson, T.A. Royal, and R.J. Hossfeld, Tumble Blending with Mass Flow Containers Improves Productivity and Quality, Powder Handl. Proc. 6 (4), Oct./Dec., 413416 (1994). J.W. Carson, T.A. Royal, and T.G. Troxel, Mix Dry Bulk Solids Properly and Maintain Blend Integrity, Chem. Eng. Prog. 11, 7280 (1996). R.C. Hwang, M.K. Gemoules, and D.S. Ramalose, A Systematic Approach for Optimizing the Blending Process of a Direct-Compression Tablet Formulation, Pharm. Technol. 22 (3), 158170 (1998). N. Harnby, An Engineering View of Pharmaceutical Powder Mixing, Pharm. Sci. Technol. Today 3 (9), 303309 (2000). D. Brone and F. Muzzio, Enhanced Mixing in Double-Cone Blenders, Powder Technol. 110, 179189 (2000). Thief sampling error T. Allen, Particle Size Measurement (Chapman and Hall, 4th ed., London, 1990). C.F. Harwood and T. Ripley, Errors Associated with the Thief Probe for Bulk Powder Sampling, J. Powder Bulk Solids Technol. 1 (2), 2029 (1977). J.T. Carstensen and C.T. Rhodes, Sampling in Blending Validation, Drug Dev. Ind. Pharm. 19 (20), 26992708 (1993). T. Garcia, B. Elesheimer, and F. Tarczynski, Examination of Components of Variance for a Production Scale, Low Dose Powder Blends and their Corresponding Tablets, Drug Dev. Ind. Pharm. 21 (18), 20352045 (1995).
78

J. Berman and J. Planchard, Blend Uniformity and Unit Dose Sampling, Drug Dev. Ind. Pharm. 21 (11), 12571283 (1995). J. Berman, A. Schoeneman, and J.T. Shelton, Unit Dose Sampling: A Tale of Two Thieves, Drug Dev. Ind. Pharm. 22 (11), 1211132 (1996). R.K. Chang, J. Shukla, and J. Bueler, An Evaluation of a UnitDose Compacting Sample Thief and a Discussion of Content Uniformity Testing and Blend Validation Issues, Drug Dev. Ind. Pharm. 22 (9,10), 10311035 (1996). J.T. Carstensen and M.V. Dali, Blending Validation and Content Uniformity of Low-Content, Noncohesive Powder Blends, Drug Dev. Ind. Pharm. 22 (4), 285290 (1996). J. Berman et al., Technical Report No. 25 Blend Uniformity Analysis: Validation and In-Process Testing, PDA J. Pharm. Sci. Technol., Supplement 51 (S3), (1997). F.J. Muzzio et al., Sampling Practices in Powder Blending, Int. J. Pharm. 155 153178 (1997). P. Gy, Sampling for Analytical Purposes (J. Wiley & Sons, West Sussex, England, 1998). T.P. Garcia, M.K. Taylor, and G.S. Pande, Comparison of the Performance of Two Sample Thieves for Determination of the Content Uniformity of a Powder Blend, Pharm. Dev. Technol. 3 (1), 712 (1998). F.J. Muzzio et al., An Improved Powder Sampling Tool, Pharm. Technol. 24 (4), 92110 (1999). T. Tanino et al.,Evaluation of Blend Uniformity: Effect of Granulation Sample Size, J. Japan Soc. Pharm. Mach. Eng. 8, 514 (1999).

Segregation J.C. Williams, The Segregation of Particulate Materials. A Review, Powder Technol. 15, 245251 (1976). H. Johansen, I.S. Andersen, and H. Leedgaard, Segregation and Continued Mixing in an Automatic Capsule Filling Machine, Drug Dev. Ind. Pharm. 15 (4), 477488 (1989). J.W. Carson, T.A. Royal, and D.J. Goodwill, Understanding and Eliminating Particle Segregation Problems, Bulk Solids Handl. 6 (1), 139144 (1986). J.K. Prescott and R.J. Hossfeld, Maintaining Product Uniformity and Uninterrupted Flow to Direct-Compression Tableting Presses, Pharm. Technol. 18 (6), 98 114 (1994). J.K. Prescott and J.W. Carson, Analyzing and Overcoming Industrial Segregation Problems, presented at the International Union of Theoretical and Applied Mechanics Segregation in Granular Flows, 1999. L. Bates, User Guide to Segregation (British Materials Handling Board, 1997). A. Alexander et al., A Method to Quantitatively Describe Powder Segregation During Discharge from Vessels, Pharm. Technol. Yearbook, 621 (2000). Analytical error G.T. Greco, Segregation of Active Constituents from Tablet Formulations during Grinding: Significance to Pharmaceutical Analysis, Drug Dev. Ind. Pharm. 9 (6), 971988 (1983). H. Egermann, Problems on Assessing Actual Content Uniformity by Spot Sample Analysis, Int. J. Pharm. Technol. Prod. Manuf. 3 (7), 5966 (1982). FDA/CDER Guidance for Industry (draft), Investigating Out
www.phar maportal.com

Pharmaceutical Technology

MARCH 2001

of Specification (OOS) Test Results for Pharmaceutical Production (September 1998). Currently at http://www.fda.gov/ cder/guidance/1212dft.pdf L.D. Torbeck, Reportable Values for Out-of-Specification Test Results, Pharm. Technol. Analytical Validation Supplement, 2123 (1999). B. Iglewicz and D.C. Hoaglin, How to Detect and Handle Outliers, ASQ(C) (Quality Press, Milwaulkee, WI, 1993). V. Barnett and T. Lewis, Outliers in Statistical Data, Wiley Series in Probablilty and Mathematical Statistics, Applied Probability and Statistics (3d ed., John Wiley & Sons, West Sussex, England, 2000).

1997). P T

Wrong mass or loss of component N. Laing, Granulated Ingredients: Reducing Drug Potency Loss, Feed Management 45 (12), 1720 (1994). Tablet weight variations W.C. Ling, Tooling as a Factor in Tablet Weight Variation and Controls, J. Pharm. Sci. 62 (12), 20072011 (1973). F.E. Sendall and J.N. Staniforth, A Study of Powder Adhesion to Metal Surfaces During Compression of Effervescent Pharmaceutical Tablets, J. Ind. Pharm. 38 (7), 489493 (1986). H. Nyquist, Prediction of Weight Variation in Tablet Production from Shear Cell Measurements, Acta Pharm. Suec. 19 (6), 413420 (1982). R.F. Shangraw, Compressed Tablets by Direct Compression, in Pharmaceutical Dosage Forms: Tablets, H.A. Lieberman, L. Lachman, and J.B. Schwartz, Eds. (Marcel Dekker, New York, 2d ed., 1989). J.K. Prescott and R.A. Barnum, On Powder Flowability, Pharm. Technol. 24 (10), 6084 (2000). Distribution of particles H.B. Pfost, Feed Mixing, Feed Manuf. Technol. 561563 (1970). S.H. Yalkowski and S. Bolton, Particle Size and Content Uniformity, Pharm. Res. 7 (9), 962966 (1990). R.R. McEllhiney, Feed Manuf. Technol. IV (American Feed Industry Association, Arlington, VA, 1994). M.M. Villiers and J.G. Watt, Measurement of Mixture Homogeneity and Dissolution to Predict the Degree of Drug Agglomerate Breakdown Achieved through Powder Mixing, Pharm. Res. 11, 15571561 (1994). Y. Zhang and K.C. Johnson, Effect of Drug Particle Size on Content Uniformity of a Low-Dose Solid Dosage Form, Int. J. Pharm. 154, 179183 (1997). Variance components and their uses Design and Analysis of Industrial Experiments, O.L. Davies, Ed. (Hafner Publishing Co., London, UK 1960). O. Kepmthorne and J.L. Folks, Probability, Statistics and Data Analysis (Iowa State University Press, 1971). G.E.P. Box and G.C. Tiao, Bayesian Inference in Statistical Analysis (Addison Wesley, Reading, MA, 1973). G.E.P. Box, W.G. Hunter, and J.S. Hunter, Statistics for Experiments, An Introduction to Design, Data Analysis, and Model Building (John Wiley & Sons, New York, 1978). J. Berman et al., Technical Report No. 25 Blend Uniformity Analysis: Validation and In-Process Testing, PDA J. Pharm. Sci. and Technol., Supplement 51 (S3) (NovemberDecember
79

Pharmaceutical Technology

MARCH 2001

Pharmaceutical Technology

JANUARY 2000

79