Ray Munden

Stability testing is a vital part of product development. It is, however, quite resource consuming and, with 12 months' stability data generally required for a new product licence, is also often on the critical path of development. Therefore, it is vital that stability protocols are designed both properly, to ensure regulatory approval, and efficiently to minimize the time and resource required. 

Fortunately, there are well-accepted procedures, namely bracketing and matrixing (B&M), to reduce the amount of stability testing required. These are procedures for reducing the number of samples of product tested for stability which, when correctly applied, should result in neither loss of data quality produced nor a significant change in the predicted shelf life. Such procedures are also theoretically applicable to drug substances, but, as there tend to be fewer presentations or batches tested than for products, the opportunities and benefits are more restricted.  

Stability samples must be manufactured, labelled and stored, so reducing the number of stability samples can also potentially save in sample production and management costs, especially since storage facilities are quite expensive to set-up and maintain. 

Bracketing, in particular, and matrixing have already been routinely applied to stability studies for some time now and are the subject of an International Conference on Harmonization (ICH) paper, Q1D issued in 2002.

  Preceding that there was a Committee for Proprietary Medicinal Products (CPMP) paper in 1997 on B&M and the Center for Drug Evaluation and Research (CDER) draft stability guidelines, issued in 1998, which also included these topics.

  Despite this, there is still much feeling in industry that the regulatory authorities will be reluctant to accept reduced stability protocols. In practice, I have had experience of submitting many reduced stability protocols with B&M without encountering any resistance from the reviewers; in fact, some were praised for their good design. I have also had conversations with regulators who have queried why more submissions do not include reduced stability designs, particularly matrixing. In fact, matrixing was described some time ago in papers authored by FDA staff.

B&M are performed across different presentations of an active in a given product type, that is across different factors such as strength, pack, fill volume and packing site. The definitions as given in ICH are as follows: 

 "The design of a stability schedule such that only samples on the extremes of certain design factors (e.g., strength and package size) are tested at all time points as in a full design. The design assumes that the stability of any intermediate level is represented by the stability of the extremes tested." 

For example, this indicates that for a product with three different strengths, say tablets at 2, 4 and 6 mg, it may be possible to omit testing of the 4 mg tablets. 

 "The design of a stability schedule such that a selected subset of the total number of possible samples for all factor combinations is tested at a specified time point. At a subsequent time point, another subset of samples for all factor combinations is tested. The design assumes that the stability of each subset of samples tested represents the stability of all samples at a given time point." 

Therefore, at a given time point (other than the initial or final ones) not every batch on stability needs to be tested. 

B&M can be applied to virtually any formulation, although for more complex delivery systems with a large number of potential drug–device interactions, such as metered-dose inhalers (MDIs), multidose dry powder inhalers (MDPIs) and transdermal delivery systems, additional justification will be required. Indeed, FDA appears reluctant to accept reduced designs for inhaled devices although other regulatory bodies have accepted them. 

ICH Q1D guidelines give some direction on the applicability of B&M. The existence of multiple strengths of a formulation is an obvious target for such reduced designs and the guidelines specifically allow this without justification when the different strengths have identical formulations. For example, tablets with identical formulations, but different compression weights, and capsules of different strengths made with identical power blends, but different fill plug sizes. Different strengths made with closely related formulations can be subject to B&M with some justification. What constitutes a closely related formulation is not defined, but examples could be: 				

Different strengths are obtained by minor changes in one excipient (say for 2, 4 and 6 mg tablets where one excipient changes to accommodate the varying active). 

Different coatings on otherwise identical tablets. 

The justification might be a statement confirming that either experiments demonstrating that varying the excipient or coating has no effect on the active, or that earlier development stability studies showed the changes are unlikely to have any effect upon stability. 

B&M can also be applied to studies of the same closure system where either container or fill size varies. For bracketing, care must be taken to ensure that the intermediate condition is correctly identified as it might not necessarily be that the largest and smallest container represent the extreme configurations. Consideration should be given to wall thickness, closure geometry, surface area to volume ratio, headspace to volume ratio, water vapour and oxygen permeation rate per dosage unit.  

Even when the closure varies bracketing and matrixing is possible with some justification. Such justification might be a demonstration that the product is not water sensitive or a discussion of the relative permeation rates of the closure systems. 

Many regulatory authorities welcome the opportunity to comment on stability protocols prior to formal submissions and I would recommend in all instances where B&M are applied (and even when they aren't) that the applicant takes advantage of this, especially with FDA.  

B&M can still be applied when all batches cannot be placed on storage simultaneously. The allowable difference in start times should be decided case by case as a business risk, because identification of any stability differences between batches will be delayed; as a guide up to 2 months' spread appears reasonable. Clearly, if studies are not started simultaneously the efficiency gains will be reduced. 

 Bracketing designs are fairly straightforward and have been around for longer than matrixed designs. A typical example of bracketing, as given in the CPMP paper, is reproduced as Table 1. This example is a product available in three different strengths, two different packs and for one of these packs, three different sizes. In this instance it would be necessary to demonstrate that the 30 and 500 high density polyethylene (HDPE) sizes bracket the 100 size as discussed earlier.  

Table 1 Example of a bracketing design.			

If, subsequent to commencing studies, it is decided not to register one of the extreme presentations, it is permissible to maintain the stability study to support any intermediate presentations that have not been placed on stability. A commitment should then be provided to place the omitted presentations on stability. 

 For matrixing, each storage condition should be treated separately under its own matrixing design. Accelerated conditions can be matrixed, but at least three time points, including initial and final, must be analysed, therefore, gains are minimal and personally I prefer full testing as accelerated conditions may provide early warning of any unexpected events. 

As far as possible a matrixing design should be balanced so that each presentation (of strength, pack, fill volume etc.) is tested to the same extent over the intended study period. Full testing must be performed at the maximum storage period at the time of submission, frequently rendering exact balancing impossible. However, the design should be approximately balanced at least. 

Examples of simple matrixed designs for a product in two strengths are shown in Table 2 (assuming submission with 12 months' data). Table 2(a) shows a one-half matrix design (or one-half reduction) and Table 2(b) — a two-thirds design (one-third reduction), giving reductions in the number of test samples of 31% and 21% respectively. These examples are for definitive stability and hence three batches are tested. Fewer batches may be permissible for follow up stability, or an abbreviated new application, but matrixing is still allowable.  

Table 2 (a) One-half matrix design. (b) Two-thirds matrix design.			

Note that these designs have assumed 36 months as the longest time point. The 36-month (and 48-month) time point could also be matrixed if one intends to test up to 48 months (or 60 months).  

				Note that in 2(b) interchanging the Xs and Os, excluding the initial, 12 month and final time points, makes it a one-third matrix (two-thirds reduction, a 42% reduction).			

With matrixing, designs can be complete (matrixing on time points alone), whereby all batches tested in all possible presentations, or incomplete (matrixing on time points and factors), whereby some batches are not tested in some presentations. The larger the number of different presentations the more likely it is that an incomplete design is acceptable. An example of a more complex design, with six factors (three strengths and three packs) using a one-half design with full testing at 12 months is given in Table 3. Table 3(a) shows a complete reduced design and 3(b) an incomplete reduced design. In 3b, while all combinations of strength and pack are tested, each individual batch of product is not tested in all strength/pack combinations.  

Table 3 (a) Complete reduced design (matrixing on time points alone) for three strengths and three packs. (b) Incomplete reduced design (matrixing on time points and factors) for three strengths and three packs.			

ICH Q1D provides other example designs. It should be noted that it is permissible to combine B&M in a single design. 

If, during the course of a matrixed stability study, it becomes apparent that the matrix is no longer appropriate, for example, if the required expiry period might not be achieved, you can revert to full testing or to a smaller reduction in testing. However, once matrixing is reduced it is not permissible subsequently to reinstate the original matrix. Therefore, it is good practice to put all samples on stability, including those that are not to be tested under the matrix design. In practice, for full stability, extra samples are usually stored to allow for any repeat analyses, so when storing samples for a reduced design you only need a sufficient amount for full testing, or a very small excess, to save on samples stored yet have enough to cover emergencies. 

It is also worth noting that should the worst happen, such as when reaching a key storage time it is found that the reduced data do not support the required shelf life, then extra replicate analyses can be performed to obtain an improved shelf life estimate. Any statistical treatment should allow for the extra data at the last time point. 

Different tests may be matrixed to varying extents. For example, if the test for active content is highly reproducible and the active is stable, then this test may be subjected to a one-half, or possibly even a one-third, matrix design. If, however, the test for a key impurity shows significant variability and there is an indication that the impurity might increase on storage, then a two third matrix, or possibly no matrix at all, may be more appropriate. In practice, I usually kept things simple and matrixed all tests to the same extent. 

The degree of reduction in testing is dependent on the variability between, and homogeneity within, batches and the precision of the test method. Generally, greater numbers of presentations support larger test reductions. The intrinsic stability of the formulation is also very important as reductions in the number of samples tested may result in wider confidence intervals and, therefore, a shorter shelf life estimate. Hence, for a product likely to have an inherently short shelf life B&M should be minimal, while greater reductions may be appropriate for a product likely to significantly exceed a 36-month shelf life. 

Table 4 gives some qualitative feel for the reductions that may be applied according to variability and anticipated expiry period (EP).  

Table 4 Degree of reduction in bracketing and matrixing.			

Stability data from bracketed and/or matrixed studies should be treated in the same way as data from nonreduced studies. If inspection of the data clearly shows that there is no significant change in a parameter then no statistical analysis is required. Where statistical analysis is required the data from different batches and presentations should first be tested for poolability. If pooling is not justified then the shortest EP obtained should be applied to all batches and presentations. If pooling is justified then the predicted EP applies to all presentations. 

The use of bracketing and matrixing for optimizing the design of stability protocols has been described. They can potentially provide considerable savings in stability studies costs as the amount of testing required is reduced, fewer samples have to be labelled and placed on storage, and consequently the space required to store them is also reduced. 

The procedures are easy to apply, well established, accepted by the regulatory authorities and are described in ICH guidelines. Therefore, there is no reason why their application should not be widespread throughout the pharmaceutical industry. 

1. International Conference on Harmonization  ICH Q1D (2002) 

Articles

This article looks at the use of bracketing and matrixing to lower the number of stability samples required and, consequently, reduce the cost of sample production, testing and management. There is a common misconception that regulatory authorities will not accept such methods, but there is actually an International Conference on Harmonization guideline (ICH Q1D) on the subject. In fact, many of these designs have already been accepted and FDA members were among the first to describe matrixing.

The role of bracketing and matrixing in efficient design of stability protocols

The validation of cleaning processes has long played a critical role in pharmaceutical manufacturing. The US Food and Drug Administration (FDA) requires firms to produce written procedures detailing the cleaning processes used for various pieces of equipment, and how those processes will be validated. Typical detection limits mentioned in the pharmaceutical literature are 10 ppm or biological activity levels of 1/1000 of the normal therapeutic dose. Most of the pharmaceutical industry currently uses high performance liquid chromatography (HPLC) to evaluate samples collected from production machines. Highly trained technicians in an analytical laboratory prepare the mobile phase and diluents and set up the HPLC instrument. Each sample collected is then prepared and, typically, through the use of an autoloader, injected and processed consecutively without operator intervention. As cleaning and sample preparation normally occupy a full working day, analysts load the instrument at the end of a one day then, on the following morning, perform any necessary calculations and enter the sample results in the company's laboratory information management system (LIMS). Production usually resumes in the middle of that day, resulting in a minimum loss of one and a half day's production.	

The ion mobility spectrometry (IMS) limit test described in this article defines a pass/fail threshold such that when the IMS signal is below the threshold, it can be proved at a defined confidence level that the measured concentration is truly below the limit. This approach takes advantage of the greatly reduced instrument set up time for IMS and the reduced sample analysis time; typically 5-20 s per sample for IMS compared with 10-20 min for HPLC (Table I). By performing multiple runs with a standard sample at the cleaning action level, the standard deviation of the measurement process is obtained which, combined with the required confidence level, is used to set the IMS signal pass/fail threshold. The result is that pharmaceutical manufacturers have the potential to save millions of dollars by dramatically reducing the time required to verify the cleaning of production equipment before switching over to another product or batch.	

In an effort to reduce the time required for cleaning validation and verification, IMS has been studied as an alternative to HPLC. IMS refers to the principles, practice and instrumentation of characterizing chemical substances based on their gas phase ion mobilities - determined by measuring drift velocities as ions move, under the influence of an electric field, through a gas at ambient pressure.1 	

Table I:  Comparison between a traditional HPLC approach and IMS limit test approach.				

Typical pharmaceutical compounds are thermally desorbed to vaporize the sample. The vaporized sample is then introduced into the IMS via a carrier gas stream before being selectively ionized, typically by an atmospheric chemical ionization (APCI) source. An electronic gate then opens periodically, to admit a finite pulse of product ions into the drift tube. The ions migrate downfield and strike a collector electrode, producing a current. The ion current is amplified and displayed as an ion mobility spectrum or plasmagram, showing ion current versus time. Figure 1 illustrates a cross-sectional view of a drift tube.	

The ratio of the drift velocity of a given ion to the applied electric field strength (ion mobility) depends on the charge, mass and collision cross-section of the ion. The cross-section, in turn, is dictated by molecular size, shape and charge distribution. As size generally scales with mass, correlations exist between the mobility and mass of an ion. Lighter, smaller ions have greater mobility values and therefore, exhibit shorter drift times. Ion mobilities, calculated from the drift times, provide information regarding the chemical identities of observed ions and their neutral precursors. The intensity of each peak in the plasmagram is related to the abundance of the corresponding ion. Thus, quantitative compound information may be obtained by analysing the amplitude or area of an ion peak derived from that compound. 		

Figure 1: A cross-section view of the drift tube.	

Most HPLC-based cleaning validation methods typically implement a one-point calibration based on the assumption that the response is linear and passes through zero (validated during the method development phase). The sample response is then converted to a concentration. The IMS method involves a one-point calibration to define the instrument response based on the assumption that samples with amounts less than the calibration level are clean (validated during the method development phase) and the sample response is converted into a simple clean/dirty response with an associated confidence level. In a sense, the HPLC method could be used as a limit test but the IMS would analyse at least 60 times faster (Table I).				

The first step is surveying the instrument mass response curve (Figure 2) to select the target mass response point, which provides a compromise between the higher response per unit mass obtained with small masses and the smaller relative standard deviation obtained with larger masses. The limit of linearity is selected because the linear region has the highest sensitivity and generally provides the lowest relative standard deviation. Once the target mass is selected, the swabbing methodology and sample extraction volume is defined using the following factors:				

sample volume used to deliver sample mass to IMS instrument (typically 1-10 mL)

target concentration (determined using target mass and sample volume)

target cleaning level on IMS instrument (typically 100-1000 ng/cm2)

assume worst case swabbing recovery (typically 60-90%)

The above factors are used so that a swab right at the action level will be diluted to the target concentration. 	

For example, if target mass 5 1 ng, sample volume 5 1 mL, target cleaning action level 5 200 ng/cm2, area to be swabbed 5 100 cm2 and swabbing recovery 5 80%, then the target concentration is 1 ng/mL (1 mg/mL) and the action level on the swab is 16 mg. Thus, the volume to extract the swab should be 16 mg/swab/1 mg/mL, which equals 16 mL per swab.	

The next step is determining the signal and uncertainty at the target concentration. This is accomplished by measuring the IMS response for several samples at the target concentration and calculating the mean (x) and standard deviation(s). To determine if the response observed from a test sample is significantly different from this value the Student's t-test can be used, as shown below in Equation 1 where x1 and x2 are the mean values associated with the target and test samples respectively, s is the pooled standard deviation of the target and test samples, n1 and n2 are the samples associated with the target and test samples respectively. 	

The tcalc value is compared with a value obtained from a standard Student's t table for the appropriate confidence level and number of observations. If then the two means are significantly different. The pass/fail threshold can be calculated based on the tabulated value of t for the required confidence level and the number of observations. The threshold value (xt), the maximum instrument response for which it is possible to state with the confidence level in question that the sample is clean, is calculated by rearranging Equation 1 as follows:	

where x1 and x2 are the mean values associated with the target and test samples respectively, s is the pooled standard deviation of the target and test samples, n1 and n2 are the samples associated with the target and test samples respectively. 

The tcalc value is compared with a value obtained from a standard Student's t table for the appropriate confidence level and number of observations. If (the following equation)  then the two means are significantly different. The pass/fail threshold can be calculated based on the tabulated value of t for the required confidence level and the number of observations. 

The threshold value (xt), the maximum instrument response for which it is possible to state with the confidence level in question that the sample is clean, is calculated by rearranging Equation 1 as follows:

Suppose that 10 analyses are run on a standard and the mean IMS response is 100 digital units (du) with a standard deviation of 5 du. Then, it can easily be determined that any solution producing an IMS signal of less than 83 du for one measurement is clean - with at least 99.5% confidence. It is important to note that the short run-time of IMS means that it is sensible to perform extra standard measurements, because they provide better predictions for the standard deviation, increase the limit threshold and, therefore, lower the chances of a false positive.	

In most cases, the swab will test as 'clean' when first measured. In some instances, the swab response will be well above the threshold, indicating the need for further cleaning. In rare cases, the swab response will be slightly above the threshold. IMS generates results so quickly that in this situation it is advisable to analyse a few more aliquots of the sample before deciding to reclean. The swab can be declared clean if the mean of replicate samples is less than the threshold. The pass/fail threshold will be higher when replicate samples are analysed because of the nature of the t-test. 				

Several important precautions should be taken during methods development to protect against false negatives. The first is to verify the consistency of swab recovery, a requirement that also applies to HPLC. The second is to check the instrument response across the range of interest to validate the system response. The third is to evaluate the effect of potential interferences (excipients, for example) which may be present. This is because, occasionally, the analyte signal may be suppressed by the presence of particular compounds, particularly detergents or excipients. 	

Instrumentation. The IonScan-LS (Smiths Detection, Warren, New Jersey, USA) ion mobility spectrometry instrument was used, and the instrumental parameters employed are listed in Table II.				

In a typical analysis, a substrate containing the sample of interest is placed on the desorber, which is maintained at a fixed programmed temperature of 290 Â°C. A carrier gas, air, transports thermally volatilized sample material into an ionization/reaction region. Volatilized compounds are selectively ionized using a 63Ni ß-source and a controlled chemical ionization environment to produce molecular ions or ion clusters. The ions are then gated into the drift chamber, at atmospheric pressure, where they are accelerated under an applied electric field toward a collector electrode.	

Identification of compounds is based on the calculation of their characteristic reduced ion mobility K0 (cm2/Vsec) values. K0 is determined using the following equation:	

where tcal is the drift time of the internal calibrant and tobs is that of the observed peak. The K0 of the internal calibrant is a known value and the drift times of the calibrant and observed peak are experimentally measured values. In the external calibration procedure, the internal calibrant K0 value to be used in this equation is set so the K0 observed for the external (primary) calibrant has the value shown in Table II.	

The polarity of the electric field applied to the drift region is either positive or negative, allowing for the analysis of positive or negative ions. Ions of the correct charge are accelerated from the reaction region towards the drift region. Each scan of the IMS spectrum starts when the gating grid opens briefly to admit a burst of ions into the drift tube, and ends just before the gating grid opens again. This interval is the 'scan period.' The data from several scans are co-added together to improve the signal-to-noise ratio and is called a 'segment.' A series of segments with characteristic ion peak patterns for the sample are obtained and can be displayed either as a series of individual segments versus desorption time in seconds (a 3-D plasmagram) or as an average of all segments obtained during the analysis (a 2-D plasmagram).			

Materials. Product X (a novel compound under development) and excipients, provided by GlaxoSmithKline, were used as supplied. Pesticide-grade acetone was used to make up the Product X samples and the excipients were made up in water or ethanol. Teflon substrate (0.45 micron porosity) was obtained from Osmonics (Minnetonka, Minnesota, USA).	

Methods. A 1.0 mL aliquot of the sample was deposited on the Teflon substrate using a 1.0 mL syringe (SGE, Melbourne, Australia) and allowed to evaporate for 15 s prior to analysis. 			

Table III:  IMS response of the target level standard.	

Eight standards containing 1 ng of Product X were analysed to determine the pass/fail threshold. Forty eight test solutions - ranging from 0.25-10 ng - were then analysed. The goal was to verify that the limit test would identify samples containing less than 1.5 ng of Product X as clean and those containing at least 1.5 ng of Product X as dirty. The target level data, listed in Table III, can be summarized as follows:				

Eight measurements of the target level produced a mean response of 398 du. The pass/fail threshold calculated from these eight data points is 339 du. That is, if one analysis of a test sample gives an instrument response of less than 339 du, we can state with at least 99.95% confidence that it contains less than the action level amount of the active pharmaceutical ingredient (API).	

Regarding the test sample data, 24 clean samples (containing less than 1.5 ng of the API) were analysed (Table IV). In 23 of these cases, the sample passed (the instrument response was below the threshold). There was, however, one false positive. For the 24 dirty test samples there were no false negatives. All 'dirty' samples tested as such.		

Figure 3:  IMS plasmagrams of Product X, with and without excipients.	

In the case of the one false positive, the prescribed course of action would be to analyse two additional aliquots of that sample and to apply the limit test to the mean of the three results. The mean would be compared with a revised threshold, which is based on more measurements and is therefore higher. As a worst case example, the three highest responses for 1 ng (329, 330 and 344 du) give a mean response of 334 du, which is less than the revised threshold of 365 du for three test samples. The test sample would therefore be declared clean.	

A limit test using ion mobility spectrometry (IMS) has the potential to dramatically reduce the time required for cleaning verification and cleaning method development. The traditional approach to cleaning verification, often using HPLC, is relatively resource intensive and can lead to significant delays in reporting results. The main advantage of IMS is that results are seen virtually instantaneously, so any necessary re-measurement can be done very quickly. If the results demonstrate cleanliness, production can resume in a matter of hours not days.