Analytical Target Profile

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282914684

“The Use of the Analytical Target Profile in the Lifecycle of an Analytical


Procedure: with an example for an HPLC Procedure”

Article · January 2015

CITATIONS READS
3 1,472

1 author:

Ron Snee
Snee Associates, LLC
403 PUBLICATIONS   6,151 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Strategies of Formulations Development: A Step by Step Guide Using JMP View project

Data Mining and Analysis - Practical Approaches View project

All content following this page was uploaded by Ron Snee on 19 November 2015.

The user has requested enhancement of the downloaded file.


Peer Reviewed: Analytical Procedure Journal of Validation Technology

The Use of the Analytical Target Profile in the Lifecycle of an Analytical Procedure
with an Example of for an HPLC Procedure
Jane Weitzel, Robert A. Forbes, and Ronald D. Snee

Introduction ufacturing processes and analytical quality by alytical procedure as described in the USP [9 actual results, are provided mainly for illustra-
The lifecycle management of an analytical design (aQbD) is that it will allow the applica- –11] and the ICH [12], will be discussed and il- tive purposes.
procedure applying Quality by Design (QbD) tion of the experience, knowledge, tools, and lustrated in the example. The terminology used In this example, we will focus on the preci-
concepts and using the analytical target profile statistics from QbD to be applied to aQbD. with both approaches will be compared and sion and uncertainty performance characteris-
(ATP) has been introduced by the USP stimu- Thus, all parties, laboratories, companies, Qual- explained. tics of the potency test method to demonstrate
li article “Lifecycle Management of Analytical ity Assurance, production, as well as regulatory The aQbD approach is illustrated using an how the gage R&R tool and experimentally de-
Procedures: Method Development, Procedure bodies can leverage their experience with QbD example of the potency test for a drug sub- signed intermediate precision studies are used
Performance Qualification, and Procedure Per- in developing an approach to aQbD. stance (DS) with specification limits of 98.0% for Procedure Performance Qualification and
formance Verification [1].” Current regulatory A key component of the aQbD paradigm to 102.0%. The potency test is one component Verification. The acceptance criterion for the
guidance around process validation is moving is the definition of the Analytical Target Pro- in the DS control strategy utilized to ensure the precision of the potency method will be set us-
away from a simple once-and-done validation, file (ATP) [1], analogous to the Quality Target safety and efficacy of a drug product (DP) man- ing the Lifecycle approach where the intended
and toward a continuous process verification Product Profile (QTPP) in the ICH [8]. The ufactured from the DS. The test ensures that the use of the method is defined using a decision
approach, also called a “lifecycle approach” [2]. ATP defines the requirements for the result of DS is of adequate potency, so that when used in rule. The decision rule defines the acceptable
The recent FDA method validation draft guid- a test method and is based on the suitability the DP, the patient will receive the appropriate level of probability of making an incorrect de-
ance also includes some discussion of a lifecycle for use of that result. This paper illustrates how dose of the active ingredient. In this way, poten- cision. For this discussion, the null hypothesis
approach for analytical methods [3]. the ATP is developed and applied to assess the cy is a critical quality attribute (CQA) of the DS is “The potencyof the sample is within specifi-
A number of articles have been written in- performance of the method at various stages of that is linked to a CQA for the DP [13]. For sim- cation,” so there are two different types of incor-
troducing the lifecycle management of analyti- its lifecycle. The use of traditional Gage R&R plicity, we will assume the limits are on the as-is rect decisions. First, a batch that has acceptable
cal procedures [4 – 7]. These articles introduced statistical analysis and control sample charting, basis, and no correction for water or solvents potency may test outside the specification limits.
the concept that analytical procedures are pro- which are commonly used in manufacturing is necessary. Obviously, there are other speci- Second, a batch with a potency truly outside of
cesses and, as such, the tools, approaches and process validation and improvement, will also fication tests (e.g. impurities, residual solvents, the specifications limits, could test within limits
statistical analyses for manufacturing process- be illustrated in the development of an analyt- etc.) that are required to ensure the DS meets its and be accepted. The first error can be consid-
es can be applied to analytical procedures. The ical procedure. The similarities, contrasts, and CQAs, but these will not be considered in this ered a false positive (positive test for unaccept-
benefit of recognizing and acknowledging the differences between the lifecycle approach, and illustration. It should also be noted that the DS able potency), or a type I error. The second er-
similarity between quality by design for man- the traditional approach to validation of an an- potency results used for this illustration, while rror would be a false negative (negative test for

www.ivtnetwork.com
Jane Weitzel, Robert A. Forbes, Ronald D. Snee
unacceptable potency) or a type II error. The the laboratory sample, continues with the treat-
consequences of each of these types of errors ment of that sample which may involve taking
are taken into consideration when developing a test-portion, comminution, extraction, disso-
the decision rule. The ATP uses the probability lution, etc. and concludes with the output, the
of each of these errors to define the maximum reportable result as shown in Figure 2.
allowed measurement variation (uncertainty)
for the reported result, which translates into an The FDA definition for process validation is

Figure 3 QbD for analytical procedures is analogous to that terms for the major steps in the QbD approach.
for production processes. This figure shows the analogous

throughout its lifecycle of use which establish- of variability.


es scientific evidence that a method is capable For the production process, the critical
of consistently delivering quality data [4].” For quality attributes are determined. The analogy
this article, the method is the same as the an- for the analytical procedure is the performance
alytical procedure and includes one full execu- characteristics required of the method. These
tion of the method, starting from the original include the accuracy/bias, linearity, limit of
sample, to produce a reportable result as de- detection, repeatability, intermediate precision,
fined in the US Food and Drug Administration etc. The critical variables are identified and
out-of-specification (OOS) guidance [14]. The characterized and a suitable control strategy is
Figure 1 shows that the intended use of the reportable result defines the decision rule, which includes the acceptable prob- reportable result is compared to the specifica- designed. For production the production de-
ability of making a wrong decision. The Analytical Target Profile uses the probability to determine the maximum allowed tion to determine compliance. sign space is defined, while for an analytical
measurement variation for the reported result or target measurement uncertainty (TMU). The TMU is an acceptance
The analogy between QbD for the produc- procedure, the performance characteristics that
criterion for the performance characteristics. The process is continuous because it can be repeated if the requirements of make up its design space are defined.
the ATP cannot be met. tion process and for an analytical procedure is
illustrated in Figure 3. QbD in both areas starts
acceptance criterion for the qualification of the “the collection and evaluation of data, from the with a clear definition of the intended use of the Decision Rules
analytical procedure as shown in Figure 1. process design stage through commercial pro- output, or product, of the process. For produc- Construction of a Decision Rule
duction, which establishes scientific evidence tion of a drug product, this considers items such Decision rule construction requires four com-
The Analytical Procedure as a Process that a process is capable of consistently deliver- as the route of administration, dosage form, ponents: the measurement result, its measure-
A process is a series of actions or steps taken ing quality products [2].” Treating an analytical bioavailability, strength, and stability. For ana- ment uncertainty, the specification limit or lim-
in order to achieve a particular end or output. procedure as a process, the analogous defini- lytical procedures, the output/product is the re- its, and the acceptable level of the probability of
The output of an analytical procedure is a re- tion is “the collection and evaluation of data portable result, and a decision rule can be used making each type of wrong decision [15 - 16].
portable result. The series of steps begins with and knowledge from the method design stage to clearly and quantifiably state the intended use The analytical quality by design approach uses
of the result [15 - 16]. risk analysis, including assessment of probabil-
For an analytical procedure, the predefined ities, to determine those four components of a
objective is the analytical target profile (ATP). decision rule. The relationship between the four
Criteria defined in the ATP refer to the quality components allows one to determine if an ana-
data attributes of the reportable result (i.e., bias lytical procedure is fit for use and to set accep-
Figure 2 shows the analytical procedure as a process that starts with the taking of the laboratory sample and concludes with [accuracy] and target measurement uncertainty tance criteria that the analytical procedure must
the output of a reportable result. (TMU) [precision]), which includes all sources meet.

Journal of Validation Technology Volume XX Issue 4


Peer Reviewed: Analytical Procedure Journal of Validation Technology
Simple Decision Rule ty (i.e., measurement variation). A type II error to a measurand (the quantity being measured), cedure validation guideline [12]. Reproducibil-
For the drug substance potency example, both occurs when the potency test result is within based on the information used” [17]. For an ity is the broadest measure of method precision
the decision rule and the acceptance limits are specifications, when the true batch potency is analytical procedure, the uncertainty includes as it includes all sources of variability in the es-
generally defined by the expectations of reg- outside of specifications. In this example, the all the random sources of variability, such as re- timate, including multiple testing laboratories.
ulatory authorities, so the manufacturer may type II error rate will be assumed to be negli- peatability, intermediate precision and the un- The ICH uses the term intermediate precision
not have much flexibility to use alternative ap- gible, which means that the process delivers certainty in the estimates of the accuracy or bias. to distinguish the precision of the method with-
proaches. A range of 98.0 to 102.0% is generally consistent, high quality DS (process variation is As such, it can be viewed as a high level, overall in a single lab from precision over multiple labs
expected for a commercial drug substance. This negligible). The DS is also assumed to be of high requirement for the total precision. (reproducibility). Normally the test for compli-
range ensures that the drug product manufac- purity, containing negligible water (non-hygro- It is important to note that there are many ance of a given batch with specification limits
tured from the DS will meet its potency specifi- scopic), solvents, and inorganic content, which definitions of different types of precision based is carried out in a single laboratory. Therefore,
cation limits, which are generally 90 to 110% in means the true DS potency is approximately on the sources of variability they include. Addi- the intermediate precision estimate is generally
the US and 95 – 105% in the EU. This approach 100%. The probability of a type I error (rejection tionally, different industries, regulatory bodies, the most appropriate precision estimate for the
is described by a simple decision rule in which of a batch when the true batch potency is within and references define the types of precision assessment of method capability. In this work,
the specification zone, the values between and specifications) is determined by the measure- differently. As a result, it is best to clearly define the term repeatability will be used to describe
including the specification limits, is the same as ment variation in combination with the width the precision terms being used by identify- short-term variability (i.e., repeat preparations
the acceptance zone and defines the values for of the acceptance zone. This probability is also ing the sources of variability that are included. within a single run) and intermediate precision
which the product is accepted. The values out- called the type I error rate. The VIM takes this approach by including the to describe long-term variability within a single
conditions of measurement in its definition of lab (i.e., preparations analyzed over different
precision: runs/setups). Note that the terminology associ-
ated with the use of the Gage R&R tool does not
“2.15 Measurement Precision make the distinction between within-lab and
closeness of agreement between indications or inter-lab precision estimates, so here the term
measured quantity values obtained by replicate “reproducibility” means the same thing as “in-
measurements on the same or similar objects un- termediate precision.”
der specified conditions … The measurement uncertainty characteriz-
Note 2 The ‘specified conditions’ can be, for exam- es the dispersion of the values being attributed
ple, repeatability conditions of measurement, in- to a measurand. This dispersion is assumed to
termediate precision conditions of measurement have a normal distribution. The basis for this as-
or reproducibility conditions of measurement” sumption is discussed in Annex G of the GUM
[18]. Even if the distribution is not normal, the
Figure 4 The figure illustrates a simple decision rule for a specification with both upper and lower limits. If the measurement The VIM definition of precision is very sim- same approach will be used.
result lies in the acceptance or specification zone, product is accepted, otherwise it is rejected. ilar to the ICH definition in the analytical pro- Uncertainty can be illustrated by imposing

side the acceptance zone comprise the rejection The consequences for a type I error will de-
zone. This simple decision rule is illustrated in termine what probability is acceptable. Since
Figure 4. there is little flexibility to increase the accep-
This simple decision rule can be stated as: tance zone, the measurement uncertainty is the
only parameter that can be adjusted in order to
The product will be considered compliant if the meet the acceptable error rate.
measurement result is within the acceptance
zone. Measurement Uncertainty
Measurement uncertainty is defined in the in-
With this decision rule, the probability for ternational vocabulary of metrology (VIM) as a
either type of error is a function of both the pro- “non-negative parameter characterizing the dis-
Figure 5 shows the normal distribution in relation to the specifications of 98.0% to 102.0%. The width of the normal distribu-
cess variability and the measurement uncertain- persion of the quantity values being attributed
tion is chosen so that the probability the true result is below98.0% is 2.5% and above 102.0% is 2.5%.

www.ivtnetwork.com
Jane Weitzel, Robert A. Forbes, Ronald D. Snee
a normal distribution over the decision rule as Target Profile. Target measurement uncertainty method precision to avoid failing to meet the able batch will fall outside this range must be
shown in Figure 5. In this example, the true drug (TMU) is defined in the VIM as the “measure- TMU. Larger studies with more runs are more NMT 5%.
substance potency is assumed to be 100.0%, ment uncertainty specified as an upper limit costly in terms of time and resources, so one
which establishes the center of the normal dis- and decided on the basis of the intended use of must weigh the benefit of greater confidence in The analytical target profile can now be de-
tribution. The width of the normal distribution the measurement results” [19]. Any analytical the precision estimate, versus the additional cost termined. The laboratory qualifies the analytical
is determined by the standard deviation which procedure that yields a reportable result with a of larger studies. procedure for a concentration range wider than
was set to 1% to coincide with a 2.5% probabili- standard uncertainty of 1.0% or less, given there the specification to allow for variability in sam-
ty of obtaining values each above and below the is no bias, will be fit for its intended use. Relationship Between Accuracy and Target Mea- ple concentration. The matrix and measurand
limits. As shown in Figure 5, 5% of the potency The probability associated with the decision surement Uncertainty are defined clearly. There is only one known
results for the sample will fall outside the limits rule is based on the theoretical normal distri- The probability of not meeting the specifica- impurity, Impurity A, and it occurs below 0.1%.
due to measurement variation (type I error rate) bution and the specified standard deviation. tion depends on where the true distribution is The ATP can now be written:
for the assumed drug substance potency and It does not take into account the variability in centered. If the true distribution is not centered
precision described above. This distribution the estimates of the uncertainty that will be at 100.0%, as in the above example, then a dif- The analytical procedure must be able to quan-
only captures the measurement uncertainty and obtained in the qualification and verification ferent proportion of the measured results are tify the drug substance in the presence of impu-
does not include process variation (variation in experiments. That is dealt with in the experi- expected to be outside the specifications. When rity A, and potential degradation products, over
the true batch to batch potency), which was as- mental designs of the development and qualifi- this occurs, the target measurement uncertain- a range of 80% to 120% of the nominal concen-
sumed to be negligible. cation experiments, and the desired confidence ty must be improved in order to maintain the tration with an accuracy and uncertainty so that
If a 5% type I error rate is acceptable, the in their resulting precision estimates. Generally, same type I error rate. However, the analytical the reportable result falls within ±2% of the true
standard deviation of 1.0% becomes the target higher confidence limits will require more ex- procedure could have a bias that cannot be value with at least a 95% probability.
measurement uncertainty for the Analytical perimental runs to provide better estimates of removed in method development. If it is not
corrected for, then the TMU will have to be ad- Performance Characteristics for the
justed. Or if the drug substance does not have a Analytical Procedure
potency of 100.0%, this will impact the degree The ATP can now be used to determine the
of precision that the method will need to attain. performance characteristics for the analytical
For example, if the DS has an actual potency procedure. The following performance char-
value of 99.0%, then the TMU must be reduced acteristics need to be included in the method
to 0.6% (Figure 6). qualification: accuracy, precision, specificity,
linearity, and range.
Analytical Target Profile The acceptance criteria for specificity (i.e.,
Based on a risk analysis that includes consider- resolution of the main peak from impurity A
ations such as the cost of OOS investigations, and potential degradation products) and linear-
the known characteristics of the product (the ity are determined based on their impact on the
drug substance is known to have few impuri- accuracy and precision. As long as their impact
ties and the potency is close to 100%), and the on the accuracy and precision is such that the
capability of the analytical technique (HPLC) a target measurement uncertainty can be met,
company can decide the probability it is willing they are acceptable. This is a key advantage of
to accept associated with making an incorrect this approach; it defines the performance that is
decision (type I error) concerning a product “good enough.” One can confidently accept the
specification. In the below example, the proba- performance because it is directly linked to the
bility selected is 5%. intended use.
The decision rule can be more detailed now: The performance characteristics and the ac-
ceptance criteria are summarized as follows:
The batch of drug substance will be considered
Figure 6 (a) shows that if the normal distribution is centered over 99.0% the probability of making a wrong decision is in-
creased to 16.3%. (b) shows the target measurement uncertainty must be reduced to 0.61% in order that the probability of compliant if the test result is within the range of The target uncertainty is NMT 1.0%. The com-
making a wrong decision remains at <5%. 98.0 to 102.0%. The probability that an accept- bination of bias and uncertainty need to be such

Journal of Validation Technology Volume XX Issue 4


Peer Reviewed: Analytical Procedure Journal of Validation Technology
that the probability of not meeting the specifica- estimate obtained. However, the uncertainty in in this study was 0.15%, with an 80% UCL of a key element of the transfer activity is to show
tion is NMT 5%. the precision estimate obtained in the qualifi- 0.20%, which met the acceptance criteria of that the precision of the method as performed
cation study should also be taken into account. NMT 1.0% SD. in the QCL is sufficient to meet the required un-
To assess the accuracy of the method when Generally a confidence interval (CI), or upper A limitation of this study was the fact that certainty as described in the ATP. The required
performed in a new laboratory, a reference stan- confidence limit (e.g., 95% UCL), is calculated only a single DS batch was tested, so potential precision would be included in the PPQ proto-
dard, or a previously tested sample with a de- for the standard deviation estimate to provide variability caused by differences in behavior of col as an acceptance criterion for the intermedi-
fined potency could be used. Results obtained an upper limit below which the actual standard different batches of the DS was not included. In ate precision study; in our example the accep-
in the new laboratory would be compared to deviation is likely to fall. The width of this inter- the case of a high-purity DS, with very consis- tance criteria would be NMT 1.0%RSD.
the defined result to test for a bias. If a bias is val is impacted by both the desired confidence tent properties from batch to batch, this source One approach to evaluate the method preci-
detected, the combination of the uncertainty for the estimate and the number of setups in the of variation is generally small but this is not al- sion after transfer to the QCL would be to per-
of the bias with the intermediate precision es- study. For more detail on the calculation of the ways the case. form an intermediate precision study similar to
timate would be compared to the allowable confidence interval, see the stimuli article in the the study performed for method qualification.
measurement uncertainty defined by the ATP. USP Pharmocopoial Forum [20]. Procedure Performance Verification An alternative approach is to use the Gage R&R
For this example, the bias (accuracy) is assumed The design and results of one such study are To further illustrate the utility of the ATP, its use tool which measures the repeatability and re-
to be negligible so that the precision component shown in Table 1, where four analysts produced in the next stage of the lifecycle of the potency producibility of the method [21]. While the
can be focused on. If the bias were significant, duplicate potency results for one batch on each method will be considered. This is the stage Gage R&R study may be new to many in the
then the analytical procedure should be im- of 8 setups using a different HPLC instrument in which the method is installed in the release pharmaceutical and biotech worlds, the method
proved to remove that bias, or the bias should and column on each setup. For this study we testing quality control laboratory (QCL). At this has been used widely in the process industries
be corrected for and the uncertainty of that cor- will choose 80% as the desired confidence for stage, the method has been fully developed and including auto, chemical and electronics. Gage
rection included in the uncertainty estimate. the precision estimate. This means that if we validated to demonstrate its appropriateness for R&R studies are also widely used in Six Sigma
Now that the ATP has been established, its obtain a standard deviation of 1.0%, we will its intended use, and some history of its perfor- process improvement studies [22]. It is not un-
use in lifecycle management of the analytical accept a 20% chance that the actual standard mance to test manufactured DS batches at a de- common to find process problems are due to
method will be demonstrated. deviation may be greater. This also means that velopment scale exists. The DS is a new chemi- poor measurement systems with the problem
the acceptance limit for the standard deviation cal entity that has been submitted for regulatory significantly reduced, if not eliminated, by im-
Procedure Performance Qualification is effectively NMT 0.74%, which corresponds to approval and the manufacturing process is proving the measurement system. Note that for
During the validation of the analytical meth- an 80% UCL of 1.0% for an 8-run study. Since being installed and validated in the long-term the Gage R&R study (as described under Mea-
od, the precision of the method is evaluated the method prescribes averaging the results for manufacturing facility. The analytical method surement Uncertainty above) we use the term
by studies to determine its repeatability, and its two sample preparations on each setup, the re- is being transferred to the long-term release lab, reproducibility to mean the same thing as the
intermediate precision. Because it includes both portable result for a batch is the average (AVG). and is undergoing performance qualification at intermediate precision, and not inter-laborato-
long-term and short-term sources of variation, The overall standard deviation (SD) obtained the new site. Some data to characterize both the ry variation as defined by the ICH. Repeatabil-
the intermediate precision estimate is consid- capability of the method and the capability of ity is the within setup precision, reproducibility
ered a representative measure of the capability the manufacturing process are available. comprises precision across multiple setups. In
of the method for ongoing batch release test- Installation of the DS potency method in the a Gage R&R Study 5-10 samples are evaluated
ing. Generally performed as one component QCL is described as a Type 3 change in the USP by 2-4 analysts (reproducibility) using 2-4 re-
of a method validation protocol, a statistically stimuli article, involving the need to operate peat tests (repeatability) sometimes involving
designed intermediate precision study is car- the method in a different environment [1]. To 2-4 test instruments (reproducibility). For ex-
ried out using multiple setups, incorporating support this activity, a procedure performance ample if 3 analysts measure each of 10 samples
multiple analysts, multiple HPLC instruments, qualification (PPQ) protocol would be written of a product in duplicate the study will produce
and HPLC columns with different packing to describe the analytical method validation or 3x10x2=60 test results.
batches. In order to meet the ATP, the accep- comparison activities needed to demonstrate A variance components analysis is then
tance criterion for the study is that the overall that the method meets its performance re- performed on the measurements to obtain
variability, the combined uncertainty, must be quirements as required by the ATP in the new estimates of the repeatability (within-setup
NMT 1.0% as a standard uncertainty or stan- environment. These activities are governed by variability) and reproducibility of the method.
dard deviation. Traditionally, the qualification Table 1, Results of Intermediate Precision Study for DS Po-
current guidance around method transfers [2, These statistics are then used to evaluate the fit-
protocol includes a criterion only for the SD tency Method. 9]. From the perspective of a lifecycle approach, ness of the method for use for product release

www.ivtnetwork.com
Jane Weitzel, Robert A. Forbes, Ronald D. Snee
and to support process improvement efforts. Samples from five API batches were tested on
The variance component estimates are of- four setups by two analysts using each of two
ten used to determine how many within-run instruments (each instrument used a unique
replicates and/or setups are required to give the column for total to two columns). New mobile
desired method intermediate precision. If the phase was prepared for each setup and each
within-run variability (repeatability) is high, analyst prepared a fresh set of standards for
then more within-run replicates maybe re- each run. Samples were prepared and analyzed
quired to generate a reportable result with the in duplicate on each run, producing a total of
desired precision level. If setup-to-setup vari- 5x2x2x2 = 40 Potency(%) test results. The data
ability (reproducibility) is high, then multiple are summarized in Table 2.
setups including one or more analysts, instru- The results of the variance components
ments, and columns maybe required to gener- analysis for the potency measurements are
ate a reportable result with the desired preci- summarized in Table 3 and shown in Figure
sion level. The reportable result consists of the 7. In this study, multiple analysts, instruments
average of multiple within run and/or setups as and columns were used to obtain more realistic
prescribed in the method and is the value that is variability estimates, and could not be separate- Figure 7. Variability Chart showing the results of the Gage R&R study to verify performance of the method in a new labo-
compared to the specification and applied to the ly evaluated for statistically significant effects. ratory.
decision rule. This kind of a replication strategy The variance components show that 84% of
can be employed to reduce the variability test variability is due to the setup and the remaining
method that currently has a higher than desired 16% due to repeatability (within-setup).
level of variability to continue to release product The standard deviation of the test method
until a better test method can be developed. In (intermediate precision for two preparations on
this way production doesn’t have to be stopped a single setup) is 0.71% with an upper 80% con- Table 3. Gage R&R Study Repeatability and Reproducibility Variance Components
(or not initiated) because of a highly variable fidence limit of1.16%. The intermediate preci- proach incorporating a confidence limit is used. ber of setups and within setup replicates.
test method. sion estimate of 0.71% meets the acceptance The method may not be good enough for its Table 4 demonstrates that increasing the
To illustrate the utility of this approach for criterion of NMT 1.0% that was defined based intended use. To address this, additional set- number of within setup measurements (re-
verification of the method transfer to the new upon the ATP. However, the upper confidence ups could be performed to obtain a better SD peatability) from one to two has little effect on
laboratory, a Gage R&R study was performed. limit exceeds 1.0%, indicating that we cannot estimate with a tighter CI. For example, if the the method intermediate precision. This is be-
say with 80% confi- results for all the batches are pooled, with the cause the within setup (repeatability) variance
dence that the standard assumption that there are no significant differ- is much smaller than the setup-to-setup vari-
deviation is not NMT ences between them, a standard deviation of ance (reproducibility). Generally, the standard
1.0%. If the verification 0.63% with an 80% UCL of 1.03% (rounds to practice for a DS potency method is to prepare
protocol utilized an 1.0%) is obtained, which meets the acceptance samples in duplicate. While the decrease in
acceptance criterion of criterion with an 80% confidence level. variation for this method is marginal, the major
NMT 1.0% for the 80% If the method still did not meet the accep- benefit of the two preparations is to control for
UCL of the SD, these tance criterion with additional runs, the test preparation errors by setting an acceptance cri-
results would not pass method variation can be reduced by employing terion on their difference. Since reproducibility
the criterion. This study a replication strategy according to Equation 1. was the larger component of variance, analyzing
provides an interest- Table 4 shows the effect of increasing the num- the samples on multiple setups would be more
ing example where the effective at reducing the method
method would pass a standard deviation, than preparing
traditional assessment, more samples for the same setup.
but does not pass when By including multiple setups (n=2)
a more rigorous ap- the method intermediate precision
Table 2. Gage R&R Study Results for DS Potency Method Transfer. could be reduced to approximately

Journal of Validation Technology Volume XX Issue 4


Peer Reviewed: Analytical Procedure Journal of Validation Technology
“…there is no better way to understand the true MR chart was 0.48, and is an estimate for the
variability of the analytical method” [23]. The intermediate precision of the method for this
control sample must be homogeneous and sta- six-year period.
ble in order to avoid out-of-control signals un- The MR chart in Figure 8 (bottom) shows
related to the measurement system. The control only one point above the control limit. The indi-
samples are typically tested 2-3 times (depend- viduals chart in Figure 8 (top) shows two points
ing on the test method) at a given point in time. outside the control limits. No trends or shifts
Table 4. Method Standard Deviation versus Replication Strategy. The sample averages are plotted on a control were observed over the 48 samples. However,
0.5%, showing that this would be an effective is used throughout the life of the method, not chart to evaluate the stability (intermediate pre- after sample 23 the method intermediate preci-
way to improve the intermediate precision if simply as a once-and-done validation, or site cision) of the method. The standard deviations sion was observed to improve. After further in-
necessary to meet the ATP requirements. These certification, exercise. of the sample replicate preparations are plotted vestigation, samples 1-23 were found to have all
results illustrate a key benefit of the R&R study Continued performance verification is ac- on a control chart to assess the stability of the been analyzed by analyst A, and the majority of
to understand the variance components. Be- complished in part by monitoring the results of repeatability of the test method [21]. the samples afterward were tested by other ana-
cause it is easier, some laboratories may be pre- the internal method system suitability require- When using control charts typically two lysts. Analyst technique appears to be the cause
paring replicates at the repeatability level thus ments. System suitability requirements ensure types of non-random patterns are observed: for the observed decrease in method variability
wasting resources and giving a false source of that the method has been set up appropriately 1. Test measurements outside of the con- following sample 23. Potential instrument and
assurance that precision has been improved. and that the equipment is functioning cor- trol limits (typically set at the process column impacts on the results were also inves-
The estimate of the standard deviation ob- rectly. Peak shape (tailing or asymmetry) and average ±3 standard deviations). Such tigated but determined to have little to no effect
tained in the QCL (0.63%, 80% UCL = 1.0%, resolution requirements ensure the peak can events are referred to as out-of-control on the observed decrease in the method preci-
pooling the batches) was larger than the stan- be precisely integrated, appropriate selectivity (OOC) signals. sion.
dard deviation estimate obtained during meth- is provided by the HPLC column, and that the 2. Non-random patterns inside the con- When the data are split into two sets: Stage
od validation (0.15%, 80% UCL = 0.20%). mobile phase has been prepared properly. Injec- trol limits; such as trends, drifts, and 1 (Samples 1-23--Figure 9) and Stage 2 (Sam-
However, the 80% UCL for both estimates met tion precision ensures that the sample injection shifts. These events are referred to as ples 24-48--Figure 10) no practical differences
the acceptance criterion of NMT 1.0% from the system is functioning properly, and surfaces of out-of-trend (OOT) signals and are de- between the average values of the two stages
ATP, demonstrating the suitability of the meth- internal parts have not become scratched and tected using the Western Electric rules were noted. But as mentioned above, the meth-
od for its intended use. There may be many worn. Precision requirements for duplicate [21]. od intermediate precision is greater in the first
reasons for the apparent difference in method standard and sample preparations ensure that Both OOC and OOT signals are based on stage than in the second stage likely due to the
performance between the two studies. For ex- the analyst has made the preparations with ad- statistical tests of significance and are included practices of analyst A. The stage 2 data (Figure
ample, the analysts performing the first valida- equate care. All these internal method require- in many statistical software packages [24]. 10) showed one sample average and one MR
tion study may have been more experienced in ments work together to ensure the measuring outside of control limits.
performing the method. system continues to be capable to meet the ATP Continued Procedure Performance Verification Another important use of the control sam-
uncertainty requirements. System suitability re- by Control Charting Example ple results is to verify that the method meets the
Continued Procedure Performance sults should be monitored and control charted The data shown in Figures 8 - 10 are from an requirements of the ATP during this period of
Verification to maintain a state of PPV, and to provide an evaluation of a DS potency test method over time. The standard deviation of all the report-
Following a successful installation of the meth- early warning of drift or changes in the method a six-year period. Periodically blind control able results for the control sample (sample av-
od in the QCL, the method moves into stage 3 capability. samples were submitted to the lab for analysis. erages) generated during the six-year period
of its lifecycle where ongoing procedure perfor- An effective way to assess the long-term sta- Figure 8 is an I-MR (Individual Moving Range) was found to be 0.62% (80% UCL = 0.83%),
mance verification (PPV) is required [1]. This bility of a test method is to periodically submit control chart. The top control chart is a plot of which meets the ATP criterion of NMT 1.0%.
is analogous to the requirement for Continued “blind control samples” (also referred to as ref- the averages of the duplicate analyses of each Furthermore, this estimate of the method inter-
Process Verification called out in the FDA Pro- erence samples, or method performance sam- sample, and the lower control chart shows the mediate precision is based on a large number
cess Validation Guidance, and is appropriate ples), from a common source, for analysis along moving range of the averages. Control limits of independent method setups consisting of
since we have argued that the execution of a test with routine production samples in a way that were calculated from the MR chart and thereby multiple analysts, instruments, and columns.
method is a process [2]. The lifecycle approach the analyst cannot determine the difference be- include both the setup-to-setup variation as well This estimate of the intermediate precision is
requires one to assess the repeatability and in- tween the production samples and the control as the within setup (repeatability) variation. The also consistent with the estimate from the Gage
termediate precision of the test method as it samples. Nunnally and McConnell conclude standard deviation estimate calculated from the R&R study (0.71%).

www.ivtnetwork.com
Jane Weitzel, Robert A. Forbes, Ronald D. Snee
Recommendations for Interpreting the observed, it may be appropriate to consider
Control Charts performing a robustness evaluation of the test
When all the results are within the control limits method as discussed above. A test method
and no OOT patterns are observed the meth- that is not robust can be expected to exhibit in-
od is performing as expected. However, over creased repeatability, intermediate precision, or
the course of time OOC and/or OOT may be both over time.
observed. The course of action will depend on
the specific situation but some general guidance Product Quality Review – Summarize
can be given. The following recommendations Test Method Performance
are similar to those regarding process monitor- The performance of the test method should be
ing when using control charts. reviewed periodically in accordance with the
First when out-of-control (OOC) or out-of- reviews for the process and product. As part of
trend (OOT) signals are obtained the data that review, the results of the test method con-
Figure 8. I-MR Control Charts for Sample Averages (n=2): All Samples collection, analysis, and control charting proce- tinued process verification system would be
dures are examined to ensure that no mistakes provided. The control charts of the test results
have occurred. If mistakes are found the data for the control samples and summary statistics
should be corrected and the analysis repeated. which document the test method performance
When mistakes are ruled out then an inves- including the repeatability and intermediate
tigation should be considered to find the cause precision standard deviations and relative stan-
of the OOC or OOT signals. The investigation dard deviations, results of recent calibration
will likely examine if that signal can be traced to studies and other information depending on
a specific change in materials, reagents, instru- the test method and situation should be includ-
ments, columns, reference standards, or ana- ed in the summary. The performance summary
lysts. The action taken will then depend on the documentation should be stored and dissemi-
cause and the specific situation. nated in accordance with the knowledge man-
However, sometimes the cause of the OOC agement expectations of the ICH [25].
or OOT cannot be determined. If this is the case,
the practical significance of the signal should be Conclusion
considered. For example, the associated results In this paper, we have illustrated the application
may be close to the control chart limits and iso- of the lifecycle management approach for an
Figure 9. I-MR Control Charts for Sample Averages (n=2): Samples 1-23 lated with no OOC or OOT points before or af- analytical procedure (a drug substance potency
ter those results. The process may be perform- test) and showed how a decision rule was used
ing at a very high level relative to specifications, to develop the analytical target profile (ATP).
suggesting that measurement variation may not The ATP was then used to define the fitness of
be a problem. the method for its intended use, which in turn
In the case when the variation has changed enables an objective assessment of the method
– increased or decreased – revised limits should performance during initial validation, transfer
be considered along with the appropriate ap- to the quality control laboratory (QCL), and its
provals and change control actions taken. Such ongoing performance in the QCL.
changes should be accompanied by appropriate The lifecycle approach is essentially applying
training of the analyst and supervision involved. Quality by Design (QbD) principles to the exe-
A control plan should also be created and put cution of the method to produce a result, in an
in place to ensure that the new procedures are analogous way that QbD is applied to a manu-
followed. facturing process producing a product. We have
Figure 10. I-MR Control Charts for Sample Averages (n=2): Samples 24-48 If unacceptable variation continues to be also demonstrated the use of the Gage R&R

Journal of Validation Technology Volume XX Issue 4


Peer Reviewed: Analytical Procedure Journal of Validation Technology
study approach to determine the repeatability 2. FDA, Guidance for Industry Process Val- 8. ICH, Q8(R2), Pharmaceutical Develop- 17. JCGM, International vocabulary of me-
and reproducibility (intermediate precision) of idation: General Principles and Practices ment – Part II: Pharmaceutical Devel- trology – Basic and general concepts
the test method within the QCL after transfer, (Rockville, MD, January 2011). opment - Annex, Step 4 version (August and associated terms (VIM), 3rd ed.,
as a means to verify its performance, versus the 2009). JCGM 200:2012(E/F), online, www.bipm.
ATP criteria has also been demonstrated. The 3. FDA, Guidance for Industry Analyti- org/utils/common/documents/jcgm/
estimate of the method precision from the QCL cal Procedures and Methods Validation 9. USP 37-NF 32 General Chapter <1224>, JCGM_200_2012.pdf, accessed Decem-
results will often be based on a larger sample for Drugs and Biologics (Rockville, MD, “Transfer of Analytical Procedures,” 1155- ber 3,2014.
size and encompass more variability than the Draft--February 2014), online, www.fda. 1157.
development studies did, leading to a more ro- gov/downloads/drugs/guidancecompli- 18. ISO/IEC, Uncertainty of measurement
bust estimate of the long-term measurement anceregulatoryinformation/guidances/ 10. USP 37-NF 32 General Chapter <1225>, – Part 3: Guide to the expression of un-
precision. The Gage R&R approach utilizes a ucm386366.pdf, accessed December 3, ‘Validation of Compendial Procedures,” certainty in measurement (GUM:1995),
greater number of test batches than a typical 2014. 1157-1162. 98-3:2008.
intermediate precision study, which provides a
higher degree of replication for assessment of 4. P. Nethercote, et al., “QbD for Better Meth- 11. USP 37-NF 32 General Chapter <1226>, 19. ICH, Q2(R1), Validation of Analytical
the measurement uncertainty. od Validation and Transfer,” PharmaMan- ‘Verification of Compendial Procedures,” Procedures: Text and Methodology, Step 4
The use of a control sample for continued ufacturing.com, online, www.phar- 1162-1163. version (November 2005).
procedure performance verification has also mamanufacturing.com/articles/2010/060.
been explained, which is an important stage html, April 13, 2010, accessed December 12. ICH, Q2(R1), Validation of Analytical 20. USP Proposed General Chapter <1210>,
in the lifecycle of a method that is often over- 3, 2014. Procedures: Text and Methodology, Step 4 “Statistical Tools for Procedure Validation,
looked. The use of an analytical method control version (November 2005). ”Pharmacopeial Forum 40 (5), 1–33.
sample to monitor performance of the method 5. M. Schweitzer, et al., “Implications and
versus the ATP requirements during its long- Opportunities of Applying QbD Princi- 13. ICH, Q11, Development and Manufacture 21. D. C. Montgomery, Introduction to Sta-
term use in the QCL has also been demonstrat- ples to Analytical Measurements,” Pharm. of Drug Substances (Chemical Entities tistical Quality Control, John Wiley and
ed. Not only does the control sample provide Tech. 34 (2), 52–59, 2010. and Biotechnological/Biological Entities), Sons, New York, NY, 7th Edition, 2013.
assurance that the ATP is being met, but it also Step 4 version (May 1, 2012).
detects special cause variation and/or shifts in 6. Phil Nethercote, Joachim Ermer, “Qual- 22. [22] R. D. Snee and R. W. Hoerl, Leading
common cause variation. This enables continu- ity by Design for Analytical Methods, 14. FDA, Guidance for Industry Investigating Six Sigma – A Step-by-Step Guide Based
al improvement of the method performance by Implications for Method Validation and Out-of-Specification (OOS) Test Results on Experience with GE and Other Six Sig-
identifying opportunities for analyst training, Transfer,” PharmTech.com, online, www. for Pharmaceutical Production (Rockville, ma Companies, FT Prentice Hall, Upper
the need for equipment repairs, or detrimental pharmtech.com/pharmtech/Feature+Ar- MD, October 2006). Saddle River, NJ, 2003.
changes in consumables such as the HPLC col- ticles/Quality-by-Design-for-Analyti-
umns or reagents. cal-Methods-Implicati/ArticleStandard/ 15. ASME, Guidelines for decision rules: con- 23. B. K. Nunnally and J. S. McConnell, Six
Article/detail/791903, October 2012, ac- sidering measurement uncertainty in de- Sigma in the Pharmaceutical Industry:
References cessed December 3, 2014. termining conformance to specifications,” Understanding, Reducing, and Con-
1. USP Stimuli Article, “Lifecycle Manage- B89.7.3.1-2001. trolling Variation in Pharmaceuticals and
ment of Analytical Procedures: Method 7. Joachim Ermer, “A Lifecycle Concept for Biologics, CRC Press, Boca Raton, FL,
Development, Procedure Performance Pharmaceutical Analysis,” European Phar- 16. Eurachem/CITAC Guide, Use of uncer- 2007.
Qualification, and Procedure Performance maceutical Review, 16 (3), 16-24, 2011, on- tainty information in compliance assess-
Verification,” Pharmacopeial Forum 39 line, content.yudu.com/Library/A1sipk/ ment, First edition (2007), online, eura- 24. L. D. Torbeck, “OOS, OOT, OOC and
(6), online, www.usp.org/usp-nf/notices/ PATampQbDSupplement/resources/in- chem.org/images/stories/Guides/pdf/ OOSC,” Pharmaceutical Technology, 35
stimuli-article-lifecycle-management-an- dex.htm?referrerUrl=http%3A%2F%2F- Interpretation_with_expanded_uncer- (10), 46-47, 2011.
alytical-procedures-posted-comment, ac- w w w. y u du. c om % 2 F ite m % 2 Fd e - tainty_2007_v1w.pdf, accessed December
cessed December 3, 2014. tails%2F345094%2FPAT---QbD-Supple- 3, 2014. 25. ICH, Q10, Pharmaceutical Quality Sys-
ment accessed December 3, 2014. tem, Step 4 version (June 2008).

www.ivtnetwork.com
Jane Weitzel, Robert A. Forbes, Ronald D. Snee

About the Authors


Jane Weitzel has been working in analytical chem-
istry for over 35 years in highly regulated, fast
paced companies with the last 5 years at the direc-
tor/associate director level. She is currently a full
time consultant and auditor.

Robert Forbes earned his PhD from Purdue Uni-


versity working with Dr. Ben Freiser in Mass Spec-
trometry. He worked as a research scientiest with
Eli Lilly working with Clinton Laboratories, Tippe-
canoe Laboratories, and Lilly's Technology Center.

Ronald D. Snee is founder and president of Snee


Associates, a firm dedicated to the successful
implementation of process and organizational
improvement initiatives. He provides guidance
to senior executives in their pursuit of improved
business performance. He worked at the DuPont
Company for 24 years prior to initiating his con-
sulting career.

Journal of Validation Technology Volume XX Issue 4


View publication stats

You might also like