Thomas A. Little is president, Thomas A. Little Consulting and BioAssay Sciences, 12401 North Wildflower Lane, Highland, Utah 84003, USA, drlittle@dr-tom.com.

Thomas A. Little, PhD

Analytical methods and bioassays require a well-thought-out manner of evaluating method validity (truthfulness) and systems suitability (a known positive control) to ensure the method is correct, reliable, and can be used to determine the value of the unknown test article. This paper recommends and explains methods for evaluating system suitability as well as validity criteria. Further, the paper discusses practical and phase-appropriate methods for setting limits for systems suitability and validity criteria.

In alignment with guidances such as International Council of Harmonisation (ICH) Q2 (1), 

) <1033>, <1032>, <1225>, and ICH Q6B (2), method development, validation and method understanding is a critical element of product measurement and control. Systems suitability should be understood to be separate from validity criteria. Systems suitability is defined as a known positive control that can be demonstrated to be within a defined set of limits. Validity criteria typically includes signal control for bioassay criteria for parallelism and linearity, and criteria for repeatability.It may further include a fixed number of outliers for data management.

There are two methods for systems suitability. The first method involves adding a positive control to the assay run. The positive control is a known standard at a fixed concentration.By adding the known standard to the plate and checking if it is within a defined limit, one can trust that the method is running correctly, and the test article may be reliably determined.The second method requires use of a standard curve. A standard curve or calibrator has a known value and is placed on the plate at three or more concentrations; a curve is then fit to the data.By measuring the concentration at a fixed position on the standard curve, one can use that data for systems suitability. The standard curve approach has a distinct advantage in reducing the analytical error of the quantitation of the positive control. It is not recommended to use a calibrator and then add the positive control to a plate, however, as this doubles the analytical error. 

 shows the systems suitability back-calculated from a fixed position on the reference standard.

 Systems suitability at a fixed position. All figures are courtesy of the authors.

Before relative potency of a test article to a reference standard can be calculated, it is crucial to verify that there is a clear signal from the dose response for the reference standard. There are two methods for verifying a reliable signal; the dose response test and curve depth.

Based upon health authority input, the dose response test is a standard approach to show signal-to-noise control. The dose test measures the effect of analyte concentration, which is used to generate a statistically significant signal (alpha = 0.05, two-sided). A regression model may be linear, square root, log, three parameter exponential model, or four parameter logistic regression (4PL), depending on the method.

The curve depth of the reference standard is another method to demonstrate a clear signal. Curve depth is the mean signal for the highest concentration subtracted from the mean signal from the lowest concentration. For assays using 4PL fit, curve depth is calculated by subtracting the lower asymptote of the standard from the upper asymptote of the standard (

). Normally the limit is set at 50% of the curve depth from qualification or validation assay runs.

 of a fitted curve is not recommended for a validity criterion, but should be for report-only. Confidence intervals (CIs) of the reportable value are a more reliable indication of the errors in quantitation due to curve fitting.

Validity criteria for bioassy parallelism

 <1032> defines parallelism as the following:

“A quality in which the concentration–response curves of the Test sample and the Reference Standard are identical in shape and differ only by a horizontal difference that is a constant function of relative potency” (3).

Parallelism must be evaluated as a validity criterion because parallelism between the test article and reference standard demonstrates that the two are similar, and, therefore, relative potency may be determined. F-tests for parallelism are not recommended because they are too sensitive and may cause a high degree of invalid results.

Demonstrating parallelism in a bioassay is done for a sigmoidal curve and may be evaluated using the upper asymptote ratio (UAR). The UAR is compared to a two-sided limit. If the ratio of the upper asymptote for the reference standard and the UAR for the test article is within defined limits, then the curves can be constrained, and relative potency can be calculated. UAR only checks parallelism at one point on the sigmoidal curve. Slope ratios may be used for a parallel line analysis of the reference and test article. A highly recommended method for determining parallelism is the absolute difference between the relative potency of the unconstrained model and the constrained model. This measures how the forcing function of constraining the curves changes the reported relative potency.

An equivalence test may be used in evaluating the UAR. If the slope ratio is outside of the equivalence bounds, the assay run fails parallelism and is invalid. 

 is a parallelism test example with a resulting slope ratio of 0.993.

 Slope ratio. UDL is upper detection limit. LDL is lower detection limit.

Linearity ratio calculation for bioassays

A bioassay also has a requirement to be linear in the dose response.The Linearity Ratio method of analysis uses a measure of curvature relative to the linear line rather than a measure of probability by comparing the effect size attributed to the quadratic term (curve) to the effect size attributed to the linear term in the full model.In practical terms, the question of linearity is, what percentage of the linear line is curvature?

A scaled estimate for the linear term is half the change in the signal over the range (distance from center). To determine the full change over the range linear effect of concentration, it must be multiplied by two. The quadratic term is curving andthe range is the full curvature, which is then divided by full change in the linear signal.

Validity criteria: assay repeatability/analytical error

In order for an analytical method to be validated, it must be shown to be repeatable at the moment of measurement (4). Before evaluating repeatability, outliers should be removed from the assay run or run of the analytical method. Jackknife z is the preferred method of outlier identification within each dose. Jackknife z is calculated independently for each concentration in the dose response for both reference and sample type. Jackknife z evaluates the influence each point is having on the curve fit where each point is removed from the model; the model is regenerated and evaluated, iteratively, for each data point. Jackknife z = (measurement - dose mean (measurement removed)) - Standard deviation (measurement removed).

CIs should be included when reporting the results for any and all test articles. CIs control for sample size, variation in the method, and risk (95%). CI range is calculated by taking the difference of the upper confidence limit of the constrained relative potency and the lower confidence limit. A narrower CI range indicates that the assay is less variable and, therefore, more repeatable. Excessive CI range may not make the assay run invalid; it demonstrates that additional assay runs are required to tighten the error.Normally, the CI range is reported as a percent of tolerance (

where USL is upper specification limit and LSL is lower specification limit. The coefficient of variance (CV) is often used to evaluate repeatability of the dose response. CV is not recommended for use because it incorrectly scales the analytical error by concentration; high concentrations appear to have low error and low concentrations seem to have high error.

Setting systems suitability and validity criteria limits

When setting specification limits based on the representative sample, risk must be considered. The wider the interval, the greater the amount of allowable transmitted variation on the Y response. K sigma is used when the sample size is 30 or more, and tolerance intervals are used when the sample size is less than 30. Before setting limits, a representative data set must be available. Systems suitability and validity criteria limits are normally set post-qualification and review and finalized post-validation of the method (5).

Tolerance intervals were developed to provide a correction for limited sample sizes and scales. The interval of risk is based on three considerations: sample size, confidence interval, and proportion of the population to be described.

Tolerance intervals should be used under the following conditions:

No transfer function is available and/or possible

No adverse nor unacceptable results are associated with the parameter or response

When n => 30, use K sigma in setting limits

Sample data are stable (use a control chart or regression analysis to evaluate stability)

Distribution is either normal or nonnormal.

The procedure for setting limits are as follows:

Make sure there is a representative sample of the assay validity or systems suitability measurement.

Check data for outliers. Jackknife z and three sigma cut point is typical.

Determine the distribution that best fits the data, normal or nonnormal distribution, use a goodness of fit or second-order Akaike’s information criterion (AICc) functions to determine the best fit.

Avoid using Johnson or Sinh-Arcsinh (SHASH) distributions when fitting nonnormal tolerance intervals as they set limits excessively wide. Normal, Gamma, Weibull, Lognormal, Normal Mixture are preferred and should all work well to set limits.

 to determine the risk and associated tolerance interval and margin. 

 Probability and risk table for setting limits. OOS is out of specification. PPM is parts per million. Cpk is process capability index.

Use a K Factor calculator (6) to determine K for the interval (sample size, confidence level andpopulation) for any tolerance interval.

Set Spec using K sigma to define the limits.

Examine results and document limits, update any standard operating procedures or calculators with the defined limits.

Using the correct methods for evaluating systems suitability and acceptance criteria is critical to ensuring that the analytical methods is fit for use. Setting specification limits as described is statistically rigorous, scientifically sound, and defendable upon regulatory review. 

 is a summary table of all criteria and recommended limits. 

 Summary table showing criteria and recommended limits. UAR is upper asymptote ratio. 4PL is four parameter logistic regression.

Validation of Analytical Procedures: Text and Methodology

, Step 4 version (November 2005).
2. ICH, Q6B 

Specifications: Test Procedures and Acceptance Criteria for Biotechnological/Biological Products

 <1032>, “Design and Development of Biological Assays,” (Rockville, Md., 2015)
4. USP, 

 <1225>, “Validation of Compendial Procedures,” (Rockville, Md., 2015).
5. USP, 

 <1033>, “Validation of Biological Assays,” (Rockville, Md., 2010).
6. NIST, “Tolerance Intervals for a Normal Distribution,” in 

NIST/SEMATECH e-Handbook of Statistical Methods

Thomas A. Little*, PhD, drlittle@thomasalittleconsulting.com, is president and principle consultant of Bioassay Sciences, and Daniel S. Harding is a principle consultant of Bioassay Sciences.

*To whom all correspondence should be addressed.

Supplement: Bio/Pharma Outsourcing Innovation 2022
February 2022
Pages: s6–s11

When referring to this article, please cite it as T.A. Little and D.S. Harding, “Establishing Systems Suitability and Validity Criteria for Analytical Methods and Bioassays,” 

Articles

Ensuring reliable analytical methods and bioassays requires a well-thought-out strategy for evaluating method validity and systems suitability.

Establishing Systems Suitability and Validity Criteria for Analytical Methods and Bioassays

	A key element of drug development and product/process characterization is the requirement to identify and control critical process parameters (CPPs) and critical material attributes (CMAs) that influence critical quality attributes (CQAs). CQAs have clinical relevance as they are defined to specifically measure drug attributes associated with safety and efficacy of the drug. In a previous article (1), a discussion of the selection of CPPs was presented. This article presents statistical methods to identify CPPs and CMAs and approaches on how to control them so they do not adversely affect the drug product or drug substance.

	International Council for Harmonization (ICH) Q8 

 (2) states in Part II, 2.5, “… These controls should be based on product, formulation and process understanding and should include, at a minimum, control of the critical process parameters and material attributes. A comprehensive pharmaceutical development approach will generate process and product understanding and identify sources of variability. Sources of variability that can impact product quality should be identified, appropriately understood, and subsequently controlled. Understanding sources of variability and their impact on downstream processes or processing, in-process materials, and drug product quality can provide an opportunity to shift controls upstream and minimize the need for end product testing. Product and process understanding, in combination with quality risk management (see ICH Q9), will support the control of the process such that the variability (e.g., of raw materials) can be compensated for in an adaptable manner to deliver consistent product quality.”

	Line-of-sight for all CQAs and CPPs/CMAs

	When working on drug development, there is a need to maintain line-of-sight (

) between CQAs and elements of drug development that may influence them. Elements of drug development that may impact CQAs are as follows:

		Formulation process for drug substance or drug product

		Processing method or sequence of operation

		Selection of normal operating ranges (NORs)

		Selection of proven acceptable ranges (PARs)

		Critical parameters in analytical methods. 

	By maintaining line-of-sight, the relationship of drug development parameters and CQAs is maintained and evaluated in a systematic manner, making knowledge gaps and oversights less likely. Quality risk management is a technique that evaluates and -manages development risks associated with CQAs.  ICH Q9 on quality risk management (3) states, “Two primary principles of quality risk management are: the evaluation of the risk to quality should be based on scientific knowledge and ultimately link to the protection of the patient; and the level of effort, formality, and documentation of the quality risk management process should be commensurate with the level of risk.”

	Quality risk management is normally used to identify development areas with risk to CQAs; however, it is not sufficient to identify the CPPs and CMAs. High-level risk assessments will indicate unit operations with risk, and low-level risk assessment will identify potential factors that may impact CQAs. Risk assessment alone is not sufficient to detect and control CPPs. It can only be based on measurement of how factors directly influence the response.

	CPPs may be found in formulation, media, upstream and downstream unit operations, and drug-product processing. Due to the large number of unit operations and media complexity, it is easy to overlook processing parameters and materials that may influence drug-substance and drug-product variation and CQAs. Failure to identify critical parameters will result in unexplainable variation during batch processing and lot acceptance.

	The key steps to CPP/CMA selection and their application to process control is as follows: 

		Identify CQAs for drug product and substance. Include responses for productivity (%viability or titer).

		Select API, excipients, processes, materials, and container closure.

		Define all unit operations and process sequence of operation.

		Define all product and process specification limits and acceptance criteria (include in-process controls and release testing).

		Qualify/validate all analytical methods.

		Complete high-level quality risk management and identify unit operations with risk to CQAs.

		For unit operations with risk, complete a low-level factor/response selection for all factors, interactions, quadratics, and materials.

		Generate design of experiments (DoEs) for unit operations or materials with risk.

		Explore the design space all key factors identified during the risk assessment using DoE or other multivariate methods.

		Determine the factor effect size and scaled estimates. 

		Determine the % of tolerance, % of margin, or % of mean, and identify all CPPs/CMAs.

	Identification of CPPs based on the scale estimate from DoEs

	DoE and multifactor experiments help to isolate the influence of every factor and interaction on the critical responses associated with the drug substance or product. Analysis of the DoE will generate the scaled estimates (one-half the change in Y relative to the change in X) also known as the half effect (

). Scaled estimates are the direct measure of how factors influence CQAs and key responses.

	Multiplying the scaled estimate by two for all main effects and two-factor interactions, and by one for all quadratics determines the full effect of each factor and or model term. Dividing the full effect by the tolerance, margin, or mean measures the factor or model term as a % of tolerance, % of design margin, or % of the mean. The reason ±20% of tolerance is considered critical, it will cause shifts in the CQA response and most likely result in out-of-specification (OOS) results (

), thus it makes the factor or material attribute critical:

	% of Tolerance = Abs(Scaled Estimates * 2)/(USL-LSL)  for two sided limits

	% of Design Margin= Abs(Scaled Estimates * 2)/(Average-LSL) for one sided LSL only

	% of Design Margin= Abs(Scaled Estimates * 2)/(USL-Average) for one sided USL only

	Average is the baseline process or product average from the DoE or other lots.

	CPPs are a function of the DoE range and their associated influence on CQAs. Increasing or decreasing the experimental range will influence CPP identification.  During characterization we want wide limits and don’t mind identification of many CPPs; however, when filing a drug with the health authorities we want to demonstrate our control of the parameters makes them not critical.

	ICH Q10 and Q11 (4,5) discuss the need for a control strategy. After the identification of CPPs/CMAs is complete, the next step is to control the influence of the factor so that it does not adversely cause uncontrolled variation in the process. The following are strategies that may be used to control CPPs/CMAs:

		Closed loop control on the factor (PAT/Statistical process control with adjust within the defined design space) (6).

		Change the process or method to another method of processing that is better controlled.

		Modify equipment and associated control loops to a tighter range.

		Purchase equipment that is better controlled.

		Institute raw material qualification on incoming materials.

		Improve vendor material control and certificate of analysis.

		Change a material usage factor based on concentration or potency to control the influence of raw material variation.

		Establish NOR and PAR associated with the product and restrict the operational range.

	Normal operating ranges and proven acceptable ranges (NOR/PAR) are determined based on evaluating normal variation from the process (typically ±3 stdev for NOR and ±4.5 stdev for PAR) and its transmitted variation from a DoE simulation (

	Edge of failure analysis is then used to evaluate and/or set the NOR/PAR limit.  Setting a NOR or PAR limit will cause restriction in the factor range, and its characterized transmitted influence will be reduced.

	Limiting the factors to the NOR and PAR range will reduce the influence of the factors, so they are now more controlled; however, does it also follow that they are now no longer critical and their variation will not result in OOS?  To determine if the factor range restriction controls the factor (

) and is no longer critical, the following procedure is recommended:

		Run a new simulation at the NOR/PAR range at the optimized set point (target) for the process. Use either uniform distribution (worst case) or random truncated (best case).

		Make sure no random variation is injected into the model except from the factors of interest to correctly evaluate their influence.

		From the simulated table with the new range, refit the original DoE model into a multiple regression analysis (Fit Model in SAS/JMP).

To protect drug product and drug substances, statistical methods to identify critical process parameters and critical material attributes, as well as and approaches to control them, are needed.

CPPs and CMAs: Identification and Control

Accelerated stability analysis is a strategy used to quickly evaluate alternative formulations, packaging, and processes. Accelerated linear studies are commonly performed and modeled; however, accelerated multiple-factor non-linear modeling has been a gap, and statistical software tools such as SAS/JMP do not directly have any provision to model multiple factor nonlinear responses. This paper outlines an approach to model and predict non-linear multiple factor stability/tablet dissolution data under accelerated and nominal storage conditions. 

There are many non-linear stability cases such as dissolution, leachables, and moisture. Being able to model these non-linear processes is crucial to proper drug development.  In addition to general non-linear modeling, there are multiple factors that may influence the non-linear curve. The following is a list of factors that may impact the asymptotes, growth rate, and inflection point of a curve:    

Stability storage temperature and humidity


Proper design of experiments for data collection and curve isolation is crucial for building non-linear models. 

 illustrates the factors that should all be square relative to all other factors and have zero correlation relative to each other. For this tablet dissolution example, multiple time points (minutes), multiple storage conditions (temperature), multiple drug substance particle sizes, and multiple weeks were measured. Percent dissolution was the response of interest.


The following is a step-by-step procedure for non-linear stability modeling and expiry determination.

Measure the data at multiple time periods, using multiple particle sizes and at multiple temperatures. Generate a plot of the data to visualize the relationship of the curves over time  (25-10-0: 25=Temperature, 10=Particle Size, and 0 = days) (see 


 Fit each curve individually. In this example, each dissolution curve was fit using a four-parameter logistics (4PL) curve. The four parameters are: upper asymptote, lower asymptote, inflection point, and slope of the dissolution curve.  R-Square should be high (typically above 0.95) and RMSE error should be low for each curve.  Outliers should be checked using the residuals.  Save the parameters of the curve.  In this example, there are four parameters, upper asymptote, lower asymptote, growth rate (slope), and inflection point (see 

 Save the parameters from the curve and the factors that influence them into a table. For this example, the growth rate and inflection point are the coefficients that may change the most based on the factors under consideration. The upper and lower asymptotes are not of concern for this problem as all of the curves have similar lower asymptotes and similar upper asymptotes, but upper and lower asymptotes could be important for other problems, so generally it is best to model all of the curve parameters (see 

 Fit the curve parameters with a least-squares multivariate regression. Growth rate, the slope of the 4PL fit, and inflection point are the most important as the dissolution starting point and the upper asymptote are essentially the same for all curves. Main effects and interaction models generally work best and p-values and F tests can be used to evaluate each model term (see 

 Save the equation from the multivariate parameter model.  An example of the inflection point model is in Equation 1.
  [Eq. 1]
  
  
  


 Substitute the growth rate (slope) and inflection point coefficients from the multivariate model into the nonlinear prediction.  The red box in 

 shows the substitution for the inflection point.

Check the model to make sure it matches the actual data. Correct any modeling errors by adding or modifying a multivariate model term. A simple YX regression plot of the model versus the measurement will indicate model quality and any systematic errors (see

Create a profiler from the equation to predict future dissolution rates.  This can be done using a modern statistical package such as SAS/JMP (see 

 Predict dissolution at any time, temperature, or particle size combination using the profiler.  For this example, particle size was set to 5 μm, temp to 25 and time in weeks to zero.  The dissolution time was fixed at 100 min with a specification of not less than 90% (see 

 From the profiler at each time point (weeks), make sure the time (min) and particle size (5 um) are fixed, determine dissolution at the nominal temperature (non-accelerated condition).  Fit the rate of degradation using either a linear or nonlinear model from the profiler predicted data.  In this case, the rate of dissolution is not linear so a non-linear curve is fit to the data and the expiry is determined based on the extrapolated curve (see 

).
The same method is used for predicting both the nominal expiry and the 95% CI expiry.

 Finally, long-term stability evaluation at nominal storage conditions will be used to confirm the early model prediction and will provide an independent secondary determination of stability and changes in dissolution.  Understanding rates of change should factor into shelf life and release specification limits (1).


  Non-linear multiple factor analysis has long been a problem in a variety of process and product modeling and prediction. The novel procedure discussed in this paper for the characterization of multiple factor non-linear product performance provides a clear, statistically defendable method for determining dissolution and accelerated stability. Long-term verification of accelerated conditions should always follow early determinations of expiry, acceleration rates, and rates of degradation.

Q6A Specifications: Test Procedures and Acceptance Criteria for New Drug Substances and New Drug Products: Chemical Substances

Mark Alasandro, PhD, is director, Allergan Irvine. Thomas A. Little PhD, is president, Thomas A. Little Consulting, drlittle@dr-tom.com. 

The right approach can provide a clear, statistically defendable method for determining dissolution and accelerated stability.

Multifactor Non-linear Modeling for Accelerated Stability Analysis and Prediction

The approach   provides the required International Conference on Harmonization   validation elements as well as information on                          interactions, measurement uncertainty, control   strategy, and continuous improvement. Despite being less resource   intensive than the traditional validation approach,   quality is not compromised. Additionally, through judicious planning,   the MVbD approach can encompass early formulation design efforts   so that a wide range of formulations is taken into consideration when   defining the method-validation design space.

Analytical method   validation is a crucial part of formulation development. It is needed   to ensure that methods provide accuracy                          and precision in detecting formulation   differences in drug dissolution, stability, active, and impurity levels.   Method validation is also key to meeting requirements of cGMP,   compendia, and the International Conference on Harmonization (ICH) for   testing and releasing clinical or commercial dosage   forms. During early development, formulations are often changed to adapt   to new                          preclinical and clinical data. Revalidating each   of these new formulations is resource intensive and affects development   timelines. 

The ICH Q2   (R1) validation guidelines (1) do not provide crucial information such   as how formulation changes affect method                          performance or what method critical validation   parameters need to be monitored and controlled. An ICH approach may not   provide                          a good understanding of a method-measurement   uncertainty, which is needed to ensure that the overall process   capabilities                          are met and that appropriate in-process controls   and specifications are set. This sort of understanding is needed to   meet                          FDA’s process validation guidelines (2). Once   the product acceptance criteria are established, the influence of assay   variation                          can be determined relative to product acceptance   rates.

This article   describes a method-validation-by-design (MVbD) approach to validate a   method over a range of formulations. It                          uses both design-of-experiment (DOE) and   quality-by-design (QbD) principles to define a design space that allows   for formulation                          changes without revalidation. The approach   provides the required ICH validation elements as well as information on   interactions,                          measurement uncertainty, control strategy, and   continuous improvement. This approach is less resource intensive than   the traditional                          validation approach without compromising   quality. Additionally, through judicious planning, it can encompass   early formulation                          design efforts so that a wide range of   formulations can be used to define the method-validation design space. 

MVbD is not   specifically addressed in ICH; however, it is supported by the   principles presented in ICH Q8, 9, 10, 11 (3–6)                          as well as ICH Q2 (R1). There are other industry   guidelines (2), publications (7–9) and presentations (10–15) that   support                          a QbD/DOE approach for analytical method   development and validation, including FDA’s 21st Century Quality   Initiative, which                          was first presented in 2001 and later updated in   2005 (15), and the outcome from FDA’s QbD pilot. 

The use of DOE is   well established for determining method robustness, such as how much   the mobile-phase composition, column                          temperature, and flow rate can vary. DOE is also   well used in formulation screening. This approach broadens DOE to   include                          method validation over a range of formulations.

The MVbD approach applies to these QbD principles: 

 of the potential critical validation parameters. Systematic analysis (e.g., fishbone diagram) is done based on historical                                knowledge. The QbD/DOE output confirms those that are critical. 

. Once the required process capability is known, the required method accuracy and precision acceptance criteria are defined                                in the ATP. 

. DOE output   defines those parameters that have the most impact on the method   performance. Monitoring these parameters throughout                                development will help define acceptable   ranges. 

. MVbD provides the acceptance criteria that must be met to move to new technologies. 

. Knowledge gained is cycled backed to accelerate the next program. 

Traditionally, each   new drug-product formulation requires validation to support clinical   testing. In early development, the                       method is validated for linearity, accuracy, and   precision. Method linearity is determined across five concentrations   from                       50% to 150% of the nominal (100%) concentrations.   Accuracy and precision can be combined with the linearity study by doing                       six replicates at the 100% (nominal) concentration   and three replicates at the other concentrations (50%, 75%, 125%, and   150%),                       which gives a total of 18 sample preparations. If   there are five new formulations during development that require separate                       validation, a total of 90 sample preparations would   be needed. This approach is resource intensive and affects project   timelines.                       In this given example, only one                       formulation component is varied. As more components   are varied, the workload increases proportionally. 

It does not statistically define a design space

A QbD/DOE approach   validates a method over a range of formulations within a defined design   space. Movement within this design                       space does not require method revalidation with   each new formulation. The DOE approach detects any excipient–API   interactions                       and critical validation control parameters. A DOE   approach is not necessarily more labor intensive than the traditional   method-validation                       approach and additional information gained is well   worth the exercise. As described, the traditional approach requires at                       least 90 sample preparations to validate five   formulations over the course of the development. Using DOE to validate   across                       a broader range that includes these five   formulations would require only three determinations for each of the   five formulations,                       resulting in a total of 15 sample preparations. 

Step 1: Define the range of formulations and required process capability

.   A formulation screening DOE may be used to determine all possible   clinical formulations for the study. Manufacturing process                             capabilities need to be considered to define   the ATP (i.e., the necessary accuracy and precision to ensure acceptable   data).                             Full factorial, fractional factorial and/or   custom (d-optimal) design all may be used in developing the study   design. 

Step 2: High-performance liquid chromatography (HPLC) method selection

.   A method should be developed such that sample preparations of the   different formulation can be diluted to the same API concentrations.                             The diluted sample from each formulation is   injected onto the HPLC column under the same chromatography conditions.   Standardizing                             these conditions will facilitate DOE and   latter method-robustness studies. 

. The set up of the DOE can be accomplished with or without using DOE software. Inputs to the DOE include a number of varying                             factors, such as the different API levels and/or excipient and preservative levels. 

.   Data are analyzed for linearity, precision, and accuracy, including   confidence intervals. Any interactions are reported                             and a design space is defined. The design   space is defined by fitting a model to the factors used in the   evaluation of assay                             concentrations and other key factors that may   interfere or influence assay precision or bias. After fitting the   model, the                             design space may be visualized using contour   plot or profilers. 

To illustrate the MVbD process, a simulated case   study is presented where three factors are varied over a range of   formulations                       (see Table I). The API concentrations are varied   from 0.005%, 0.01%, 0.015%, 0.02%, and 0.025% (w/w). Preservative   concentrations                       are either 0.01% or 0.03% (w/w) and concentrations   of Excipient 3 are either 0.5% or 0.8% (w/w).

. The DOE   pattern is presented as 1, 2, 3, 4 or 5 --, -+, +- or ++. The numbers 1   through 5 represent the five API concentrations                       and the “+” or “–” represent with or without the   two possible preservative and excipient concentrations. Three replicates                       are done for each combination of API, preservative,   and excipient concentration. A total of 60 sample preparations is   required.                       The DOE typically suggests a randomized sample   preparation to avoid any sample preparation variability; however,   because the                       variability is low, a standard sample-preparation   scheme can be followed.

Figure 1 also presents the percent recovery data.   The required ICH statistics can be obtained from these data. For   example, 

 shows linearity data along with   typical linearity statistics. One key output is the root mean square   error (RMSE); this parameter                       identifies method variability that is critical for   understanding the method contribution to the overall process   variability. 

 also shows linearity plots of   percent recovery versus the two excipient concentrations. The difference   in slopes indicates                       interaction (i.e., a lower API recovery at a higher   level of Excipient 3 concentration), possibly due to interaction of the                       excipient with the API. Such information is   important for the formulator and method developer to consider if higher   preservative                       levels are needed.

 presents the mean percent recovery   for each combination of API, preservative, and excipient along with the   corresponding                       confidence limits. In the example describe, there   was a somewhat lower API recovery with the higher excipient levels,   which                       is further illustrated in 

.   It is obtained by mathematically modeling the primary main effects and   secondary two-factor interactions as well as possible                       polynomials using a DOE software. The operating   range is the white space shown in Figure 4; it is where the method   accuracy                       and precision meet the acceptance criteria. Adding   formulations outside this range would require confirmatory testing.

System linearity, quantitation limits (QL), and detection limits (DL) 

.   To ensure the sample amount injected is within the linear range of the   detector, a separate experiment can be done where                       the sample concentration is varied from 50% to 150%   of the nominal (100%) concentration. If the potency method is also used                       to quantitate impurities, a wider linearity study   can be performed from 100% to 0.03%. The QL and DL for the impurities   can                       be determined by the ratio of RMSE to the slope, as   described in ICH Q2 (R1). 

 shows there is less precision (i.e., higher standard deviation and   confidence values) at the higher API levels for lower                       amounts of Excipient                       3. This lower precision is approaching the   acceptance limit. Similarly, accuracy is approaching the acceptance   limit as the                       excipient and API levels increase. These parameters   need to be monitored in case the method needs to be optimized prior to                       transfer to the final manufacturing site(s). 

 shows a graph of the balance between precision and bias and its   influence on product acceptance/failure rates based on the                       DOE data. Combinations of method precision and bias   must fall within the designated white space to meet the acceptance   criteria.                       The equation for how precision and bias influence   product acceptance rates is as follows:

LSL = lower acceptance criteria, USL = upper acceptance criteria, bias = mean percent recovery - 100%, nominal = 100%, and                       precision = mean standard deviation. 

The prediction profiler (bottom graphs) in 

 allows the precision and bias to be varied to determine the impact on %   acceptance rates. This tool can be used to justify                       moving to new technologies and new specifications   by showing that the acceptance criteria will still be met. 

. The lower recovery of API observed at high levels of excipient concentrations adds to the formulator’s knowledge base. If                       higher excipient concentrations are needed in future formulations, this aspect needs to be considered. 

Regulatory strategy and potential hurdles

Since the MVbD approach is not specified in ICH regulations, not all regulatory agencies may accept this approach. Some filing                       strategies are briefly discussed. 

: The MVbD approach can be   presented in an investigational new drug (IND) application and discussed   at the end of Phase II                             chemistry, manufacturing, and controls (CMC)   meeting. The details can be present in a MVbD design protocol along with   a rationale                             for moving within the design space without   revalidation. Moving outside the design space could be justified through   the use                             of comparability protocol. 

:   Submission of the MVbD rationale, justification, and protocol early in   the development process is warranted since a face-to-face                             conversation may not be possible.   Alternatively, this information can be supplied when responding to any   agency questions. 

The authors describe a method-validation-by-design (MVbD) approach to validate a method over a range of formulations using both design-of-experiment and quality-by-design principles to define a design space that allows for formulation changes without revalidation.

Method Validation by Design to Support Formulation Development

				The systematic approach for CPP selection and use discussed in this article was developed in line with the International Conference of Harmonization (ICH) Q8, Q9, Q10, and Q11 guidelines, which recommend quality risk management and the identification of CPPs as part of drug quality and process control development (1–4).			

Specifically, ICH Q8(R2) Section 2.5 on Control Strategy (1), states: 

"…. These controls should be based on product, formulation and process understanding and should include, at a minimum, control of the critical process parameters and material attributes…. A comprehensive pharmaceutical development approach will generate process and product understanding and identify sources of variability. Sources of variability that can impact product quality should be identified, appropriately understood, and subsequently controlled. Understanding sources of variability and their impact on downstream processes or processing, in-process materials, and drug product quality can provide an opportunity to shift controls upstream and minimize the need for end product testing. Product and process understanding, in combination with quality risk management (see ICH Q9), will support the control of the process such that the variability (e.g., of raw materials) can be compensated for in an adaptable manner to deliver consistent product quality."

CPP selection has traditionally been difficult because of a lack of a systematic approach to the problem. CPPs can be found in media, upstream and downstream unit operations, and drug-product processing. Due to the large number of unit operations and media complexity, it is easy to overlook processing parameters and materials that may impact drug-substance and drug-product variation and CQAs. Failure to identify critical parameters can result in unexplainable variation during batch processing and lot acceptance.  

The key steps to selecting CPPs and their application to process control are as follows:  					

Identify critical quality attributes (CQAs) for drug-product and substance

Select API, excipients, materials, and container closure

Define all unit operations and process flow

Define all product and process specification limits

Achieve acceptable results for method validation of all analytical methods

Complete quality risk management for factor/response selection for all critical unit operations and materials

Explore the design space for all key factors identified during the risk assessment using design of experiments (DOE) or other multivariate methods

Determine the factor effect size and select all CPPs

Evaluate CPPs for ease of control and practical application to process control.

These steps are considered in detail below. 

 CQAs are those attributes that are important to the quality of the drug product and that remain consistent with those used in clinical studies. The industry generally associates them with ICH parameters, such as identity, purity, potency, stability, and safety. CQAs provide the justification and rationale of what is critical to function and what ultimately needs to be controlled to assure compliance and fit for use. CQAs are the foundation upon which the CPPs must be associated. Line-of-site between CPPs and CQAs is considered a major component of the drug-development strategy. 

Ingredient, materials, and container closure.

Critical process parameters (CPPs) and their associated process controls are crucial to drug development and process validation and to the evaluation of every manufacturing unit operation.