Trends in Analytical Chemistry 101 (2018) 24e33

Contents lists available at ScienceDirect

Trends in Analytical Chemistry

journal homepage: www.elsevier.com/locate/trac

‘Quality by Design’ approach for the analysis of impurities in

pharmaceutical drug products and drug substances
Amandine Dispas a, b, Hermane T. Avohou a, Pierre Lebrun c, Philippe Hubert a,

dric Hubert a, *
Ce
a
University of Li
ege (ULiege), CIRM, Laboratory of Pharmaceutical Analytical Chemistry, Li ege, Belgium
b
University of Li
ege (ULiege), CIRM, Laboratory for the Analysis of Medicines, Li
ege, Belgium
c
Arlenda S.A., Li
ege, Belgium

a r t i c l e i n f o a b s t r a c t

Article history: The pharmaceutical industry is highly regulated by quality policies. The concept of risk management is
Available online 14 November 2017 strongly integrated into the quality assurance system to ensure pharmaceuticals' quality and patients'
safety. In the context of quality control, the detection of impurities in raw materials and finished products
Keywords: is a major concern. It can be challenging for analytical scientists to meet specificity/selectivity and
Quality by Design (QbD) sensitivity requirements. Obviously, separation techniques are widely used for the detection of impu-
Analytical Quality by Design (AQbD)
rities but the method development required to achieve Analytical Target Profile (ATP) concerns is often
Method Operable Design Region (MODR)
challenging. Therefore, to ensure pragmatic and systematic methods development and simultaneously
Design Space (DS)
Risk management
manage the risk associated with analytical methods, the principles of Quality by Design (QbD) should be
Pharmaceutical impurities applied. This paper provides an overview of QbD principles and statistical strategies (mainly DoE-DS
Separative analytical techniques approach) which can be applied to impurity detection methods, as well as a review of the literature
where QbD has been applied to these types of analytical methods.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction [1]. This can be achieved by the introduction of control procedures

in line with the risk incurred throughout the process.
Quality assurance is a major concern in the pharmaceutical in- Consequently, an in-depth scientific knowledge of the quality of
dustry, as described by Good Manufacturing Practices (GMP) re- the product is essential. Such a level of understanding is only ac-
quirements. This concept should be present through the whole quired by the application of a systematic approach to pharmaceu-
pharmaceutical product lifecycle to ensure product quality and tical development, as facilitated by the Quality by Design (QbD)
GMP compliance. This includes the management of (among others) strategy [2].
environment, equipment, procedures and staff, as well as all kinds The QbD concept, initially introduced for manufacturing pro-
of materials/reagents/references or data and deliverables. None- cesses, can be described in four steps [3]:
theless, it is now largely recognized that one important component
of quality assurance system is the management of risk associated - determination of patient requirements, namely, the Quality
with the pharmaceutical product. As stated in ICH Q9, “An effective Target Product Profile (QTPP);
quality risk management approach can further ensure the high quality - design and development of the manufacturing process;
of the drug (medicinal) product to the patient by providing a proactive - risk assessment and definition of the manufacturing Design
means to identify and control potential quality issues during devel- Space (DS); and
opment and manufacturing. Additionally, use of quality risk man- - implementation of a control strategy.
agement can improve the decision making if a quality problem arises.”
An important step in the manufacturing process lifecycle con-
sists of quality control activities either during or after production.
Depending on the manufactured product, analysts may face chal-
lenges in developing analytical methods fit for their intended
* Corresponding author. purposes, especially in the field of impurities control. Over the
E-mail address: [email protected] (C. Hubert).

https://doi.org/10.1016/j.trac.2017.10.028
0165-9936/© 2017 Elsevier B.V. All rights reserved.
 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33 25

course of many years, scientists have acquired detailed knowledge during the method lifecycle are managed. This refined DS is also a
about analytes and their chromatographic/electrophoretic behav- robust area where the desired quality is achieved. Consequently,
iour. Retention or migration mechanisms for liquid, gas, super- QbD strategy may be considered as a learning process [5].
critical fluid chromatography or electrophoresis are deeply The Analytical Quality by Design (AQbD) concept is defined
understood today, facilitating the management of complex sam- similarly to the QbD approach for manufacturing processes.
ples. Two distinct approaches are observed in analytical develop- Therefore, AQbD also includes four main steps:
ment [4,5]:
- determination of the required analytical method performances
- the use of empirical strategy to select a suitable experimental (the ATP);
condition as quickly as possible; or - determination of relevant method parameters and quality at-
- management of the development through a risk-based strategy, tributes (the screening phase);
facilitating an extensive knowledge of the analytical method. - optimization of the method and assessment of risk by defining
the analytical DS (the robust optimization); and
Prior knowledge is the fundamental keystone of both analytical - implementation of a control strategy for the continuous
method development strategies. Development commonly relies on improvement of the method.
One-Factor-At-a-Time (OFAT) or the trial-and-error approach,
expressed by Quality-by-Testing (QbT). This methodology is wide-
spread because it is (wrongly) believed to yield a suitable and faster 2. Discussion
answer. Although such an approach could be applied to relatively
simple problems, it can be inappropriate in cases of complex 2.1. Overcoming barriers to implement Quality by Design (QbD) in
samples or difficult separations (i.e., impurities analysis). Indeed, it analytical method development
has been repeatedly demonstrated that a QbT approach does not
provide any proper characterization of analytical methods. An ac- The starting point for the development of a separative method
curate understanding of how factors such as mobile phase, back- is the selection of method parameters. For readability purposes
ground electrolyte composition, pH, gradient time (and so on) the discussion below will focus on the liquid chromatography
affect the peak retention/migration time, and how method un- technique, but the same could also apply to other separation
certainties affect the peaks' separation are not managed [6]. techniques. Specific problems exist for the screening of chro-
Moreover, such a development does not allow systematic assess- matographic factors, leading to some reluctance to use DoE in the
ment of robustness throughout the development process or even to industry. One of these issues is that some early-studied principal
meet USP requirements [7,8] or pharmaceutical guidelines [1,2,9] factors, such as stationary phase and solvent, are qualitative.
regarding risk management. Consequently, the number of parameters to be considered in
The second approach is the one recommended by the authors. statistical modelling explodes, preventing the elaboration of
This approach can be defined as Analytical Quality by Design sparse and efficient experimental designs. Another issue is that,
(AQbD) strategy, which is not explicitly discussed by ICH Q8. In this for different stationary phases the chromatographic interactions
context, EMA and FDA introduced some guidelines to implement of continuous factors (pH, etc.) might be so different that simple
the concept defined in ICH Q8 in the field of analytical methods models cannot cope with all column-to-factor interactions.
[10,11]. The Analytical Target Profile (ATP) and Method Operable However, in the case of models with higher order interactions
Design Region (MODR) were introduced as parallel analytical (complex models) it is not possible to keep the number of ex-
concepts to QTPP and DS, defined specifically for manufacturing periments low.
processes. In the present paper, MODR is described as the combi- Therefore, during screening, the authors advise the fixing of
nation of Design of Experiments (DoE) and computation of a qualitative factors such as the type of stationary and mobile phases
probabilistic DS (that has quantifiable risk). The DoE-DS strategy is based on scientific knowledge of the molecules and any relevant
systematically used in this paper to refer to MODR, in order to be impurities. Indeed, it is very unlikely that these parameters will
coherent with referenced research papers. The DS represents a change during routine use. Then, during method optimization for
specific area of the experimental domain gathering a set of exper- robustness, quantitative factors such as pH, mobile phase compo-
imental conditions where the desired quality is achieved while sition or gradient time can be studied.
heeding inevitable uncertainties, i.e., a robust experimental If necessary, pre-tests on qualitative factors can be conducted.
domain. From a quality assurance point of view, working within the Usually, the chosen column(s) will be those optimizing peak
DS does not represent change of method [2]. One of the key con- shapes, time of analysis and selectivity. Knowledge of the in-
cepts of this strategy is the continuous improvement of the chosen teractions between the stationary phase, the mobile phase and the
method by means of experience acquired or new data collected molecule of interest should of course be included in such studies.
throughout its lifecycle. Consequently, AQbD strategy is increas-
ingly being adopted since it allows an earlier understanding of 2.2. Difficulties related to impurities analysis
method and guarantees the determination of a wider set of
experimental conditions [12e15]. In the analytical method development, impurities have always
Fig. 1 contrasts QbT and QbD approaches. In this representation, been a matter of particular importance because of several diffi-
“Method understanding?” is presented as the link between QbT and culties discussed below.
QbD methodology. Indeed, having an understanding of the method
is the key concept that differentiates both approaches. In the latter, 2.2.1. Availability of material
the knowledge space resulting from a first optimization phase Some impurities (i.e., degradation and manufacturing impu-
(called “DS 1”) could be further explored. Indeed, despite the fact rities) might be difficult to obtain in sufficient quantities to run a
that working within “DS 1” already ensures that all separations DoE. Forced degradation of raw materials or finished products is an
obtained are acceptable, a subsequent optimization phase could be option, but it should be noted that this does not provide proof that
performed using a second DoE (called “DoE 2”) resulting in a sec- all possible impurities observed in the future will be present in the
ond DS (called “DS 2”). In this refined DS, new constraints arising sample.
26 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33

Fig. 1. Comparison of Quality by Testing (QbT) approach (Left) and Quality by Design (QbD) approach (Right).

2.2.2. Degradation of the material during experiments 2.2.5. Concentration levels

Degradation of the material during the several runs that One of the main issues is the concentration levels and especially
comprise the DoE can occur. In this case, peak width and height the difference between API and impurities concentration (i.e., fac-
could evolve during the experiments, adding difficulties in com- tor 100e1000) is usual). Furthermore, the impurity level might be
pound identification, as well as noise in the tracking of the reten- very low. It means that a stability indicating method must be able to
tion times. track them. Nevertheless, it is the very purpose of the DoE to
develop such a method. Spiking the sample to increase impurities
2.2.3. Change of formulation concentration can be an option, when possible (see “availability of
During the lifecycle of a drug, it is often necessary to adapt its material”).
formulation. This is typically done to improve long-term stability,
hence potentially reducing the number of impurities. This can also 2.3. Analytical Quality by Design (AQbD) and risk management
lead to new compounds being present in the chromatogram,
increasing the risk of co-elution at a fixed chromatographic setting. In order to obtain a flexible method that can overcome the
difficulties described above, it is necessary to develop DS that ex-
2.2.4. Active Principle Ingredient (API) and impurities behaviour press a high probability of meeting specification(s) with a high
Because impurities generally have molecular structures that are robustness. This means that the range of chromatographic settings
close, or related to, the API, they also tend to show similar chro- defining the DS must be large enough to allow the method to be
matographic behaviour. It is of paramount importance in the choice updated easily when needed. Though a large DS might seem costly
of the stationary phase to use all possible knowledge to choose a to develop, the cumulative cost of OFAT development (and with it,
column that will eventually allow a full separation. However, these completely redeveloping an assay due to a new impurity or a
separations often remain limited, i.e., they do not appear as change of formulation) is tremendously higher when all multivar-
robustly as needed. Fortunately, as their behaviour remains corre- iate combinations and interactions of input variables are not known
lated, they tend to stay separated even when undergoing separative precisely.
process variability, as illustrated in Fig. 2. On the left, the two
compounds are fairly different, thus inducing independence in 2.3.1. Analytical Target Profile
their chromatographic behaviour. On the right, the API and impu- As described above, AQbD starts with the definition of the ATP,
rity dependencies allow the separation to remain under analytical which is analogous to the QTPP in a product's QbD approach [16].
process variability. This shows how critical it is to account for these Furthermore, Critical Methods Attributes (CMAs) could be intro-
correlations between peaks when attempting to model the chro- duced as an analytical equivalent to the Critical Quality Attributes
matographic behaviour, and to meet the ICH Q8 definition of DS [2]. (CQAs) defined in ICH Q8. The ATP describes the purpose and scope
 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33 27

Fig. 2. Impact of peak retention correlation when two peaks behave independently (Left) or not, as in the case of API and impurity (Right).

of the method along with a set of performance criteria including: definitive screening designs [18]. Given their very low cost in term
the parameters to be measured; the CMAs of the reportable results; of experiments, such designs generally enable only limited under-
their specifications and quality levels [3,15]. For example, in chro- standing of interactions among CMPs, and thus provide insufficient
matography, CMAs could be the separation criterion (defined as the method optimization. Consequently, optimization designs such as
difference between the end of a first peak and the beginning of the the central composite, the BoxeBehnken and I-optimal designs are
second peak of the critical pair) [14]. then used to find the combination of relevant CMPs that predicts
the optimal CMAs with good precision [14,19]. The interested
2.3.2. Risk assessment reader is invited to read specialized textbooks [20e22] for details
Based on the ATP, the most appropriate analytical techniques are on adequate DoE for DS.
selected and then risk-assessed to identify and prioritize potential
factors that can affect their performances. To do so, devoted tools
such as flow-charts and Ishikawa diagrams may be used to split the 2.3.3.2. Analytical Design Space (DS). The DS concept is intimately
method into key sequences, and to identify and characterize po- linked with the QbD approach. ICH Q8 defines the DS for phar-
tential risks associated with them. These risks can be prioritized maceutical processes as the “multidimensional combination and
with appropriate tools such as Failure Mode and Effects Analysis interaction of input variables that have been demonstrated to provide
(FMEA), and then grouped into controlled factors, potential noise assurance of quality”. Applied to analytical methods, the DS may be
factors and CMPs [3,14,15]. Controlled and uncontrolled noise fac- defined as the set of all combinations of a method's input variables
tors should be identified in order to perform a method ruggedness that have been proven to provide assurance of the quality of the
study, focused on uncontrolled factors [17]. data produced by the method [3,14,15]. In practice, the analytical DS
corresponds to the range of operating conditions where future
CMAs of the analytical method are within acceptance limits, with a
2.3.3. Design of Experiments (DoE) and Analytical Design Space
high level of assurance. If the DS is large, the method can be
(DS)
considered robust; this is because changes of operating conditions
Potential CMPs such as method and instrumental factors are
within the DS will not degrade the quality of the results. It must be
further investigated for robustness using the statistical DoE and
stressed that the concept of quality assurance underscores the need
multivariate analysis. DoE is a critical step in understanding the
to indicate the likelihood of the method producing acceptable re-
method's operation and defining control strategies. Contrary to the
sults [6,23,24], for instance expressed as a joint probability of
OFAT approach, DoE is a structured, cost-effective and cost-efficient
meeting specifications for all CMAs.
method to organize experiments and to determine the simulta-
A mathematical formalism to compute a DS that is fully
neous effects and interactions of multiple CMPs on the CMAs. The
compliant with this requirement of quality assurance, as stated in
ultimate objective of DoE is the definition of the analytical DS,
ICH Q8, has been proposed by Peterson [6] as:
which is the operating ranges of CMPs that guarantee quality re-

   
sults. This can ideally be derived from the multivariate statistical e
analysis of the DoE outputs. DS ¼ xe 2c : Pr Ye 2Ax; data  p0

2.3.3.1. Design of Experiments (DoE). DoE involves first the choice of where x~ is a vector of CMPs, c is the experimental domain, DS is the
appropriate experimental design(s). Generally, two types of designs design space, Prð$Þ stands for the posterior probability of an event,
might be used, depending on the number of CMPs to be tested and ~ is a vector of predicted CMAs, A is the subspace defined by the
Y
the complexity of the mathematical relationships between the acceptance limits for the CMAs, p0 is the minimum quality level,
CMAs and the CMPs. Although prior scientific knowledge should and data denotes the data used to build the model. In other words,
allow selection of the riskiest factors, if a large set of CMPs are the DS includes any point of the experimental domain whose
identified, screening designs might first be needed to screen out predicted CMAs meet the specifications with at least a pre-specified
those having negligible effects on the CMAs. The most commonly probability p0, given the data.
used screening designs are the popular PlacketteBurman or frac- An effective approach to obtain predictions Y ~ of the CMAs is to
tional factorial designs. Another common class of designs are the generate Monte Carlo samples from the joint posterior predictive
orthogonal designs or D-optimal designs and the more recent distribution of the CMAs [23,24]. This predictive distribution is
28 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33

derived from a Bayesian (multi-response) multiple regression of determination of impurities. As explained previously, it is impor-
the CMAs as functions of the CMPs as proposed by Peterson [6] and tant to note that method development that does not take predic-
Lebrun et al. [24]. Thus, this distribution accounts for correlations tion error and its propagation into account does not allow the
among CMAs and uncertainties related to both model parameters performance of QbD compliant optimization. Indeed, this approach
and common cause variation. In cases where the modelled re- does not manage the risk. Consequently, methodology involving
sponses differ from the CMAs, the predictive distribution of the mean response surfaces obtained using DoE to define optimal
CMAs can subsequently be computed as function of the multivar- condition could not be considered as a QbD compliant approach.
iate predictive distribution of the modelled responses based on the Consequently, some methods described in the literature as ‘QbD
Monte Carlo samples in a Monte-Carlo error propagation scheme. approach’ are not truly QbD methodology and are not discussed in
The Bayesian approach for multi-response optimization of methods the present paper. Several research groups focus their work on QbD
is strongly recommended, due to is recognized high performance. implementation; their main papers related to detection of impu-
The reader is referred to the textbook of del Castillo [21] for details rities are summarized in Table 1.
on its rationale and technical implementation. An illustrated flowchart of QbD compliant methods for detec-
A widespread alternative method to compute the DS is the tion of impurities is presented in Fig. 3. The first step of a QbD
overlapping mean responses, which determines the DS as the compliant method is the definition of method objectives by means
subspace of CMPs' domain where the estimated mean responses of of ATP. When dealing with detection of impurities, the ATP is
individual CMAs are all within specifications. However, this mainly focused on method selectivity to ensure a complete sep-
approach has several flaws that have been extensively demon- aration between API, related and unknown impurities, and
strated by many studies [6,14,19,23,24]. In brief, firstly, the models eventually excipients. A second objective is also to reach the
used do not account for correlations among multiple CMAs and required method sensitivity regarding the tolerance (e.g. 0.1%
uncertainties about unknown model parameters. Secondly, the of API). Regarding published methods [5,25e42], method
predicted CMAs are mean values. Consequently, even if mean re- selectivity is the main target extensively studied during method
sponses meet specifications, it is well-established that individual development.
future runs of the method will not necessarily be within acceptance It is now well known that QbD encompasses DoE strategy; the
limits. Therefore, the DS-based mean responses may include definition of DoE is therefore also a crucial step in reaching the
operating conditions with low assurance of quality results, and method objective. To determine DoE factors (CMP), a screening
therefore this approach does not produce a DS compliant with QbD step using adequate screening design is generally used and rec-
expectations. However, such approaches are widely used by the ommended in the field of drug manufacturing (e.g. process opti-
scientific community. mization, etc.). Regarding analytical method development, the use
of screening design could be considered a marginal practice.
2.3.4. Control strategy (CS) Indeed, empirical experiments are generally mandatory to provide
Following the DS computation, it is recommended to define a to the analyst some preliminary results/information. In the specific
control strategy (CS). The goal of the CS is to ensure that the chosen case of chromatography, stationary phases (column) selection is
method performs as intended in routine use. Performance param- performed before the definition of the DoE for two main reasons
eters for routine monitoring can be derived from the outcomes of [34,42]. Firstly, stationary phase is a qualitative factor requiring the
the DS analysis. These parameters are known as validity tests or DoE to be performed integrally on each column and leading
system suitability tests (SST). quickly to a huge number of experiments. Secondly, performing
exactly the same DoE on different columns is not always relevant.
2.4. Quality by Design (QbD) compliant methods for the It is also important to notice that some software [37] extrapolates
determination of drug impurities: a critical review data from one stationary phase to another, which is not chro-
matographically relevant. In the field of electrophoretic tech-
The control of impurities during the manufacture of drugs in- niques, the selection of separation mode (e.g., CZE, MEKC, MEEKC,
volves two main steps: the control of raw materials before drug etc.) should also be determined before establishing method opti-
manufacturing; and the control of finished products before batch mization design [25e32]. Afterwards, continuous factors are
release. This two-stage control process requires selective, specific investigated, such as pH, organic content, gradient time, temper-
and sensitive analytical techniques. In this context, separation ature, voltage, etc. When the analyst selects too many factors
techniques are generally used (chromatography and electropho- (CMP), a screening design is really interesting to help opt for the
resis) with UV or mass spectrometer detectors. Liquid chromatog- most relevant/influential ones before starting method optimiza-
raphy (LC) is generally considered the gold-standard analytical tion with maximum 3e4 factors [25e32]. In the literature, some
technique involved in the raw material control, as described in the applications dealing with Fusion software proposed a screening
Pharmacopeias (EP, USP, BP). However, some of the standardized step with totally different CMP than those selected for the opti-
methods coming from the Pharmacopeia could be considered mization step. A two-steps optimization denomination should be
obsolete and several alternatives are now proposed in the litera- more adequate but shows also that this approach is not fully QbD
ture. Among those techniques, optimized LC, capillary electropho- compliant [37].
resis (CE) and its related techniques and, more recently, The response(s) to be studied should also be selected before
supercritical fluid chromatography (SFC) present several advan- starting the DoE. This corresponds to the CMAs, making it possible
tages such as efficiency, speed, sensitivity, etc. to fulfil the method objective. Method selectivity is the core busi-
Separation of API and its related impurities is highly challenging ness for the determination of impurity, and various CMAs could be
because of the similarity between the chemical structures of the selected: resolution Rs [25e28,30e32,37e40], selectivity a [33,35]
targeted analytes. Moreover, the concentration level of API in or separation S [5,29,34,36,41,42]. A discussion about the advan-
comparison with impurities could lead to a really wide peak. In this tages and drawbacks of these responses (Rs, S) regarding mathe-
context, method development using DoE is now generally used. matical modelling is beyond the scope of this review and can be
Furthermore, as described above, the pharmaceutical regulatory found in more specific papers [43]. Nevertheless, as mentioned in
requirements and advantages of QbD methodology have led to an Table 1, analysts use several strategies: the CMA is applied to one of
increase in use of the QbD compliant analytical method for the several critical pair(s) or to the whole panel of analytes. Moreover,
Table 1
An overview of QbD compliant method developments for detection of impurities.

Ref. Analytes Technique Methodology to select CMP Method optimization DoE CMP CQA Statistical approach to define DS Software
and CQA

[25] Amitriptyline and 4 impurities Solvent-modified Ishikawa diagram Doehlert design Voltage Rs of two critical Sweet spot plots to fix one factor Modde
MEKC Screening design 23 n-butanol concentration pairs (V) and compute DS on the three
ACN concentration Analysis time others
Urea concentration Monte Carlo simulations
[26] Drugs formulation: Captopril and 1 Cyclodextrin- and Preliminary empirical Central composite design BGE composition: Rs of two critical Sweet spot plots and Monte-
impurity e Hydrochlorothiazide solvent-modified experiments 29 - pH pairs Carlo simulations
and 3 impurities MEKC Screening symmetric matrix - sodium cholate Analysis time
concentration
- n-butanol concentration
Voltage
[27] Metformin and 5 impurities Cyclodextrin- Preliminary empirical Doehlert design Injection time Rs of two critical
modified MEKC experiments 23 CD concentration pairs
Screening asymmetric matrix Buffer concentration Analysis time
pH
[28] Glibenclamide and 2 impurities CE Preliminary empirical BoxeBehnken design Voltage Rs of one critical

A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33

experiments 27 Injection time pair
Screening symmetric matrix Buffer concentration Efficiency
pH Analysis time
[29] Almotriptan and 3 impurities MEEKC Preliminary empirical D-optimal Temperature S of two critical
experiments 62 Voltage pairs
Screening asymmetric matrix Buffer concentration Analysis time
Buffer pH
Microemulsion composition: %
aqueous phase, % oil phase, %
surfactant
[30] Zolmitriptan and 5 impurities CZE Preliminary empirical BoxeBehnken design Buffer concentration Rs of two critical
experiments 15 pH pairs
Screening symmetric matrix Temperature Analysis time
Efficiency
[31] Diclofenac and 5 impurities Cyclodextrin- Preliminary empirical D-optimal Voltage Rs of one critical
modified MEEKC experiments 36 Buffer pH pair
Screening design (D-optimal CD concentration Analysis time
16 exp.) Microemulsion composition: %
aqueous phase, % oil phase, %
surfactant
[32] Levosulpiride and enantiomeric Dual- Preliminary empirical Doehlert design pH Rs
impurity cyclodextrin experiments 23 sCD concentration Analysis time
chiral CE Screening asymmetric matrix nCD concentration
Voltage
[33] Bilastine and 2 degradation HILIC Preliminary empirical study BoxeBehnken design ACN content (%) Selectivity of
impurities HILIC theoretical knowledge 15 Ammonium acetate critical pair
concentration Analysis time
pH
[34] Olanzapine and 7 impurities HILIC Chromatographic screening Rechtschaffen design Temperature S between two
Preliminary empirical study 10 Initial aqueous phase content critical pairs
(%)
Gradient time

(continued on next page)

29
Table 1 (continued )

30
Ref. Analytes Technique Methodology to select CMP Method optimization DoE CMP CQA Statistical approach to define DS Software
and CQA

[35] Pramipexole and 5 impurities RP-HPLC Preliminary empirical study D-optimal Stationary phase Retention of first Retention factor modelling to Design-
29 Salt type and last eluting predict the CQAs and Monte- Expert
Salt concentration peaks Carlo simulations and
% ACN Selectivity Matlab
Temperature
[36] Dabigatran etexilate mesilate and RP-HPLC Preliminary empirical study BoxeBehnken design Gradient time S Retention times modelling to
10 impurities 16 % ACN initial predict the CQAs and Monte-
% ACN final Carlo simulations

[37] Drugs formulation: Naproxen and 6 RP-UHPLC Fractional factorial design Fractional factorial design Flow rate Rs Process capability index using Fusion
impurities e Sumatriptan and 5 (pH e stationary phase e 20 Temperature Number of Monte-Carlo simulations
impurities organic modifier) Gradient time peaks
(44 experiments)

[38] HIV-tritherapy: lamivudine, RPLC Previous methods used in the Full factorial pH Rs of 6 critical Overlay of response surface and Devize
abacavir, dolutegravir and 26 lab design þ complementary Temperature pairs Monte-Carlo simulations
impurities experiments to reach central Mobile phase part B % RT of 2

A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33

composite design compounds
19 S of 2 critical
pairs

[39] Pramipexole and 4 impurities RPLC Based on compendial method Full factorial design Gradient time Rs Retention times modelling DryLab
(European Pharmacopeia) 21 pH Rs prediction
Temperature
Independent eluent study:
ionic strength and ion pairing
concentrations
[40] Terazosin and 10 impurities RP-UHPLC-MS Preliminary empirical study Full factorial design Gradient time
12 Temperature
Ternary eluent composition

[5] Confidential API and 4 impurities RPLC-MS Preliminary empirical study D-optimal mixture design Mobile phase composition: S Retention times modelling JMP and
33 MeOH/ACN/buffer Bayesian model and Monte- R
Temperature Carlo simulations to obtain the
[41] Cholecalciferol (vitamin D3) and 3 SFC Preliminary empirical study Central composite design Temperature distribution of S
impurities 35 % EtOH final
Gradient time
[42] Salbutamol sulphate and 5 SFC Preliminary empirical study Central composite design % MeOH initial
impurities Chromatographic screening 27 Gradient time
Pressure
Temperature
 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33 31

Fig. 3. Flowchart: QbD compliant methodology for the determination of impurities.

32 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33

one or more CMA could be added, such as the analysis time, the are reference techniques for detection of impurities. The method
peak efficiency, etc. optimization is most often focused on separation of peaks. The
Following the definition of CMP and CMA, the DoE should be research literature presents independent studies of different CMP.
selected according to the method objective. As expected, the refer- This methodology is not totally relevant to developing method
enced methods [5,25e42] are most often used in design optimiza- knowledge and appropriate DS computation. Joint prediction is a
tion. However, data treatment to model and compute the DS is one really useful tool to simultaneously model all required CMAs, as
of the most interesting topics to discuss. Different strategies are described in several papers. Furthermore, AQbD compliant method
published: the most popular one is to overlay contour plot (or sweet optimization has to take into account the prediction error, and its
spot plot) for each CMA and then perform Monte-Carlo simulations propagation, in order to manage the risk. In this context, both DoE
to define the DS [25e34,38]. Using this strategy, the CMA modelling and the statistical approach used to compute the DS should be
is performed independently by neglecting the interactions and selected carefully.
could lead to method misunderstanding. Furthermore, the inde- Finally, API and impurities separation is not enough to properly
pendent study of different CMPs is not relevant in reaching method fulfil the ATP. The method's quantitative performance should be
knowledge/understanding and adequate DS computation [20]. Joint evaluated by means of analytical method validation. The total error
prediction is a really useful tool to simultaneously model all required approach must be promoted with reference to pharmaceutical risk
CMAs, as described in several papers [5,41e45]. This statistical management requirements.
strategy is mandatory to reach an in-depth method understanding
compliant with ICH Q8 requirements. Acknowledgements
Regarding the definition of DS, the method robustness is directly
assessed during method optimization. In this context, robustness Research grants from the Walloon Region of Belgium and EU
testing is no longer required. Some papers propose robustness Commission (project FEDER-PHARE) to Amandine Dispas are
study not as a mandatory analytical method lifecycle step but as a gratefully acknowledged.
demonstration of the reliability of the DS methodology. Even if an Research grants from the Walloon Region (project Vibra4Fake)
extensive robustness study is not required, it is a good practice to to Hermane T. Avohou are likewise gratefully acknowledged.
verify several DS conditions to ensure that the CMAs are met. It also
important to note that the majority of published QbD compliant
methods focus on qualitative CMAs (i.e., separation, resolution, References
etc.), so the method should be robust regarding qualitative criteria.
[1] ICH, Q9, Quality Risk Management, 2005.
Consequently, robustness of quantitative performances should be [2] ICH, Q8(R2), Pharmaceutical Development, 2009.
studied before or during method validation. More extensive dis- [3] P. Borman, K. Truman, D. Thompson, P. Nethercote, M. Chatfield, The appli-
cussion about method robustness study in the context of a risk- cation of quality by design to analytical methods, Pharm. Technol. 31 (10)
(2007) 142e152.
based approach can be found in a specific reference dealing with [4] R.M. Maggio, S.E. Vignaduzzo, T.S. Kaufman, Practical and regulatory consid-
the determination of impurities [46]. erations for stability-indicating methods for the assay of bulk drugs and drug
Following the optimization step using AQbD strategy, the formulations, Trends Anal. Chem. 49 (2013) 57e70.
[5] C. Hubert, P. Lebrun, S. Houari, E. Ziemons, E. Rozet, Ph. Hubert, Improvement
quantitative performance of the method should be evaluated by of a stability-indicating method by Quality-by-Design versus Quality-by-
means of method validation. In this context, the SFSTP commis- Testing: a case of a learning process, J. Pharm. Biomed. Anal. 88 (2014)
sion proposed a validation strategy focused on risk management 401e409.
[6] J.J. Peterson, A Bayesian approach to the ICH Q8 definition of design space,
and using a clear decision tool as recommended by ICH Q9 J. Biopharm. Stat. 18 (2008) 959e975.
[1,19,20,47e50]. Using accuracy profile as a powerful decision tool [7] U.S. Pharmacopeial Convention, New Chapter 1224, 1225, 1226, USP panel
has already been demonstrated for detecting impurities with LC or expert.
[8] FDA, Code of Federal Regulations, Title 21, Section 211, Current Good
SFC methods [5,41,42]. The full integration of optimization and
Manufacturing Practice for Finished Pharmaceuticals (Rockville, MD, 2012).
validation phases was also proposed in order to perform an [9] ICH, Q10, Pharmaceutical Quality System, 2008.
evaluation of the quantitative performances of the whole [10] EMA-FDA Pilot Program for Parallel Assessment of Quality-by-Design Appli-
cations: Lessons Learnt and Q&A Resulting from the First Parallel Assessment,
robustness area [44]. This first demonstration is a really inter-
2013.
esting perspective regarding pharmaceutical impurities applica- [11] Quality by Design Approaches to Analytical Methods e FDA Perspective, AAPS,
tions. Considering this approach, the CMA could also be the 2011.
quantitative performances expressed as trueness, precision or [12] S. Karmarkar, R. Garber, Y. Genchanok, S. George, X. Yang, R. Hammond,
Quality by Design (QbD) based development of a stability indicating HPLC
accuracy, with the ATP describing the minimal requirements for method for drug and impurities, J. Chromatogr. Sci. 49 (2011) 439e446.
these criteria to achieve a method usable for product character- [13] M.H. Rafamantanana, B. Debrus, G.E. Raoelisen, E. Rozet, P. Lebrun, S. Uverg-
ization and release. Ratsimamanga, Ph. Hubert, J. Quentin-Leclercq, Application of design of ex-
periments and design space methodology for the HPLC-UV separation opti-
mization of aporphine alkaloids from leaves of Spirospermum penduliflorum
3. Conclusion Thouars, J. Pharm. Biomed. Anal. 62 (2012) 23e32.
[14] E. Rozet, P. Lebrun, P. Hubert, B. Debrus, B. Boulanger, Design Spaces for
analytical methods, Trends Anal. Chem. 42 (2013) 157e167.
The detection of impurities is a major concern for quality control [15] F.G. Vogt, A.S. Kord, Development of quality-by-design analytical methods,
of raw materials and finished products. Separation techniques are J. Pharm. Sci. 100 (2011) 797e812.
widely used for detection of impurities regarding selectivity and [16] M. Pohl, K. Smith, M. Schweitzer, M. Hanna-Brown, J. Larew, G. Hansen,
P. Borman, P. Nethercote, Implications and opportunities of applying QbD
sensitivity requirements. To deal with close chemical structures principles to analytical measurements, Pharm. Tech. 34 (2010) 52e59.
and different concentration levels between API and impurities, a [17] P. Borman, M.J. Chatfield, I. Damjanov, P. Jackson, Method ruggedness studies
systematic approach to method development is advised. In this incorporating a risked base approach: a tutorial, Anal. Chem. Acta 703 (2011)
101e113.
context, the authors proposed the AQbD strategy, which encom-
[18] J. Bradley, C.J. Nachtsheim, A class of three-level designs for definitive
passes the use of DoE and the computation of a probabilistic DS. screening in the presence of second-order effects, J. Qual. Technol. 43 (2011)
AQbD strategy facilitates in-depth understanding of method and 1e15.
risk management. [19] J.J. Peterson, S. Altan, Overview of drug development and statistical tools for
manufacturing and testing, in: L. Zhang, M. Kuhn, I. Peers, S. Altan (Editors),
Different AQbD approaches are available in the literature and Nonclinical Statistics for Pharmaceutical and Biotechnology Industries,
discussed above. As expected, electrophoresis and chromatography Springer International Publishing AG, Cham, Switzerland, 2016, pp. 383e414.
 A. Dispas et al. / Trends in Analytical Chemistry 101 (2018) 24e33 33

[20] P. Goos, J. Bradley, Optimal Design of Experiments: a Case Study Approach, [36] J. Pantovic, A. Malenovic, A. Vemix, N. Kostic, M. Medenica, Development of
John Wiley & Sons, Chichester, United Kingdom, 2011, p. 287. liquid chromatography method for the analysis of dabigatran etexilate
[21] E. Del Castillo, Process Optimization: a Statistical Approach, Springer Science mesilate and its ten impurities supported by quality-by-design methodology,
& Business Media, New York, USA, 2007, p. 459. J. Pharm. Biomed. Anal. 111 (2015) 7e13.
[22] R.H. Myers, D.C. Montgomery, C.M. Anderson-Cook, Response Surface Meth- [37] P.N. Patel, V.S. Karakam, S. Ragampeta, Quality-by-design based ultra high
odology: Process and Product Optimization Using Designed Experiments, John performance liquid chromatography related substances method development
Wiley & Sons Inc., New Jersey, USA, 2016, p. 825. by establishing the proficient design space for sumatriptan and naproxen
[23] J.J. Peterson, K. Lief, The ICH Q8 definition of design space: a comparison of the combination, J. Sep. Sci. 38 (2015) 3354e3362.
overlapping means and the Bayesian predictive approaches, Stat. Biopharm. [38] T. Tol, N. Kadam, N. Raotole, A. Desai, G. Samanta, A simultaneous determi-
Res. 2 (2010) 249e259. nation of related substances by high performance liquid chromatography in a
[24] P. Lebrun, B. Boulanger, B. Debrus, P. Lambert, Ph Hubert, A Bayesian design drug product using quality by design approach, J. Chromatogr. A 1432 (2016)
space for analytical methods based on multivariate models and predictions, 26e38.
J. Biopharm. Stat. 23 (2013) 1330e1351. [39] A.H. Schmidt, M. Stanic, I. Molnar, In silico robustness testing of a compendial
[25] S. Furlanetto, S. Orlandini, B. Pasquani, M. Del Bubba, S. Pinzauti, Quality by HPLC purity method by using a multidimensional design space build by
Design approach in the development of a solvent-modified micellar electroki- chromatography modelling e case study pramipexole, J. Pharm. Biomed. Anal.
netic chromatography method: finding the design space for the determination 91 (2014) 97e107.
of amitriptyline and its impurities, Anal. Chim. Acta 802 (2013) 113e124. [40] R. Kormany, I. Molnar, J. Feteke, Renewal of an old European Pharmacopeia
[26] B. Pasquani, S. Orlandini, C. Caprini, M. Del Bubba, M. Innocenti, G. Brusotti, method for terazosin using modelling with mass spectrometric peak tracking,
S. Furlanetto, Cyclodextrin- and solvent-modified micellar electrokinetic J. Pharm. Biomed. Anal. 135 (2017) 8e15.
chromatography for the determination of captopril hydrochlorothiazide and [41] B. Andri, P. Lebrun, A. Dispas, R. Klinkenberg, B. Streel, E. Ziemons, R.D. Marini,
their impurities: a Quality by Design approach, Talanta 160 (2016) 332e339. Ph Hubert, Optimization and validation of a fast supercritical fluid chroma-
[27] S. Orlandini, B. Pasquini, R. Gotti, A. Giuffrida, F. Paternostro, S. Furlanetto, tography method for the quantitative determination of vitamin D3 and its
Analytical quality by design in the development of a cyclodextrin-modified related impurities, J. Chromatogr. A 1492 (2017) 171e181.
capillary electrophoresis method for the assay of metformin and its related [42] A. Dispas, V. Desfontaine, B. Andri, P. Lebrun, D. Kotoni, A. Clarke, D. Guillarme,
substances, Electrophoresis 35 (2014) 2538e2545. Ph Hubert, Quantitative determination of salbutamol sulfate using achiral su-
[28] S. Furlanetto, S. Orlandini, B. Pasquani, C. Caprini, P. Mura, S. Pinzauti, Fast percritical fluid chromatography, J. Pharm. Biomed. Anal. 134 (2017) 170e180.
analysis of glibenclamide and its impurities: quality by design framework in [43] P. Lebrun, B. Govaerts, A. Ceccato, G. Caliaro, Ph Hubert, B. Boulanger,
capillary electrophoresis method development, Anal. Bioanal. Chem. 407 Development of a new predictive modelling technique to find with confi-
(2015) 7637e7646. dence equivalence zone and design space of chromatographic analytical
[29] S. Orlandini, B. Pasquani, M. Stocchero, S. Pinzauti, S. Furlanetto, An integrated methods, Chemom. Intell. Lab. Syst. 91 (2008) 4e16.
quality by design and mixture-process variable approach in the development [44] C. Hubert, S. Houari, E. Rozet, P. Lebrun, Ph. Hubert, Towards a full integration
of a capillary electrophoresis method for the analysis of almotriptan and its of optimization and validation phases: an analytical-quality-by-design
impurities, J. Chromatogr. A 1339 (2014) 200e209. approach, J. Chromatogr. A 1395 (2015) 88e98.
[30] S. Orlandini, B. Pasquani, C. Caprini, M. Del Bubba, S. Pinzauti, S. Furlanetto, [45] J.J. Peterson, M. Yahyah, A Bayesian design space approach to robustness and
Analytical Quality by Design in pharmaceutical quality assurance: develop- system suitability for pharmaceutical assays and other processes, Stat. Bio-
ment of a capillary electrophoresis method for the analysis of zolmitriptan pharm. Res. 1 (2009) (2009) 441e449.
and its impurities, Electrophoresis 36 (2015) 2642e2649. [46] G. Okafo, A. Laures, P. Jackson, P. Borman, M. Chatfield, Reduced method
[31] S. Orlandini, B. Pasquani, C. Caprini, M. Del Bubba, L. Squarcialupi, V. Colotta, robustness testing of analytical methods driven by a risk-based approach,
S. Furlanetto, A comprehensive strategy in the development of a cyclodextrin- Pharm. Tech. 34 (2010) 72e86.
modified microemulsion electrokinetic chromatographic method for the assay [47] Ph Hubert, J.J. Nguyen-Huu, B. Boulanger, E. Chapuzet, P. Chiap, N. Cohen,
of diclofenac and its impurities: mixture-process variable experiments and P.A. Compagnon, W. Dewe
, M. Feinberg, M. Lallier, M. Laurentie, N. Mercier,
quality by design, J. Chromatogr. A 1466 (2016) 189e198. G. Muzard, C. Nivet, L. Valat, Harmonization of strategies for the validation of
[32] S. Orlandini, B. Pasquani, M. Del Bubba, S. Pinzauti, S. Furlanetto, Quality by quantitative analytical procedures e a SFSTP proposal e part I, J. Pharm.
design in the chiral separation strategy for the determination of enantiomeric Biomed. Anal. 36 (2004) 579e586.
impurities: development of a capillary electrophoresis method based on dual [48] Ph Hubert, J.J. Nguyen-Huu, B. Boulanger, E. Chapuzet, P. Chiap, N. Cohen,
cyclodextrin systems for the analysis of levosulpiride, J. Chromatogr. A 1380 P.A. Compagnon, W. Dewe
, M. Feinberg, M. Lallier, M. Laurentie, N. Mercier,
(2015) 177e185. G. Muzard, C. Nivet, L. Valat, E. Rozet, Harmonization of strategies for the
[33] J. Terzic, I. Popovic, A. Stajic, A. Tumpa, B. Jancic-Stojanovic, Application of validation of quantitative analytical procedures e a SFSTP proposal e part II,
Analytical Quality by Design concept for bilastine and its degradation impu- J. Pharm. Biomed. Anal. 45 (2007) 70e81.
rities determination by hydrophilic interaction liquid chromatographic [49] Ph Hubert, J.-J. Nguyen-Huu, B. Boulanger, E. Chapuzet, N. Cohen, P.-
method, J. Pharm. Biomed. Anal. 125 (2016) 385e393. A. Compagnon, W. Dewe
, M. Feinberg, M. Laurentie, N. Mercier, G. Muzard,
[34] A. Tumpa, A. Stajic, B. Jancic-Stojanovic, M. Medenica, Quality by Design in the L. Valat, E. Rozet, Harmonization of strategies for the validation of quantitative
development of hydrophilic interaction liquid chromatography method with analytical procedures e a SFSTP proposal e part III, J. Pharm. Biomed. Anal. 45
gradient elution for the analysis of olanzapine, J. Pharm. Biomed. Anal. 134 (2007) 82e96.
(2017) 18e26. [50] Ph Hubert, J.-J. Nguyen-Huu, B. Boulanger, E. Chapuzet, N. Cohen, P.-
[35] A. Vemic, T. Rakic, A. Malenovic, M. Medecina, Chaotropic salts in liquid A. Compagnon, W. Dewe
, M. Feinberg, M. Laurentie, N. Mercier, G. Muzard,
chromatography method development for the determination of pramipexole L. Valat, E. Rozet, Harmonization of strategies for the validation of quantitative
and its impurities following quality-by-design principles, J. Pharm. Biomed. analytical procedures e a SFSTP proposal e part IV, J. Pharm. Biomed. Anal. 48
Anal. 102 (2015) 314e320. (2008) 760e771.