Professional Documents
Culture Documents
DOI 10.1007/s12247-012-9132-z
RESEARCH ARTICLE
combination and interaction of input variables (e.g., material the means to compensate for it. This study demonstrates the first
attributes) and process parameters that have been demon- step of a feed forward quality control system.
strated to provide assurance of quality” [3].
The establishment of a design space should rely on the
use of a design of experiment (DOE) for a single or multiple Material and Methods
unit operations where the relevance of process parameters is
investigated. Several DOEs might be necessary for complex Formulation and Manufacturing
processes. To understand the process, CPPs must be varied
and their associated product properties or critical quality A 2003 Excedrin® Tension Headache-like formulation contain-
attributes (CQA) determined. The information provided by ing acetaminophen (APAP) and caffeine was used. A design of
the DOE constitutes a knowledge space. The design space is experiment was created to optimize the formulation for
established as a subset of the knowledge space based on a direct compression. As shown in Table 1, four lactose mono-
CQA acceptance range [3, 4]. The QbD paradigm offers hydrate:microcrystalline cellulose ratios, two levels of croscar-
flexibility in process control as product quality will be mellose sodium (disintegrant), four sets of blend parameters,
ensured throughout a range of CPPs defined by the design and four tablet compression force targets were tested. The latter
space. It consequently offers the potential for reduced rou- was established by tablet force to failure testing at-line. To limit
tine quality testing and real-time release. the number of experiments, a D-optimal criterion was
Authors assert that over time the design space will need employed, reducing the number of runs from 128 to 16.
to be adapted in light of new knowledge gathered on the The formulation comprised 31.25 % of APAP (Mallinck-
product or due to adaptations in the manufacturing process. rodt Inc., Raleigh, NC, USA) with a median particle size of
For instance, the modification of a unit operation or the 100 μm, 4.05 % caffeine (anhydrous; Spectrum, Gardena,
selection of a different supplier for excipients or drug sub- CA, USA), from 12.44 to 50.56 % of microcrystalline
stance would require the derivation of a new knowledge cellulose (MCC; Avicel PH 102, FMC Biopolymer,
space and the adaptation of the original design space. Such
modifications would require appropriate supplementary
submissions and regulatory approvals.
Table 1 Design of experiment for the initial knowledge space
However, other subtle changes should be considered.
Lot-to-lot variability and physical form changes can signif- Run Run Excipient Croscarmellose Load RMSNV
icantly impact manufacturing processes and clinical perfor- # ordera ratiob sodium level force weightsd
mance [5, 6]. While pharmaceutical companies attempt to (%) (pound)c
include multiple sources of variability in their original 1 16 4:1 2 12,000 210
DOEs, it is typically not possible to envision all of them
2 6 4:1 2 6,000 111
(especially those as unpredictable or uncontrollable as raw
3 14 4:1 1 10,000 201
material variations). In such situations, the robustness of the
4 12 4:1 1 8,000 300
design space is challenged. When changes are significant
5 13 3:2 2 10,000 210
and adversely impact CQAs, the design space may need
6 1 3:2 2 8,000 111
to be adapted. In the present document, a two-step
7 5 3:2 1 12,000 201
approach was taken to study the potential use of an
8 9 3:2 1 6,000 300
adaptive design space. After initially developing a de-
9 15 2:3 2 8,000 201
sign space for a direct compressible formulation (blend-
10 11 2:3 2 12,000 300
ing followed by tableting), the impact of varying raw
11 10 2:3 1 10,000 111
materials on the design space was studied. The ability
12 3 2:3 1 6,000 210
to compensate for variations in raw material properties
13 7 1:4 2 10,000 300
by modifying CPPs and consequently adapting the de-
14 2 1:4 2 6,000 201
sign space as a function of the properties of the input
raw material was demonstrated. 15 8 1:4 1 8,000 210
A comprehensive feed forward control system must be 16 4 1:4 1 12,000 111
implemented to ensure patient safety and product efficacy. a
Order in which runs were manufactured
The possibility to adapt the manufacturing process to not only b
Ratio microcrystalline cellulose/lactose anhydrous
raw material properties but also equipment aging and environ- c
Tablets compression force target at the press (right after compression)
mental changes can provide pharmaceutical companies with d
Weights used in the RMSNV calculation. The first corresponds to the
relevant quality control tools and modeling systems establishing summed weight given to the APIs, the second is the summed weight
the relationship between a new factor affecting the process and for the two excipients, and the third is for the disintegrant
J Pharm Innov
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
u Tablets were analyzed after aging for various attributes (as
u w Yb Y 2 þ w Yb Y 2 þ . . . þ w Yb Y 2
t 1 1 1 2 2 2 n n n
model CQAs). Aging is typically attributed to mechanical
RMSNV ¼ P
ð wÞ
relaxation following uniaxial compression. The magnitude
ð1Þ of these changes varies as a function of the relative ratio of
material having elastic and viscoelastic behaviors under
where w is the weight given to a component, Yb is the compression. A minimum of 2 weeks separated tablet com-
predicted value from NIR spectra and Y is the target pression and data collection. Friability (18 tablets per design
value for the nth component, i is the number of data point (1 per time bin ) — TA series, Erwera, Heusenstamm,
points used in a single window of data collection, t is Germany), disintegration time (18 tablets per design point
the latest collected data point, and j is an individual randomly selected from each time bin — Van Kel Inc., Cary,
prediction time. The RMSNV allows the evaluation of NC, USA), and dissolution profiles (six tablets per design
the trends and distance from target concentrations for point; sampled at 0.5, 5, 10, 15, 30, 45, and 60 min, ran-
not only the active ingredient but also for the major domly selected from each time bin — Dissolution system
excipients of the formulation. 2100B, Distek Inc., North Brunswick, NJ, USA) were mea-
To allow control of the blend endpoint when two or more sured following USP standards for APAP and caffeine tab-
sensors are used, a pooled RMSNV (RMSNVp) statistic was lets [11]. Additionally, radial tensile strength (RTS) was
J Pharm Innov
calculated from force to failure measurements made on 18 using APAP crystal (Rhodapap, Rhodia Organique, Rous-
tablets per design point randomly selected from each time sillion, France) and the physical form of lactose was mod-
bin with a PTB 311 (Pharmatest, Hainburg, Germany). ified by mixing lactose monohydrate with lactose anhydrous
Process parameters and formulation factors were then (1:1 ratio) (51808TT, Sheffield Bio-Science, Norwich, NJ,
related to quality attributes to first determine the critical USA).
quality attributes and second to build a design space. Note JMP 8 (SAS Institute Inc., Cary, NC, USA) was used to
that no clinical data were available in this study and conse- generate the D-optimal design, and knowledge and design
quently, CQAs were determined based on manufacturing spaces were drawn with MATLAB 2010a (The Mathworks,
feasibility and USP standards [11, 12]. An optimal formu- Natick, MA, USA). Calibration models used to calculate the
lation (excipient ratio and disintegrant level) was chosen blend endpoints were developed with the PLS_Toolbox v.
along with the best process parameters (compression force 5.8 (Eigenvector Research Inc., Wenatchee, WA, USA) and
target at the press and blend RMSNV weights). That optimal scripts written in-house.
set of parameters was subsequently used to test the robust-
ness of the design space.
Results
Design Space Robustness
Prior to tableting and the determination of quality attributes,
the RMSNV was used to stop blending. Blend prediction
To test the robustness of that optimal system with respect to
models had the following calibration and block cross-
raw material properties, three additional runs were created.
validation model errors: root mean square error of calibra-
The effects of APAP particle size and/or the ratio of lactose
tion/cross-validation: 2.71/2.93, 1.44/1.52, 2.64/2.68, 4.32/
anhydrous to lactose monohydrate on the CQAs and ulti-
4.36, and 1.36/1.45 % w/w for APAP, caffeine, MCC, lac-
mately on the design space were investigated. For each of
tose, and croscarmellose sodium, respectively.
the three runs, a 3-by-2 full factorial design was generated
where the impact of tablet press speed and compression
force were evaluated. Table 2 presents the additional DOE. The Knowledge and Design Spaces
Note that in the knowledge space DOE, the compression
speed was left constant (30 rpm). A larger particle size Table 3 presents quality attributes for each design point of
APAP (median particle size of 600 μm) was obtained by the knowledge space experiments. All design points passed
1 0.31 34 0.89 5 10
2 0.39 45 0.97 10 10
3 0.23 35 0.84 5 5
4 0.26 381 1.92 15 45
5 0.12 129 1.65 15 15
6 0.26 17 0.78 30 30
7 0.17 191 1.24 30 10
8 0.22 66 1.10 5 10
9 0.14 20 0.72 5 5
10 0.25 120 1.47 10 15
11 0.13 203 1.68 5 10
12 0.16 27 1.01 10 10
13 0.24 116 1.50 5 10
14 0.12 36 1.23 45 10
15 0.28 60 1.22 5 10
a
Time at which USP criterion 16 0.19 76 1.99 10 30
was reached
the friability and dissolution tests based on USP standards were set to span median CQAs values and consequently
[11, 12]. However, disintegration time and RTS demonstrat- specify tablets that are neither too soft nor too hard and
ed significant variability. These two parameters were signif- have a reasonable disintegration time. Note that these limits
icantly influenced by tablet compression force target and should be set based on desired drug release patterns, and
excipient ratio. The RMSNV weights and the level of dis- authors realize that in many situation, a tablet disintegrating
integrant used during processing were not demonstrated to as quickly as in 80 s might not be suitable. Figure 2 presents
have a statistically significant effect on the quality attributes the design space generated when acceptable CPP ranges
(α00.05). Consequently, RTS and disintegration time were (defined in Table 4) were overlapped onto the knowledge
used as critical quality attributes, compression force target spaces.
as critical process parameter, and excipient ratio as a critical
formulation factor (the latter will be included in the CPPs in Effect of Raw Material Properties on the Robustness of the
later discussion). Note that CQAs and CPPs were selected Design Space
solely based on manufacturing considerations and not clin-
ical relevance. This is important as the dissolution profile The robustness of the design space with respect to raw
would have been taken into account if a certain pharmaco- material variability was tested based on the design of exper-
kinetic/pharmacodynamic model or release profile was of iment shown in Table 2. The formulation chosen had an
interest. excipient ratio of 2:1 (41.3 % of MCC and 20.7 % of
Figure 1 illustrates knowledge spaces created from the lactose) and 2 % of croscarmellose sodium. Compression
information presented in Table 3 and represents the relation- force target at the press was set to 11 kp and RMSNV
ship existing between the critical process parameter (com- weights were 1-1-1 for APIs, excipients, and croscarmellose
pression force target and excipient ratio) and each of the two sodium, respectively. This formulation and set of parameters
CQAs:disintegration time (Fig. 1a) and RTS (Fig. 1b). Both were chosen because they were near of the center of the
knowledge spaces show areas of significant variations in design and constituted typical target points for routine man-
CQAs with a nonlinear combination and interaction of the ufacturing. Given these CPPs, the corresponding CQAs
critical formulation factor and critical process parameter. It were 1.53 MPa and 104 s for RTS and disintegration time,
is interesting to note that tablets made with an excipient ratio respectively.
of 4:1 (MCC/lactose) were the hardest tablets, but with very Design points 1B, 2B, and 3B correspond to the resulting
limited resistance to disintegration due to the swelling effect CQAs for the above defined CPPs (11 kp and 30 rpm).
that water has when disrupting particle–particle bonds when While the RTS of the tablets was slightly reduced for all
MCC is present in high concentration [13]. conditions, CQAs remained within the limits of the design
Table 4 summarizes CQAs acceptance criteria used in the space. However, the disintegration time of design points 1
present study. The limits for disintegration time and RTS and 2 (corresponding to change in APAP and change in
J Pharm Innov
Fig. 1 Two-dimension knowledge spaces. Each critical quality attribute is expressed as a function of the critical process parameters (tablet compression
force target and formulation excipient ratio)
lactose respectively) was outside the design space. Thus, the space developed from the knowledge space (for excipient
variability in raw materials that may be present when raw ratio of 2 only); a green square corresponds to a batch that
materials are inadequately characterized prior to entering the met the specification and a red square represents a batch out
manufacturing process can produce tablets that do not meet of specification. When the particle size of APAP was mod-
CQAs and, consequently, would not deliver the desired ified, only the use of a higher compression force (outside of
therapeutic effect. the design space) at low compression speed generated ap-
propriate tablets. Data for design point 1F are not available,
Adaptive Design Space to Adjust for Raw Material but based on RTS for 1E, it would not have met the spec-
Characteristics ifications. When the ratio of lactose anhydrous to monohy-
drate was modified, increasing the compression speed
The potential to adapt the CPPs to compensate for an unan- resulted in in-specification tablets. Finally, as seen before,
ticipated variance in raw materials and consequently adapt- the combination of variability in APAP and lactose raw
ing the design space was explored. As described in Tables 2 materials compensated each other and original settings were
and 5, tablet compression force target at the press and
compression speed were modified and the corresponding
CQAs determined. The use of higher compression forces
and higher compression speed generated harder tablets that
were also more resistant to disintegration. However, based
on the limits set on Table 4, some tablets were too hard and
thus would fall outside of the design space. Note that fria-
bility and dissolution time were not taken into account.
Figure 3 illustrates how each design point falls with
respect to tablet specifications (Table 4) and the design
Table 5 Critical quality attributes for each design point of the second- formulation and process parameter changes were fortuitous
ary design of experiment
and were not an intentional design characteristic.
Run # Sub- Disintegration Radial tensile Results
run # time (s) strength (MPa)
Discussion
1 A 21 1.06 Fail
1 B 63 1.45 Fail
When developing knowledge and design spaces, phar-
1 C 195 1.91 Pass maceutical companies cannot envision all the sources of
1 D 54 1.14 Fail variability that will be encountered throughout the drug
1 E 244 1.75 Fail product life cycle. Thus, despite the creation of an
1a F – – – appropriate design space, robustness issues will appear
2 A 41 1.14 Fail when unplanned variations emerge, such as instances
2 B 68 1.40 Fail where excipient suppliers modify their production con-
2 C 200 1.75 Fail ditions, a new supplier is employed, storage conditions
2 D 72 1.20 Fail change (relocation of a production line), etc. The same
2 E 131 1.45 Pass approach could have been taken to simulate the effect
2 F 299 1.83 Fail of other process perturbations. Process drifts, due to
3 A 25 1.05 Fail equipment aging and environmental changes, can have
3 B 98 1.40 Pass similar effects and adaptation.
3 C 437 2.26 Fail While only a limited number of critical quality attrib-
3 D 85 1.26 Pass utes were considered in this study, the general idea that
3 E 330 1.83 Fail one can achieve consistency in outcome by modifying
3a F – – – critical process parameters with respect to incoming raw
a
material variability is an important concept that should
Combination not compressible with considered tablet press
be addressed by both pharmaceutical companies and
regulatory bodies as process parameters considered not
sufficient to obtain good tablets. Increasing the compression critical in the original design space might become crit-
speed and reducing the compression force yielded the same ical to maintain product quality. The general consensus
results (outside of the design space). Note that the results of that tends to lock the design space should be revised
given that an appropriate process monitoring and control
system is employed to validate the new settings. One
can easily envision a system in which batches of vari-
ous excipients are characterized prior to entering the
manufacturing line. This characterization would trigger
the selection of CPPs that would allow the production
of consistent drug products, robust to lot-to-lot variabil-
ity, to manufacturer-to-manufacturer production proto-
cols, and changes in environmental conditions.
The generalization of such a dynamic system would
heavily rely on process analytical technologies. One can
envisage their use to not only characterize raw material
variability but also to track the state of a process (con-
tinuous or batch based) in an online fashion. For in-
stance, one could generalize this adaptive design space
approach to changing drying parameters based on parti-
cle size, moisture content of a granulation to reach
Fig. 3 Graphical representation of raw material challenges to the
consistent powder properties throughout the drug manu-
design space. Green squares represent a batch that met specifications
while red squares point to out-of-specification trials. White crosses are facturing life cycle no matter the numerous sources of
trials that could not be completed. Not all design points belonged to the variability that can impact the resulting material from
design space generated for a microcrystalline cellulose/lactose ratio of various unit operations. Such process control approaches
2:1. Trial 1 corresponds to large particle of APAP, trial 2 to 1:1 lactose
have already been developed [14, 15] and will play an
anhydrous/monohydrate, and trial 3to both modifications to the origi-
nal design. A to F represents variations in compression force and speed. increasing role in the development of safe and efficient
See Table 2 for more details drug products.
J Pharm Innov