Beruflich Dokumente
Kultur Dokumente
DOI 10.1007/s12247-014-9194-1
RESEARCH ARTICLE
Abstract
Purpose This study describes the validation of a near infrared
spectroscopic method for monitoring a continuous
manufacturing system. The achievement of this goal requires
determining the near infrared (NIR) methods accuracy, precision, obtaining an estimate of the sample volume analyzed,
and the frequency with which measurements should be
obtained.
Methods Five calibration blends were prepared spanning a
concentration range from 7 to 13 % w/w ibuprofen to prepare
a partial least squares (PLS) calibration model for a five
component formulation. NIR spectra were obtained after powders passed through feeders and a custom-made tumble mixer.
The calibration models precision and accuracy were determined with the prediction of three validation blends.
Results NIR concentration predictions obtained from spectra
collected during the validation runs showed relative standard
errors of predictions of 2.8 % at target concentration with
standard deviations NMT 0.2 % w/w. At certain time points,
blend samples coming out of the blender were collected for
UV analysis. Results from the NIR and UV analysis were
comparable with the largest difference being 0.36 % w/w
between the methods. A continuous blending process was
monitored for 3 min after steady-state conditions were
achieved. All individual predictions were within 3 standard
deviations of the average reference value. The standard deviation for the NIR concentration predictions was 0.49 % w/w.
Conclusion The use of the standard normal variate (SNV)
transform significantly reduced the effect of differences in
Y. M. Coln : R. J. Romaach (*)
Department of Chemistry, University of Puerto Rico at Mayagez,
PO Box 9000, Mayagez Campus, Mayagez 00682, Puerto Rico
e-mail: rodolfoj.romanach@upr.edu
M. A. Florian : D. Acevedo : R. Mndez
Department of Chemical Engineering, University of Puerto Rico at
Mayagez, Mayagez, Puerto Rico
Introduction
Current pharmaceutical manufacturing processes primarily
involve batch processes. These batch processes require that
considerable development efforts be dedicated to product
scale-up and transference of the process from development
to manufacturing facilities [14]. The use of a continuous
manufacturing process, as described in this work, will eliminate scale-up efforts as the same mixing unit used in development could be used in product manufacture.
The continuous manufacturing process described in this
paper uses a small custom-made mixer that offers significant
flexibility at a maximum of 80 kg/h at 70 rpm [5]. The
excipients and active pharmaceutical ingredient (API) are
added to the mixer through volumetric feeders. The size of a
lot will not be restricted by the manufacturing equipment
available at a pharmaceutical site. Lot size will be adjusted
by the amount of time employed during manufacture. Thus,
the system will be used for a certain time to fulfill a customers
order of 100,000 tablets and for a longer period for a 500,000
tablet order. The continuous manufacturing system is also a
very compact system, thereby reducing product hand-offs and
transfers, reducing the potential for product cross-
292
293
294
RMSEP
v
2
u Xm
u
t i1 byi yi
N
Fig. 2 Raw NIR spectra for the three major components in the pharmaceutical blend: ibuprofen, lactose, and microcrystalline cellulose. The
shaded area corresponds to the spectral region associated with the API
spectral bands
Blend Sampling
Samples were retrieved at the end of the conveyor belt
for each test sample set and continuous mixing run. The
flowing powder coming out at the end of the conveyor
belt was sampled every 5 s using a plastic cup to
sample the flowing stream obtaining approximately 3 g
of powder per time point [16]. Afterward, API quantification of each sample was performed off-line by the
in-house validated UV method.
Fig. 3 PCA score plot for the calibration sample set in the 1,0901,428nm wavelength region
295
Spectral range
(nm)
Data pretreatment
Percent of variation
R2Ycum
RMSEC
(%)
RMSEP
(%)
1
2
3
4
1,0901,605
1,0901,428
SNV+1st derv
1st derv
1st derv
SNV+1st derv
1
1
2
1
87.1
99.0
99.5
93.6
0.77
0.22
0.15
0.54
0.79
0.27
0.24
0.55
SNV+1st derv
1st derv
1st derv
SNV+1st derv
SNV+1st derv
2
1
2
1
2
97.9
95.0
98.9
94.6
98.5
0.31
0.48
0.22
0.50
0.26
0.37
0.59
0.40
0.59
0.42
5
6
7
8
9
1,0901,255
NIR near infrared, RMSEC root mean standard error for calibration, RMSEP root mean standard error of prediction, derv derivative, SNV standard normal
variate
0.79
0.83
0.90
0.97
1.10
1.19
1.29
SD (n=6)
SD (n=6)
8.90
9.04
9.24
9.51
9.89
10.20
10.56
0.04
0.05
0.09
0.08
0.09
0.12
0.04
8.29
8.27
8.41
8.48
8.40
8.38
8.48
0.16
0.17
0.07
0.08
0.16
0.16
0.14
296
UV
8.89
9.61
11.17
0.24 2.81
0.20 2.82
0.21 3.93
0.02
0.02
0.02
NIR near infrared, API active pharmaceutical ingredient, RSEP relative standard error of prediction
8.78
9.81
10.81
CI (95 %)
0.47 0.27
0.43 0.25
0.49 0.28
297
Fig. 5 NIR predictions obtained from the monitoring of three validation blends. UV measurements of API content for each run are also illustrated
Fig. 6 NIR predictions obtained from real-time monitoring of a continuous blending process. UV measurements of API content are also
illustrated
298
UV
SD
CI (95 %)
SD
CI (95 %)
10.5
0.49
0.04
9.3
0.71
0.23
Fig. 7 Variographic analysis of the validation blends and the real-time monitoring of the continuous blending process
299
N
2
1 X
aN j aN
2N j N1
Fig. 8 FFT analysis of the validation blends and the real-time monitoring of the continuous blending process
300
process could be taken every 1.4 s, more than half the observed frequency of 3.1 s. In future experiments, the sampling
rate will be set to the optimized parameters found. Thus, the
variogram is a useful tool for determining how frequently to
sample.
FFT analysis was performed to study the variability observed in the continuous manufacturing results. Figure 8
shows the frequencies associated to the measurements of
NIR API prediction concentration obtained through time. In
this continuous manufacturing setup, the screw feeders [24]
and the mixer influence in the observed cyclic behavior. Two
dominant frequencies at approximately 0.98 and 1.32 Hz were
observed in the validation runs. These frequencies could be
associated to the mixer and screw feeder speed cyclic behavior, respectively. These units were operated under the same
conditions in all validation runs. Three dominating frequencies were observed in the continuous blending run. One of
dominating frequencies (0.98 Hz) was also observed in the
validation runs. This observation is not unexpected since the
mixer conditions were the same as for those runs. The other
two dominating peaks (0.62 and 0.96 Hz) were unique for the
continuous run. This could probably be associated to the use
of two feeders in the continuous blending process. Moreover,
both peaks are at lower frequencies compared to the 1.32 Hz
associated with the screw feeder speed in the validation runs.
This is in agreement with the fact that during the continuous
blending process, the screw feeder speed in feeder #2 was set
at a lower RPM to maintain the total feed rate since an
additional feeder (feeder #1) was used. Therefore, the two
0.62 and 0.96 Hz peaks can be associated to the two different
screw feeder speeds used in the continuous blending run.
Thus, the use of FTT complemented the variogram results in
identifying the root causes of the cyclic behavior observed.
Conclusions
The sources of error in the NIR method used to monitor a
continuous mixing process have been investigated. Repeatability studies measured the instrumental error that ranged
from 0.07 to 0.17 % (w/w). The precision and accuracy in
determining well-mixed blends were also determined by monitoring previously mixed blends as they passed through the
continuous manufacturing setup. The standard deviation of
the well-mixed blends was about 2.5 times greater than the
variation observed in the repeatability study. This increase in
standard deviation is due to variations in the composition of
the powder mixture and powder flow. Powder flow differences affect the level of powder over the conveyor belt and
vary the distance between the powder and the NIR instrument.
NIR predictions were affected by the probe to powder distance
in the initial calibration models. However, the use of the SNV
References
1. Bell TA. Challenges in the scale-up of particulate processesan
industrial perspective. Powder Technol. 2005;150(2):6071.
2. Leuenberger H. New trends in the production of pharmaceutical
granules: the classical batch concept and the problem of scale-up.
Eur J Pharm Biopharm. 2001;52(3):27988.
3. Mort PR. Scale-up of binder agglomeration processes. Powder
Technol. 2005;150(2):86103.
4. Rodriguez-Justo O, Morales M. Analysis of process parameters on
the characteristics of liposomes prepared by ethanol injection with a
view to process scale-up: effect of temperature and batch volume.
Chem Eng Res Des. 2011;89(6):78592.
5. Florian M, Velzquez C, Mndez R. New continuous tumble mixer
characterization. Powder Technol. 2014;256:18895.
6. Jamrogiewicz M et al. Determination of API content in a pilot-scale
blending by near-infrared spectroscopy as a first step method to
process line implementation. Acta Pol Pharm. 2013;70(3):41929.
7. Shi Z et al. Process characterization of powder blending by nearinfrared spectroscopy: blend end-points and beyond. J Pharm
Biomed Anal. 2008;47(45):73845.
8. Liew CV, Karande AD, Heng PW. In-line quantification of drug and
excipients in cohesive powder blends by near infrared spectroscopy.
Int J Pharm. 2010;386(12):13848.
9. Sekulic SS et al. On-line monitoring of powder blend homogeneity
by near-infrared spectroscopy. Anal Chem. 1996;68(3):50913.
10. Momose W et al. Process analytical technology applied for end-point
detection of pharmaceutical blending by combining two calibrationfree methods: simultaneously monitoring specific near-infrared peak
intensity and moving block standard deviation. Powder Technol.
2011;210(2):12231.
11. Vanarase AU et al. Real-time monitoring of drug concentration in a
continuous powder mixing process using NIR spectroscopy. Chem
Eng Sci. 2010;65(21):572833.
301
18. Sulub Y et al. Real-time on-line blend uniformity monitoring using
near-infrared reflectance spectrometry: a noninvasive off-line calibration approach. J Pharm Biomed Anal. 2009;49(1):4854.
19. Koller DM et al. Continuous quantitative monitoring of powder
mixing dynamics by near-infrared spectroscopy. Powder Technol.
2011;205(13):8796.
20. El-Hagrasy AS et al. Near-infrared spectroscopy and imaging for the
monitoring of powder blend homogeneity. J Pharm Sci. 2001;90(9):
1298307.
21. Berthiaux H, Marikh K, Gatumel C. Continuous mixing of powder
mixtures with pharmaceutical process constraints. Chem Eng Process
Process Intensif. 2008;47(12):231522.
22. Bakeev KA. Process analytical technology: spectroscopic tools and
implementation strategies for the chemical and pharmaceutical industries. Wiley; 2005.
23. Petersen L, Esbensen KH. Representative process sampling for reliable data analysisa tutorial. J Chemom. 2005;19(1112):62547.
24. Engisch WE, Muzzio FJ. Method for characterization of loss-inweight feeder equipment. Powder Technol. 2012;228:395403.