European Journal of Pharmaceutical Sciences 75 (2015) 160–168

Contents lists available at ScienceDirect

European Journal of Pharmaceutical Sciences journal homepage: www.elsevier.com/locate/ejps

A study on the applicability of in-line measurements in the monitoring of the pellet coating process Grega Hudovornik a, Klemen Korasa a, Franc Vrecˇer a,b,⇑ a b

KRKA, d.d., Novo mesto, Šmarješka cesta 6, 8000 Novo mesto, Slovenia University of Ljubljana, Faculty of Pharmacy, Aškercˇeva 7, 1000 Ljubljana, Slovenia

a r t i c l e

i n f o

Article history: Received 21 November 2014 Received in revised form 4 April 2015 Accepted 6 April 2015 Available online 28 April 2015 Keywords: In-line near infrared spectroscopy Spatial filtering technique Process analytical technology Controlled release coated pellets Real time monitoring

a b s t r a c t Special populations including paediatric and elderly patients often need advanced approaches in treatment, such as one-a-day dosing, which is achieved with modified release formulations or alternative routes of applications such as nasogastric route. Pellets are a dosage form that is frequently used in such formulations. The aim of the present work was to study the applicability of two in-line techniques, namely, Near Infrared Spectroscopy (NIR) and Spatial Filtering Technique (SFT) in the pellet coating process. The first objective of our work was to develop a prediction model for moisture content determination with the in-line NIR and to test its robustness in terms of sensitivity to changes in composition of the pellets and performance in wide range of moisture content. Secondly, the in-line SFT measurement was correlated with different off-line particle size methods. The third objective was to evaluate the ability of both in-line techniques for the detection of undesired deviations during the process, such as pellet attrition and agglomeration. Finally, the ability to predict coating thickness with the in-line NIR probe was evaluated. Results suggested that NIR prediction model for moisture content was less robust outside the calibration range and was also sensitive to changes in composition of the film coating. Nevertheless, satisfactory prediction was achieved in the case when coating composition was partially altered and adequate calibration range was used. The SFT probe results were in good correlation with off-line particle size measurement methods and proved to be an effective tool for coating thickness determination during the coating, however, the probe failed to accurately show the actual amount of the agglomerates formed during the process. In experiment when pellet attrition was initiated, both probes successfully detected abrasion of the pellet surface in real time. Furthermore, a predictive NIR model for coating thickness was made and showed a good potential to measure coating thickness in-line, suggesting that the NIR probe can be used as a single tool to monitor water content, coating thickness, and deviations in the coating process. Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction Multi particulate systems offer many advantages over single unit matrix systems or coated tablets in terms of safety and efficacy, such as lower occurrence of dose dumping and lower inter-subject variability when modified drug release is needed (Bechgaard and Nielsen, 1978). Furthermore, alternative routes of administration such as nasogastric tubes can be used with such formulations (Toedter Williams, 2008). This is especially important in case of special populations such as elderly or paediatric patients. Coated pellets are the most common type of multi particulate systems used in modified release formulations. The essential step of pellet production is coating of pellets with functional coating. ⇑ Corresponding author at: KRKA, d.d., Novo mesto, Šmarješka cesta 6, 8000 Novo mesto, Slovenia. Tel.: +3867 331 3608; fax: +3867 331 3751. E-mail address: [email protected] (F. Vrecˇer). http://dx.doi.org/10.1016/j.ejps.2015.04.007 0928-0987/Ó 2015 Elsevier B.V. All rights reserved.

This is usually performed by bottom spray fluidized bed coating technology. Strict control of the coating process is required since the performance of modified release coating is directly correlated with in-vivo performance of the formulation. In the recent years a number of new techniques enabling real-time monitoring of the process emerged as a consequence of new approach in development and production of pharmaceutical products called Quality by Design (QbD). QbD is described in Quality guidelines (Q8, Q9, and Q10) of The International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH, 2009, 2005, 2008). Near Infrared Spectroscopy (NIR) is one of techniques that is most commonly researched in the field of Process Analytical Technology (PAT) (FDA, 2004), which is an essential part of QbD. Numerous publications are published covering the use of in-line

161

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

NIR in almost all production steps of solid dosage forms these being from blending process (Berntsson et al., 2002), granulation process (Rantanen et al., 2000), tableting process (Karande et al., 2010), and tablet coating process (Perez-Ramos et al., 2005). Several publications also investigate important pellet characteristics in the pellet coating process, such as moisture content (Mantanus et al., 2009), coating thickness (Lee et al., 2011; Avalle et al., 2014), and drug release profile (Pomerantsev et al., 2011). However, there still remains a need to further investigate the ability of in-line NIR to predict multiple pellet characteristics and to evaluate the robustness of NIR models. The other frequently studied area of PAT are in-line particle size measurement techniques. These include optical techniques which measure chord length of particles, such as Focused Beam Reflectance Measurement (FBRM) and Spatial Filtering Technique (SFT), and different image analysis approaches (Silva et al., 2013; Kadunc et al., 2014). Use of passive acoustic emission technique to measure particle size is also described in the literature (Poutiainen et al., 2012). In contrast to the above mentioned techniques which measure particle size, Terahertz Pulsed Imaging (TPI) can be used to directly determine coating thickness as well as coating surface roughness. Most publications describe off-line use of TPI to determine coating thickness on pellets (Haaser et al., 2012) or tablets (Ho et al., 2009, 2008, 2007), although one publication also describes an in-line use of TPI to measure coating thickness during tablet film coating process (May et al., 2011). Most publications evaluate the use of FBRM and SFT in fluid bed granulation process in comparison to traditional off-line techniques. Good correlation can be obtained although absolute values of particle size differ due to different measuring principles (Närvänen et al., 2009; Hu et al., 2008). Recently an article comparing FBRM and SFT performance has been published. Authors concluded that both techniques show similar trends in particle size growth (Kukec et al., 2012). In another recent publication authors study the use of SFT in pellet coating. A correlation with Dynamic image analysis is made showing good correlation between techniques. The influence of process parameters and key SFT measurement setting (Particle buffer) on measurement results is discussed. Authors concluded that Particle buffer setting has an important impact on measured particle size values (Folttmann et al., 2014). Although in-line particle size measurements could be very useful in pellet coating process for evaluating coating thickness and also undesirable processes, such as agglomeration and pellet attrition, there is still very little research in this area. The objective of the present work was to study the applicability of in-line measurements in the process of pellet film coating. In-line NIR data were correlated with different off-line methods

for moisture content determination, models for moisture content prediction were set, and later moisture content in pellets with different film coating compositions was predicted. Furthermore, NIR data were correlated with in-line particle size measurements in order to predict coating thickness using a PLS model. Robustness of NIR models was investigated by changing the composition of the film coating and by simulating process deviations that could occur in the production environment. The second objective of the present work was to correlate SFT in-line particle size measurements with different off-line particle size determination methods and to use in-line results for evaluation of coating thickness, pellet agglomeration, and pellet attrition. The study of SFT and NIR in-line tools should contribute to a better understanding of the PAT applicability in the monitoring, controlling, and designing of modified release pellet coating processes. Firstly, such tools enable efficient coating thickness control and hence better active ingredient dissolution repeatability. In addition, better stability and consistent final product quality is achieved through real time moisture content management. Paediatric and geriatric patients are especially vulnerable to different health issues, thus, manufacturing of medicine at the highest quality level, with the help of PAT tools, can improve the quality of their lives. 2. Materials and methods 2.1. Manufacturing of the film coated pellets In our study six pilot scale batches of film coated pellets were produced and phase of controlled release film coating application onto the drug layer pellets was precisely examined. Drug layered pellets comprised hydroxypropyl cellulose (Klucel™ EF, Ashland Inc., USA), diclofenac sodium (Krka, d.d., Slovenia), and talc (Imerys Talc, Italy) applied on the neutral sugar pellets (Hanns G. Werner, GmbH + Co., Germany). Drug layered pellets were further film coated. Controlled release coating compositions of batches No. 1, 2, 5, and 6 were identical, while compositions of the third and fourth batch were modified in comparison to the other batches. Quantitative composition of the film coating dispersion was changed in the third batch and polymer type was altered in the fourth batch, i.e. qualitative variation. Compositions of different experiments are shown in Table 1. All of the coating dispersions comÒ prised polymethacrylic coating polymer (Eudragit RL, RS or L 30D, Evonik Industries, Germany), triethyl citrate (Vertellus Performance Materials, USA), and talc (Imerys Talc, Italy). Purified water was used as a dispersing medium. Pellets were coated in the Aeromatic-Fielder™ MP 3/2/4 fluid bed processor (GEA Pharma Systems, Switzerland), on which the bottom spray

Table 1 Compositions of the individual batches. Component

Function

Batches No. 1, 2, 5, 6

Batch No. 3

Batch No. 4

Eudragit RS 30 D

Controlled release polymer Controlled release polymer

Eudragit L 30 D

Delayed release polymer

2.37 kg (16.3%) 0.59 kg (4.1%) –



Eudragit RL 30 D

1.0 kg (4.8%) 1.0 kg (4.8%) –

Talc

Anti-tacking agent

Triethyl citrate

Plasticizer

API layered pellets

Active ingredient carrier

Purified water

Dispersing medium

0.30 kg (1.4%) 0.078 kg (0.38%) 5.0 kg (24.1%) 13.39 kg (64.5%)

1.15 kg (7.9%) 0.17 kg (1.2%) 4.0 kg (27.5%) 6.26 kg (43.1%)

– 2.0 kg (9.6%) 0.30 kg (1.4%) 0.078 kg (0.38%) 5.0 kg (24.1%) 13.39 kg (64.5%)

162

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

Table 2 Process parameters of the film coating spraying phase. Batch No.

Spray rate (g/ min)

Product temperature (°C)

Inlet air temperature (°C)

Inlet air flow (m3/h)

Atomising air pressure (bar)

1 2 3 4 5 6

50–125 30–100 40–80 50–100 30–85 30–80

25–30 25–30 25–30 25–30 25–30 25–30

70–90 50–90 60–80 55–80 50–75 50–80

170–270 170–250 230–240 290–320 300–310 230–250

2.5 2.5 2.5 2.5 2.5 2.5

Precision-Coater™ unit was mounted. Coating dispersion was applied with high spray rates during the manufacturing in order to achieve strong wetting of the pellets, rapid particle size increase, and emergence of agglomerates. More detailed description of the applied process parameters range is shown in Table 2. During the manufacturing of the fifth batch the spraying was started with the delay of approximately 20 min, with the goal to achieve attrition of the drug layer and to evaluate if this phenomenon can be detected with the in-line PAT tools. 2.2. Moisture content determination with the in-line NIR probe All of the batches were monitored with the NIR diffuse reflectance probe (Lighthouse probe™, GEA Pharma Systems, Belgium) equipped with the diode array spectrometer (J&M, Germany). The probe collected spectral data in the range from 1100 nm to 2200 nm with 256 pixel resolution (4 nm). The sinTQ™ Lite version 3.4 (Optimal Industrial Automation, UK) data-management system was used for managing the NIR probe and for spectral data logging. The spectra were collected every 10 s and automatic window wash was performed every 600 s to prevent fouling of the measuring windows. The probe was installed 9 cm above the bottom spray chamber’s distribution plate. NIR spectral data were correlated with water activity (aw), loss on drying (LOD), and Karl Fischer (KF) methods. aw measurements were Ò carried out using the Rotronic PA20 Hygrometer (Rotronic, Switzerland) device. Measurements were performed on 5 g of pellets at 25 °C until water activity equilibrium was reached. LOD was determined with the HR73 Halogen Moisture Analyzer (Mettler Toledo, Switzerland) at 85 °C for 20 min. Karl Fischer titrations were made in methanol medium with the volumetric Karl Fischer titrator (V30, Mettler Toledo, Switzerland). All of the off-line moisture content measurements were carried out in two replicates. Spectral data pre-treatment, data analysis, regression, and preÒ diction calculations were carried out using The Unscrambler X 10.2 and 10.3 (CAMO Software, Norway) multivariate data analysis software. All of the spectral data were pre-treated with standard normal variate (SNV) transformation, which is a well-established pre-treatment correction for quantification of moisture content with the NIR (Lam Yip et al., 2012; Kauppinen et al., 2014). 6 individual SNV pre-treated spectra were then averaged to reduce the effect of the spectral deviation and examined with the principal component analysis (PCA) to verify the suitability of the in-line data. After the PCA, regression with the off-line methods was calculated by the partial least square (PLS) regression using the entire wavelength range, i.e. 1100–2200 nm, and the optimum number of Ò factors as suggested by The Unscrambler . The effect of first and nd second derivative (Savitzky-Golay, 2 polynomial, 15 smoothing points) transformations on the in-line–off-line correlation was also investigated. In this case, three factors were used for all PLS models, thus the observed model quality was a result of the applied pre-treatment and not due to a different number of factors. It was our goal to calibrate and validate the PLS model for moisture content determination with the first two batches and to use this model for moisture content prediction in the pellets with

changed coating dispersion composition, i.e. the third and fourth batch. Samples for off-line moisture content determination were collected every 5–15 min. 2.3. In-line particle size monitoring with the SFT probe In-line particle size measurements were performed with the SFT probe (Parsum IPP-70, Parsum GmbH, Germany). The probe was installed at a height of 9 cm above the distribution plate and 1.5 cm from the sidewall of the process chamber. Data recording and real-time data display was done with the In-line Parsum Probe Software (IPPS) v7.13. Internal airflow was set to 20 L/min and external air flow to 3 L/min. Particle buffer size for the particle size parameters calculations was set to 100.000 particles. SFT probe measures chord length (CL) of particles passing the measuring laser beam and average CL in specific time point is calculated from all of the values in the pre-set particle size buffer. Individual in-line measurements were logged every 30 s. The SFT measurements were compared with the off-line sieve analysis (Air Jet Sieve 200 LS-N, Hosokawa Alpine, Germany) and static image analysis (Morphologi G3, Malvern Instruments Ltd., UK) measurements. Sieve analysis (SA) was made using 250, 500, 710, 1000, and 1250 lm mesh size sieves. The same sieves were selected in the IPPS before the coating had started, which enabled comparison of the SFT’s particle size distribution with the off-line SA results. Static image analysis (SIA) was carried out on 200– 500 pellets with the help of integral Sample Dispersion Unit. SIA calculates diameter of a circle with the same area as 2D image of the measured particle (circle equivalent). We compared median of volume distribution (Dv50) of SFT and SIA measurements. Comparison of this parameter enabled the evaluation of the probe’s suitability for the determination of average pellet size and average film coating thickness in real time. Captured SIA images also allow calculation of different particle shape parameters, which enabled us the evaluation of pellet circularity. Furthermore, comparison between the in-line and at-line SFT measurements was done. At-line SFT measurements were carried out with the use of special measuring adapter, which enables particle size determination outside the process chamber. At-line measurements were performed on 20 g of sample using cumulative measurement setting. Batches No. 1, 2, 3, and 5 were used to compare in-line results with the results of off-line measuring techniques, while the fourth batch was used to evaluate the suitability of the SFT probe for detection of the agglomeration phenomena. In case of agglomeration evaluation, the ninetieth percentile of volume distribution (Dv90) measured by the SFT probe was compared with the actual number of agglomerates in 50 g of sample. When determining the actual quantity of agglomerates, agglomerated pellets were firstly isolated from the rest of the sample with the help of 1250 lm mesh size sieve and later the agglomerates were manually counted. Samples for off-line particle size evaluation were taken every 15– 30 min of the film coating process. 2.4. Suitability of the NIR probe to evaluate coating thickness changes We also evaluated the suitability of the NIR probe to evaluate coating thickness changes of the drug layer, i.e. decreasing of the coating thickness due to the attrition process, and of the film coating, i.e. increasing of the film coating thickness during the normal course of the coating process. For this purpose, the NIR spectral data were compared with the in-line SFT measurements. Before the comparison, average was calculated for a set of six SNV pre-treated spectra and for two SFT measurements, resulting in one measurement per minute for individual in-line technique. Afterwards, correlation was determined with the PLS regression, where the NIR spectra were set as predictive variables and change

163

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

4 PC-1 3

Score (PC-1, 95 %)

2 1 0 -1 -2 -3 0

10

23

33

45

55

68

78

90

100

Time (min) 0.2 PC-1 0.15 0.1

Loadings

0.05 0 -0.05 -0.1 -0.15 1094 1143 1191 1240 1289 1337 1385 1433 1481 1529 1577 1625 1672 1720 1767 1814 1861 1908 1955 2001 2048 2094

of thickness (DT) as a response data set. DT was calculated as (Dv50s Dv50e)/2, where Dv50s presents Dv50 at the time when coating thickness change started and Dv50e presents Dv50 at the time after the coating thickness change ended. The optimum number Ò of factors, as suggested by The Unscrambler , was applied for the PLS regression calculation. Identification of characteristic peaks of a specific substance helped us understand in-line NIR data changes and was carried out at-line by measuring spectra of individual components using the Antaris™ Target Analyzer (Thermo Scientific, USA). Definition of the wavelength range was done prior to the calibration of the PLS model for film coating thickness determination, since water has wide and strong absorption bands with maxima at 1450 nm and 1940 nm (Blanco et al., 2000), which are not appropriate for coating thickness model calibration. After the wavelength range was identified, it was used as a predictive data set for the PLS regression calculation. In-line measurements in the period before the pellet size increase was detected were not used for the PLS regression calculation, since minor, but detectable abrasion of the drug layer was present in the first few minutes of the coating process. Such measurements do not correlate with film coating NIR signal and would worsen predictive potency of the model. First two batches were used to calibrate the PLS model for coating thickness determination, which was later used to predict polymer film thickness of the sixth batch.

Wavelength (nm)

3. Results and discussion 3.1. In-line NIR data correlation with off-line moisture determination techniques

10

0.9

9

0.8

8

0.7

7

0.6

6

0.5

5 0.4

4

0.3

3

LOD

2

KF

1

Water acvity

Moisture content (%)

Results of aw, KF, and LOD during the first batch are shown in Fig. 1. It is evident that measured values were increasing with time and that aw responded differently to rising moisture content than LOD and KF. Such results are not surprising since aw is not in linear dependence with moisture content (Reid, 2006). PCA analysis of the in-line NIR data showed strong and wide positive bands at 1450 nm and 1940 nm on the PC-1 loadings plot and changing of PC-1 consistent with the off-line results (Fig. 2), which confirmed that the NIR data contained moisture content information. Thus, PLS regression between the NIR data and off-line measurements was calculated. Table 3 shows that all methods had R2 of calibration (R2c ) higher than 0.94 and R2 of cross-validation (R2v ) higher than 0.91. Considering that the models were calibrated in a wide range of measurements and calibrations were done with only 10 off-line samples, such R2 values confirmed good correlation between the off-line and in-line measurements. It is noteworthy that R2 values were high for all calculated regressions although it

0.2 0.1

aw

0 0

10

20

30

40 50 60 70 Coang me (min)

80

90

100

0 110

Fig. 1. Off-line moisture content measurements during the coating of the first batch.

Fig. 2. First principal component (PC-1) during the film coating process of the first batch (scores and loadings plots).

Table 3 PLS model calibration parameters for different off-line reference methods (R2 of calibration (R2c ), R2 of cross-validation (R2v ), root mean square error of calibration (RMSEC)). Reference method

No. of factors

R2c

R2v

Slope

RMSEC

Aw KF LOD

3 2 2

0.9815 0.9469 0.9550

0.9297 0.9148 0.9151

0.9815 0.9469 0.9550

0.0174 0.4479 0.3832

is clear (Fig. 1) that there were significant differences between the individual off-line methods.

3.1.1. Prediction of moisture content in the pellets with changed coating composition It was shown above (Section 3.1) that all off-line moisture determination methods had good correlation with the NIR in-line probe. However, we decided to calibrate and validate the PLS model for later moisture content predictions only with KF and LOD since these methods are most widely used and also listed as moisture determination methods in the U.S. Pharmacopeia (The United States Pharmacopeia 37, 2013) and European Pharmacopoeia (European Pharmacopoeia, 2013). Additional batch of film coated pellets was produced and used together with the first batch results for the PLS model calibration. While calculating the regression, the effect of 1st and 2nd derivative on final characteristics of the model was evaluated, since such spectral pre-treatment is a very common transformation step in moisture content determination by NIR (Zheng et al., 2008; Markovic´ et al., 2014). Parameters of the PLS models calculated with different pre-treatment transformations are presented in Table 4. It is evident that the use of derivatives had positive effects on the PLS model properties. Derivative application increased R2c and R2v , moved slope closer to 1, and decreased root mean square error of

164

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

Table 4 The effect of different spectral pre-treatments on PLS model characteristics. Method

Pre-treatment

No. of factors

R2c

R2v

Slope

RMSEC

KF

SNV SNV + 1st derivative SNV + 2nd derivative

3 3

0.9515 0.9598

0.9214 0.9335

0.9515 0.9598

0.4660 0.4245

3

0.9685

0.9396

0.9642

0.3757

SNV SNV + 1st derivative SNV + 2nd derivative

3 3

0.9654 0.9722

0.9348 0.9563

0.9654 0.9722

0.3519 0.3152

3

0.9804

0.9702

0.9804

0.2652

LOD

calibration (RMSEC). It is also obvious that higher derivative order additionally improved the correlation, therefore, models calibrated with 2nd derivative pre-treated spectra were chosen. The reason for improved characteristics of the model transformed with 2nd derivative is probably the increased relative intensity of water bands. Before using the models for moisture content prediction, one outlying off-line measurement in every model was identified and removed to improve the predictive potency. Final R2c ’s were 0.9834 and 0.9931, R2v’s were 0.9578 and 0.9878, and RMSECs were 0.2795 and 0.1610 for KF and LOD, respectively. The LOD model was calibrated with three factors and KF model with two factors. Hereinabove PLS models were used for KF and LOD prediction during the coating of the third batch. Average absolute errors of prediction were 0.469% and 0.319% and average relative errors of prediction were 19.5% and 14.8% for KF and LOD, respectively. Most of the predicted values were consistent with the measured values, however some very high deviations were noticed (up to 1% of moisture content). Changed quantitative composition could be one reason for the outlying predictions. Different ratio of components in the coating dispersion effects spectrum characteristics and consequently model calculated with the results of the first two batches was not appropriate for the prediction of the third batch’s results. Secondly, the percentage of dry components in dispersion of the third batch was higher than in the first two. As a result KF and LOD results were lower during the third batch and some of the measured values were outside the calibration range, i.e. from 2.5% to 10%. All in all, predicted values were moderately consistent with the measured ones, which mean that such approach is potentially useful for application in the pellet coating processes. However, further studies are needed to identify what were the reasons for outlying predictions and what difference in the ratio of dispersed components is still allowed for accurate moisture content prediction. All off-line measurements of the fourth batch were in the range from 2% to 5%, so the measured values were lower than values used for the PLS model calibration. In order to achieve better prediction accuracy and reduce the effect of very high off-line measurements, additional models were calibrated in the range of fourth batch’s measured values. Each model was calibrated with 11 off-line measurements resulting in R2c values of 0.9605 and 0.9219 and RMSEC values of 0.1248 and 0.1628 for KF and LOD, respectively. Both PLS models were calibrated using one factor. These two models were used for the prediction of off-line measurements of the fourth batch (Fig. 3) and very good correlation between the predicted and measured values was observed. Average absolute errors of prediction were 0.21% and 0.18% and average relative errors of prediction were 6.2% and 5.8% for KF and LOD, respectively. Not only that average errors were low, also no apparent outliers were identified among the predicted values. Results of the fourth batch suggest that moisture content prediction in the pellets with moderately

Fig. 3. Comparison of predicted and measured moisture content during the coating of the fourth batch (LOD above and KF below).

changed composition is possible. Such models would be very useful for process monitoring, controlling, and would enable its user better process understanding. 3.2. In-line particle size measurement with the SFT probe Firstly, we analysed how the SFT probe responds to changes of the process parameters and if measured values are consistent with the process flow. Changing of the tenth (Dv10), fiftieth (Dv50), and ninetieth (Dv90) percentile of volume distribution of the first batch was compared with the changing of coating dispersion’s spray rate. The particle size growth rate was calculated by dividing the particle size (diameter) change (DD) with the spraying time at a specific spray rate (Table 5). Results showed that Dv10 and Dv50 were responding to particle size change and the growth rate was increasing by enhancing the coating dispersion’s spray rate. On the other hand, Dv90 was not consistent with the spray rate and did not give any beneficial data for process evaluation. The reason for the Dv90 inconsistency was formation of slight aggregates in the dynamic fluid bed environment, the phenomena explained more precisely in Section 3.2.1. Such results suggested that the probe

Table 5 Pellet size growth rates (DD) calculated from the SFT measurements. Spray rate (g/ min)

Time (min)

DDv10 (lm)

DDv50 (lm)

50 80 100 125

25 27 30 27

0.25 8.7 14.5 16.5

0.2 9.4 19.3 18.3

DDv90 (lm)

56.4 19.3 4.2 102.2

Growth rate Dv10 (lm/ min)

Growth rate Dv50 (lm/ min)

0.01 0.32 0.48 0.61

0.008 0.35 0.64 0.68

Growth rate Dv90 (lm/ min) 1.4 0.7 0.1 3.8

165

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168 Table 6 Comparison of SFT and SA particle size distribution of the third batch.

SFT in-line 20 40 60 80 100 120

1250 lm

4.0 3.9 3.5 3.1 2.7 2.4

48.5 48.4 42.7 36.7 30.3 25.3

23.1 24.5 29.3 35.7 43.5 48.7

23.8 22.5 23.9 23.9 23.0 23.2

1100 Batch 5

Dv50 [μm]

Sampling time (min)

1200

1000 900 Batch 1

Batch 2

Batch 3

800 700

SFT SIA

600 0.5 0.5 0.5 0.5 1.0 0.5

0.5 0.5 0.5 1.0 0.5 0.5

28.0 18.5 6.5 2.5 0.5 0.5

70.5 80.0 92.0 95.0 97.5 98.0

0 0 0 0.5 0 0

Sampling me (min) Fig. 5. Comparison of Dv50 measured by the SFT and SIA.

was a responsive and useful tool for monitoring the particle size in real time, but it lacked accuracy for measurement of upper particle size fraction. In the next phase, correlation of the probe with off-line particle size determination techniques was evaluated to clarify if in-line measurements were an indicator of the actual particle size. Comparison of the SFT particle size distribution with the SA particle size distribution showed that SFT assessed smaller particle size and wider particle size distribution (Table 6). However, both methods responded to the increasing particle size during the process, which was observed by increasing fraction of particles in the range from 1000 to 1250 lm. The particle size increase was less pronounced with the SA because of narrower particle size distribution. Afterwards, fraction of particles bigger than 1000 lm (% > 1000 lm) was compared (Fig. 4). It was evident that particle size determined by the SA was significantly higher, but response to particle size changes of both methods was comparable indicating good correlation of real time measurements with the off-line results. SIA measures size of each particle individually in static environment and is capable to determine particle size from 0.5 lm to several mm. Therefore, this method is regularly used in pharmaceutical industry for pellet size measurement and indirect coating thickness determination. Comparison of the SFT results with the SIA measurements allowed us to evaluate what is the suitability of in-line measurements for average particle size and pellet coating thickness determination. Changing of Dv50 values during the process was investigated. All of the samples collected during the coating process of four batches (batches No. 1, 2, 3,

and 5) were compared (Fig. 5). It was clear that particle size values determined by the off-line method (SIA) were again higher than in-line particle size measurements. However, response to particle size changes was very similar with both measuring techniques. Linear regression with Dv50 values of both methods was calculated to quantify the correlation between the SIA and SFT results. R2 and slope of the regression line were 0.9933 and 1.044, respectively, indicating very high correlation. Such results suggested that although SFT measures particle size in the dynamic fluid bed environment, it was capable to determine the pellet size with high precision and gave results correlating with the SIA. The results confirm the findings of Folttmann et al. in a recently published article comparing SFT to Dynamic image analysis (Folttmann et al., 2014). Results presented hereinabove showed that the SFT measurements correlated with both off-line techniques (SIA, SA), but particle size measured with the in-line approach was in both cases smaller than off-line measurements. The reason for such results is different approach of characterising the particle size. SA measures particle size by passing the particle through the mesh and SIA calculates circle diameter from the area of particle’s 2-D projection. In case that a pellet has a sphere-like shape, both off-line approaches measure dimension close to the actual pellet diameter. Circularity of drug layered pellets and polymer coated pellets was close to 1 (Fig. 6) suggesting that used pellets had spherical shape and off-line measurements were a good approximation of the actual particle size. On the other hand, SFT characterises particle by measuring CL perpendicular to the direction of the measuring beam (Fischer et al., 2011). Consequently, SFT measures actual pellet diameter only in case when chord is passing through the centre

100 90

% > 1000 μm

80 70 60

Batch 1

Batch 2

Batch 3

50 SFT

40

SA

30

Sampling me (min) Fig. 4. Comparison of particle size fraction bigger than 1000 lm measured by the SFT and SA.

Fig. 6. Circularity values of drug layered pellets with the highest circularity (A), drug layered pellets with the lowest circularity (B), polymer coated pellets (second batch) with the highest circularity (C), and polymer coated pellets (second batch) with the lowest circularity (D).

166

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

Table 7 Comparison of the second batch in-line and at-line SFT particle size distribution. 1250 lm

Dv50

SFT in-line 20 60 100 140

0.8 0.7 0.6 0.5

4.8 4.2 3.7 3.0

53.6 47.9 41.0 29.8

27.5 30.1 36.7 49.3

13.3 17.1 18.0 17.4

969.8 988.2 1011.9 1055.3

SFT at-line 20 60 100 140

1.3 0.8 0.8 0.6

5.7 6.3 4.1 3.7

61.7 56.2 46.2 33.3

31.3 34.5 48.9 54.6

0 2.2 0 7.8

956.5 974.3 988.3 1041.2

1800

400

1700

350

1600

300 250

1400 200 1300 1200 1100

Dv90

150

Number of agglomerates

100

180

160

120

100

80

60

40

0 20

50

900 0

1000 140

Dv90 [μm]

1500

Number of agglomerates

Sampling time (min)

Time (min) Fig. 7. In-line Dv90 compared to the actual number of agglomerates during the fourth batch coating process.

1700 1600

Dv90 [μm]

1500 1400 1300 1200 Batch 1 (50 g/min)

1100

Batch 2 (30 g/min) 1000

Batch 5 (0 g/min) 20

15

10

5

0

900

Time [min] Fig. 8. Dv90 on the beginning of the spraying phase of different batches (spray rates are stated in the brackets).

of the particle. In case that chord is not passing the centre, i.e. collinear to the longest CL, the measured size is smaller than actual pellet diameter. For the same reason the SFT’s particle size distribution is wider than SA’s distribution. In-line SFT measurements were also compared with the results of the same probe measured in the at-line environment (Table 7). Smaller particle size and narrower particle size distribution were determined by the at-line approach, however strong correlation with the in-line measurements was observed. Linear regression calculated with Dv50 values of four batches (No. 1, 2, 3, and 5) resulted in R2 of 0.9891 and regression line slope of 1.067. Bigger particle size and wider distribution of the in-line results were a consequence of larger particle size fraction above 1250 lm. There were two possible reasons for such results. Firstly, particle loading of in-line measurements was from 5% to 15%, which was significantly higher comparing to approximately 1% during the at-line

determination. Higher particle loading could lead to higher particle size values, since the probe was not able to distinct two or more overlapping particles. Secondly, slight reversible aggregates could be formed during the process spraying phase and disaggregate after the process. More about the second possibility is written in the following section of this article (Section 3.2.1.). All in all, different results measured with the same probe in distinct environment showed that location and position of the probe could have significant effect on determined particle size (Roßteuscher-Carl et al., 2014). 3.2.1. Suitability of the SFT probe for agglomeration detection To evaluate the SFT’s suitability for agglomeration detection, correlation of Dv90 with the actual number of agglomerates in 50 g of sample was examined (Fig. 7). The actual number of agglomerates was gradually increasing through the process, while Dv90 rapidly increased at the beginning of the spraying phase and did not change significantly anymore. Such results suggest that detection of the agglomeration phenomena was not possible, but they do not disclose the actual reason for such behaviour. Dv90 reached very high particle size values at the very beginning of the process, therefore, we considered that slight reversible aggregates were formed immediately after the spraying phase initiation and they blurred the detection of actual irreversible agglomerates. Spraying rate was different at the beginning of individual batches, therefore Dv90 of different batches was compared (Fig. 8). The analysis showed that Dv90 value was the lowest at the start of the fifth batch when spraying of coating dispersion was delayed on purpose. The second batch’s Dv90 was significantly higher although spray rate was only 30 g/min. Dv90 of the first batch, when spray rate reached 50 g/min, was additionally higher. These results supported the slight aggregates formation hypothesis and showed that the SFT probe was not able to detect or quantify the presence of actual agglomerates. 3.3. Applicability of the NIR probe during the pellet attrition process The spraying of the coating dispersion was deliberately delayed when manufacturing the fifth batch in order to achieve drug layer attrition and evaluate what is the applicability of the probes for monitoring complex systems of more simultaneously ongoing processes. When attrition of the drug layer was detected in real-time by the SFT probe (Fig. 9), spraying of coating dispersion was initiated. In the second part of the process, attrition slowed down and later minor particle size increase was observed. PCA results of SNV pre-treated spectra were clearly different from the results of the first two batches. Instead of two wide bands indicating water, loadings plot showed many narrow peaks and PC-1 values were consequently inconsistent with the off-line moisture content measurements (Fig. 10). NIR measurements of individual components of the first (drug layer) and second (polymer) coating

167

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

1000

45

Pellet size [μm]

900

Dv50 850 800

Coang thickness (μm)

40 950

35 30 25 Predicted

20

Measured

15 10 5

750

0 700

0 0

10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170

15

30

45

60

75

90

105

120

135

150

165

Time (min)

Time [min] Fig. 9. Dv50 during the attrition process of the fifth batch (vertical line marks time point of the spraying phase beginning).

3

Score (PC-1, 94 %)

2 PC-1

1 0 -1 -2 -3 -4 0

10

23

33

45

55

68

78

90

100 113 123 135

Time (min)

Fig. 11. Comparison of the predicted film coating thickness from the NIR spectral data with the SFT measured coating thickness.

initiation and the extent of attrition in-line would be very useful in the manufacturing of film coated pellets. The correlation of the NIR data with drug layer thickness reduction measured by the SFT was determined. The PLS regression between thickness variation and spectral data for the time period of the attrition was calculated, i.e. time from Dv50max to Dv50min. The PLS was calculated using a training set of 63 measurements and random cross-validation with 20 segments was done simultaneously. The entire NIR spectral range, i.e. 1100–2200 nm, was used as a PLS predictive data set. R2c , R2v , RMSEC, and slope of the regression line for the PLS regression calculated with 3 factors were 0.9912, 0.9901, 2.905, and 0.9912, respectively. Such results showed very high correlation of the NIR data with the layer thickness reduction and suggested that NIR probe is not only suitable tool for pellet attrition detection but also for its quantification.

0.2 0.15 PC-1 loadings

Loadings

0.1

3.4. Polymer film coating thickness determination with the NIR probe

0.05 0 -0.05 -0.1 -0.15 1094 1143 1191 1240 1289 1337 1385 1433 1481 1529 1577 1625 1672 1720 1767 1814 1861 1908 1955 2001 2048 2094

-0.2

Wavelength (nm) Fig. 10. First principal component (PC-1) during the attrition process (scores and loadings plots).

revealed that the maximum value on PC-1 loadings plot corresponded to the neutral pellet core peak at around 1440 nm and the minimum loadings value corresponded to the active ingredient peak at around 1670 nm. In the beginning PC-1 rapidly increased and got steady after approximately 90 min of the coating process. Considering the loadings values, increasing PC-1 indicated growing of the neutral pellet core peak and reducing of the active ingredient peak. Such results were consistent with the SFT observations and suggested that PC-1 described the attrition of the first coating (drug layer). Attrition is an unwanted phenomenon in the pellet coating process and can lead to unwanted final product characteristics, such as lower active ingredient content, drug release variation, reduced product stability, reduced gastro-resistance, etc. If attrition of the pellet surface occurs in the beginning of a coating process, which happened in our case, this can be difficult to prove with off-line methods after the process, since additional coating covers the damaged surface underneath. Therefore, a system for detecting

First objective of the PLS model calibration for polymer film thickness prediction was to identify the wavelengths which are the most suitable for this purpose. During the PCA analysis evident decreasing of the peak at 1670 nm was observed. It was shown hereinabove (Section 3.3) that this peak corresponded to the active ingredient present in the drug layer. Decrease of this peak was indirectly linked with the polymer coating thickness increase, since polymer film coating was gradually covering the first layer causing reduction of the active ingredient peak’s intensity. Furthermore, the 1670 nm peak was visible during the whole time of the coating phase and thus it presented a good basis for the PLS model calibration. Therefore, the model was calibrated with the in-line results of the first two batches in the wavelength range from 1600 to 1751 nm. R2c , R2v, RMSEC, and slope of the regression line of the PLS model calculated with 5 factors in the range from 0 to 41 lm coating thickness were 0.9762, 0.9746, 1.765, and 0.9762, respectively. The model was used for coating thickness prediction of the sixth batch (Fig. 11). The predicted thickness values correlated well with the measured ones. Slightly higher deviation, i.e. up to 5 lm, was observed in the middle part of the process, but average absolute error of prediction was less than 2.5 lm. R2 of prediction, root mean square error of prediction (RMSEP), and slope of the regression line between predicted and the reference values were 0.9441, 2.9759, and 0.9240, respectively, indicating good correlation between the NIR and SFT in-line measurements. Such results suggested that the NIR probe was capable to predict sustained release polymer coating thickness. Thickness of sustained release polymer film coating is directly linked to the dissolution rate of active ingredient, thus the results also suggested that the NIR probe has a potential to predict this pellet characteristic.

168

G. Hudovornik et al. / European Journal of Pharmaceutical Sciences 75 (2015) 160–168

4. Conclusion Both in-line techniques proved as very valuable tools for in-process monitoring of the pellet coating process in a pilot scale fluid bed coater. We showed that the NIR was able to predict moisture content, detect deviations in the process, and to predict coating thickness of the sustained release film coating. These are critical attributes of a film coating process, which have a pronounced effect on the quality of the end product. Results indicate that the NIR probe is a promising tool to monitor the coating process on the regular basis in the industrial scale. However, the sensitivity of the PLS models for moisture content determination to changes in the composition suggests usefulness of the probe in research phase is somehow limited. Moreover, a large sample base needs to be used in order to verify the robustness of the model and to provide validity over a broad moisture content range. The SFT technique proved as a useful tool to measure film coating thickness with good correlation to the traditional off-line methods. Its ability to detect process deviations such as agglomeration was however limited, since formation of slight aggregates in the process was faulty detected as agglomeration. On the other hand, very good response was obtained during the pellet attrition process. A PLS model using the in-line data of both probes was calculated, correlating the measured coating thickness of the SFT probe to the NIR data. Results showed that the NIR model was able to predict the film coating thickness with good precision. This suggests possibility to use a single technique in industrial scale for monitoring multiple variables in the film coating process. Acknowledgment The authors would like to thank KRKA, d.d., Novo mesto for providing support during this study. References Avalle, P., Pollitt, M., Bradley, K., Cooper, B., Pearce, G., Djemai, A., Fitzpatrick, S., 2014. Development of process analytical technology (PAT) methods for controlled release pellet coating. Eur. J. Pharm. Biopharm. 87, 244–251. Bechgaard, H., Nielsen, G.H., 1978. Controlled-release multiple-units and single-unit doses a literature review. Drug Dev. Ind. Pharm. 4, 53–67. Berntsson, O., Danielsson, L.-G., Lagerholm, B., Folestad, S., 2002. Quantitative inline monitoring of powder blending by near infrared reflection spectroscopy. Powder Technol. 123, 185–193. Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., Pages, J., 2000. NIR calibration in non-linear systems: different PLS approaches and artificial neural networks. Chemometr. Intell. Lab. 50, 75–82. European Pharmacopoeia, 2013. 8th Edition. Council of Europe, Strasbourg, France. Fischer, C., Peglow, M., Tsotsas, E., 2011. Restoration of particle size distributions from fiber-optical in-line measurements in fluidized bed processes. Chem. Eng. Sci. 66, 2842–2852. Folttmann, F., Knop, K., Kleinebudde, P., Miriam, P., 2014. In-line spatial filtering velocimetry for particle size and film coating thickness determination in fluidized-bed pellet coating processes. Eur. J. Pharm. Biopharm. 88, 937–938. Guidance for Industry, 2005. Q9 Quality Risk Management. ICH Harmonised Tripartite Guideline, Step 4, November. Guidance for Industry, 2008. Q10 Quality Systems Approach to Pharmaceutical CGMP Regulations. ICH Harmonised Tripartite Guideline, Step 4, June. Guidance for Industry, 2009. Q8 (R2) Pharmaceutical Development. ICH Harmonized Tripartite Guideline, Step 4, August. Haaser, M., Youness, K., Velghe, C., Cuppok, Y., Gordon, K.C., Pepper, M., Siepmann, J., Strachan, C.J., Taday, P.F., Rades, T., 2012. Evaluating Critical Film Coating Characteristics of Sustained Release Coated Pellets with Different Size Using Terahertz Pulsed Imaging, ECPS2012, (3/9/2015): . Ho, L., Müller, R., Römer, M., Gordon, K.C., Heinämäki, J., Kleinebudde, P., Pepper, M., Rades, T., Shen, Y.C., Strachan, C.J., Taday, P.F., Zeitler, J.A., 2007. Analysis of

sustained release tablet film coats using terahertz pulsed imaging. J. Control. Release 119 (3), 253–261. Ho, L., Müller, R., Gordon, K.C., Kleinebudde, P., Pepper, M., Rades, T., Shen, Y., Taday, P.F., Zeitler, J., 2008. Application of terahertz pulsed imaging to sustained release tablet film coating quality assessment and dissolution performance. J. Control. Release 127 (1), 79–87. Ho, L., Müller, R., Gordon, K.C., Kleinebudde, P., Pepper, M., Rades, T., Shen, Y., Taday, P.F., Zeitler, J.A., 2009. Terahertz pulsed imaging as an analytical tool for sustained release tablet film coating. Eur. J. Pharm. Biopharm. 71 (1), 117–123. Hu, X., Cunningham, J.C., Winstead, D., 2008. Study growth kinetics in fluidized bed granulation with at-line FBRM. Int. J. Pharm. 347, 54–61. Kadunc, N.O., Šibanc, R., Dreu, R., Likar, B., Tomazˇevicˇ, D., 2014. In-line monitoring of pellet coating thickness growth by means of visual imaging. Int. J. Pharm. 470, 8–14. Karande, A.D., Wan Sia Heng, P., Valeria Liew, C., 2010. In-line quantification of micronized drug and excipients in tablets by near infrared (NIR) spectroscopy: real time monitoring of tabletting process. Int. J. Pharm. 396, 63–74. Kauppinen, A., Toiviainen, M., Lehtonen, M., Järvinen, K., Paaso, J., Juuti, M., Ketolainen, J., 2014. Validation of a multipoint near-infrared spectroscopy method for in-line moisture content analysis during freeze-drying. J. Pharm. Biomed. Anal. 95, 229–237. Kukec, S., Dreu, R., Vrbanec, T., Srcˇicˇ, S., Vrecˇer, F., 2012. Characterization of agglomerated carvedilol by hot-melt processes in a fluid bed and high shear granulator. Int. J. Pharm. 430, 74–85. Lam Yip, W., Gausemel, I., Arne Sande, S., Dyrstad, K., 2012. Strategies for multivariate modeling of moisture content in freeze-dried mannitolcontaining products by near-infrared spectroscopy. J. Pharm. Biomed. Anal. 70, 202–211. Lee, M.-J., Seo, D.-Y., Lee, H.-E., Whang, I.-C., Kim, W.-S., Jeong, M.-Y., Choi, G.J., 2011. In line NIR quantification of film thickness on pharmaceutical pellets during a fluid bed coating process. Int. J. Pharm. 403, 66–72. Mantanus, J., Ziemons, E., Lebrun, P., Rozet, E., Klinkenberg, R., Steel, B., Evrard, B., Hubert, P., 2009. Moisture content determination of pharmaceutical pellets by near infrared spectroscopy: method development and validation. Anal. Chim. Acta 642, 186–192. Markovic´, S., Poljane, K., Kercˇ, J., Horvat, M., 2014. In-line NIR monitoring of key characteristics of enteric coated pellets. Eur. J. Pharm. Biopharm. 88, 847–855. May, R.K., Evans, M.J., Zhong, S., Warr, I., Gladden, L.F., Shen, Y., Zeitler, J.A., 2011. Terahertz In-line sensor for direct coating thickness measurement of individual tablets during film coating in real-time. J. Pharm. Sci. 100 (4), 1535–1544. Närvänen, T., Lipsanen, T., Antikainen, O., Räikkönen, H., Heinämäki, J., Yliruusi, J., 2009. Gaining fluid bed process understanding by in-line particle size analysis. J. Pharm. Sci. 98, 1110–1117. PAT Guidance for Industry, 2004. A Framework for Innovative Pharmaceutical Development, Manufacturing and Quality Assurance, US Department of Health and Human Services, Food and Drug Administration (FDA), Center for Drug Evaluation and Research (CDER), Center for Veterinary Medicine (CVM), Office of Regulatory Affairs (ORA), September. Perez-Ramos, J.D., Paul Findlay, W., Peck, G., Morris, K.R., 2005. Quantitative analysis of film coating in a pan coater based on in-line sensor measurements. AAPS PharmSciTech 6. Article 20. Pomerantsev, A.L., Rodionova, O.Y., Melichar, M., Wigmore, A.J., Bogomolov, A., 2011. In-line prediction of drug release profiles for pH-sensitive coated pellets. Analyst 136, 4830–4838. Poutiainen, S., Matero, S., Hämäläinen, T., Leskinen, J., Ketolainen, J., Järvinen, K., 2012. Predicting granule size distribution of a fluidized bed spray granulation process by regime based PLS modeling of acoustic emission data. Powder Technol. 228, 149–157. Rantanen, J., Räsänen, E., Tenhunen, J., Känsäkoski, M., Mannermaa, J.-P., Yliruusi, J., 2000. Moisture measurement during granulation with a four-wavelength near infrared sensor: an evaluation of particle size and binder effects. Eur. J. Pharm. Biopharm. 50, 271–276. Reid, D., 2006. Water determination in food. In: Meyers, R.A. (Ed.), Encyclopedia of Analytical Chemistry. John Wiley & Sons Ltd. Roßteuscher-Carl, K., Fricke, S., Hacker, M.C., Schulz-Siegmund, M., 2014. In-line monitoring of particle size in a fluid bed granulator: investigations concerning positioning and configuration of the sensor. Int. J. Pharm. 466, 31–37. Silva, A.F., Burggraeve, A., Denon, Q., Van der Meeren, P., Sandler, N., Van Den Kerkhof, T., Hellings, M., Vevaet, C., Remon, J.P., Lopes, J.A., De Beer, T., 2013. Particle sizing measurements in pharmaceutical applications: comparison of inprocess methods versus off-line methods. Eur. J. Pharm. Biopharm. 85, 1006– 1018. The United States Pharmacopeia 37, 2013. The National Formulary 32 (NF32), United States Pharmacopoeia Convention, Rockville, MD. Toedter Williams, N., 2008. Medication administration through enteral feeding tubes. Am. J. Health-Syst. Pharm. 65, 2347–2357. Zheng, Y., Lai, X., Wrang Bruun, S., Ipsen, H., Nedergaard Larsen, J., Lowenstein, H., Sondergaard, I., Jacobsen, S., 2008. Determination of moisture content of lyophilized allergen vaccines by NIR spectroscopy. J. Pharm. Biomed. Anal. 46, 592–596.

A study on the applicability of in-line measurements in the monitoring of the pellet coating process.

Special populations including paediatric and elderly patients often need advanced approaches in treatment, such as one-a-day dosing, which is achieved...
1MB Sizes 0 Downloads 6 Views