4.2 Materials and Methods
4.2.2 Continuous manufacturing process description
Continuous direct compression tableting line consisted of two loss-in-weight feeders (K-Tron, Types K-ML-D5-KT20 and K-CL-SFS-KT20, Niederlenz, Switzerland), continuous mixer (Modulomix, Hosokawa Micron B.V., The Netherlands) and tablet press (PTK-PR1000, South-Korea).
Due to the varying amount of lubricant in the formulation, the feed rate of in-coming materials (pre-blend and lubricants) were set as shown in Table 1. (See Figure 15.)
48
Table 1. The production feed rates (kg/h) of set points at each formulation due to the different lubricant concentrations.
Total production feed rate 5 kg/h
Total production feed rate 10.5 kg/h
Total production feed rate 16 kg/h
Lub.
(%) Pre-blend Lubricant Pre-blend Lubricant Pre-blend Lubricant
0.5 4.975 0.025 – – 15.920 0.080
1.0 – – 10.395 0.105 – –
1.5 4.925 0.075 – – 15.760 0.240
Figure 15. Equipment integrated into the top-down direct compression manufacturing process.
Powder feeders fed the continuous mixer through the mass inlet port A or inlet port B (Figure 16). The inlet port A was used in all set points for the pre-blend. Feed rates and compression forces were monitored and analyzed during the process. The feed rate of all of the excipients and paracetamol was acquired from the feeders at one second intervals.
In this study, one of the qualitative process factors was the inlet port of the mixer (A or B) for lubricants. The second mixer process factor was the mixer rpm (500 rpm, 850 rpm, 1200 rpm).
feeders
blender
tablet press
49
Figure 16. Continuous mixer with two inlet port A and B.
The outlet port of the mixer was directly connected to the tablet press hopper. This simple top-down direct compression manufacturing process was adjusted, monitored and the data was recorded using in-house software.
The speed of the turret (51 rpm) was kept constant on each run. Due to the constant tableting rotation speed, at each feed rate (kg/h), the number of punches was changed (i.e.
when the feed rate decreased, a specified number of punches was removed from the tableting machine). The tablets were compressed using flat shaped 10 mm diameter punches and the target weight of tablets was 330.0 mg. Tableting parameters were kept constant throughout the study except for the die filling depth, which was increased with the formulation containing StAc.
The continuous top-down manufacturing process was started by turning on the feeders and mixer. The tablet press was turned on when approximately 1 kg of the powder blend was fed into the tablet press hopper. The duration of the tabletting process of each run was 20 min. During the compaction process, the average compression force was recorded in all of the punches for each rotation.
4.3 QUALITY TARGET PRODUCT PROFILE (QTPP)
The QTPP was defined for the final mass and the final product (Table 2). The main target was that the continuous manufacturing process would produce a product that fulfilled all the predetermined quality attributes.
50
Table 2. The QTPP of the final product and process.
Response Objective and target Flowability (s/60 g) Minimize: 2.5 s (max. 12 s) Ejection force (N) Minimize: 425 N (max. 500 N) Tablet strength (N) Maximize: 147.1 N (min. 49.0 N) Dissolution 2.5 min (%) Maximize: 90% (min. 80%)
4.4 DESIGN OF EXPERIMENT
In this study, a Design of Experiment (DoE) with two levels was used (Frac Fac Res V +, MODDE Pro 11, Umetrics MKS AB, Umeå, Sweden). The regression models for responses were created by using multiple linear regression (MLR). The design of experiment consisted totally of 19 study points (runs), including 3 center points. Two qualitative factors were studied: lubricant type (MgSt and StAc) and mixer inlet port for lubricants (A and B). The three quantitative factors were total production feed rate (5 kg/h, 10.5 kg/h, 16 kg/h), mixer rpm (500 rpm, 850 rpm, 1200 rpm) and amount of lubricant (0.5%, 1.0%, and 1.5% (w/w)). The design of experiment with factors and their settings are listed in Table 3. The model with the main factors as well as all interaction terms was used in the initial model. Later, all statistically non-significant interaction terms were excluded individually from the final model starting from the least significant term, until the criteria (r2 > 0.780 and q2 > 0.420) were achieved.
51
Table 3. Factor settings in the design of experiment.
Due to the formulation settings of the lubricant, the amount of pre-blend (consisting of paracetamol, microcrystalline cellulose and sodium starch glycolate) in formulations and the consequent amount of paracetamol at each lubricant levels are shown in Table 4.
52
Table 4. The amount of paracetamol at different lubricant levels.
Lubricant (%) Pre-blend (%) Amount of paracetamol in pre-blend (%)
0.5 99.5 76.0
1.0 99.0 75.6
1.5 98.5 75.2
4.5 DESIGN SPACE
The design space was created using Modde (Modde 11, Sweden). Response objectives and targets for the optimizer are presented in Table 2. The optimizer desirability setting was set to the target. The optimizer suggested alternative set points where the initial set point was selected based on DPMO (Defects per million opportunities outside of specification) and Log(D) (normalized distance to the target). Design Space was created from the selected initial set point using the find robust set point function. The robust set point function runs the Monte-Carlo simulation with the following settings: a resolution of 32, an iteration of 50,000, and a DPMO limit of 10,000. Resolution describes to how many sections each factor range is divided. Iteration defines how many simulations are performed in each section. DPMO is used as a stopping criterion in the setpoint analysis.
The limit of 10,000 equals to 1% outside of specification. DPMO is calculated using following equation. (2)
where H is the number of hits outside of the specifications and Ns is the number of iteration in the simulation. The model error and factor precision were included into the simulation. The interval estimation was set to confidence. Interval estimation is used in statistics as an uncertainty measure of a population parameter computed from sample data. Interval estimate is used for two purposes. Firstly, to assess if model parameters are statistically different from zero (null-hypothesis). Secondly, to state an interval within which we are confident that we find future predictions. A common statement is to say that with a confidence of 95% we will find a future sample in this region enclosed the interval, i.e. average ± interval. In MODDE, there are several interval types available.
Confidence interval estimation encloses average of populations, with some confidence (95%) and is mainly used to illustrate the variance of the model coefficients.
DPMO =1 000 000∗H
Ns (2)
53
4.6 CHARACTERIZATION OF MATERIALS AND FINISHED PRODUCT
According to the pre-determined QTPP parameters, flowability, ejection force, tablet strength and dissolution profile were studied. In addition, the pre-blend and the powder blend after the mixer were assayed and the surface morphology of the tablet was determined.
While testing mass homogeneity, samples of approximately 3 g of the pre-blend were collected after mixing. After mixing the mass was poured into two mass conveyer and two samples of each conveyer from three levels (top, middle, down) were collected with a single point sampler. After each continuous run, the powder blend was collected into the plastic bottles. 3 samples of approximately 3 g of the powder blends of one level (middle) were collected. The amount of the paracetamol was analyzed with HPLC.
The ejection force of the tablets was analyzed with a compaction simulator (PCS-1, Puuman Oy, Kuopio, Finland). Ten mm diameter and flat shape punches were used. Six replicate measurements of each powder blend were performed. The samples were weighed to the targeted weight of 330 mg by balancer (Mettler AG245, Mettler-Toledo GmbH, Greifensee, Switzerland). The simulator die was filled manually.
The flowability (s/60 g) was measured from each excipient and powder blends from each design points by using flow rate apparatus (Erweka GDT, Erweka Gmb, Heusenstamm, Germany). The 60 g samples (n = 3) when there were no lubricants and microcrystalline cellulose (did not flow at all) were weighed (A&D HF-6000, A&D Instruments Ltd) prior to the flow rate measurements. The flow rate (sec) of a pre-weighed sample was determined (Cielo, Professional Stopwatch) by delivering the sample into the stainless steel hopper. The measurement was started with unit vibration while the mass was entering through the outlet nozzle. The measurement was stopped when the whole mass had exited through the nozzle.
Tablet strength (N) was measured by using a tablet strength apparatus (n = 3) (Tablet Tester Model 54, Dr. Schleuniger, Pharmatron, Switzerland). Tablet samples collected at the tableting time points of 0, 5, 10, 15, 20 min and were measured immediately.
The dissolution profile of paracetamol was determined from the tablets (n = 3) collected at the tableting time point of 20 min with dissolution paddle apparatus (SOTAX AT 7 smart, Sotax AG, Switzerland). In the dissolution, 900 ml of phosphate buffer solution (pH 5.8), the rotation speed of paddles of 100 rpm and temperature 37.0 ± 0.5 °C were used. The dissolution samples (5 ml) were collected at time points of 2.5, 5, 10, 15 and 30 min and replaced with fresh medium. Samples were filtered immediately using filter paper (Whatman No 4). Paracetamol concentrations were measured with UV/Vis spectrophotometer (UV-1700, Shimadzu, Japan) at 243 nm.
The surface morphology of powder samples and the tablets (collected at the tableting time point of 20 min) were visualized with a scanning electron microscope (SEM, Zeiss Sigma HD VP, Carl Zeiss NTS, Cambridge, UK). Samples were attached to aluminum 54
stubs using colloidal graphite glue. Samples were coated with a gold layer using sputtering time of 120 s (Automatic Sputter Coater B7341, Agar Scientific Ltd., UK). Each sample was examined using a 1500 × magnification.
4.7 RESULTS AND DISCUSSION
4.7.1 Homogeneity of the pre-blended mass and the powder blends after mixing
The mass homogeneity of the pre-blended mass showed that paracetamol was well mixed in the mass. The assays of the samples varied between 98.71% and 101.16% from the average. The assay analyzed from the powder blends collected after tabletting varied between 95.36% and 103.88% (from the average), meeting the requirements of 95–105%., except the blends of N10 (93.36%) and N15 (106.02%), thus the mean assay of paracetamol did not meet the quality requirements of 95–105%. Some errors might have occurred when analyzing the blend N4, as the SD value differs extensively from the other blends.
The amount of paracetamol was not analyzed from the tablets, thus these results are only indicative. However, as the pre-blend was exactly the same in each run, the slight variation in the drug concentration is mainly due to the different process parameters (feed rate, mixing speed).
4.7.2 Monitoring of continuous direct compression process 4.7.3 Feed rate
The feed rate of all of the excipients and paracetamol was acquired from the feeders at one second intervals. The feed rate of pre-blended mass reached a steady state after 5 min and the feed rate was extremely constant during the whole process at each runs. This result shows that starting materials with good flow properties exert an influence on the feeding accuracy of the feeders. The mass flow of each total feed rate of pre-blended mass is presented in the supplementary data. The feed rate of the lubricant (setpoint 0.240 kg/h and 0.105 kg/h) is presented in Fig 17. (N4 was missing from the Figure 17, because of problems with data collection).
55
Figure 17. The mass flow of total feed rate of the lubricants with setpoint of A: 0.240 kg/h and B: 0.105 kg/h (N4 was excluded from the data because of problems with data collection).
The feed rate of lubricants did not reach steady state with runs at the set points of 0.075 kg/h, 0.025 kg/h (total feed rate 5 kg/h) and 0.105 kg/h (total feed rate 10.5 kg/h). As seen from the results, incorporating the lubricants from their own feeders, results challenges to maintain a constant feed rate with low feed rates because such low feed rates are at the very low end of feeder capacity. At the set point of 0.08 kg/h (total feed rate 16 kg/h), the lubricant flow of runs N2 and N14 was rather constant. This was also the case with the set point of 0.240 kg/h, the lubricant flows of N10 and N16 were rather constant (total feed rate 16 kg/h).
The variation in the feed rate was extremely low with pre-blends (N1-N13, N15-N19, RSD% varied between 0.01 and 0.05). However, the variation in N14 (total feed rate 15 kg/h) differed from the others (RSD% 2.51). The variation (RSD%) in the feed rate was extremely high with lubricants with a total feed rate of 5 kg/h (N1: 60.48; N3: 322.52; N9:
141.44). Furthermore, the RSD% value was rather good with N2 (9.61), N10 (6.53), N16 (2.39) with the total feed rate of 16 kg/h, and N17 (9.09) with the total feed rate 10.5 kg/h.
The results demonstrate that with extremely low feed rates of lubricants, it was rather challenging to have consistent feed rates.
4.7.4 Compression force
The mean compression forces are shown in Fig 18. It can be seen that the main compression force (kN) remained rather constant, except with N6; approximately at the 7 min point, the mean compression force radically decreased. This was due to the change in the filling depth in order to ensure a constant tablet weight. Initially, tablet press settings were kept constant as lubricant type changed in the formulation from MgSt to StAc. However, the target tablet weight was not achieved and it was necessary to change the filling depth.
56
Figure 18. The mean compression force (kN) of each run.
Interestingly, no such similar phenomenon was observed with the N17 run when the lubricant type was changed in the formulation from StAc to MgSt. The feed rate of lubricant with the run N6 was rather high at the first 10 min (0.5 kg/h) whereas the setpoint was 0.240 kg/h. It could be speculated that an extremely high variation in feed rates could influence the further processing steps such as tableting. In other runs, the mean compression force was approximately between 15.0 and 20.0 kN. It can be seen that in the runs with MgSt, the mean compression force was slightly higher (16.53–19.35 kN) than in the runs of StAc (14.67–18.47 kN).
57
4.7.5 Tablet structures
The properties of the lubricants were observed on the structure of the tablet surface (Figure 19). Tablets containing StAc had rougher surfaces than tablets containing MgSt.
Figure 19. SEM images of surface of the tablet (mixer speed 1200 rpm, feed rate 16 kg/h) A: StAc 1.5% B: MgSt 1.5%.
In addition, it was observed that the powder blends lubricated with StAc stuck to the punches during tableting. This sticking tendency of StAc and MgSt during tableting has been reported to be dependent on the melting point of lubricant and the main material components (Roberts et al., 2004). Furthermore, it has been shown only with MgSt that the sticking tendency was increased when the concentration was increased. This differs from the current study, as there was no sticking to the punches when the powder blends were lubricated with MgSt. This might be due to the different active ingredients used and their different melting points (ibuprofen 75 to 77 °C and paracetamol 168–171 °C (Open Chemistry database)). Furthermore, based on visual observations, the changes in process parameters (mixing speed and feed rate of in-coming materials) exerted no influence on the structure of blends, surface of tablets or sub-divided tablet. This indicates that it is the attributes of the lubricant that have the greatest influence on the final product with a continuous manufacturing process. The additional scanning electron microscope images of continuous mixed mass, surface of the tablet and the sub-divided tablets are shown in the supplementary data.
4.7.6 Results from design of experiment and design space 4.7.7 Flowability
Based on visual observations, paracetamol (7.61 s/60 g) and sodium starch glycolate (14.44 s/60 g) exhibited a good and uniform flow rate. Microcrystalline cellulose formed a hole in the middle of sample and the flow stopped immediately. The type of microcrystalline cellulose was Vivapur®102 (bulk density: 0.30 g/ml, PSD: d50: 136 μm, d90: 255 μm according to the manufacturer). According to previous experiments, the flow rate of microcrystalline cellulose can be improved when the particle size is increased. In
A B
58
addition, increasing the relative humidity has been shown to decrease the flow rate of microcrystalline cellulose (Crouter and Briens, 2014). In this current study, relative humidity was not recorded. MgSt and StAc are both cohesive and sticky materials, thus the flow rate could not be measured with lubricants. One of the main functions of a lubricant is to increase the flow rate of powder blends.
The flow rate was rather high (2.81–7.51 s) with the powder blends containing StAc (N5–N7, N13–N19). One exception was N8 (StAc 0.5%, mixer speed 1200 rpm, feed rate 16 kg/h), where the flow rate was (11.59–14.72 s). The flow rates of the powder blends containing MgSt (N1–N4, N9–N11) were slower (8.05–17.82 s), compared to the powder blends with StAc. The flow rate of N12 (MgSt 0.5%, mixer speed 1200 rpm, feed rate 16 kg/h) was exceptionally faster (5.41–6.98 s) than other blends containing MgSt.
According to the flow rate investigation of lubricants conducted by Morin and Briens (2013), MgSt was reported to improve the flow rate of spray-dried lactose than StAc. This might be explained by MgSt's ability to fill the cavities (Perrault et al., 2011) between spray-dried lactose particles by creating more spherical and smoother particles (Roblot-Treupel and Puisieux, 1986). Thus, MgSt reduces surface irregularities of the excipients.
However, in this study, the opposite results could be explained by the different physical properties of the particles in the studied blends i.e., particle size and distribution, surface morphology, density and particle shape. However, there was rather extensive variation in the blends containing MgSt (the SD values of the blends with StAc varied between 0.27 and 1.73, whereas the blends with MgSt SD varied between 0.43 and 1.88), which might be evidence of a greater influence of input variables for the mixed blends. In summary, StAc has a higher impact than MgSt on achieving good flow rate properties for the mixed blends. The MLR model was developed to determine the effects of input variables on flowability of the mixed blends (see supplementary). The accuracy of the model was R2 (0.988), Q2 (0.843), model validity (> 0.25), and reproducibility (> 0.5), indicative of an excellent model. The 4D response contour plots of the flow rate properties are presented in Figure 20. From the coefficient plot (Supplementary data), the main significance terms were lubricant type and inlet port.
59
Figure 20. 4D Contour plot (MLR) of the effects of lubricant and process variables on flow rate. Plot A: lubricant inlet port A, Plot B: lubricant inlet port B.
From the 4D contour plots of flow rate, it can be seen that the flowability of powder blends with MgSt is lower than with StAc (Figure 20). With MgSt, flowability was relatively independent of process variables and concentration from inlet port A, whereas from inlet port B, process factors and concentration did exert some effect on flowability.
In the case of StAc, with feeding through the inlet port B, the flowability was fast and virtually independent of any variables whereas with feeding through the inlet port A, the flowability seemed to be dependent on the concentration and feed rate, even those variables were not statistically significant according to the coefficient plot (Supplementary data). Based on these observations, the most robust and best flowability could be achieved with StAc and inlet port B, a finding in agreement with the ejection force results (will be shown later). Based on the ejection force and flowability results, it seems that inlet port B achieved better results. In practice, when feeding the lubricant through the inlet port A, it is exposed to a more intense and longer blending time, which might cause over-lubrication.
4.7.8 Ejection force
One of the main functions of the lubricant (in addition to enhancing flowability) is to reduce the friction between the tablet and the die wall. The ejection force can be used as a measure of lubrication efficiency. The lower the ejection force needed to eject the tablet from the die, the more efficient the lubricant. The highest ejection force was needed with N3 (724.8 N), whereas the lowest ejection force was observed with N13 (369.8 N). An MLR model was developed to determine the effects of input variables on ejection force. The model for ejection force R2 (0.960), Q2 (0.833), model validity (> 0.25), and reproducibility (> 0.5), was excellent. As seen in the coefficients (Supplementary data), the main significance terms were lubricant type, concentration, inlet port and total feed rate.
A B
60
As Figure 21 shows, the main effects on ejection force were lubricant type and concentration, with StAc being a slightly more efficient lubricant than MgSt. In addition, the inlet ports A and B have also a significant effect on ejection force. The influence of feed rate was observed clearly with MgSt 0.5%. The ejection force was decreased when increasing the feed rate up to 16 kg/h. With inlet port B, the ejection force was found to be lower than when using inlet port A.
Figure 21. The 4D contour plots (MLR) of the effects of lubricants and process variables on ejection force. Plot A: lubricant inlet port A, Plot B: lubricant inlet port B.
Figure 21. The 4D contour plots (MLR) of the effects of lubricants and process variables on ejection force. Plot A: lubricant inlet port A, Plot B: lubricant inlet port B.