Implementation of quality by design approach in manufacturing process optimization of dry granulated, immediate release, coated tablets – a case study

Abstract The aim of this study was to optimize the process of tablets compression and identification of film-coating critical process parameters (CPPs) affecting critical quality attributes (CQAs) using quality by design (QbD) approach. Design of experiment (DOE) and regression methods were employed to investigate hardness, disintegration time, and thickness of uncoated tablets depending on slugging and tableting compression force (CPPs). Plackett–Burman experimental design was applied to identify critical coating process parameters among selected ones that is: drying and preheating time, atomization air pressure, spray rate, air volume, inlet air temperature, and drum pressure that may influence the hardness and disintegration time of coated tablets. As a result of the research, design space was established to facilitate an in-depth understanding of existing relationship between CPPs and CQAs of intermediate product (uncoated tablets). Screening revealed that spray rate and inlet air temperature are two most important factors that affect the hardness of coated tablets. Simultaneously, none of the tested coating factors have influence on disintegration time. The observation was confirmed by conducting film coating of pilot size batches.


Introduction
In the contemporary pharmaceutical industry, the implementation of quality by design (QbD) principles becomes a necessity. According to the ICH Q8 (R2) guideline, the quality of medical product should be designed and built into the product [1]. To ensure that products will deliver predefined performance, the development should be carried out according to a complex, science-based approach. It could be based on collected knowledge, results of risk analysis, use of knowledge management throughout the lifecycles, and/or statistical methods [1][2][3]. The interaction between process inputs, defined as material attributes and process parameters, and outputs (critical quality attributes -CQAs) can be presented as a Design Space. It enables to identify the range of process parameter values which does not compromise the dosage form's quality and ensures its adherence to specification. It can be developed for new products, as well as for marketed ones. Its applicability can cover single processes or even entire manufacturing process, enabling its better understanding. Furthermore, determination of design space is essential to define the control strategy of product quality [1].
The statistical tool used to mathematically define relationships is known as Experimental Design. Among these highly organized experimental plans are those dedicated to the identification of the most significant impacting variables on CQAs (screening designs i.e. Plackett-Burman design [4]) or those which are intended to determine in a precise manner existing interaction (factorial designs [5]).
The aim of this work was to optimize the manufacturing process of the product available on the market in line with the QbD concept. The content is organized into two parts. The first one describes the optimization of tablets compression process with the step of dry granulation performed by slugging of powder mixture containing model API. The influence of CPPs, established in preliminary studies, i.e. slugging and tableting compression forces, was related to tablets' quality parameters, such as resistance to crushing, disintegration time, and thickness. The relationship between independent variables and quality output was defined by means of design space established with the application of statistical methods.
To the authors' knowledge, in the available literature so far there have not been published any studies on the use of the principles of DOE in the process optimization of slugging and tableting with regard to the product's mechanical properties, although there has been extensive research on reduction in tablets' hardness related to slugging process, e.g. [6][7][8][9][10][11][12][13]. Beli c et al. applied artificial neural networks and fuzzy logic in the process optimization leading to reduced tablet capping; however, process parameters of slugging were not included in the final model [14].
The second part of the article focuses on identification of coating CPPs affecting selected CQAs of the final product. Plackett-Burman screening design was employed to identify laboratory size batch coating process variables which may have influence on the resistance to crushing and disintegration time of coated tablets. The aim of the research at laboratory scale batches is to present how to identify process variables that could potentially affect the pilot size batch CQAs during scale-up in an easy and inexpensive way. Afterwards, the impact of previously screened factors was evaluated during coating of pilot size batches.

Tablets manufacturing process via dry granulation
Powder mixture premix was prepared by combining and sieving API, croscarmellose sodium and colloid silica anhydrous through 0.8 mm screen (GS 100 rotosieve, Glatt GmbH Systemtechnik, Dresden, Germany) into a container where part of microcrystalline cellulose had been loaded previously. The premix was blended for 5 min at 11 rpm (Canguro Tumbler, Zanchetta, Italy) and sieved through 0.8 mm screen. Then the second part of microcrystalline cellulose was added into the container with premix and blending for 10 min at 11 rpm was performed. Before compression stage magnesium stearate was added as a lubrication substance, which was spread uniformly upon the whole volume by blending. The prepared powder mixture was subsequently compressed (slugged) using a rotary tablet press (Fette 1200, Schwarzenbek, Germany) equipped with Ø 13 mm, round flatfaced punches. Three different values of compression force were used (4.5, 8.0, and 11.5 kN). Slugs obtained with the application of different compression forces were collected and processed separately. The slugs were calibrated to obtain granules, using conical screen mill (Quadro Comill, Quadro Engineering Inc., Waterloo, Canada) with a 1.0 mm screen at 2500 rpm. Before tableting, the granules were mixed for 4 min at 11 rpm with a previously sieved amount of croscarmellose sodium. The same tabletpress machine, equipped with 10 mm round, concave punches was used to compress the granulate mixture into tablets within the tested range of main compression force values (9.0, 13.0, and 17.0 kN). Subsequently, an evaluation of the uncoated tablets CQAs was carried out and the obtained cores were used for further laboratory size batch coating trials.

Tablet film coating
The coating dispersion was prepared by mixing Lustre Clear LC 103 with purified water of room temperature for 60 min using a mixer working with rotation speed of 200-400 rpm. The coating process was performed in a Glatt GMPC II pan coater (Glatt GmbH, Binzen, Germany) equipped with changeable perforated pan until weight gain of 7 mg per tablet was achieved. Small perforated drum equipped with one nozzle (1.5 mm) was used for laboratory scale batches (1.5 kg). The number of the same diameter nozzles was doubled during the coating of pilot size batches. Also bigger perforated drum was used in order to enable coating of 25-fold larger quantity of tablets.

Quality by design approach
The entire manufacturing process was split into two parts: tablets compression with the step of dry granulation and coating of tablets. The optimization was not performed in terms of CMAs (critical material attributes). The product is already available on the market. Thus, the change of formulation (raw materials or/and its quantities) is connected with the necessity of making changes in registration dossier, which would generate expenses. Moreover, it means that probably the product would have to be developed from scratch. The objective of our optimization was to avoid submitting any changed documentation to regulatory authorities. This is the reason why we omitted CMAs and any changes in the formulation. Our work focused on in-depth understanding of existing relations between CPPs and CQAs with the aim to define the design space of compression process.
In the case of coating process, the intention was to identify these process parameters that have the most significant influence on CQAs of coated tablets. The coating trials were performed on small laboratory scale batches according to the Plackett-Burman screening design. We tried to answer the question whether coating parameters identified at the laboratory scale find its applicability at the pilot scale. Additionally, it would be a proof that identification of CPPs during laboratory phase might be helpful during scale-up. Moreover, the knowledge gathered at the laboratory scale can improve process understanding and minimalize the risk of failure.
Tableting design spacepredictive models development To analyze existing relations between process parameters and their output the 3 2 full factorial design and multivariate regression method were employed. The influence of two processing variables (independent variables), namely slugging compression force (A) and tableting compression force (B) on response output (dependent variables), represented by uncoated tablets' (cores) CQAs, such as their resistance to crushing, disintegration time and thickness, was investigated. These two variables were chosen based on failure mode and effects analysis (FMEA) risk analysis, whose discussion is beyond the scope of this manuscript.
The application of 3 2 full factorial experimental design means that both input variables values are set on three levels, that is low, medium, and high, labeled as À1, 0, and þ1, respectively. In total, 11 experiments were performed, including two replications of the center point (10C and 11C) with the aim to enable the assessment of pure error (Table 1). This type of matrix enables to investigate quadratic relationship existing between independent variables and dependent ones. Although two-level design with center points also enables to identify the presence of curvature and in addition it requires fewer trials, it was decided that threelevel design will be chosen. On the one hand, conducting trials on three levels instead of two will not increase the overall cost of research in our case. On the other hand, it will deliver more data to our analysis.
Statistical analysis was performed using STATISTICA V R 10 software (StatSoft Inc., Tulsa, OK). The least-squares method was used to estimate the coefficients of polynomials. The analysis of variance at p < .05 was used to find the terms that are statistically significant. The factors found to be insignificant were gradually eliminated to achieve final mathematical model that on the one hand would be simple, and on the other hand would reflect the influence of all major factors. The normality of residuals was confirmed by calculation of Shapiro-Wilk statistic (a ¼ .05). The acceptability of fitted polynomial equations was assessed by the calculation of adjusted coefficient of determination (R 2 adj ) values.
Plackett-Burman designlaboratory batch size coating process parameters screening The Plackett-Burman screening design was chosen to identify the process parameters which may have significant impact on CQAs of the final product. All trials were performed using laboratory size batches to minimalize the overall cost of research. This statistical tool enables the evaluation of the significance and magnitude of effects of many variables affecting CQAs in a relatively small number of runs. This type of design identifies the main effects based on trials according to the matrix where low and high levels of independent variables are tested (Table 2). Therefore, it is helpful in indicating which factors should be deeply investigated using other statistical tools which enable to define existing relations between CPPs and CQAs in a mathematical manner. In the case of our experiment seven potentially critical variables (spray rate, air volume, inlet air temperature, atomization air pressure, drying time, preheating time, drum pressure) were tested in eight runs in order to find out whether they influence product CQAs, that is, resistance to crushing and disintegration time. The trials were carried out in a randomized order. Additionally, three trials were conducted at medium settings (central point with three repetitions -9C, 10C, and 11C - Table 2), which enabled to estimate model pure error.

Pilot size batch assessment
The objective of this part of research was to conduct pilot size batches coating using modified process parameters. Two most significant parameters established previously in the screening experiment performed on laboratory size batches were subjected to changes. The intention was to examine whether or not these parameters play a crucial role during coating of pilot size batches and, thus, are able to influence the CQAs of film-coated tablets.
The fixed values of the remaining process parameters applied during pilot size batch coating were established in preliminary studies performed during the optimization of pilot size production. These values differ from those applied during coating of laboratory scale batches because 25 times higher quantity of tablets was loaded into a bigger drum. Table 3 presents the process parameters values adjusted to pilot size batch and drum that were kept constant during our trials. Therefore, the environment of coating processes was controlled by different levels of the most powerful effects, i.e. spraying rate and inlet air temperature. Batches manufactured at the pilot plant were used to evaluate the effect of coating pattern on the tablets' resistance to crushing and disintegration time in comparison with its value for tablet cores. Tablets were coated either by intermittent spraying (spraying: 2:30 min, drying break: 20 s), which increases cores wetting, or by continuous spraying, which reduces the wetting due to the longer time of the process and consequently the lower total rate at which the coating dispersion comes into contact with core surfaces (Table 3).

Tablet evaluation
Weight, thickness, resistance to crushing Weight, thickness, and resistance to crushing were measured simultaneously by a Multicheck TU III tester (Erweka GmbH, Germany) for a sample of 20 tablets.

Disintegration time
Disintegration time was assessed using a PTZ Auto 1 disintegration tester (Pharma Test, Germany). Tablets (n ¼ 6) were placed in basket tubes without disks. The assembly was immersed in purified water (37 ± 0.5 C) at the rate of 30 ± 1 cycles per min. The disintegration of dosage form was detected visually and the disintegration time of the last tablet was recorded.    Results and discussion

Tableting optimization
Design of experiment A three-level full factorial design was employed to study how two critical process parameters (CPPs) influence the response variables. The compression forces applied during slugging and tableting are the CPPs. Resistance to crushing, thickness, and disintegration time of tablets are response variables. The input values and response results are presented in Table 4. Model building was performed by backward elimination of statistically insignificant terms to give the most simplified form of the predictive equations. The normality of residuals was confirmed by calculation of Shapiro-Wilk statistic (a ¼ .05) ( Figure S1). In the case of the disintegration time model, logarithmic transformation was applied with two objectives. The first one was to ensure that residuals come from a population which has normal distribution as indicated by Shapiro-Wilk test. The second one was to stabilize the variance of residuals ( Figure S2supplementary materials). Moreover, equations describing disintegration time and thickness contain an insignificant term represented by independent variable Aslugging compression force. It means that despite its appearance in mathematical model, its influence is negligible. The rationale for the inclusion of the insignificant term is the same as the reason for the logarithmic transformation described above. The power of effects is presented by Pareto charts. Each of the listed effects in equations and their magnitudes are represented by bars. Every effect whose standardized value oversteps the red line (p value <.05) on the graph significantly influences the value of response ( Figure 1).
The lack of fit was tested for all models and in all cases was insignificant, which is desirable (Table S1supplementary materials). High values of R 2 adj , presented in the sub-chapters below, show that the variability of response is described in most extent by the changes of factors included in predictive equations.
Resistance to crushing. The model developed for resistance to crushing (R 2 adj ¼ .99) is given by the equation: Acompression force applied during slugging [kN], Bcompression force applied during tableting [kN], and resistance to crushing is expressed in newton (N).  The 3D plot presents the response parameter: hardness by surface curvature as a function of A and B input variables (Figure 2(I)). The most significant factor that influences the tablets' hardness is the compression force applied during tableting step. The impact is positive: the higher compression force is applied during tablet compression, the higher value of resistance to crushing is achieved. However, the strength of the effect is modified by the value of compression force used during slugging. In other words, the lower the compression force used at the slugging step (lower hardness of slug), the lower force needs to be used during tablet compression in order to manufacture tablets with certain hardness. The effect is nonlinear.
Disintegration time. The model developed for disintegration time (R 2 adj ¼ .97) is given by the equation: Acompression force applied during slugging [kN], Bcompression force applied during tableting [kN], and disintegration time is expressed in seconds (s).
The main factor influencing the tablets' disintegration time is the compression force used during tableting (Figure 2(II)). With an increase in compression force during tableting the disintegration time also increases. Depending on the applied compression force during slugging, unit change of pressure during tableting will result in different changes in the disintegration time. It means that it is easier to adjust the process of tableting to achieve the desired value of disintegration time when lower compression force is used for slug production.
Thickness. The model developed for thickness (R 2 adj ¼1.00) is given by the equation: Acompression force applied during slugging [kN], Bcompression force applied during tableting [kN], and thickness is expressed in millimeters (mm).
Tablets' thickness is mainly affected by compression force applied during tableting (Figure 2(III)). It is described by a second order polynomial. The higher the value of impacting factor, the bigger decrease in tablets' thickness is observed. However, at lower ranges of compression force the decrease in thickness is larger than at higher range. The impact of slugging compression force is negligible.
Design space. The latest production batch record was the source of parameters limits applied in our research. The choice of investigated CQAs was arbitrary. Three CQAs, namely: resistance to crushing, disintegration time, and thickness were chosen. The purpose was to improve the understanding of CPPs influence on selected CQAs by building design space.
Thus, a 2D contour plot was constructed by overlaying relationships between independent variables and dependent ones taking into consideration their constraints: resistance to crushing of all tablets should be not less than 150 N and not more than 200 N. Therefore, the assumption was made that the average value of this parameter should be kept from 162 to 188 N. It should ensure that all tablets will be within the wider range set for individual tablets, disintegration time should be not less than 40 s but not more than 90 s. Disintegration time is a parameter that affects the product's dissolution profile. Additionally, the API has irritating properties to the oral tissues. This is the reason why coated tablets should get from mouth to stomach in an unchanged form. To achieve that, their disintegration time must be neither too long nor too short. This is the reason why these limits were defined, thickness of tablets should be between 4.4 and 4.9 mm. The thickness of final tablets is important from manufacturer point of view because during the last production step bulk product is packed into blisters and format parts dictate the tolerance ranges. In our study, it was assumed that uncoated tablets are covered with the same quantity of film. Therefore, the layer is uniform and in each case has the same thickness. This assumption allowed us to optimize the thickness of cores (uncoated tablets) as an essential parameter from the packaging point of view. Figure 3 depicts graphical interpretation of the influence of two process parameters on all the studied CQAs. Green marked area shows the values of both factors that enable to produce tablets with all quality attributes within the specified limits. This common area is the design space. Red marked area presents the values of factors whose application will result in achieving product with at least one of the CQAs out of the limits.
Within tested ranges of CPPs, thickness exceeds the upper border only if tableting compression force is lower than 9.4-9.6 kN. However, the line marking the lower limit of disintegration time (40 s) on the graph cuts off the area of tableting compression force below 14.2-15.3 kN. Therefore, the importance of the upper thickness limit can be neglected. Additionally, the constrained resistance to crushing limits marks the strip of area that fulfills all quality requirements and is highlighted in green. It means that applying parameters values that cross in the green zone will give the product in compliance with our requirements.
Design space area, established by means of 3 2 full factorial design is largely limited by independent variables values which were assumed in the early planning phase. Therefore, to investigate the influence of tableting compression force whose values exceed the already tested upper limit, extra trials were performed and regression models were applied. The decision to expand the investigation to higher values of compression force was made after performing DOE evaluation, which revealed that higher values of this parameter might enable to achieve tablets fulfilling our requirements. Moreover, we had some remaining granulates from the previous experiment to perform extra trials. Multivariate forward stepwise regression and nonlinear estimation methods were utilized to create models describing the relationship between the same input and output as described above, however, the analyzed range of independent variables was expanded by adding some points. The slugging compression force values remained the same as in DOE, while additional value of tableting compression force (21.0 kN) was covered by the model. In addition, our experimental matrix was enriched by three random values of tableting compression force chosen from the investigated area (one value for each granulate) to produce samples to obtain more data for analysis (Table 5).

Regression
The values of tested input parameters and achieved responses of resistance to crushing, disintegration time, and thickness of uncoated tablets are shown in Table 5.
Based on collected data, mathematical equations presenting relations among them were computed (Equations 4, 5, and 6). Residuals analysis of variance and its histogram with calculated Shapiro-Wilk statistic is shown in the supplementary materials ( Figures S3 and S4).
Resistance to crushing ¼ À 7:428Ã A þ 11:161 Ã B þ 49:301; Taking into account predefined constraints joint 2D contour plot was drawn by overlaying plots for each of tested responses (Figure 4). It is a graphical presentation of process parameters values permitted to use during production to obtain tablets whose quality will be in line with the defined limits.
It is not surprising that disintegration time depends only on the applied tableting compression force. Resistance to crushing strongly depends on both slugging and tableting compression forces. Thus, it is recommended that in further routine production the hardness of slugs (slugging compression force) should be restricted to specific values to diminish the variability of applied tableting compression force. In other words, by using the same hardness of slugs in the process we increase the robustness of tableting process. Although virtually we can produce tablets using high-and low-process parameters values, the lower setups will be recommended. Acceptable thickness ranges are so wide that the  tablets with valid resistance to crushing and disintegration time also present proper thickness. Both approaches, i.e. DOE and regression led to the establishment of a similar design space connecting process parameters and CQAs. However, the shape of design space borders is slightly different due to different methods of statistical analysis, which were based on altered number of data and ranges.

Models assessment
During the production of 23 commercial batches divided into seven campaigns the correctness of estimations and applicability of the developed predictive models were tested. Based on the independent variables values, responses were estimated for each of the compressed batches. Additionally, the differences between observed and predicted values and prediction error (PE%) were calculated ( The ranges of slugging and compression forces were limited to 5.0-8.0 and 14.0-16.7 kN, respectively. The reason for doing so was the fact that it was the production of commercial batches. Thus, we have sought to obtain tablets complying with the intermediate product requirements. Our superior intention was to obtain tablets whose hardness would be close to 170 N so as to maintain the value of this parameter on the same or similar level during further film-coating step. Resistance to crushing predictive models showed the following ranges of PE: À0.7 Ä 10.2% and À1.6 Ä 8.8% for the narrower and wider range of tableting compression forces, respectively. High values of R 2 adj coefficients (.99 in both models describing resistance to crushing) indicate that the variability in response (resistance to crushing) to major extent should be explained by the changes of the process parameters values included in our predictive equations. However, after performing the analysis of data collected during the production of 23 batches divided into seven campaigns, we concluded that relative error of prediction is higher than expected. It indicates that not all the factors affecting the response were taken into account. The source of higher deviations than it would have appeared from the properties of predictive models should be seen in the fact that the evaluated data come from the production that was performed in campaigns with a few months of space between each other. This kind of approach entails some consequences, e.g. each stage of production was carried out by employees who were randomly assigned a task to complete, each device was assembled separately for production and after completed step was disassembled, annual calibration of measuring devices took place in the meantime, the quantity of slug inside the roto-sieve and its speed (wide limits of rotation speed of impeller) was not strictly controlled, which may affect the granulate PSD, etc.
However, the most likely explanation for the existence of the differences between observed and predicted values is raw material variability [16]. The flow and order of used raw materials not only in pharmaceutical, but also in other industries, is organized according to the rule described as 'first in, first out'. It means that among many batches of the same raw material the oldest one in stock is assigned for the production of a particular batch. In consequence, the batches produced in the beginning of the year may be manufactured from different batches of raw materials than these at the end. We assume that focusing on the compression of slugs of the same hardness, which then will be used for the production of granules, would provide less variability into the process. The variability of raw materials among different batches will be offset by adjusting the pressure applied during the slugging so as to obtain slugs of tightly specified hardness. The issue of raw materials influence on process robustness will be the scope of our future research.
Nevertheless, the knowledge gained about tableting compression force required to obtain tablets of desired resistance to crushing in relation to the slugging compression force used indicates that slugging compression force and related slug hardness are critical parameters and require special control. Due to this fact, we decided to exert strict control of this parameter by narrowing the acceptance limit of slug resistance to crushing in order to ensure stability and robustness of the process on the subsequent production steps.
Disintegration time models present the following PE values: from 1.3 to 36.5% and from À21.9 to 21.2% for the narrower and wider range of tableting compression forces, respectively. Additionally, the PD values are correlated with resistance to crushing ones. R 2 value is .57 and .56 for DOE and regression models, respectively. It means that the source of prediction error may have the same origin as described above. However, data collecting of  disintegration time is the main source of error because it is based on visual evaluation whether tablets disintegrated or not. To mitigate the importance of this factor it is necessary to implement the method of disintegration time measurement that is not related to the human evaluation, but performed automatically. Both thickness models present small PE value (from À1.4 to 0.6% and from À1.5 to 0.6% for the narrower and wider range of tableting compression forces, respectively), which makes them an accurate tool to predict tablets thickness.

Influence of coating variables on CQAs
The Plackett-Burman screening design was employed to identify the main factors affecting the CQAs (resistance to crushing and disintegration time of coated tablets) and to find process parameters that should be controlled during coating. Out of financial reasons, all runs according to the Plackett-Burman screening design were performed on small laboratory scale batches. All values of input and output are presented in Table 7.
The power of effects is presented by Pareto charts ( Figure 5). The analysis of coating process parameters that influence tablets' hardness revealed three significant factors: inlet air temperature, spray rate, and atomization air pressure. The most powerful is the first of the mentioned. The positive value of the effect suggests that with an increase of inlet air temperature tablets tend to be harder. The second effect, that is spray rate, acts negatively. With an increase in spray rate, the strength of tablets is expected to be diminished. Both factors influence the coating environment providing more or less humid conditions inside the drum.
Our findings are in agreement with several studies, where high inlet air temperatures and low spray rates resulted in tablets of higher resistance to crushing, which is related to high water removal efficiency and lack of overwetting [17][18][19]. None of the investigated process parameters significantly affect disintegration time.
The data and knowledge collected in the preliminary screening study, performed on laboratory batches, became the basis of our decision to coat pilot scale batches using three different approaches (Table 3) to find out whether the observations made at laboratory scale would also be valid for pilot size that is 25-fold bigger. The first batch (I) was coated using intermittent spray pattern. It means that after 2 min and 30 s of spraying, 20 s of drying was applied. The coating lasted until proper weight gain was achieved ( Figure S5 supplementary materials depicts spray rate as a function of time used during our trial.) Despite drying breaks applied during spraying, the coating of batch I is the fastest process among those carried out in the study due to high spray rate. Compared to the second batch (II) and the third (III), coating conditions inside the drum result in the most intensive wetting of the first batch.
Box plots present resistance to crushing values of individual tablets before and after coating ( Figure 6). The top and bottom of each box represent the 25th and 75th percentile of the samples, respectively. The square in the middle of the box shows the sample median that was calculated based on the sample of 20 tablets. The whiskers, i.e. the lines extending above and below each box, represent the non-outlier range. It is a range below the upper and above the lower outlier limits calculated as a 1.5 of box height.  All outliers are marked as circles. Batch I showed the highest difference of resistance to crushing between uncoated and coated tablets. This observation is connected with the most humid conditions and water influence on cores. Lesser drop in hardness was detected in the case of batch II. A small increase in hardness was reported in the case of batch III. The last trial can be considered as the longest coating, however, guaranteeing the driest coating conditions.
It can be concluded that the difference in hardness between cores and coated tablets is connected with the coating conditions, namely inlet air temperature and spray rate. In our case, the most significant factors affecting the coating process determined with the screening design experiment at laboratory scale were confirmed at pilot scale. Additionally, it was shown that proper adjustment of coating conditions may influence tablets' CQAs. As it is depicted in the Figure 6, in the case of batch I resistance to crushing was out of the limit. Only by changing process parameters as it was done in the case of batch II and III, the tablets' hardness can be kept within the previously defined limits.

Conclusion
The present case study demonstrates the successful optimization of tablets production from granulates manufactured by dry granulation process (slugging) and identification of significant process parameters affecting CQAs at coating step. The first part shows the application of 3 2 full factorial design and regression methods to describe relations between two process variables (slugging and tableting compression force) and each of uncoated tablets CQAs (resistance to crushing, disintegration time, and thickness) by means of mathematical equations. These outputs were overlaid together with defined constraints to create design spacethe range of process parameters that guarantees that quality attributes of uncoated tablets will be within predefined ranges. The applicability of developed models was confirmed during the manufacturing of commercial batches. However, higher variability of results can be expected because not all of the factors may be included into prediction models. The second part contains the application of Plackett-Burman screening design at the laboratory scale to identify the coating process factors that may have impact on the CQAs of final dosage form. Inlet air temperature and spray rate were the most significant parameters that influence resistance to crushing. By coating pilot scale batches, it was confirmed that 'dry conditions' (lower spray rate and higher inlet air temperature, which means longer coating process) cause that tablets will present tendency to increase their resistance to crushing compared to uncoated tablets. Simultaneously, 'wet conditions' cause the decrease in tablets hardness. None of coating process parameters was detected to influence the disintegration time. Thus, it was demonstrated how proper adjustment of coating conditions bears on quality attributes of final dosage form. Therefore, the identification of CPPs at the development stage of the formulation in the laboratory with the use of statistical methods is justified, as it increases the understanding of the process and minimalizes the risk of failure during scale-up.
The results of our work were used to optimize the manufacturing process and have been successfully implemented for routine production.