Quality risk assessment and mitigation of pharmaceutical continuous manufacturing using flowsheet modeling approach

Quality risk assessment and mitigation of pharmaceutical continuous manufacturing using flowsheet modeling approach

Computers and Chemical Engineering 0 0 0 (2019) 106508 Contents lists available at ScienceDirect Computers and Chemical Engineering journal homepage...

3MB Sizes 0 Downloads 4 Views

Computers and Chemical Engineering 0 0 0 (2019) 106508

Contents lists available at ScienceDirect

Computers and Chemical Engineering journal homepage: www.elsevier.com/locate/compchemeng

Quality risk assessment and mitigation of pharmaceutical continuous manufacturing using flowsheet modeling approach Geng Tian∗, Abdollah Koolivand, Nilou S. Arden, Sau Lee, Thomas F. O’Connor∗ Office of Pharmaceutical Quality, Center for Drug Evaluation Research, Food and Drug Administration, Silver Spring, MD, USA

a r t i c l e

i n f o

Article history: Received 28 January 2019 Revised 17 April 2019 Accepted 30 June 2019 Available online xxx Keywords: Continuous pharmaceutical manufacturing Direct compression Sensitivity analysis Flowsheet modeling Residence time distribution

a b s t r a c t Integrated flowsheet modeling is an engineering approach that can provide a framework for understanding the impact of process dynamics on drug quality and associated risks during production, thereby facilitating the development of robust continuous processes. In this investigation, flowsheet modeling of continuous direct tablet compression process is exploited to perform sensitivity analysis in the risk assessment and identify potential process parameters and material attributes that affect critical quality attributes of the tablet. In addition, the dynamic response of the model in the presence of a variety of flow disturbances is demonstrated by the use of the residence time distribution analysis for the risk mitigation of the high-risk areas in the continuous direct compression process, which result in out-of-specification products. The outcome of the present work shows the potential of process models in quality risk assessment and mitigation for continuous pharmaceutical manufacturing of drug products and scientific considerations regarding model development. © 2019 Published by Elsevier Ltd.

1. Introduction Continuous manufacturing (CM) is an emerging technology that offers several advantages over the traditional batch processes, including production rate flexibility, quality, robustness, and economic improvements (Badman and Trout, 2015; Lee et al., 2015; Nasr et al., 2017). Major pharmaceutical and equipment manufacturing companies across the world are embracing and integrating CM into their technology portfolios for both legacy and new drug products (Markarian, 2018). Even though the stage appears to be set for increasing use of continuous pharmaceutical for manufacturing tablet products, challenges remain in the implementation of continuous manufacturing technologies for such applications. These challenges include, for example, integrating the continuous manufacturing system and understanding disturbance propagations throughout the integrated system in relation to material properties, process conditions and equipment design. Moreover, continuous pharmaceutical processes are designed to operate under a state of control in which a set of controls consistently provide assurance of continuous process performance and product quality (ICH, 2009). Thus, proper determination of nominal operating conditions requires a thorough understanding of the risks to product quality and the implementation of risk mitigation controls.

Corresponding authors. E-mail addresses: [email protected] (G. Tian), [email protected] (T.F. O’Connor). https://doi.org/10.1016/j.compchemeng.2019.06.033 0098-1354/© 2019 Published by Elsevier Ltd.

Quality risk management is a broad system of considerations and practices, applied across the drug product lifecycle, that encompass the assessment, control, and communication of risks to product quality. Risk assessment consists of the identification of hazards and the analysis and evaluation of risks associated with exposure to those hazards. Hazards identified during the risk assessment should be controlled. Risk control focuses on approaches for mitigating or avoiding a quality risk when it exceeds an acceptable level. Risk communication is the sharing of information about risk and risk management between the decision makers and other stakeholders. Various types of models can be used to support quality risk management. In many cases, process models can identify potential hazards by estimating the impact of potential variations in the process, incoming raw materials or equipment conditions (i.e. model inputs) on product quality attributes (i.e. model outputs). The control strategy adopted to mitigate the identified hazards can be incorporated into the process model and the impact of potential variations in model inputs can be re-examined, enabling an evaluation of the control strategy’s effectiveness in ensuring product quality. Furthermore, process models can provide a formal structure for capturing process knowledge and assumptions, facilitating risk communication among various stakeholders, managing prior knowledge for its use in other product and process development. Sensitivity analysis is the study of how the variability in the output of a model can be apportioned to different sources of vari-


G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508

ability in the model inputs. Sensitivity analysis can be a tool for performing quantitative risk assessments as it evaluates the relationships between process parameters, material attributes and product quality attributes described by the model. In the application of sensitivity analysis, model parameters can be treated as model inputs along with process parameters and material attributes. Sensitivity analysis has been utilized for continuous manufacturing of pharmaceutical drugs to unravel the associated nonlinearities and interactions in integrated processes, to facilitate the development of control strategies, and to ensure product quality by identifying important input factors and reducing the dimension of the model by screening out less important input factors (Boukouvala et al., 2012; Rogers et al., 2013; Wang et al., 2017). Sensitivity Analysis could be used to examine the impact of parameter uncertainties on model predictions. The presence of parameter uncertainties caused by various sources such as measurement noises can propagate along integrated processes and alternate product quality. The range for the parameter uncertainties could be the confidence interval considered during parameter estimation. Thus, it is of interest to quantify how an uncertainty in the product quality is related to process parameter’s uncertainty. One common approach to incorporate the uncertainty effects in model development is to first determine the nominal value for the parameters and then set them as the mean of normal distributions, while the standard deviations represent the parameter’s uncertainties (Boukouvala et al., 2012a). In many cases, flowsheet modeling is computationally expensive and thus, reduced order models, i.e. surrogate models, are alternatively used to explore the uncertain parameter space and feasibility region (Wang et al., 2017; Wang & Ierapetritou, 2017). Mitigating potential hazards and process disturbances identified during the risk assessment is an important component in the development of an appropriate control strategy. Control strategy implementations can consist of establishing ranges for material attributes and process parameters that provide assurance of product quality. Given the dynamic nature of the continuous manufacturing process, it is important to consider the amplitude and duration of any upstream disturbances. Process models can be a useful tool to evaluate the ability of parameter limits that account for the amplitude and duration of the disturbance to detect the production of out-of-specification drug. In the present study, we develop an integrated process model for a common continuous drug product manufacturing process and demonstrate how the model can be used to support quality risk assessment. An integrated process model for a continuous manufacturing process is typically labeled flowsheet model as the flow of information between the unit operation models, resembles the flow of material(s) between unit operations. Flowsheet models have been recently developed for continuous pharmaceutical manufacturing processes and have been used as a tool for process design, optimization, risk assessment, control strategy development and analysis, and monitoring of a continuous pharmaceutical process (Barrasso and Bermingham, 2018; Boukouvala et al., 2013; Garcia-Munoz et al., 2018; Rogers et al., 2013; Sen et al., 2013; Singh et al., 2015; Singh et al., 2014, Singh et al., 2014). Flowsheet models have been shown to effectively capture integrated process dynamics (Boukouvala et al., 2012, 2011; Sen et al., 2013; Wang et al., 2017). For example, process models that are based on the residence time distribution (RTD) can be used to examine the process’s ability to mitigate the impact of disturbances (i.e. variation in the feeder’s flowrate). RTD-based process models have mostly been used as a tool for material tracking (Bhaskar and Singh, 2018; Gao et al., 2011; Kruisz et al., 2018; Manley and Shi, 2018; Martinetz et al., 2018). Tian et al. (2017) and GarciaMunoz et al. (2018) have demonstrated the use of RTD process models in the risk assessment and evaluation of control strategies.

To demonstrate how flowsheet models capture the interconnected effects of process units during continuous operation, how sensitivity analysis can be utilized for identifying critical process parameters (CPPs) and critical material attributes (CMAs), and how models that describe the process dynamics (e.g. RTD models) can be used to evaluate risk mitigation strategies, a case study has been conducted. The case study considers quality risk assessments of a continuous direct compression (CDC) process for a pharmaceutical tablet. Direct compression is a common process requiring less equipment and is less labor intensive in comparison to traditional batch systems (Çelik, 2016). The remainder of the article is organized as follows. First, a brief description of a CDC process is presented. The model including integrated unit operations is validated with a set of experimental data. After model validation, the Morris and Sobol methods are utilized to perform sensitivity analysis to support the quality risk assessment of the process. The sensitivity analysis illustrates the effects of operating variable and material properties on API concentration, tablet hardness, and thickness. Finally, the process model is used to simulate the feeder disturbances and to evaluate control limits intended to mitigate the impact of these disturbances on product quality. A summary of results that provides insights for quality risk assessment and mitigation of a CDC process for a pharmaceutical tablet is presented as conclusions. 2. Process description and mathematical formulation A flowsheet model is an approximate representation of an actual plant operation, which supports the risk analysis, failure mode, and performance evaluation of a continuous manufacturing system (Boukouvala et al., 2012, 2013 Rogers et al., 2013; Schwier et al., 2010). The first goal of this study is to develop a dynamic flowsheet model that considers process dynamics with detailed characterization of involved unit operations. The developed flowsheet model then leads to quantifying the impacts of variation in CPPs and CMAs on the critical quality attributes (CQAs) and capable of simulating a variety of different dynamic scenarios such as introducing disturbances into the process. The models used for each unit operation range from simple data-driven models to more complex first-principle based models, depending on available knowledge for each unit operation. In integrated flowsheet modeling of pharmaceutical processes, individual unit operation models are integrated by taking the results from a preceding unit and introducing them as the inputs of subsequent units (Ramachandran et al., 2011; Rogers et al., 2013, Inamdar; Vanarase and Muzzio, 2011). The present study considers a common and popular process in tablet manufacturing, i.e. CDC process, which includes several unit operations, i.e. raw material feeding, blending, and tablet compression as shown in Fig. 1. The process begins with constant feeding of individual raw granular materials (input materials) into the blender using loss-in-weight (LIW) feeders. A total of six LIW feeders are used to feed the input materials into two blenders connected in series and a tablet press. The feeders are operated by a twin screw to convey the material into the blenders using a gravimetric flowrate control scheme. The feeders are refilled volumetrically from the bulk containers via an intermediate hopper. Four of the six feeders feed the API and three excipients (filler, diluent, and disintegrant) to the first blender, while the remaining two feeders have the option of delivering the other two excipients directly to the second blender. Excipients 4 and 5 are sent directly to the second blender to avoid over lubricating the blend. The material exiting the second blender is sent directly to the tablet press where the blend is compacted into tablets. The composition of the formulation is not given due to the protection of proprietary information. In short, the excipient 1 and excipient 2 can take up to

G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508


where ωscrew and ωimpeller are the feeder screw speed and impeller speed, respectively, τ is the time constant for the screw speed, FRout is the flowrate at the outlet of the feeder, ff(t) is the feed factor, ρ effective is the effective density, Vscrewpitch is the volume of the screw pitch, σ V is the vertical stress, β is the density constant. Mblade and Rblade are the mass and radius of the impeller blade, Afeeder and Ascrewport are the area of the feeder and the area of the port where screws enter the bowl, g is the gravity constant, W is the powder weight in the feeder and α is the impeller ratio between the impeller speed and the screw speed. 2.2. Blender Two continuous blenders are used to mix the materials and homogenize the composition fluctuations caused by the feeders through axial dispersion. The corresponding model captures mass balance, mixing, and residence time distribution (RTD) in the blenders. The RTD represents the amount of time a representative solid particle could spend inside the unit operation. The RTD models are capable of characterizing the non-ideality in material mixing within the blender (Gao et al., 2011). The RTD depends upon the design and operating parameters of the blender as well as the physiochemical properties of the powder blend (Gao et al., 2012; Marikh et al., 2008, 2005; Portillo et al., 2009, 2010). It is typically determined experimentally by injection of a tracer pulse into the blender inlet and monitoring the outlet concentration. For simplicity and adequacy, a series of continuous stirred tank reactor (CSTR) model is employed to predict measured experimental RTD as follows:

Fig. 1. Schematic of process flow diagram for the current CDC tablet.

90% of the tablet and the weight fraction of excipient 3, excipient 4, and excipient 5 is less than 10%. In what follows, a detailed description of involved unit operations along with corresponding mathematical formulation required for the flowsheet modeling is briefly presented. More detail on the development of these models can be found elsewhere (Boukouvala et al., 2012; Rogers et al., 2013; Wang et al., 2017) 2.1. Loss-in-weight feeder A LIW feeder consists of a hopper that holds a certain amount of powder material and an adjustable-speed rotating screw which feeds the powder to the blender. The mass flowrate setpoint of a LIW feeder is controlled based on the weight of the material remaining in the feeder at any time and by manipulating the screw speed. A simplified semi-empirical first-order model is utilized for modeling the feeder that accurately predicts the relationship between screw speed and mass flowrate of the feeder. The corresponding equations for the LIW feeder module are given as follows:


dωscrew (t ) F Rout + ωscrew (t ) = dt f f (t )

f f (t ) = ρe f f ective (t )Vscrewpitch

ρe f f ective (t ) = ρsat + e σV (t ) =

−σV (t )

(1) (2)

(ρini − ρsat )


  Mblade Rblade ωimpel l er 2 W (t )g + cos 2π ωimpel l er t A f eeder Ascrewport



dW (t ) = F Rin − F Rout dt


ωimpel l er (t ) = αωscrew (t )


E (t ) =

(t − t0 )n−1 ( −n(tτ¯−t 0) ) ne (n − 1 )!( τn¯ )


where t0 is the time delay, τ¯ is the blender mean residence time, and n, the number of CSTRs are determined by fitting the model to the experimental data. Accordingly, the blender outlet concentration can be determined by the following convolution formulation:

Cout (t ) =

t 0


Cin t − t˜ E t˜ dt˜


The total flowrate is determined by considering the following equations:

F Rtotal out =


F Rtotal − in

dM (t ) dt

dM (t ) + M (t ) = Mss dt



where M and Mss are the dynamic and steady-state total mass holdup in the blender. 2.3. Tablet press A tablet press consists of an integrated hopper and feed frame. The hopper guides the powder blend into the press and the feed frame moves the blend into the die that defines the area and shape of the tablet. A high-pressure punch compresses the powder into a tablet, which is then discharged via cam tracks. Poor tablet hardness and variability in the tablet weight are linked to the tablet press performance, which itself is significantly affected by the powder properties including bulk density, compressibility and flowability (Kuentz and Leuenberger, 20 0 0; Mehrotra et al., 2009). Thus, to capture these effects, empirical correlations between powder properties and the resulting hardness properties of the tablets are utilized to predict tablet thickness and hardness (Gentis and Betz, 2012; Patel et al., 2007).


G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508

Tablet thickness is commonly calculated as a function of precompression and main compression forces, initial porosity of the powder, powder fill depth in the die, and the tablet area. Tablet hardness can be calculated as a function of maximum hardness, relative density and critical relative density (Kawakita and Lüdde, 1971; Kuentz and Leuenberger, 20 0 0). A combination of two empirical models, i.e. Kuentz and Leuenberger model (Kuentz and Leuenberger, 20 0 0) which describes the indentation hardness as a function of relative density, and the Kawakita model (Kawakita and Lüdde, 1971) that considers a correlation between compression pressure and powder volume are used as follows (Singh et al., 2010):

Pcomp =

b(Vprecomp − V ) V − Vs

Pprecomp =


b(V0 − Vprecomp ) Vprecomp − Vs


  1 − ρ  r ρr − ρcr + ln 1 − ρcr

H = Hmax 1 − exp


2 Vprecomp = L precomp π Dtablet /4


2 V0 = Ldie π Dtablet /4


2 V = Ltablet π Dtablet /4



Vs = V


Vs = (1 − a )V0


where H and Hmax are the tablet hardness and its maximum hardness, ρ r is relative density and ρ cr is the critical relative density, b is the kawakita parameter, V0 , Vprecomp , Vs, and V are the die feed volume, pre-compression volume, solid volume, and tablet volume, respectively. Ltablet , Lprecomp , and Ldie are the tablet thickness, pre-compression thickness, and die feed fill depth, respectively. Dtablet is the tablet diameter, a is the initial porosity of the material, Pcomp is the main compression pressure, and Pprecomp is the pre-compression pressure. Maximum hardness and critical relative density are material properties (given constant values estimated from the experimental data), while the relative density depends on the initial porosity of the powder, fill depth in the die, area and thickness of the tablet. Finally, it is of interest to briefly discuss the approach for parameter estimation and model validation. A least-square based regression method is considered for the parameter estimation of each unit operation and the following objective function is minimized with 95% confidence interval:

min O.F. =


yˆk − yk




where yˆk and yk represent the model responses and corresponding experimental data, respectively. Additionally, standard error of residual (SER), and similarity factor f2 are considered as additional metrics for evaluation of estimation. The following conditions are considered as criteria of good estimation:



(k=1 )

f2 = 50 · log10

yˆk − yk


⎧ ⎨ ⎩


1 n


where n and m represent total number of data points and number of model parameters, respectively. The gPROMS parameter estimation feature and customized MATLAB scripts were utilized for the determination of parameters. For model validation, additional experimental data, which was not used during the model calibration, was compared to the output responses of each unit operation models as presented in the next section. 3. Flowsheet modeling: simulation validation A dynamic flowsheet model consists of integrated unit operations described in Section 2 is developed within gPROMS, Process Systems Enterprise software. In flowsheet modeling, individual unit operation models are combined by taking the results from a preceding model and using it as the inputs of a subsequent one. A combination of unit operations available in the gPROMS platform and custom-built models coded via gPROMS language are utilized for the development of the flowsheet model. MATLAB was used for post processing of the data obtained from the simulations. The built-in Global Sensitivity Analysis of gPROMS has been applied to the model for the sensitivity analysis. The comparison of experimental and model predicted results of each unit operation is provided next. 3.1. Feeding process The feed factor shows a characteristic profile as a function of time which is well predicted by the model. The feed factor increases by refilling the material and decreases as the material in the feeder decreases, requiring an increase in the screw speed to maintain the target flowrate. The flowrate generally remains constant while a small variation due to refills can be observed within a specified acceptable range. A good agreement between the model prediction results and the experimental data is observed as shown in Fig. 2, where the impact of refill on the API mass flowrate, feed factor, net weight of API, and feeder screw speed are presented. 3.2. Continuous powder blending process In this case study, a near-infrared spectroscopy (NIR) probe was employed to measure local blend uniformity at the discharge of the second blender. The NIR technique is a non-destructive tool that can be used for in-line monitoring and control of tablet process. The experimental RTD is measured for the series of two blenders by introducing a pulse of tracer material at the entrance of the first blender and measuring the response using an in-line NIR at the second blender discharge. The total RTD representing the RTD of the entire line including two blenders and the tablet press is measured in a similar way. The RTDs representing blender 1 and blender 2 are convoluted to calculate the RTD of two blenders. As shown in Fig. 3(a,b), the predicted RTDs are in good agreement with the experimental concentration data before and after entering the tablet press, respectively. The mean API concentration at the discharge of the second blender is then calculated by convolution of total RTD and frequency variability of the API concentration at the feeder outlet, as shown in Fig. 3c. 3.3. Tablet compression process

≈0 100 n  k=1


⎫ ⎬

yˆk − yk

2 ⎭ ≈ 100


A direct result of variation in compression forces is the variation in the tablet hardness and thickness. Using the tablet press model and the experimental tablet hardness data, the Kawakita constant, maximum hardness, and critical relative density are simultaneously calculated via optimization. As shown in Fig. 4, an overall

G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508


Fig. 2. Steady-state operation of a feeder over time; predicted API mass flowrate, feed factor, net weight, and screw speed.

Fig. 3. Comparison of experimental and predicted RTD models at (a) discharge of the blender 2, (b) after the tablet press, and (c) predicted mean API concentration at the discharge of the blender 2.

Fig. 4. Comparison of experimental and model predicted results for (a) tablet thickness and (b) tablet hardness.

good agreement between the model-predicted results and the experimental measurements is found, except for the experiment #8 where the highest tablet hardness was measured. The predicted value of tablet hardness for the experiment #8 is observed to be similar to those predicted for the experiments #7, #10, #11, and #13 under the same main compression force, while the precompression force varies among those experiments. The results suggest that the hardness of this tablet may be more sensitive to change in the main compression force than the pre-compression force. 4. Sensitivity analysis: a risk assessment tool For the sensitivity analysis, a total of 28 inputs are varied within the specified range. It should be noted that the mass fraction of excipients 3–5 are small and close to each other. Therefore, to reduce the computational cost of the sensitivity analysis, the excipi-

ent 4 is considered as the representative of the two other feeders (the excipients 3 and 5). The input factors are considered to vary within ±10% of their nominal values for consistency with other studies available in the literature (Rogers et al., 2013; Wang et al., 2017), and the overall integrated process remains close to its nominal steady-state condition. The output factors of interest include material properties such as mean bulk density, operation parameters such as mass holdup, mixing characterization such as mean residence time, and tablet product qualities such as composition. The 44 outputs of interest are the final properties of the produced tablets as well as intermediate product properties. A summary of the varied input parameters with uniform or normal distributions and the output responses is listed in Table 1. The Morris and Sobol methods are utilized for the sensitivity analysis. A detailed description of these methods can be found elsewhere, e.g. (Morris, 1991; Saltelli et al., 2010; Saltelli et al., 2009; Saltelli et al., 2000; Saltelli et al., 2004; Sobol, 1993).


G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508

Table 1 Input factors, their range of variation and the output responses used in sensitivity analysis.

The provided density values represent the order of magnitude for each material. Actual values are masked due to protection of proprietary information.

G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508


Fig. 5. Intensity plots representing steady-state sensitivity analyses that capture the effects of input factors on the output responses for (a) mean elementary effect determined by the Morris method and (b) the total effect determined by Sobol Method.

Briefly, the Morris method is based on a global “one-at-a-time” method, where the elementary effect is calculated with only one input factor perturbation at a time. It is an effective way of screening important input factors with a small sampling cost (Morris, 1991; Saltelli et al., 2009). The mean elementary effect represents the significance of an input factor on the output response. The utilized Sobol method is a variance-based Monte Carlo method, which accounts for the total contribution to the output variation due to an input factor, i.e. its first-order effect plus all higher-order effects due to interactions. A comparison between the results of the Morris and the Sobol method in shown in Fig. 5. The good agreement observed between the Morris and Sobol sensitivity analysis indicate that the results are consistent and sufficient for the purpose of identifying critical model inputs for this case study. The column bar provided on the right-hand side of each plot in Fig. 5 is a metric that quantifies the effects of different input factors on a given output response. The input factors with values close to unity (dark red) are more important than the ones with values close to zero (light red). Thus, one may exclude the less important input factors in subsequent reduced-order model development. The drug product in the case study has a low API content and thus factors that impact assay and content uniformity are a focus area for the quality risk assessment. As shown in Fig. 5, the flowrates of excipients 1 and 2 in addition to the API flowrate strongly affect the blend composition. Tablet weight is a strong function of the material bulk densities, their flowrate and the tablet die fill depth. Furthermore, the blender speed and throughput directly affect the process dynamics as evidenced by the impact on the mean residence time of the blend materials, fill level, mass holdup and outlet flowrate. It is also crucial to identify the important parameters that affect the feed factor since the feed factor is used by the controller during refill cycle to effectively adjust the motor speed to counterbalance the effects of density increase. Fig. 5 shows that the sweeping volume of screw pitch and the bowl area for each feeder highly affect its corresponding feed factor. The API bulk density also shows strong effects on the feeder’s feed factor. The bulk density of excipients 1 and 2 are important material attributes that affects the tablet quality attributes (e.g. tablet density) as well as the intermediate properties such as mean bulk density of the blend material at the blender outlet and the blender fill level. Material properties of the blends such as bulk density highly

depend on their particle size distributions, and thus, stock variation due to different suppliers should be minimized. In summary, the material properties of the API and the excipients, operating parameters including flowrates and blenders’ rpm, and design parameters such as the tablet die fill depth and main compression force are shown to be the most influential factors in the current CDC process. The process is relatively insensitive to several designs and operating parameters, indicating that either they should be excluded from subsequent analyses or the model needs improvement by introducing new experimental data. For instance, minimal sensitivity to excipient 4 feeder parameters is observed in the existing flowsheet. However, it has been experimentally demonstrated that excipient 4 concentrations can affect powder flowability and tablet properties such as hardness. To account for the effects of excipient 4 on tablet hardness, a modified hardness model that considers a correlation between the maximum hardness and tablet composition may be developed empirically and added to the existing tablet press model. It is worth to emphasize on the fact that the flowrates of the API and excipient are the main input factors that impact the content uniformity and the API dosage in the low-dose tablet that is the subject of the current case study. Thus, it is important to appropriately understand and determine the acceptable range of disturbances that does not cause out-of-specification drug tablet, and subsequently design an effective strategy to control the process and minimize the need for end-product testing. Thus, a question that should be addressed is the effects of flow disturbances on the content uniformity of the tablet. Below, we discuss the dynamic response of the system on the quality attributes of the tablet for a variety of disturbances at different flowrates. The present study focuses on the determination of critical process parameters and material attributes that significantly affect the critical quality attributes of low dosage tablets. Thus, the application of sensitivity analysis is limited to identifying potential areas that may lead to quality risks during continuous tablet manufacturing. Notably, the cases where the flowsheet modeling is computationally expensive, the outcome of sensitivity analysis can be used for ranking the input factors, and the development of reducedorder models that facilitates feasibility and flowsheet optimization studies.


G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508

Fig. 6. Impact of feeding disturbances on the final concentration of the API in the tablet as a function of time; disturbance in the feeder flowrate of (a) API, (b) excipient 1, (c) excipient 2, (d) excipient 3, (e) excipient 4, and (f) excipient 5.

5. RTD models: a tool for the evaluation of risk mitigation controls Parameter ranges such as feeding limits can be established as a component of the controls intended to ensure product quality. In continuous manufacturing the parameter range limits can consider the combination of the parameter value as well as the duration of the deviation from the target value. The predictive capability of the current flowsheet model provides informative insights for when the feeding disturbances are introduced to the system. The RTD models can be used to evaluate the impact of a disturbance of certain duration in different feeders on the API concentration after the tablet press (Tian et al., 2017). Fig. 6 illustrates the out-ofspecification blend region as a function of disturbance magnitude and duration. Note that two different RTD models are used for this calculation. The funnel plots corresponding to the excipients 4 and 5, only includes the RTD of 2nd blender, whereas the RTD corresponding to the rest materials (i.e. API and the excipients 1, 2, 3) includes both blenders. It is important to note that the domain-space presented in Fig. 6 only reflects the combinatorial effects of mass holdup, residence time, and flowrate of the components on the API dose in the tablet and does not consider effects of composition on the operability of the process. Any change in the flowrate of API feeder directly affects the final API dosage in the tablet. As a result, introducing a small disturbance on the API feeder as shown in Fig. 6a, even for a short duration (e.g. API feeder with 70% LC upset for 60 s) causes a tablet with out-of-specification API dosage (<95%). A large fraction of the final tablet is made of excipient 1 and excipient 2. Thus, as shown in Fig. 6(b, c), feed disturbances with high amplitude in the

flowrate of excipient 1 or 2 for a short duration of time or low magnitude for a long duration causes an out-of-specification tablet. On the other hand, owing to smaller weight fractions of excipients 3,4, and 5, the API dose of the tablet remains in the pre-specified domain (95% to 105%) even for long disturbances with high amplitude (e.g. 200% LC upset for 300 s), as shown in Fig. 6(a,b,c). Moreover, it is worth to note that although the existence of a disturbance in the flowrate of one feeder may not cause out-ofspecification tablet production, the cumulative effects of flow disturbances in multiple feeders should be taken into the consideration, especially for designing an in-process control strategy. Fig. 7 examines the impact of multiple flow disturbances in excipient 1 and 2 feeders. It is of interest to consider the feeders interactions on the tablet composition and design a flow rate control strategy that take into consideration such cumulative effects rather than designing fully-decoupled control-loop for each individual feeder. In the case of individual control of each feeder, the overall process may maintain in the state of control, while the produced tablet has an out-of-specification API dosage. For example, when an excipient 1 feeder disturbance of +20% for 100 s does not result in out-ofspecification tablet as shown in Fig. 6(b), the addition of +20% increase of excipient 2 feeder flowrate causes an out of specification API dosage as can be seen from Fig. 7. In another case, a decrease in the flowrate of one excipient is compensated by the flowrate increase in another excipient, and thus tablet may still contain an acceptable API dosage while other manufacturing downstream product quality attributes (i.e. tablet dissolution and disintegration) may be impacted. For instance, Fig. 6(b) suggests that an excipient 1 feeder disturbance of +50% for 100 s results in out-of-specification tablet. However, if this disturbance is accompanied with an excipient 2 feeder disturbance

G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508


significant factors that highly affect the process operability and tablet quality. Finally, the domain space of tablet with desired API dosage (95–105 LC%) in the presence of disturbances in the feeders’ flowrates are quantified. The current study provides a discussion on the potential of utilizing models in supporting quality risk assessment of continuous pharmaceutical manufacturing processes. The approach is versatile and can be implemented for other integrated CM processes. The outcome of the present work supports the development and utilization of process models for identifying uncaptured phenomena and reducing the knowledge gap between the theory and experiment in continuous manufacturing of drug substances and drug products. Disclaimer This article reflects the views of the author and should not be constructed to represent FDA’s views or policies. Acknowledgment Fig. 7. Impact of simultaneous disturbance in the flowrate of feeders 1 and 2 on the API in the tablet as a function of time.

of −40%, the blend composition still does not show any out-ofspecification API dosage. Therefore, single monitoring and control of the API concentration may not be adequate. In fact, to achieve a blend uniformity with proper downstream product quality attributes, it may be crucial to simultaneously monitor and control the API composition as well as all the excipients at the outlet of the blenders and the tablet press. 6. Considerations for using models to support quality risk assessment Models can be classified based on their contribution in assuring the quality of the product: low-impact, medium-impact, and highimpact models. Models used to support quality risk assessment are typically low impact. Models used as the primary evidence to justify control strategy limits may fall into the medium-impact classification. It is important to fully understand the assumptions underlying the individual models. If the physics related to a potential failure mode (e.g. over lubrication) is not included in the model, factors impacting this risk will not be identified by the sensitivity analysis. These risks may need to be investigated through experimental studies. Also it is important to completely understand the parameter space over which the models are validated. The sensitivity analysis should consider the validated parameter space for the models when selecting input ranges to be investigated. 7. Conclusions The current work considers the simulation of a CDC process using an integrated flowsheet model. The simulation results are compared against experimental data and good agreement is observed. Quality risk assessment and mitigation strategies based on sensitivity analysis and dynamic response of the model in the presence of flow disturbances are discussed. Two common sensitivity analysis approaches, namely the Morris and Sobol methods, are utilized to identify the key material attributes and process parameters of the CDC process. The results of both methods suggest that the API and the excipients density, their flowrates, the blenders’ rpm, tablet die fill depth, and main compression force are the most

We thank Dr. Marianthi Ierapetritou and Dr. Manuel Sebastian Escotet Espinoza from Rutgers University for assistance with the model development under the Cooperative Agreement (U01) FD005295-01. The second author would like to acknowledge the funding provided by the Oak Ridge Institute for Science and Education (ORISE) fellowship. References Badman, C., Trout, B.L., 2015. Achieving continuous manufacturing may 20–21 2014 continuous manufacturing symposium. J. Pharm. Sci. 104, 779–780. Barrasso, D., Bermingham, S., 2018. Design space characterization and risk assessment through mechanistic modeling. Pharm. Technol. 42, 46–49. Bhaskar, A., Singh, R., 2018. Residence time distribution (RTD)- based control system for continuous pharmaceutical manufacturing process. J. Pharm. Innov. 1– 16. doi:10.1007/s12247- 018- 9356- 7. Boukouvala, F., Chaudhury, A., Sen, M., Zhou, R., Mioduszewski, L., Ierapetritou, M.G., Ramachandran, R., 2013. Computer-aided flowsheet simulation of a pharmaceutical tablet manufacturing process incorporating wet granulation. J. Pharm. Innov. 8, 11–27. Boukouvala, F., Niotis, V., Ramachandran, R., Muzzio, F.J., Ierapetritou, M.G., 2012. An integrated approach for dynamic flowsheet modeling and sensitivity analysis of a continuous tablet manufacturing process. Comput. Chem. Eng. 42, 30–47. Boukouvala, F., Ramachandran, R., Vanarase, A., Muzzio, F.J., Ierapetritou, M.G., 2011. Computer aided design and analysis of continuous pharmaceutical manufacturing processes. In: Pistikopoulos, E.N., Georgiadis, M.C., Kokossis, A.C. (Eds.). In: Computer Aided Chemical Engineering, 29. Elsevier, pp. 216–220. Çelik, M., 2016. Pharmaceutical Powder Compaction Technology. CRC Press. Gao, Y., Muzzio, F.J., Ierapetritou, M.G., 2012. A review of the residence time distribution (RTD) applications in solid unit operations. Powder Technol. 228, 416–423. Gao, Y., Vanarase, A., Muzzio, F., Ierapetritou, M., 2011. Characterizing continuous powder mixing using residence time distribution. Chem. Eng. Sci. 66, 417–425. Garcia-Munoz, S., Slade, D., Butterbaugh, A., Leavesley, I., Francis Manley, L., Bermingham, S., 2018. A flowsheet model for the development of a continuous process for pharmaceutical tablets: An industrial perspective. AIChE J. 64, 511–525. Gentis, N.D., Betz, G., 2012. Compressibility of binary powder formulations: investigation and evaluation with compaction equations. J. Pharm. Sci. 101, 777–793. ICH, 2009. Guidance for industry: Q10 pharmaceutical quality system. https://www. fda.gov/media/71553/download. Kawakita, K., Lüdde, K.-H., 1971. Some considerations on powder compression equations. Powder Technol. 4, 61–68. Kruisz, J., Rehrl, J., Faulhammer, E., Witschnigg, A., Khinast, J.G., 2018. Material tracking in a continuous direct capsule-filling process via residence time distribution measurements. Int. J. Pharm. 550, 347–358. Kuentz, M., Leuenberger, H., 20 0 0. A new model for the hardness of a compacted particle system, applied to tablets of pharmaceutical polymers. Powder Technol. 111, 145–153. Lee, S.L., O’Connor, T.F., Yang, X., Cruz, C.N., Chatterjee, S., Madurawe, R.D., Moore, C.M., Yu, X.L., Woodcock, J., 2015. Modernizing pharmaceutical manufacturing: from batch to continuous production. J. Pharm. Innov. 10, 191–199. Manley, L., Shi, Z., 2018. Characterizing drug product continuous manufacturing residence time distributions of major/minor excipient step changes using near infrared spectroscopy and process parameters. Int. J. Pharm. 551, 60–66.


G. Tian, A. Koolivand and N.S. Arden et al. / Computers and Chemical Engineering 000 (2019) 106508

Marikh, K., Berthiaux, H., Gatumel, C., Mizonov, V., Barantseva, E., 2008. Influence of stirrer type on mixture homogeneity in continuous powder mixing: a model case and a pharmaceutical case. Chem. Eng. Res. Des. 86, 1027–1037. Marikh, K., Berthiaux, H., Mizonov, V., Barantseva, E., 2005. Experimental study of the stirring conditions taking place in a pilot plant continuous mixer of particulate solids. Powder Technol. 157, 138–143. Markarian, J., 2018. Modernizing pharma manufacturing. Pharm. Technol. 42, 20–25. Martinetz, M., Karttunen, A., Sacher, S., Wahl, P., Ketolainen, J., Khinast, J., Korhonen, O., 2018. RTD-based material tracking in a fully-continuous dry granulation tableting line. Int. J. Pharm. 547 (, 1–2 ), 469–479. Mehrotra, A., Chaudhuri, B., Faqih, A., Tomassone, M.S., Muzzio, F.J., 2009. A modeling approach for understanding effects of powder flow properties on tablet weight variability. Powder Technol 188, 295–300. Morris, M.D., 1991. Factorial sampling plans for preliminary computational experiments. Technometrics 33, 161–174. Nasr, M.M., Krumme, M., Matsuda, Y., Trout, B.L., Badman, C., Mascia, S., Cooney, C.L., Jensen, K.D., Florence, A., Johnston, C., Konstantinov, K., Lee, S.L., 2017. Regulatory perspectives on continuous pharmaceutical Manufacturing: moving from theory to Practice: september 26-27, 2016, international symposium on the continuous manufacturing of pharmaceuticals. J. Pharm. Sci. 106, 3199–3206. Patel, S., Kaushal, A., Bansal, A., 2007. Effect of particle size and compression force on compaction behavior and derived mathematical parameters of compressibility. Pharm. Res. 24, 111–124. Portillo, P.M., Ierapetritou, M.G., Muzzio, F.J., 2009. Effects of rotation rate, mixing angle, and cohesion in two continuous powder mixers—A statistical approach. Powder Technol. 194, 217–227. Portillo, P.M., Vanarase, A.U., Ingram, A., Seville, J.K., Ierapetritou, M.G., Muzzio, F.J., 2010. Investigation of the effect of impeller rotation rate, powder flow rate, and cohesion on powder flow behavior in a continuous blender using PEPT. Chem. Eng. Sci. 65, 5658–5668. Ramachandran, R., Arjunan, J., Chaudhury, A., Ierapetritou, M.G., 2011. Model-based control-loop performance of a continuous direct compaction process. J. Pharm. Innov. 6, 249–263. Rogers, A.J., Hashemi, A., Ierapetritou, M.G., 2013a. Modeling of particulate processes for the continuous manufacture of solid-based pharmaceutical dosage forms. Processes 1, 67–127. Rogers, A.J., Inamdar, C., Ierapetritou, M.G., 2013b. An integrated approach to simulation of pharmaceutical processes for solid drug manufacture. Ind. Eng. Chem. Res. 53, 5128–5147. Saltelli, A., Annoni, P., Azzini, I., Campolongo, F., Ratto, M., Tarantola, S., 2010. Variance based sensitivity analysis of model output. Design and estimator for the total sensitivity index. Comput. Phys. Commun. 181, 259–270.

Saltelli, A., Chan, K., Scott, E.M., 2009. Sensitivity Analysis. Wiley. Saltelli, A., Tarantola, S., Campolongo, F., 20 0 0. Sensitivity anaysis as an ingredient of modeling. Stat. Sci. 15, 377–395. Saltelli, A., Tarantola, S., Campolongo, F., Ratto, M., 2004. Sensitivity Analysis in Practice: A Guide to Assessing Scientific Models. John wiley & sons. Schwier, D., Hartge, E.-U., Werther, J., Gruhn, G., 2010. Global sensitivity analysis in the flowsheet simulation of solids processes. Chem. Eng. Process. Process Intensif. 49, 9–21. Sen, M., Chaudhury, A., Singh, R., John, J., Ramachandran, R., 2013a. Multi-scale flowsheet simulation of an integrated continuous purification–downstream pharmaceutical manufacturing process. Int. J. Pharm. 445, 29–38. Sen, M., Rogers, A., Singh, R., Chaudhury, A., John, J., Ierapetritou, M.G., Ramachandran, R., 2013b. Flowsheet optimization of an integrated continuous purification-processing pharmaceutical manufacturing operation. Chem. Eng. Sci. 102, 56–66. Singh, R., Escotet-Espinoza, M.S., Lee, S.L., Sen, M., O’Connor, T.F., Ierapetritou, M., Chatterjee, S., Ramachandran, R., Muzzio, F., 2015. Flowsheet models modernize pharmaceutical manufacturing design and risk assessment. Pharm. Technol. 39. Singh, R., Gernaey, K.V., Gani, R., 2010. ICAS-PAT: a software for design, analysis and validation of PAT systems. Comput. Chem. Eng. 34, 1108–1136. Singh, R., Sahay, A., Karry, K.M., Muzzio, F., Ierapetritou, M., Ramachandran, R., 2014a. Implementation of an advanced hybrid MPC–PID control system using PAT tools into a direct compaction continuous pharmaceutical tablet manufacturing pilot plant. Int. J. Pharm. 473, 38–54. Singh, R., Sahay, A., Muzzio, F., Ierapetritou, M., Ramachandran, R., 2014b. A systematic framework for onsite design and implementation of a control system in a continuous tablet manufacturing process. Comput. Chem. Eng. 66, 186–200. Sobol, I.M., 1993. Sensitivity estimates for nonlinear mathematical models. Math. Model. Comput. Exp. 1, 407–414. Tian, G., Lee, S.L., Yang, X., Hong, M.S., Gu, Z., Li, S., Fisher, R., O’Connor, T.F., 2017. A dimensionless analysis of residence time distributions for continuous powder mixing. Powder Technol. 315, 332–338. Vanarase, A.U., Muzzio, F.J., 2011. Effect of operating conditions and design parameters in a continuous powder mixer. Powder Technol. 208, 26–36. Wang, Z., Escotet-Espinoza, M.S., Ierapetritou, M., 2017. Process analysis and optimization of continuous pharmaceutical manufacturing using flowsheet models. Comput. Chem. Eng. 107, 77–91. Wang, Z., Ierapetritou, M., 2017. A Novel surrogate-based optimization method for black-box simulation with heteroscedastic noise. Ind. Eng. Chem. Res. 56, 10720–10732.