A perspective on Quality-by-Control (QbC) in pharmaceutical continuous manufacturing

Abstract

The Quality-by-Design (QbD) guidance issued by the US Food and Drug Administration (FDA) has catalyzed the modernization of pharmaceutical manufacturing practices including the adoption of continuous manufacturing. Active process control was highlighted recently as a means to improve the QbD implementation. This advance has since been evolving into the concept of Quality-by-Control (QbC). In this study, the concept of QbC is discussed, including a definition of QbC, a review of the recent developments towards the QbC, and a perspective on the challenges of QbC implementation in continuous manufacturing. The QbC concept is demonstrated using a rotary tablet press, integrated into a pilot scale continuous direct compaction process. The results conclusively showed that active process control, based on product and process knowledge and advanced model-based techniques, including data reconciliation, model predictive control (MPC), and risk analysis, is indispensable to comprehensive QbC implementation, and ensures robustness and efficiency.

1. Introduction

Pharmaceutical manufacturing has traditionally operated in the batch mode, which consists of processing a defined amount of input material, called a batch or lot, through various unit operations, such as reaction, crystallization, filtration, drying, blending, granulation, tableting, etc., to obtain final drug products, for example, oral solid dosage. Since the quality attributes of in-process material or final product are tested at the end of each batch processing step, this approach to quality control is known as Quality-by-Testing (QbT) (Hubert et al., 2014). One example is in the blending step, in which powders of active ingredients and excipients are mixed in a rotating Y- or V-shaped vessel and the degree of mixing uniformity is tested at the conclusion of the batch blending process (Igne et al., 2014).

Various process monitoring and control strategies have been proposed for batch manufacturing in pharmaceutical or other manufacturing industries to achieve desired batch-end product quality, for example, multivariable partial least squares for batch process monitoring (Nomikos and MacGregor, 1995; Su and Chiu, 2016), shrinking horizon nonlinear MPC for batch-end product quality control (Su et al., 2016), etc. However, most of these strategies rely on a priori determined optimal or nominal process operation trajectories or recipes. It is the optimization and tracking of these operational trajectories during the execution of the batch that leads to the batch-end product quality. Rework of the material at the end of a batch, particularly in pharmaceutical secondary manufacturing processes, should be carefully evaluated before implemented in such a highly regulated industry (US Food and Drug Administration CDER, 2016). Hence, the loss of a whole batch is possible after testing since remedial control actions can only be implemented for subsequent batches. As a result, there is a large time delay before product quality improvements are realized: this is usually known as the batch-to-batch control strategy (Mockus et al., 2015; Su et al., 2017).

The US FDA recognized that increased testing does not necessarily improve product quality and thus quality must be built into the product (Yu et al., 2014), following the QbD concept (Juran, 1992). Over the years, pharmaceutical QbD has evolved with the issuance of the guidance documents ICH Q8 (R2) (Pharmaceutical Development), ICH Q9 (Quality Risk Management), ICH Q10 (Pharmaceutical Quality System), and ICH Q11 (Development and Manufacture of Drug Substance). These documents provide highlevel directions concerning the scope and definitions of QbD as it applies to the pharmaceutical industry (Yu et al., 2014).

Though the QbD principles have and are being implemented in conventional pharmaceutical batch manufacturing (Potter, 2009), over the past decade there have been significant advances in the implementation of continuous manufacturing of pharmaceuticals. The potential to improve agility, flexibility, and robustness in the development and manufacture of pharmaceuticals with the adoption of the QbD paradigm and process analytical techniques (PAT) have accelerated the scientific and regulatory readiness for continuous manufacturing (Lee et al., 2015). For example, FDA has already identified pharmaceutical continuous manufacturing as an emerging technology and commercial adoption is now progressing from conceptual designs to pilot or production implementation. Recently, several products produced via continuous manufacturing facilities have been approved by the FDA, specifically, Orkambi (lumacaftor/ivacaftor) from Vertex in 2015, Prezista (darunavir) from Janssen in 2016 (Yu L., 2016), Verzenio (abemaciclib) from Lilly and, Symdeko (tezacaftor/ivacaftor) from Vertex in early 2018, and Daurismo (glasdegib) from Pfizer in late 2018. These approvals of continuously manufactured drug products are the outcome of systematic integration of product and process knowledge, instrumentation and automation systems, quality control protocols and real-time process management of the solids processing operations and reflect the long-term commitment of the respective organizations and the regulatory agency to continuous manufacturing.

Besides the flexibility and cost-saving features of pharmaceutical continuous manufacturing, a unique advantage of this technology compared to batch manufacturing is that the critical quality attributes (CQAs) and critical process parameters (CPPs) identified under the QbD concept can be continuously monitored and controlled in real time. Real-time monitoring and control pave the way for the next stage in real-time product quality assurance, namely real-time release (RTR). Published case studies demonstrate that continuous manufacturing, implemented using online/inline-PAT tools and efficient process control systems, can accelerate the comprehensive implementation of the QbD principles for the next generation of pharmaceutical products (Singh et al., 2013; Yu et al., 2014). Of course, active process control strategies are not new; they have been employed in other industries employing continuous manufacturing, e.g., bulk chemicals and petrochemical industries (Darby and Nikolaou, 2012). However, despite the demonstrated feasibility of active process control, in both batch and continuous crystallization processes for drug substance (DS) manufacturing (Yang and Nagy, 2015a,b; Su et al., 2015; Nagy et al., 2013; Nagy and Braatz, 2012) as well as in continuous manufacturing for drug product (DP) in the form of tablets, there remains a hesitation to the widespread adoption of active process control strategies (Diab and Gerogiorgis, 2018), in part due to the lack of experience in technology transfer from development to manufacturing. Most pharmaceutical continuous unit operations are still controlled one operation at a time with limited integration of unit operations using advanced active process control systems. Recently the concept of QbC has been introduced and is gaining traction as a means of addressing this gap (Nagy, 2016; Sommeregger et al., 2017; de Barros et al., 2017; Mesbah et al., 2017). The objective of this paper is to review the concept of QbC and demonstrate its utility as a framework for the implementation of efficient active process control of the CPPs and CQAs of continuous pharmaceutical processes.

The QbC paradigm takes the QbD approaches to a higher level by putting the application of active process control approaches as the focal point (Nagy, 2014). The concept originally was applied to enhance the efficient design of pharmaceutical crystallization processes (Saleemi et al., 2012; Saleemi et al., 2012; Simone et al., 2015; Yang et al., 2015; Su et al., 2015). The position of the QbC concept in the evolution of systematic approaches to quality assurance in pharmaceutical manufacturing is shown in Fig. 1, including the extension of the QbC paradigm to the next-generation smart manufacturing, also referred as, Industry 4.0. The key proof-of-concept studies that introduced the concept of QbC for both DS and DP manufacturing were: (1) the design of the batch crystallization of a biopharmaceutical compound (Simone et al., 2015), (2) the development of continuous crystallization processes (Yang et al., 2015; Su et al., 2015) and (3) design of a continuous secondary manufacturing process using drop-on-demand technology (Içten et al., 2015). In addition to the aforementioned studies that have shown the benefits of QbC versus conventional QbD approaches in process design to achieve targeted CQAs, it has also been shown that QbC may be indispensable in the comprehensive QbD implementation to handle operating problems related to process disturbances and uncertainties, and that it ensures more robustness and efficiency than conventional techniques used in QbD approaches in reducing product quality variance (Koswara and Nagy, 2017; Sommeregger et al., 2017; Su et al., 2017). While there has been an increasing number of studies on the application of the QbC concept, there is a lack of systematic understanding of the extension and further development of the QbC framework, specifically, to the emerging continuous manufacturing in downstream oral solid dose processing.

Fig. 1.

The systematic progression in quality assurance via QbT, QbD, and QbC.

The remainder of this paper is organized as follows. The concept of QbC is discussed in detail in the next section, including summarizing the key background literature, defining the QbC concept, and providing a brief review of the recent developments in research and industrial applications of the QbC. This will be followed by the demonstration of QbC on a rotary tablet press including the use of data reconciliation, active process control, and risk analysis. The paper will close with a perspective on the challenges of QbC implementation and a summary of key learnings and future directions.

2. Quality-by-Control

2.1. Background

The QbD guidance has promoted the systematic generation of the essential product and process knowledge required to implement continuous operation by identifying the critical material/quality attributes, process parameters, and the control strategies required to maintain the process operation and the quality of the product under a state of control. Recently, a three-level control strategy based on pharmaceutical QbD guidance was proposed, in which the idea of active process control was highlighted (Yu et al., 2014). This quality control strategy was then further elaborated by Lee and co-workers (Lee et al., 2015) with an emphasis on modernizing pharmaceutical manufacturing by transitioning from batch to continuous production. These developments have laid the foundations for the idea of QbC (Simone et al., 2015; Yang et al., 2015; Içten et al., 2015; Nagy, 2014).

A three-level quality control strategy for a continuous manufacturing process (Yu et al., 2014) was proposed to maintain the quality of the product in response to potential variations or disturbance in the process, equipment conditions, incoming raw materials, or environmental factors over time. For example, an intuitive Level-3 quality control strategy imposes tight constraints on material attributes and process parameters that affect product quality and relies on extensive end-product testing at each processing step to ensure final product quality. This level of control is commonly used in batch manufacturing, viz., the QbT, by strictly tracking a recipe during operation to ensure that those parameters are maintained within constrains. This level of control requires limited understanding, particularly at the early stage of drug development, on how raw material and process variability affects product quality. It is too conservative and is neither feasible to be effectively implemented in continuous manufacturing processes nor adaptable to achieve the benefits of continuous manufacturing (Lee et al., 2015).

The demands on end-product testing can be reduced by using a Level-2 control strategy under which variations of raw material attributes and process parameters are maintained within a design space. The design space established under the QbD guidance (ICH-Q8, 2009) requires the identification of potential sources of raw material and process variability that can impact product quality, as well the understanding of the impact that variability from these sources has on in-process materials, downstream processing, and drug product quality. Hence, drug development at the late stage or pilot manufacturing within a design space allows some flexibility in raw material and process parameters and reduces the reliance on extensive end-product testing. Intrinsically in continuous manufacturing, the process is operated in such a manner as to be consistently within the design space. As a result, Level-2 controls which employ an established design space have been implemented in most reported continuous manufacturing facilities using multivariate statistical process control methods (Almaya et al., 2017).

However, the operation within the limited design space established during product and process development can result in a lack of effectiveness in responding to process disturbances or variations that are commonly seen in continuous manufacturing process. For example, when a process disturbance leads to a departure of a CQA variable from its targeted setpoint or acceptable range, the adjustment of a CPP variable within the limited range allowed by the design space could take a long time to bring the CQA variable back to acceptable criteria, resulting a long period of generation of non-conforming material (Su et al., 2015; Koswara and Nagy, 2017). On the other hand, a more aggressive manipulation of the CPP variable, e.g., with an intentional overshoot, which may extend outside of the design space, is more likely to return the CQA variable within the design space much sooner (Su et al., 2017). Furthermore, the concept of design space is also too rigid to adapt to mismatches in product and process understanding, e.g., material compressibility changes due to powder moisture content or particle size variations (Gupta et al., 2013).

By contrast, a Level-1 quality control strategy would feature the use of an active process control system to monitor and control the quality attributes of materials in real-time. In response to a disturbance, process parameters would be automatically and optimally adjusted to ensure these quality attributes consistently conform to the established acceptance criteria. This level of control represents a high degree of product and process understanding that can also be identified under the QbD guidance. The enhanced process understanding, which includes identification of dynamic relationships linking critical material attributes (CMAs) and CPPs to CQAs, will enable the design of an engineering control system with quantitative and predictive capabilities. As a result, the impact of upstream disturbance on downstream processing can be minimized through optimal control adjustments and any mismatches in product and process understanding can be mitigated with an adaptive and predictive control strategy. The design of such a quantitative and predictive control system at Level 1, which is based on the QbD guidance, minimizes the risk of producing off-spec product and enables achievement of real-time release, is the foundational component of QbC.

As defined by the QbD guidance, the application of the QbD framework also includes the design and implementation of a suitable control system, subsequent to the design of the operating space. However, the QbD guidance does not suggest a systematic quantitative procedure for the design of the suitable control architecture and methodology. In this context, the original QbC idea can be extended and enriched as the framework for systematic design of an active process control system that enables both the robust design and operation of the pharmaceutical manufacturing system. Thus, QbC is a logical extension of QbD which is backward compatible with the QbD guidance.

The recent progress in mechanistic understanding of processes and advancements in PAT tools can also be incorporated within this holistic QbC framework to further the adoption of model-based process automation and quality assurance in process operations (Pantelides and Renfro, 2013). Interestingly, a QbC application which highlighted the use of MPC in pharmaceutical bioprocesses was reported in (Sommeregger et al., 2017). Moreover, as the pharmaceutical industry begins to adopt smart manufacturing practices, a modular approach to systems integration and process operations is crucial for optimal asset utilization and knowledge management. A systematic integration of process equipment, sensors and control systems in accordance with process automation standards enables the effective implementation of the real-time data analytics and associated knowledge management which are required to achieve RTR testing in pharmaceutical manufacturing.

2.2. Definition

The definition of QbC can be restated as follows: QbC consists of the design and operation of a robust manufacturing system that is achieved through an active process control system designed in accordance with hierarchical process automation principles, based on a high degree of quantitative and predictive product and process understanding. QbC in general enables reliable batch and continuous process operations, especially the real-time release in continuous manufacturing of pharmaceutical products.

2.3. Recent development towards QbC

Systematic approaches to the design and implementation of active process control systems for continuous manufacturing processes have been proposed, mainly for drug substance manufacturing systems, in particular, batch and continuous crystallization processes (Fujiwara et al., 2005; Abu Bakar et al., 2009; Nagy, 2009; Mesbah et al., 2012), including several integrated systems, such as continuous crystallization and wet mill processes (Yang et al., 2015). Among these early studies were also the first papers that introduced the term and concept of QbC (Simone et al., 2015; Yang et al., 2015). For continuous drug product manufacturing, systematic approaches for the design and implementation of feedback control strategies have been proposed and progressively improved in the past few years, specially, by the Engineering Research Center for Structured Organic Particulate Systems (ERC-SOPS) for a drop on demand additive manufacturing system (Içten et al., 2015; Içten et al., 2018), continuous direct compaction and dry granulation processes (Singh et al., 2014, 2009; Gupta et al., 2013; Hsu et al., 2010a,b; Ierapetritou et al., 2016; Su et al., 2017), and importantly, the integration of PAT sensors into the control system (Singh, 2018a,b).

Moreover, a range of control techniques, from heuristic control algorithm, such as concentration control for crystallization (Su et al., 2014, 2015), simple proportional-integral-derivative (PID) controllers (Yang and Nagy, 2015a,b), to advanced model-based control and real-time optimization for pharmaceutical batch and continuous crystallization and manufacturing processes via direct compaction, roller compaction, and wet granulation (Singh et al., 2015a,b; Ramachandran et al., 2011; Yang and Nagy, 2015a,b; Singh, 2018c; Nagy and Braatz, 2012; Nagy et al., 2013), have been tested. Specifically, the simulated plant-wide MPC for an end-to-end continuous manufacturing process at the Novartis-MIT center was also demonstrated recently (Mesbah et al., 2017). Performance in setpoint tracking and disturbance rejection was systematically evaluated to finalize these control system designs. The Novartis-MIT center also proposed a plant-wide control strategy, e.g., involving a bottom-up and hierarchical top-down approach, which stabilizes the underlying process control layers upward and prioritizes the control objectives downward (Lakerveld et al., 2013; Lakerveld et al., 2015). This work has also encouraged attempts to categorize control techniques and systematically evaluate their comparative performance with respect to aspects that are of concern to regulators. It is also recognized that PAT sensors/controls are indispensable to achieving superior control performance and to ensuring the consistent production of quality solid-dosage under the range of process disturbances encountered in practice (Gupta et al., 2013; Bondi and Drennen, 2011).

To rigorously establish continuous manufacturing processes that will consistently produce quality products, the efforts of regulators, academia, and industry are also focused on issues of risk assessment and management of quality control, e.g., the ICH Q9 Quality Risk Management and Q10 Pharmaceutical Quality System (ICH-Q9, 2006; ICH-Q10, 2007). However, the risk-based assessments of the degree of process automation and control system design, which would ensure robust operation and real-time release in continuous manufacturing has not been thoroughly developed (Ierapetritou et al., 2016; Singh et al., 2012). Such assessments ought to also include the implementations of ISA-88 Batch Control Standard (Cao et al., 2018), ISA-95 Enterprise-Control System Integration Standard, or the ISPE GAMP 5.0 Good Automation Manufacturing Practice, with an aim to achieve a resilient and fault-tolerant plant-wide active process control system design and implementation for pharmaceutical continuous manufacturing processes and advancement to smart manufacturing practices.

Furthermore, there are existing frameworks for fault tolerant control, which have been applied in other continuous manufacturing industries over the last three decades that can accommodate faults among system components automatically while maintaining system stability along with a desired level of overall performance (Blanke et al., 1997; Jiang and Yu, 2012). Generally, these use two main approaches to deal with faults. The first approach responds to a failure by re-organizing the remaining system elements in real-time to carry out necessary control functions. The other is to make the system failure-proof for certain well-defined risk/fault sets at the design stage (Jiang and Yu, 2012). However, for the rapidly emerging pharmaceutical continuous manufacturing technology, the practical implementation of fault-tolerant control, or even the basic process control, has been largely under-investigated. In addition to system stability, product quality under system component failures is even more important under the ICH Q9 Quality Risk Management guidance. Hence, these features provide the motivation for a systematic framework for combining existing fault-tolerant control practice with product quality concerns for process control design and risk analysis in pharmaceutical continuous manufacturing processes (Singh et al., 2014; Lakerveld et al., 2013), as well as for effective alarm management frameworks (Gupta et al., 2013).

Recently, with an aim towards integrating design and operations along the QbC paradigm, a systematic framework employing appropriate process systems engineering tools was proposed to develop and evaluate feasible active process control strategies (Su et al., 2017), as shown in Fig. 2. Additionally, the framework interfaces with additional supporting knowledge and tools that facilitate systems integration for process control strategy design and implementation.

Fig. 2.

A systematic framework for fault-tolerant process control design & risk analysis.

Specifically, the hierarchical process control structure, as shown in Fig. 3, structured according to the ISA-95 Enterprise-Control System Integration Standard, is focused more on the implementation with the levels classified according to the scale of their control objectives, not to be confused with that in (Yu et al., 2014) for quality control strategy. For example, in a continuous direct compaction process, the Level 0 control is often implemented via the programmable logic control (PLC) system that is built into the unit operation equipment to control single/multiple CPPs. The Level 1 control relies on the use of PAT tools to measure and control CQAs and may encompass multiple unit operations. Hence, the Level 1 control supervises the Level 0 control typically using cascaded single input and single output (SISO) control loops with the aim of achieving desired setpoints for CQAs. Level 1 control systems often span across unit operations and are designed using efficient feedback/feedforward control algorithms to reduce the impact of disturbance that otherwise may propagate downstream. A distributed control system (DCS) is employed in this regard for integrating process equipment such as the feeders and tablet press and the instrumentation for measuring material properties. The distinguishing feature of the more advanced approaches applied at Level 2 is the use of mathematical models for validating process measurements, predicting the effects of disturbances and changes in the CPPs on the CQAs, fault detection, and intensifying process operations. The functionalities provided at Level 2 may include data reconciliation (DR) and gross error detection (GED), MPC, and real-time optimization (RTO), among others.

Fig. 3.

A hierarchical implementation of control systems for a continuous direct compaction process.

A risk map for the manufacturing process can be presented in the form of a matrix, which characterizes the likelihood that a risk event will occur and describes its impact on the manufacturing system. Only the nominal risks that are acceptable to continuous manufacturing are investigated at the control design. An acceptable risk is a risk that is understood and tolerated usually because the cost or difficulty of implementing an effective permanent counter-measure exceeds the expected impact of the risk event on process operations. For instance, reduced flowability of powders may occur due to increased humidity in the environment during the rainy seasons of the year and this could have an adverse effect on the mixing uniformity of the API and excipient blend. However, it might be too costly to monitor moisture content in feed materials and to add a unit operation for reducing the water content to ensure specified flowability measures. Continuous improvements in the manufacturing process will be pursued either by improving the process understanding or by enhancing the product formulation and process designs.

3. Case study: QbC implementation in continuous tableting

3.1. Continuous rotary tablet press

Tablets are the most common oral solid dosage form. They are manufactured by direct compression or by dry/wet granulation, based on material properties and formulation requirements. The processing steps involved in direct compression consist of powder feeding, blending and tableting unit operations. The case study presented in this work was performed in Continuous Solids Processing Pilot Plant at Purdue University. This integrated continuous manufacturing line begins with two Schenck AccuRate PureFeed^® AP-300 loss-in-weight feeders. These feeders continuously feed the API, Acetaminophen (APAP, Grade 0048), and the excipient, Microcrystalline Cellulose Avicel PH-200 (MCC 200), into a Gericke GCM-250 continuous blender, wherein the two components are mixed. A Schenck AccuRate PureFeed^® DP-4 disc feeder feeds silicon dioxide as a glidant into another Gericke GCM 250 continuous blender. The blended material is conveyed to feed a Natoli BLP-16 rotary tablet press featuring a total of 16 punch-die stations.

The tablet press is a multi-stage process, in which each station undergoes the following main steps: die filling, metering, pre-compression, main-compression, tablet ejection and take-off from lower punch, as shown in Fig. 4. After the blend is fed into the die, the metering stage is adjusted to achieve the dosing position, i.e., the volume of powder inside the die. The powder is then locked between upper and lower punches during pre-compression and main-compression until the tablet ejection and take-off stages are reached. The pre-compression stage serves to remove air trapped in the die and to rearrange the particle packing, while the main-compression stage compacts and transforms the powder bed into a tablet. The tablet weight can be controlled by changing the dosing position subject to variations in powder bulk density, and in filling time due to changes in turret speed, or in filling efficacy due to changes in powder flow properties. The in-die tablet thickness is determined by the punch displacement which is manually set before the tableting operation for the tablet press used in this study. Hence, the maximum main-compression force dependents on the amount of powder in the die or, equivalently, on the tablet weight.

Fig. 4.

Major steps in Natoli BLP-16 rotary tablet press.

3.2. QbC implementation

The continuous direct compaction process was integrated with PAT sensors to monitor the process operation within the design-space and process control strategies to maintain consistent product quality. For example, the API mass fraction was measured in situ using a Near-Infrared spectrometer (Control Development, Inc.) at the exit of the first continuous blender (Vanarase et al., 2010). The powder flow rate was measured using an X-ray based mass flow meter (SETXvue XP-300, En’Urga, Inc.) (Ganesh et al., 2017).

An Emerson DeltaV 13.3 distributed control system is used to integrate process equipment and develop the automation system in this pilot-plant-scale facility. A modular and hierarchical network architecture has been implemented following ISA 95 and DeltaV Security Manual recommendations for systematic implementation of QbC. The network diagram of the pilot plant is shown in Fig. 5. Relevant firewalls are set up based on DeltaV Area Control Network, DeltaV and non-DeltaV machine interfacing and access to Purdue’s Network for software licenses. The DeltaV workstations are set up as virtual workstations using VMWare Type 1 hypervisor. The loss-in-weight feeders and blender communicate using a Profibus network, while the Yasakawa controller on the Natoli BLP-16 tablet press equipment communicates with DeltaV via a VIM2 card, configured using VIMNet explorer in the Engineering Station. Control modules for the process equipment are implemented using DeltaV Control Studio in the DeltaV ProPlus Workstation. Relevant process variables are recorded in the DeltaV historian. The DeltaV data access server and historian are accessed using the Application Station. The data from PAT Tools that are interfaced with the process are acquired in laptop computers consisting of the PAT specific hardware and software. Execution of the data reconciliation algorithm is performed in the PAT-Main laptop. Tools such as KepServerEX, LinkMaster (both Kepware, PTC Inc.) and Matlab’s Instrument Control Toolbox (MathWorks Inc.) are used to interface the PATs, the laptops and the control system.

Fig. 5.

Network setup in continuous tablet manufacturing pilot plant at Purdue University.

Specifically, the critical-to-quality variables in the tablet press were identified as tablet weight, relative density, tensile strength and main compression force (Su et al., 2018). The weight of the tablet ultimately determines the API potency within a dose. It also determines the main compression force at the pre-set punch displacement, or in-die tablet thickness, and thus the relative density and tensile strength of the tablet, which in turn affect the final product attributes such as tablet dissolution behavior. Commercial at-line instruments, such as Sotax AutoTest 4 tablet tester, are often employed to measure the tablet weight, as well as tensile strength and dimensions, at a frequency of several minutes. However, destructive and time-consuming measurements cannot be efficiently integrated with existing process control system to maintain consistent quality production in real time. Therefore, an in-house design for real-time tablet weight measurement based on a Mettler Toledo ME 4001E balance was employed in this study, as discussed in the next subsection. Though a similar design was also used in a recent work for tablet weight control (Bhaskar et al., 2017), neither the tablet weight measurement reliability and accuracy were thoroughly verified therein, nor its validity in enhancing the real-time tablet weight control was confirmed. For example, the effect of introducing extra variations due to measurement data imperfection or process control into the tablet quality attributes compared to the conventional open-loop or the Level 0 control operation was not demonstrated. Another drawback of a sampling time of 20 s in their tablet press data acquisition system was also found, which may impede capturing process dynamics and thus downgrade the expected process control performance (Shardt and Huang, 2013). Hence, in the following sections, a QbD understanding of the material properties and tablet press performance is combined with a data reconciliation strategy for tablet weight measurement to enhance the QbC implementation for the continuous rotary tablet press.

3.3. Data reconciliation for tablet press

An in-house design for real-time tablet weight measurement based on a Mettler Toledo ME 4001E balance was used in this study. As shown in Fig. 6, tablets exiting the tablet press chute are collected in a container placed on top of the electronic balance, which records the cumulative weight of the tablets produced. The balance is connected to a laptop computer via RS232 cable to transfer the measurement data of total tablet weight via Matlab (MathWorks) Instrument Control Toolbox. The corresponding tablet production rate is determined by taking the first-order derivative of the cumulative tablet weight measurement. A tablet weight measurement (W_t) is then calculated by dividing the tablet production rate by turret speed and number of stations, as shown in Eq. (1),

$$W_t=\frac{\text{Tablet}\text{production}\text{rate}}{\text{Turret}\text{speed}\times \text{Number}\text{of}\text{stations}}$$ (1)

where the turret speed determines the total counts of tablet produced per time unit. The proposed in-house design is efficient and accurate; however, it lacks in precision due to the noisy measurement generated by the electronic balance. Hence, a moving time window averaging was adopted during which roughly 100 tablets are collected for purposes of computing the data needed in order to estimate the first-order derivative used to calculate the tablet production rate and tablet weight. However, measurement gross errors can also occur when tablet samples are pneumatically diverted to the at-line Sotax Auto Test 4 device or when the container is replaced. Therefore, a data reconciliation strategy is needed to reduce the measurement uncertainty and, if needed, to eliminate the gross errors with this tablet weight measuring device.

Fig. 6.

Tablet weight measurement real-time monitoring and control.

First, to capture the QbD understanding of the material, the powder compressibility is modeled by using the relationship between main compression force and the resulting in-die tablet relative density. In this study, the Kawakita model was employed to characterize the main compression force (CF) at a given in-die tablet relative density (ρ_r) (Ludde, 1966), i.e.,

$$\frac{CF}{1-{\rho }_c/{\rho }_r}=\frac{CF}{a}+\frac{\pi D^2/4}{ab}$$ (2)

where ρ_c is the critical relative density, parameters a is interpreted as the maximum degree of compression, b is interpreted as the reciprocal of the pressure applied to attain the maximum degree of compression, and D is the diameter of the die. In the above equation, the in-die relative density is computed from the tablet weight and it is given by

$${\rho }_r=\frac{4W_t}{\pi D^2{\rho }_{t}H}$$ (3)

where the true density of the powder blend ρ_t is known a priori, and the in-die tablet thickness H is the pre-set distance between the two-flat round face, B-type tooling punches at compaction force. It is worth mentioning that the critical relative density ρ_c is an important property which can be used to characterize the material and the compression process, which should be monitored in real-time and periodically re-estimated. The initial estimates for these parameters were obtained using offline characterization of the formulation and the tablet press (Su et al., 2018).

A generalized data reconciliation framework was implemented for the tablet weight measurement to correct for the imperfect measurement data and estimate the critical relative density based on the process knowledge of material compressibility. The generalized reconciliation problem is posed as an optimization, as defined in Eq. (4),

$$\begin{gathered} \underset{z_t}{\min }\sum _{t\in T}{w}_z^T\rho \left(z_t-z_{m,t}\right)+\sum _{t\in T}{w}_m^T{\left(z_t-z_{t-1}\right)}^2 \\ \text{subject}\text{to}\text{z}\subseteq [y,x,u,\theta ] \\ f(y,x,u,\theta ,t)=0 \\ g(y,x,u,\theta ,t)\le 0 \end{gathered}$$ (4)

where z_m,t is the vector of measured process variables z at time t; z_t is the corresponding reconciled measurement at time t; y, x, u are process output, state, and input variables, respectively; and θ the model parameters in the vector function f (·) and g (·) which represent the process model, consisting of equality and inequality constraints, respectively; a robust estimator ρ(·) is incorporated to eliminate gross errors; w_z is the weight vector for the measurement error e_t = (z_t − z_m,t); and w_m is the penalty weight vector for successive move of the reconciled measurement or estimated process variables and model parameters; T is a moving window within which the process data set are used for dynamic data reconciliation (Liu et al., 2018).

Since it is difficult to determine what distribution the measurement error follows given the limitations of the in-house tablet weight measurement (the impact of dropping tablets on the measurement error of the dynamic mass measurement changes as the total tablet mass builds up in the container) as well as because of the occurrence of other gross errors, a Welsch robust estimator ρ_w was used in this study (Dennis and Welsch, 1978; Prata et al., 2010), as shown below.

$${\rho }_w(\epsilon )\frac{c_w^2}{2}\left\{1+\exp \left[-{\left(\frac{\epsilon }{c_w}\right)}^2\right]\right\}$$ (5)

$$\epsilon =\frac{e}{\sigma }$$ (6)

where ε is the standardized residual; c_w is a tuning parameter; e = (z − z_m) is the measurement error; σ is the standard deviation of measurement error e, which can be estimated from historical data for the main compression force and tablet weight measurements (Wu et al., 2016). To jointly estimate and update the Kawakita model parameter, the critical relative density ρ_c is included in the reconciled vector variable z = [CF, W_t, ρ_c]. Instead of a measured parameter, a reference density estimated from the historical distribution of ρ_c can be considered herein (Camara et al., 2016). In this manner, both the uncertainty with the tablet weight measurements and the model-plant mismatch or variations in powder compressibility can be tackled using a simple, but not too simple, steady-state real-time optimization problem. Note that although the turret speed is also a measured variable, its measurement error is inherently included in determining the measured value of the tablet weight in Eq. (1), hence is not included in the data reconciliation.

Thus, the reconciliation application in this case amounts to reconciling CF and W_t and re-estimating ρ_c subject to the Kawakita equation. Fig. 7 shows the result of the data reconciliation which was performed on a run with the built-in Level 0 control in which step changes were imposed on turret speed and dosing position. The Sotax AT4 tester was employed to analyze the tablet weight when the process reached steady-state after each step change. The Kawakita model with a previously determined critical relative density ρ_c of 0.250 was used as a soft sensor to predict the tablet weight from the main compression force, by way of example, in Fig. 7(a) the solid line shows the tablet weight predicted by the soft sensor with the critical density parameter unchanged. Note the offset from the tablet weight measurements produced by the Sotax. The noisy measurement of tablet weight obtained from the balance is shown in Fig. 7(b). As is evident, although noisy, the weight agrees quite well with the Sotax measurement. When the two measured variables, CF and W_t were reconciled, together with a re-estimation of the critical relative density parameter, the reconciled tablet weight measurements show good agreement with the at-line measurements provided by the Sotax, as shown in Fig. 7(c). Fig. 7(d) shows the trajectory of updates of the model parameter, which are obtained over time. It should be noted that the Sotax measurement is not used in the reconciliation itself, rather, it is only employed for validation purposes.

Fig. 7.

Data reconciliation with Level 0 control experiment with at-line Sotax AT4 sampling.

The reconciled tablet weight measurement obtained by data reconciliation was integrated into the tablet press control system on the DeltaV DCS system, along with the tablet production rate calculated by Eq. (1) using the reconciled values of tablet weight and turret speed.

3.4. Active process control of continuous tablet press

The Natoli BLP-16 tablet press has a built-in PLC panel to manipulate process parameters of dosing position and turret speed, which is regarded as a Level 0 control in this context. A Level 1 control with decoupled PID control loops was designed for a cascaded control of tablet weight, tablet production rate, and main compression force by manipulating the setpoints of dosing position and turret speed at the Level 0 control. A Level 2 MPC was also designed, in which the main compression force was constrained and monitored as it is closely related to the tablet CQAs of hardness, tensile strength, and dissolution rate. Emerson DeltaV Control Studio and DeltaV Predict toolbox were utilized for Level 1 and 2 control development and implementation, respectively, details of the control loops can be found in Fig. 8. Note here that both Level 1 PID control and Level 2 MPC control used the reconciled tablet weight measurement from Level 2 data reconciliation. Details of system dynamic responses under step changes in dosing position, turret speed, and system interactions under Level 1 and Level 2 closed-loop control of the studied tablet press can be found elsewhere (Su et al., 2018). For QbC demonstration purpose in this case study, continuous tableting experiments were performed in three different scenarios: (i) Level 0 control, (ii) Level 1 control, and (iii) Level 2 control to validate the online data reconciliation, as well to compare the control system performances, as shown in Fig. 9. The at-line Sotax AT4 tester was implemented to sample the tablets during the run to independently verify the final tablet quality.

Fig. 8.

The hierarchical Level 1 PID (top) and Level 2 MPC (bottom) control for continuous tablet press at DeltaV DCS system.

Fig. 9.

Online integrated data reconciliation and process control of Level 1 and 2 for continuous tablet press.

A total of 16 tablets were analyzed at each sampling instant with the Sotax AT4 tester, viz., by collecting all the tablets produced in one rotation of the turret. In such a way, variation among punch stations and variation with processing time can both be characterized, (see the box plot of sampling data points and the zoomed-in inset in Fig. 9(a)). The Level 0 control operation at the beginning of the tablet press run in Fig. 9(a) was to confirm that the reconciled tablet weight measurement agreed well with the at-line measurement or to allow the data reconciliation to automatically update the critical relative density in order to reduce possible model-plant mismatch due to material variation. Note that the reconciled tablet weight measurement started to match the Sotax AT4 measurement at the third sample and the updated critical relative density at the beginning of the operation in Fig. 9(c). Furthermore, during the control closed-loop operation, the data reconciliation continued updating the model parameter and reached a plateau under Level 1 and Level 2 control set-point changes. Even after a reinitialization of data reconciliation at time 3600 s in Fig. 9(c) by setting the Kawakita model parameter of critical density ρ_c to its initial value of 0.250, offset between reconciled tablet weight measurement and at-line Sotax AT4 measurement was observed but was then gradually reduced with the update of critical relative density. Hence, the proposed data reconciliation demonstrated an important feature of on-line automatic calibration for tablet weight measurement and was not interfering with the control system design.

The control performance of the tablet press was good using both the Level 1 and Level 2 strategies, viz., the tablet weight reached the setpoints steadily under both control strategies except that the Level 2 MPC control showed a more aggressive and promising control performance. During the setpoint changes of tablet weight, the tablet production rate was maintained the same to adjust to the campaign production or processing capability upstream, e.g., the feeding and blending, see Fig. 9(b). More importantly, variations of the tablet weight among 16 stations at steady-state remained the same as the control open-loop operation with current experiment runs, (see the box plots of each sampling point in Fig. 9(a)). Moreover, these variations along the processing time were also well preserved and verified under steady-state operation, see the at-line tablet weight measurements at setpoint of 234 and 260 mg, except during the time when data reconciliation was deliberately reinitialized. Overall, the control system design was shown capable of achieving the process automation to reach the targeted tablet weight setpoint automatically and steadily, which is important during process startup or product switch. Another benefit of the active process control system is to maintain the tablet weight under common risk of process disturbances or material property variations, thus attaining a real-time release strategy. Specifically, the performance improvement by the Level 2 MPC is significant in shortening the period of diversion of off-spec product during setpoint changes or process disturbance.

3.5. Risk analysis of tablet press process control

Several risk scenarios arising from common cause errors in the rotary tablet press were identified as acceptable risks at the current stage according to the risk scoring scheme for the real-time release strategy in pharmaceutical continuous manufacturing (Potter, 2009), as shown in Table 1. These risks were considered and further classified, according to the product of risk severity and risk probability, into R0 low, R1 medium, and R2 high-risk levels in this study for evaluating the performance, e.g., in controllability, of the three control designs for the tablet press, schematically shown in Fig. 10. For example, the sensor noise of tablet weight measurement is not entering the Level 0 control, therefore it is not affecting the Level 0 control of dosing position or turret speed. However, if this noise measurement as shown in Fig. 7(b) is entering the Level 1 PID that supervises Level 0 control, it will have adverse effect on the process operation, posing the risk of increased variation in final quality attributes. However, with a hierarchical three-level control design, the Level 2 data reconciliation will minimize this effect, as well as the effect of raw material property change, e.g., on compressibility. Another example is the risk of system interaction. A change of turret speed will increase or reduce the die filling time thus affect the tablet weight, even with a constant dosing position as is the case of Level 0 control only. This risk is indicated in red in Fig. 10. This can be mitigated to some extent with the cascaded PID control loops of the Level 1 design, as indicated with a yellow highlight. However, the decoupled nature of PID control loops results in comprised tuning of each PID loop and thus the conservative performance in setpoint tracking, as shown in Fig. 9(a). The multiple input and multiple output (MIMO) capability of MPC at Level 2, however, can handle this system interaction more robustly based on the process dynamics model, resulting in a superior control performance, as also shown in Fig. 9(a). To summarize, it is thus demonstrated that a high degree of QbD understanding will help achieve the QbC implementation.

Table 1.

Common cause errors risk scoring for the real-time release strategy in pharmaceutical continuous manufacturing.

Rate	Severity		Probability		Controllability
1	No effect	None	<1 occurrence during the continuous production in 5 years (e.g., control instrument failure)	Very rare	All the applicable measurements & controls are in place	Always under control
3	No patient effects Process performance decreasing (e.g., divert of non-conforming product)	Low	1 occurrence during the continuous production in 1 to 2 years (e.g., calibration errors)	Rare	Most of the applicable measurements & controls are in place	High credibility of under control
5	No patient effects Process performance decreasing (e.g., large amount of non-conforming product, process shut-down)	Medium	<1 occurrence during a single continuous production campaign (e.g., variance in raw material properties)	Sometimes	Some of the applicable measurement & controls are in place	Moderate credibility of under control
7	Potential patient effect (e.g., large variance in critical quality attributes)	High	>1 occurrence during a single continuous production campaign (e.g., raw material reloading)	Frequent/No information	Few of the applicable measurements & controls are in place	Remote credibility of under control
9	Significant patient effect (e.g., large variance in target product quality profile, loss of clinical performance)	Very high	At any time during the continuous operation (e.g., machinery vibration)	High frequent	None of the applicable measurements & controls are in place	No control

Fig. 10.

Risk assessment for three level control designs (top: Level 0 control only; center: Level 1 control; bottom: Level 2 control).

4. Challenges of QbC in continuous pharmaceutical manufacturing

First, systematic frameworks based on sound control engineering theories for process control development in continuous pharmaceutical manufacturing have not yet achieved a common understanding and wider application in the industry. Classical process control engineering theory has matured and been extensively employed in continuous fluid-based petroleum and chemical industries (Darby and Nikolaou, 2012). In these applications, process dynamics is often driven by chemical reaction or mass transport, which have response times measured in minutes to hours. Thus, this process control experience may or may not be directly transferable to the more challenging solid-based unit operations typically employed in pharmaceutical secondary manufacturing, where physical changes usually occur within seconds or minutes and therefore faster response may be required of the control system (Singh et al., 2015a,b). Furthermore, in pharmaceutical continuous manufacturing there may be limited hold-up in each unit operation to mitigate segregation and thus the buffering provided by material inventory is limited. Additionally, stream recycling or substantial back mixing in the process must be avoided in highly-regulated pharmaceutical secondary manufacturing due to the necessities of maintaining lot identity for material tracking purposes (Lee et al., 2015). Thus, variability in raw materials upstream has a rapid and direct impact on downstream processes, which affect the in-process materials and final drug product qualities. In this regard, control system with QbC design should be able to respond to the disturbance rapidly in a predictive or combined feedback and feedforward manner (Singh et al., 2015a,b), rather than the classical feedback control design, making the consistent production of quality solid dosage challenging.

Secondly, deployment of PAT tools in real-time remains challenging given the complexity of sensor calibration, and model validations (Ierapetritou et al., 2016). The sensor positioning, sampling concerns and fouling result in measurement drifts and bias, thereby affecting real-time process data accuracy. Sensor network design and maintenance for reliable CQA measurements have not been systematically studied in continuous manufacturing in pharma industry. It is worth noting that the robust mechanical design of traditional manufacturing equipment (such as rotary tablet press, roller compactor, etc.) has resulted in minimum variation of CPPs and/or CQAs during operation and thus allowed batch pharmaceutical manufacturing to assess quality using post-batch statistical quality control (SQC) methods. A QbC active process control system, by contrast, is challenged to use possibly noisy and biased CQA measurements to effectively supervise the control of CPPs and minimize the need for batch-end SQC (Su et al., 2018). These issues impose the need for some degree of redundancy in sensor network so as to allow application of methodology such as data reconciliation and gross error detection (Ganesh et al., 2018). The data reconciliation strategy has been recently shown to be able to reduce the measurement noise in a continuous feeding-blending system and to detect measurement errors in CQA variables (Liu et al., 2018; Moreno et al., 2019; Su et al., 2017). Further studies on using data reconciliation combined with joint state and parameter estimation are needed for QbC implementation to address issues associated with the often uncertain measurements of CQA’s provided by some spectroscopy-based PAT tools (Bagajewicz and Cabrera, 2003; Câmara et al., 2017; Wu et al., 2016; Weiss et al., 1996; Rafiee and Behrouzshad, 2016). An important aspect to investigate is the extent to which this integration imposes additional dynamics on the process, and how this could potentially amplify variations in CPPs and thus in CQAs.

Thirdly, process performance monitoring and continuous improvement in continuous manufacturing are seldom reported in the pharma industry. Continuous improvement is pursued in most manufacturing sectors to exploit the deeper understanding of the manufacturing system and its components, which naturally develop as manufacturing experience with a product and process is accumulated, as also identified in QbD guidance. Despite its potential, continuous improvement has not been pursued aggressively in pharmaceutical manufacturing, given the real and perceived regulatory burden of approvals required for changes. Hence, the advent of continuous pharmaceutical manufacturing with the proposed QbC paradigm opens the door to continuous improvement at multiple levels, including predictive maintenance, control performance monitoring, control structure re-organizing, etc., since such improvements can be targeted to achieve tighter tracking of CQA and more robust plant-wide control, which will maintain the process within its designed operating space. The direct impact is to allow longer continuous runs without forced interruption, reduced frequency and duration of periods during which nonconforming materials are generated, and reduced risk that a product lot released may include nonconforming material. Herein, three research challenges to be addressed under QbC are centered on the use of high frequency sampled data, reduction of process-plant mismatch and closed loop model identification (Tyler and Morari, 1996; Dumont et al., 2002; Shardt et al., 2015; Jiang et al., 2009), which are captured in Fig. 11. For example, process model identification is one of the important steps in systematically accounting for process uncertainties or model-plant mismatch in model-based control strategies in pharmaceutical continuous manufacturing. Progress in the areas of robust or stochastic MPC, has recently been reviewed in (Mesbah, 2016).

Fig. 11.

Continuous improvement in QbC system in pharmaceutical continuous manufacturing.

5. Conclusions

Based on the product and process understanding gained through the QbD studies, effective active process control can be attained in pharmaceutical continuous manufacturing through the proposed QbC concept. Compared to rigid process operation within a predefined design space, active process control response to common risk scenarios due to process variations, disturbances, or uncertainties can be automatically and optimally obtained in a quantitative and predictive way. A case study of QbC in continuous pharmaceutical secondary manufacturing was demonstrated with a pilot scale direct compaction process. The case study highlighted the benefit of implementing MPC integrated with steady-state data reconciliation in a rotary tablet press. The framework mitigates risk of measurement uncertainty in tablet weight measurements and uncertainty in powder compressibility. The systematic implementation of the integrated model-predictive control and data reconciliation framework in continuous tableting was demonstrated, leveraging QbD understanding of the process to achieve robust and efficient process operations and real-time quality control of tablet weight.

A new paradigm in pharmaceutical manufacturing is evolving under which quality should not only be originally designed using product and process understanding based on QbD, but more robust processes can also be implemented using active process control approaches and thus quality can be ultimately controlled in real-time based on the QbC principle. Further research efforts in systematic sensor network maintenance, control performance monitoring and continuous improvement are being pursued in our research group to further advance QbC implementation. The integration of equipment and sensors along with the hierarchical control system implementation based on QbC also sets the stage for introducing smart manufacturing practices into pharmaceutical processing. Furthermore, the concept of QbC can also be further extended to the final administration of a drug, e.g., the control of blood sugar level by adjusting insulin dosage, highlighting the importance of QbC over the lifecycle of a pharmaceutical product.