Chris O’Callaghan is senior product manager at Innopharma Technology, Dublin, Ireland, ocallaghanc@innopharmalabs.com.

Chris O’Callaghan

	Globally, there is an increasing trend toward the use of Industry 4.0 principles, with the Industrial Internet of Things (IIoT) being a key component, while regulators are actively encouraging pharmaceutical companies to modernize their approaches to drug development and manufacturing to deliver higher quality products. Better process understanding, drug product development, and manufacturing throughout the commercial lifecycle of drug products will lead to faster time-to-market and a more reliable, predictable supply chain (1).

	Adopting several of the tools and technologies which are part of the current Industry 4.0 revolution (e.g., process analytical technology [PAT], big data analytics, manufacturing intelligence, in-process control, and cloud architecture) into everyday pharmaceutical product development and commercial manufacturing may provide an effective solution to many manufacturing quality challenges. Adoption of these technologies would also dramatically improve productivity while maintaining competitive advantage and reducing costs for the manufacturer (2,3).

	This article presents a practical application of Industry 4.0 architecture with commercially available technology solutions and demonstrates how the system can be implemented to reduce risks associated with traditional fluid-bed granulation manufacturing processes.

	Fluid-bed wet granulation involves agglomerating a mix of dry primary powder particles (APIs and excipients) by the addition of a granulating solution in a fluid-bed granulator. In the subsequent drying phase, control is crucial because over-drying can lead to increased attrition and fracture of the product, while insufficient drying can result in bed stalling, poor flow, and product stability issues (4). The traditional control approach is recipe driven and largely operator dependent, with minimal provisions for the impacts of raw material or atmospheric variations, both of which are known to affect final granule properties (5).

	The automated approach described in this article resulted in greater in-process control and repeatability as well as less batch-to-batch variation. The controller design presented here is intended as a novel example to highlight the flexibility and potential when developing this type of automated, control-driven approach.

A placebo formulation was used for all batches. It consisted of a mixture of lactose (1 kg Pharmatose 200M, DFE Pharma) and microcrystalline cellulose (0.5 kg AvicelPH-101 NF, DuPont). The liquid binder was an aqueous solution of polyvinylpyrrolidone (1 L, 5.8% w/w, Plasdone K-90, Ashland). Materials were supplied by IMCD Ireland.

Fluid bed granulation was performed in a granulator (Glatt GPCG2) equipped with a particle analyzer (Eyecon

, Innopharma Technology) and near infrared (NIR) spectrophotometer (Multieye

, Innopharma Technology) measuring particle size distribution and product moisture content, respectively. The equipment is shown in 

. The automated process control platform (SmartX, Innopharma Technology) provided time-aligned data aggregation of process parameter data, PAT data, and environmental sensor data.

	Controller development is complex and requires a thorough understanding of the process, including critical process parameters (CPPs), their impact on critical quality attributes (CQAs), and the required process specifications. In this case, information on the process design space and optimum control was derived from retrospective analysis of more than 160 batches run on the test-bed system (SmartX Innopharma Technology), while further detailed experimentation was performed to quantify the differences in end-product quality between the results of this advanced dynamic process control (ADPC) approach and the results using a traditional control approach.

	The first step in the development was to clearly define the control logic for each process phase. This included identification of key dynamic control relationships, establishing fixed setpoints as well as phase and process endpoint criteria. Once configured, this flexible control logic was then implemented and executed via a process-centric scripting environment within the integrated ADPC module.

	Throughout the process, real-time PAT data and process sensor data from the fluid-bed system and environmental systems provided a continual input feed to the controller. The controller used this information to make scenario-based decisions on how to respond to process deviations as well as required process changes, including phase changes and endpoint detection.

	For the ADPC example presented in this article, five process phases were defined: empty heating, material heating, spraying I, spraying II, and final drying. 

describes the five process phases and their corresponding key set-points and endpoint criteria.

	Spraying is divided into two phases to demonstrate how PAT measurements may be implemented to achieve in-process control. Additionally, the two phases are designed with the intention to help mitigate against product attrition as typically observed during final drying, thus delivering more consistent endpoint particle size with less batch-to-batch variation. Spraying I is defined by rapid wetting and maximum growth, while Spraying II is defined by further hardening of the granules through reduced spray rate and increased moisture removal to mitigate against product attrition during the drying phase.

	A specific moisture-content reduction rate was empirically determined to achieve a quasi-stable median volume distribution (D

50) particle size while allowing for faster control reaction and, therefore, minimized process deviations as compared to controlling directly based on particle size.

The series of CQA and CPP profiles shown in 

are taken from one of the fluid-bed granulation processes executed by the ADPC controller that was developed; these profiles demonstrate the control method’s capabilities. 

The key relationship between spray rate and D

50 particle size can be observed between 

50 particle size target to 450 µm for the duration of spraying and uses real-time particle size data, as measured by the Eyecon

, to monitor the growth profile. During Spraying I, a fixed spray rate is maintained for rapid moisture addition and growth until the target particle size is reached. On entering Spraying II, the target particle size is maintained by following the empirical target moisture-content profile.

	This profile is maintained by dynamic control of the spray rate based on real-time moisture content data. Comparing 

, modulation of the spray rate after a brief delay can be observed in response to small deviations of the moisture content trend either above or below the target moisture content profile (see 

, dashed line labeled ideal moisture overlay). This process slowly dries the granulate to 5%, which is the trigger to transition to the final drying phase.

	Another novel aspect of this control approach can be observed in 

, where the effect of linking air flow rate to moisture content during the Spraying I phase can be seen. This approach allows optimum fluidization to be maintained while the bed becomes heavier and more cohesive, avoiding both the attrition and efficiency impacts of over-fluidizing, and the under-fluidizing risk of bed-stalling.

50 particle size values from a number of granulation batches manufactured with the ADPC controller were compared to the endpoint D

50 particle size values from earlier batches manufactured using a non-ADPC controlled, recipe-driven approach. A significant difference in endpoint product consistency is apparent between the two approaches.

illustrates a significantly wider distribution of endpoint Dv50 particle sizes for batches manufactured via the non-ADPC controlled approach, with variation of 171 µm from the smallest to largest Dv50 value. Comparing batches manufactured with the ADPC controller, a tighter distribution in endpoint Dv50 particle size values is evident, with variation of only 46 µm reported from smallest to largest D

50 value. These results demonstrate the consistency in batch-to-batch particle size that can be achieved by implementing such a control approach within a fluid-bed granulation process. The ability to achieve greater particle size control via the ADPC controller approach leads to more consistent endpoint particle size and less variation between batches.

	Endpoint moisture content values analyzed using the at-line loss-on-drying (LOD) methodology were compared for both control approaches. There is a significant difference in the endpoint LOD values for both of these approaches, primarily due to the non-ADPC controlled approaches using product temperature as an indication of endpoint rather than in-line moisture measurement. The resulting over-drying of the non-ADPC batches is a source of energy waste and possible attrition of the end-product material. Additionally, the ability to reliably fall within, but at the upper end, of a moisture specification helps to improve overall product yield.

clearly demonstrates this variation with a much wider distribution of final LOD values evident for the non-ADPC controlled batches. The total spread of moisture content values is 0.48% for these batches, compared to only 0.16% for the ADPC-controlled batches, which demonstrate much tighter control. These results demonstrate the benefit of the in-line NIR moisture-content endpoint detection method.

	Endpoint moisture content of the fluid bed granulation process is critical to final product quality and process performance and must be tightly controlled to avoid issues with downstream processing, product dissolution, and stability as well as drug absorption rates in the body. Implementing an ADPC approach can reduce batch-to-batch variation and improve batch repeatability and quality.

	The ADPC-controlled approach to fluid bed granulation was shown to produce more consistently sized granules with less batch-to-batch variation when compared to granules produced from a non-ADPC controlled process. In addition, endpoint LOD analysis for the ADPC batches showed significantly less variation and greater consistency. Overall, high process repeatability and reproducibility were demonstrated across multiple, successfully manufactured fluid-bed granulation batches.

	The real-time measurements of particle size and moisture content allowed the ADPC controller to effectively determine phase-end criteria. It was further shown to be possible to dynamically manage spray rate, thus ensuring a predetermined moisture content profile was followed by leveraging the NIR moisture-content data.

	Finally, the addition of PAT and its integration into the process control strategy dramatically reduces the need for at-line sampling and testing associated with more traditional granulation approaches, as well as reducing the risks associated with human error.

		L. Kiernan, “The Role of Smart Manufacturing in Enabling QRM and KM to Realize Safer and More Affordable Products for Patients in the 21st Century,” in 

An Audience with Pharmaceutical Regulators, Academia and Industry 

(TU Dublin Academic Press, Dublin, 2019), pp. 65–74.	

		T. Dedeurwaerder, et al., “How Data is Changing the Pharma Operations World,” 

		R. Gaertner, “Pharma 4.0-Time to Rethink Manufacturing and Quality,” 

		R. Mattes et al., “In-line Process Analysis of Residual Moisture in a Fluid Bed Granulator-Dryer Using NIR Spectroscopy,” 

	Caroline McCormack is process laboratory manager; Chris O’Callaghan is head of engineering, 

; and Gareth Clarke is Multieye Product Manager, all at Innopharma Technology, Dublin, Ireland. Prof. Ian Jones, PhD, is CEO of Innopharma and Innopharma Education, Dublin, Ireland; Luke Kiernan is a PhD student at the Technological University Dublin in Ireland; and Gavin Walker, PhD, is professor and director of the Synthesis and Solid State Pharmaceutical Centre at the University of Limerick in Ireland.


	Vol. 44, No. 1
	January 2020
	Pages: 42–45

	When referring to this article, please cite it as C. McCormack et al., “Self-Guided Control of a Fluid Bed Granulation Process," 

Articles

Advanced dynamic process control of a fluid-bed granulation process using PAT data improves product quality.

Self-Guided Control of a Fluid Bed Granulation Process

Drug-layered multiparticulates are a common dosage form for extended or modified-release pharmaceutical formulations. Delivered either in capsules, tablets, or as food additives in pediatric or geriatric applications (1), these formulations typically feature a functional coating designed to delay dissolution of the drug in the body.

	Wurster coating, using bottom-spray fluid-bed technology, is commonly used to manufacture these formulations, in a multi-phase process. Manufacturing is typically controlled by spraying a fixed quantity of coating factor on the substrate. For a well-developed coating process, spray efficiencies can be highly consistent. However, variability in product quality can often result from raw material variations in the substrate.

	This article will discuss research into ways to improve control of the overall process, to minimize substrate raw material and final product variability. In this work, microcrystalline cellulose (MCC) multiparticulates were used as a substitute for a drug-layered substrate and were layered with an aqueous-based enteric coating. 

	Because coating thickness is the primary critical material attribute influencing dissolution rate (2–4), the goal of this research was to minimize the impact of varying substrate raw material particle size and surface area on the resultant coating thickness. A smart process control system was used in conjunction with process analytical technology (PAT) to monitor and dynamically control key process parameters in order to improve consistency in measured coating thickness at the end of spraying. The automated fluid-bed control system was designed so that the spraying process would stop once a pre-determined coating thickness had been reached that would provide the required dissolution profile.

	The approach was demonstrated in application to two different substrate materials, with marginally different particle size characteristics in order to represent real-world raw material variability. Results showed that even a small variation in median diameter can have a significant influence on the total surface area. Experiments documented the differing quantities of coating factor that were required to achieve target growth in each case.

Cellets 500 (MCC) (Ingredient Pharm) were used as a substrate material for coating. No API layer was applied for this development study due to processing limitations. A 15% w/w aqueous suspension of 80:20 Surelease:Opadry (Colorcon Inc.) was used to coat the particles. Surelease (aqueous dispersion of ethyl cellulose) was applied as a barrier membrane coating on the Cellets while Opadry (a hypromellose-based coating system) acted as a pore former in the coating formulation.

	The Cellets 500 (approximately 500–710 µm) were screened with a 600-µm sieve to create two populations of marginally different sizes (approximately 67-µm difference in Dv50, which is a measure of the volumetric median particle diameter). Both populations fall within the material specification for Cellets 500 and may be considered to represent a batch-to-batch variation for this application.

	Three batches of each size were coated to establish the repeatability of results. The three batches of larger size pellets are referred to as L1, L2, and L3, and the three batches of smaller material are referred to as S1, S2, and S3.

Wurster coating was conducted in a Glatt GPCG2 lab-scale fluid-bed system with a six-inch, PAT-compatible, bottom-spray product container. A Schlick 0.8-mm nozzle was used to spray the coating solution with a 4.5-mm air-collar spacing. A type-B orifice plate was used for appropriate fluidization, with a Wurster column height of 25 mm.

Direct Optical Imaging Particle Analyser (Innopharma Technology) was used for real-time measurement of the particle size distribution inline, through the lowest window of the product container, as shown in 

. Direct imaging involves capturing images of the particles in-process through the window/inspection port, and running these through a series of image analysis steps to measure the size and shape of each particle present. Analysis parameters were set to fluid-bed coating defaults, with a results integration period of 120 seconds to optimize data for smooth process control.

In-line particle size data and all GPCG2 sensor data were aggregated during processing in real time and used for process control by the SmartFBx (Innopharma Technology) advanced development and manufacturing platform for fluid-bed systems.

	Rather than determining the end point of the process’ spraying phase based on when a fixed quantity of coating factor has been added, the SmartFBx controller was configured to monitor particle size growth and continue spraying until a target growth had been achieved. 

	This target was determined by inline, real-time measurement of the Dv50 of the fluidized pellets during the material preheating phase prior to the start of spraying and comparing Dv50s reported throughout the spraying phase with this baseline value to determine growth. For these experiments the target Dv50 growth was 32.5 µm, equating to a coating thickness of 16.25 µm. This value was chosen because it was the approximate growth achieved in prior experiments during which coating factor had been added to reach a predicted 10% weight gain.

	Other key equipment parameters of the GPCG2 were also controlled by SmartFBx within optimum ranges as determined during an earlier design of experiments study. The controller also automatically stepped through process phases when the appropriate conditions for each had been met. A flow diagram for the process phase logic used is presented in 

	Two approaches were used to explore the effects of the variation in substrate particle size on the coating process:

	1. Measurement of growth in coating thickness for a fixed quantity of coating factor, equivalent to a constant projected weight gain.

	2. Control of the total quantity of coating factor sprayed based on Eyecon data to achieve a precise target coating thickness. 

aids in visualizing the particle size data used by the process controller by showing an example of the Dv10, Dv50, and Dv90 trends measured by the Eyecon

, in this case from batch S2. Dv50 is the volumetric median particle diameter, while Dv10 and Dv90 define the 10th and 90th percentiles; together these three values provide a simple description of the particle size distribution (PSD). Spraying took place between 16:05 and 18:22 minutes from cycle start, during which time a steady increase in particle size across all three parameters was seen. During the course of each batch, the Eyecon

made approximately 500,000 particle measurements.

. Before examining results from the coating-thickness-driven control strategy, it is important to consider what effect the ~67-µm variation in Dv50 would have had on product quality under a traditional fixed-spray-quantity control regime. To assess this, the measured particle growth for a given spray quantity (1050 g of solution or a predicted weight gain of 9%) was compared across all batches.

, two groupings of points can clearly be seen: batches S1, S2, and S3 with initial size of 570–575 µm, and batches L1, L2, and L3 with initial size of approximately 640 µm. The coating thickness value was derived from half the difference in Dv50s between the start of spraying and the point at which 1050 g of coating factor was added. It can be seen from 

that the coating thickness for the smaller batches is approximately 3 µm thinner than that of the larger pellets. This result demonstrates that there is a discernible impact on coating thickness due to variation in particle size of substrate under a fixed-spray-quantity control regime.

To assess the influence that coating thickness would have on end-product dissolution (assuming that the process target were 9% weight gain), the authors developed a simple mathematical model (4), correlating coating thickness against dissolution for the same Surelease/Opadry coating applied to a chlorpheniramine maleate (CPM) drug-layered particle. The relevant equations from the model (4) are:  

indicates dissolution percentage after the given time in the water bath, and T

is the measured thickness of the coating at the time the sample was extracted from the process. 

shows the variability that would be expected for these functional coating thicknesses applied over a CPM-coated bead. 

	Due to the relatively thin coatings and function of Opadry as a pore former in this formulation, these dissolution rates are representative of a relatively fast extended/modified-release product, such as one targeting a specific area of the gastrointestinal tract. For a slower release coating, however, similar variations in dissolution but over longer timeframes would be expected. Variation in this case is >10% at the 30-minute dissolution time point and 9% at 1 hour.

Variation in required coating quantities. 

Using the control strategy of spraying until a target particle growth is reached resulted, as expected, in consistent coating thicknesses with a variation in the total amount of coating solution required for each batch. This result can be linked to the variation in the particle size of the substrate, and may also be influenced by other parameters, such as fluidization patterns, surface porosities, and spray efficiencies (5,6). 

demonstrates a clear downward trend in coating solution requirement from the “small” runs (left-hand cluster) to the “large” runs (right). This behavior is as expected due to the greater total surface area present in the smaller particle size batches. The behavior demonstrates the control strategy’s ability to effectively compensate for these variations without the need for formulation-specific empirical models.

The benefit of basing these control decisions on PAT measurements can be demonstrated by examining the results of a simple prediction model relating coating factor requirements to initial substrate particle size. 

displays the experimental results compared to the predicted quantities for a range of starting Dv50s based on a calculation of the ratio of final coating layer volume to substrate volume.

shows an agreement in overall trend between the experimental and predicted results, although there is considerable variation. Because each batch was coated to a constant coating thickness, the variation in quantities sprayed could be attributed to variations in other processing factors such as spray efficiency, substrate porosity, particle mass effects, PSD width, and any agglomeration or attrition present in the process. 

	Making control decisions based on real-time process measurements to track true particle growth eliminates the need for complex, formulation-dependent empirical models to calculate and compensate for such sources of variability. This traditional approach would be necessary if spray quantities were being controlled solely based on off-line raw material size measurement.

	With traditional process control methods, variation in substrate particle size impacts coating thickness on a meaningful scale. Dissolution model results indicate that the tested size difference of ~67 µm in Dv50 would have resulted in more than 10% variability in quality control dissolution test results in which pellets were coated to 9% weight gain.

	A control strategy using real-time particle size distribution measurement to calculate growth, however, has been shown to provide consistent coating thickness results for varying substrate sizes. Similar results could not be achieved using offline measurement without the use of complex empirical models and variability due to other process attributes. 

	Additional work is planned to apply this control methodology to an API-coated substrate and validate the predicted lower variability in quality control dissolution testing.

	The authors thank Colorcon inc. for their supply of the Surelease and Opadry materials used in the coating solution, and Glatt GmbH/Ingredient Pharm for the supply of Cellets.

	4. P. Patel et al., “Solid Dosage Drug Development and Manufacturing” Supplement to 

Eur. J. Pharmaceutics & Biopharmaceutics 

	When referring to this article, please cite it as C. O'Callaghan et al., "Exploring a Modern Control Strategy for Wurster Coating," 

is senior product manager at Innopharma Technology, Dublin, Ireland, ocallaghanc@innopharmalabs.com; 

is founder and CEO of Innopharma Technology and Innopharma College of Applied Sciences, Dublin; 

is formulation technologies manager at Colorcon, PA, ppatel@colorcon.com; and 

is manager, Process Technology at Glatt Air Techniques, NJ, info.gat@glatt.com.

	*To whom all correspondence should be addressed

A process control system based on PAT can compensate for variations in particle size, resulting in more consistent coating thickness in a Wurster coating process.

Exploring a Modern Control Strategy for Wurster Coating

Process analytical technology, based on monitoring particle size distribution and tracking coating thickness measurements in real time, can be used to predict the dissolution of polymer-coated multiparticulates.Process analytical technology (PAT) describes the use of analytical instruments to measure specific product attributes during manufacturing, substantially minimizing the need to sample product for off-line analysis. This approach offers several key advantages including the ability to take process measurements in situ, with instant access to data, facilitating rapid decision making during product development and manufacturing.

	Although the time between product sampling and obtaining off-line results may range from minutes to days, depending on the test being performed and the analytical structures in place, many PAT systems can measure results in real time. This capability enables process control decisions to be made, not just based on a recipe, but on the critical quality attributes (CQAs) of the material at that point in time. Access to data in real time can compensate for variabilities in raw material or mechanical wear in processing equipment components, and supports compliance with newer quality assurance (QA) initiatives such as continuous verification.

	Many PAT instruments support the measurement of several physical and chemical quality attributes. One CQA that cannot be directly measured inline, however, is dissolution, due to the length of time required for testing. This is particularly true for modified- or extended-release formulations, for which dissolution testing can take several days or even weeks. If a method were available to generate dissolution data in real time by accurately predicting the dissolution test results of a modified-release product, development and production cycle times could be reduced significantly.

	This article discusses tests that were conducted to show the feasibility of predicting dissolution drug release profiles on multiparticulates in a Wurster coating process. This approach required measuring coating thickness inline, based on growth in the material’s particle size distribution.

	The goal of this work was not only to predict dissolution, but to explore the use of PAT for particle size distribution to measure coating thicknesses across a wide range of formulations (as shown in 

	Process settings were chosen based on the functional coating material. In each case, coating was added to achieve a predicted 20% weight gain. Samples were extracted from the process at time points corresponding to a predicted weight gain of 2.5%, 5%, 7.5%, 10%, 12.5%, 15%, 17.5%, and 20%, based on the quantity of coating solution sprayed.

	Additionally, for the aqueous-based functional coat (Surelease, Colorcon), samples were taken at 30 minutes and 1 hour of curing, and analyzed to determine dissolution. Separately, particle size distribution was measured.

	Chlorpheniramine maleate (CPM) and propranolol hydrochloride (PRP) were layered onto sugar spheres (Suglets, Colorcon) with mesh sizes of 18/20 (850-1000 μm) and 20/25 (710-850 μm), respectively. CPM and PRP layered pellets (1.5-2 kg) were coated with aqueous ethylcellulose dispersion (Surelease, Colorcon) as a barrier membrane coating and Opadry hypromellose-based coating system (YS-1-19025-A, Colorcon) as a pore former, at 80:20 ratios.

	The coating dispersion was prepared by dissolving Opadry in deionized water and then added to coating to obtain total solid content of 15% w/w. A ethylcellulose (EC) organic coating system (Opadry EC, Colorcon) was used as an alternative fully formulated barrier membrane organic coating to evaluate the performance on CPM- and PRP-loaded pellets.

	The coating solution was prepared in a 90:10 ratio of ethanol to water. The targeted coating weight gain was 18-20%, and samples were taken at every 2.5% of weight gain.

	Coating system. A lab-scale fluid-bed dryer (GPCG-2, Glatt Air Technologies) with a six-inch Wurster product container was used for experiments. Critical process parameters are spray rate, atomizing air pressure, air volume, product temperature, orifice plate configuration, and partition height. Understanding and control of these parameters are crucial for a robust and reproducible process.

	The product container used for these tests featured multiple suitably specified viewing windows to allow for installation of non-product-contact PAT, to avoid having the analytical equipment come into direct contact with product. The particle size analyzer (Eyecon2, Innopharma Technology) was installed on the lowest positioned window, as shown in 

, for optimal measurement of pellets, in real time, during Wurster processing.

	In general, the use of particle size, as well as moisture and API content measuring devices, can be utilized to gain full process understanding at an economic scale. Process understanding gained from designs of experiments (DOEs) at this scale can be translated into a robust commercial process with integrated real-time in-process product measurement and process control.

	Analytical instrument. An non-product-contact direct-imaging particle analyzer (Eyecon, Innopharma Technology) was used to measure particle size change in real time, in-line, during processing, providing D Value and shape data as well as images of the particles. A typical lab type instrument using the principle of dynamic image analysis (Camsizer, Retsch Technology GmbH) was used to measure particle size distribution off-line, to assess the correlation of in-line and off-line measurements. Drug release was measured using a UV spectrometer from 1 gram of CPM and PRP barrier membrane-coated pellets in a dissolution bath using US Pharmacopeial Convention (USP) apparatus I (baskets) at 100 rpm. USP purified water was used as a dissolution media (1000 ml) at 37.0 ± 0.5 °C.

 demonstrate the data from the in-line particle size analyzer, tracking two of the coating processes. While data are captured on a continuous basis, only data points corresponding to every 2.5% weight gain are shown here for clearer presentation and later comparison to off-line samples taken.

	Dv50 is the volumetric median particle diameter, while Dv10 and Dv90 define the 10th and 90th percentiles. Together these three values provide a simple description of the particle size distribution.

	A clear growth can be seen in each graph, though the overall size of the materials differs by approximately 100μm. This corresponds with the differing mesh sizes of the CPM and PRP pellets used. It is also evident that the final two data points in 

 show negligible growth. These correspond to the curing process applied to the aqueous-based coating, during which no further material is sprayed, thus causing no weight gain at this point. Because the size does not appear to change either, it can be concluded that minimal density, abrasion, or film shrinkage effects are at play during the curing step.  To allow for dissolution model building and establishment of the repeatability of the overall process and measurement techniques, two CPM experiments were run with identical process parameters. 

 shows the Dv50s of each of these experimental runs. Minimal variance between the two processes can be seen. Because the offset between the two is relatively consistent across the duration of coating, the cause of the variance can most likely be attributed to minor variability in the starting material.

	Next, the data measured with the in-line particle analyzer were compared to measurements using the off-line image analyzer.Because  some variation is always present between different particle size measurement techniques (e.g., sieve, laser diffraction, back-light imaging), the primary goal was to establish a strong correlation between these two methods rather than to seek direct agreement.

 shows measurements from the in-line particle size analyzer graphed against those from the off-line instrument. While a clear offset is present between the methods, a similar trend can be seen. 

 plots the in-line and off-line results against each other for D10, D50, and D90 from the CPM-SR-1 experiment. It is evident from the fit lines that a high quality of correlation was present for all three values, with R2 for the Dv10 and Dv50 greater than 0.98.  To ensure that this correlation holds across other experiments and material sizes, 

 graphs the Dv 50s derived from inline measurements against those from off-line measurements for all sample points during the DoE.

	The two populations visible are due to the distinct size ranges of pellets used in the CPM and PRP experiments. A strong correlation is still present with an R2 for all points of almost 0.99. The D10 and D90 results correlate similarly.

	Calculating film thickness from measured PSD

	Although particle size and distribution can be measured using many off-line methods, dissolution performance is related more to the thickness of the functional coating, or the film thickness applied, than the overall size of the pellets. As such, the film thickness must be determined from measured size data.  While the principle involved is simple (diameter increase during coating/2), there are different ways to define diameter increase. 

 explores three methods:  difference in the Dv50s, difference in the average of the Dv10, Dv50, and Dv90, and difference in the average of all the volumetric percentiles made from in-line measurements. In practical terms, the results of all three of these methods match closely, so Dv50 was chosen as the value used for further analyses. 

 shows the calculated film thickness for the CPM coating experiment. Although a similar trend is evident, the total film thickness is considerably lower due to differing densities of the functional coating.

	Predicting dissolution using in-line measurement

	Because a number of factors affect dissolution beyond functional coating thickness, it was necessary to build a formulation-specific model for prediction based on the in-line-measured particle size. In the CPM-SR experimental runs, this was accomplished using data from CPM-SR-1 to build a correlated model against film thickness growth, which was then used to predict the dissolution results for the samples taken from CPM-SR-2. Although more data would result in a more robust prediction mechanism, this approach was considered sufficient to demonstrate a proof of concept.

	To build a prediction model from CPM-SR-1, the film thickness at each sampling point was first calculated as in 

. Results were then graphed against the dissolution result, divided into data sets for each dissolution sampling time-point (as shown in 

 shows the result of this process, applying best-fit polynomials to each of the data sets. The equations of the best-fit polynomials shown in 

effectively form the basis of predicting dissolution performance based on a measured film thickness. For a given thickness, an equation exists to describe the expected dissolution percentage for each time point measured in CPM-SR-1. To apply this to CPM-SR-2, the measured film thicknesses for each sample point were substituted into the polynomial equations from 

. Data from any point in the coating process could be used for this step, enabling dissolution to be predicted for any moment, but only sample points could be compared to off-line results, to validate the method.

	These data, when graphed, predict the dissolution curves. 

 overlays the analytical measured dissolution data, denoted (A), over the predicted (P) dissolution performance.

	Generally, the predicted dissolution curves overlap well with the measured results, showing the viability of the prediction method. Based on the limited size of the data set, better prediction could almost certainly be achieved by expanding the model data set from repetition of the experiment. For future experiments, the results of CPM-SR-2 can also be integrated into the predictive model to improve the accuracy and robustness of the prediction algorithms.

	In short, these studies showed that a strong correlation exists between functional coating thickness and dissolution profile. In addition, research proved predicting the dissolution of a coating process in real time using particle size data and a formulation-based model is a viable control method. The work also illustrated the benefits of PAT in achieving greater process understanding and faster, more efficient process profiling. This approach could be used during process development, optimization, scale up and transfer, and showed the potential for using PAT particle size distribution trending as a process control or troubleshooting tool.

	Piyush Patel, is formulation technologies manager at Colorcon, PA, ppatel@colorcon.com; Edward Godek is manager of process and technical operations at Glatt Air Techniques, NJ, info.gat@glatt.com; Chris O’Callaghan* is senior product manager at Innopharma Technology, Dublin, Ireland, ocallaghanc@innopharmalabs.com, and Ian Jones, PhD, is founder and CEO of Innopharma Technology and Innopharma College of Applied Sciences, Dublin.

olid Dosage Drug Development and Manufacturing
	

	When referring to this article, please cite it as P. Patel et al., "Predicting Multiparticulate Dissolution in Real Time for Modified- and Extended- Release Formulations," 

olid Dosage Drug Development and Manufacturing

Process analytical technology, based on monitoring particle size distribution and tracking coating thickness measurements in real time, can be used to predict the dissolution of polymer-coated multiparticulates.