J Pharm Innov DOI 10.

1007/s12247-012-9119-9

RESEARCH ARTICLE

PAT within the QbD Framework: Real-Time End Point Detection for Powder Blends in a Compliant Environment
Geir Rune Flten & Ana Patricia Ferreira & Luke Bellamy & Paul Frake

# Springer Science+Business Media, LLC 2012

Abstract Introduction A process analytical technology (PAT) method using a near infrared (NIR) spectrometer attached to a drum blender was developed for real-time end-point detection for an extragranular blend. Methods and theory The method and corresponding documentation are part of the control strategy for a recently filed and approved pharmaceutical product, and the method is currently in place and in use for manufacturing of this pharmaceutical product at GlaxoSmithKline. Results and discussion The immediate benefits are a dramatic reduction in blend time as well as real-time verification of blend homogeneity for every batch. The development of the blending end-point method is presented and a set of fundamental requirements for developing any PAT method is proposed. Conclusion It is discussed how the reported PAT end-point method fits with the proposed PAT framework, and the validation aspects required to achieve a compliant method for use in drug product manufacturing are presented. Keywords QbD . PAT . Blend . Chemometrics . NIR . Compliant

process, and process and product understanding are essential in order to achieve this goal. The process and product understanding provided by a successful development process can be used to devise a control strategy to ensure the manufacture of consistent right quality products. One of the possible elements in a control strategy is process analytical technology (PAT) [1]. PAT can be defined as a mechanism to design, analyse, and control pharmaceutical manufacturing processes through the measurement of critical process parameters (CPPs) which influence critical quality attributes (CQA). PAT can also be used to measure CQAs directly and thus enable real-time release testing. Usually, the PAT concept is implemented using a relevant sensor for in-process measurements and a model translating the measured signal into the property of interest, e.g., assay or in-specification/out-of-specification. To facilitate the implementation of a PAT method, there are further requirements with respect to software, data transfer/automation, and business processes for operating the method as well as acting on the method output. Generally there are five fundamental requirements to PAT methods intended for use in regulated environments: 1. Basic understanding of process Before starting the development of a PAT method, there must be some knowledge about what a good process looks like, i.e., what are the CQAs, and preferably some understanding about what the CPPs are, i.e., how to achieve satisfactorily CQAs. A complete understanding of the process is however not required as PAT can be used to refine or expand the process understanding. 2. Ability to measure After identifying what is important in the process, it is necessary to ensure that the considered measurement

Introduction In a quality by design (QbD) environment, the objective is to build quality into the product through the development
G. R. Flten (*) : A. P. Ferreira : L. Bellamy : P. Frake (*) GlaxoSmithKline, Technical, Ware GMS, Priory Street, Ware, Hertfordshire SG12 0DJ, UK e-mail: post@grflaten.net e-mail: paul.2.frake@gsk.com

J Pharm Innov

technology is able to measure this parameter or attribute, directly or indirectly. If this is not feasible, any developed PAT method will not be value adding no matter the effort. 3. Ability to translate measured signal into relevant information Ability to measure the parameters of interest is the fundamental requirement for PAT in QbD, but in order to support a control strategy it is also necessary to translate the measurement into relevant information with a resolution appropriate for the chosen control strategy. The two main approaches are: (a) To verify quality on target This is a reactive approach in the sense that the performance is only monitored. With this approach, one can assure quality but quality cannot be controlled or changed. (b) To control/adjust process This is a proactive approach which requires a link between the PAT application and control systems to establish a feedback (or feed forward) solution. Right quality is achieved through adjustment of the process in [near] real-time. 4. Method feasibility System understanding, appropriate measurements and relevant translation are of no benefit if the proposed method is not feasible to implement. Feasibility links to technical constraints such as scale considerations and pacing of process and measurements, but also area limitations such as ATEX certification as well as business considerations. The benefits must obviously outweigh the costs in order to have a viable business case for implementation. 5. Compliance In regulated manufacturing environments such as pharmaceutical industries, compliance is another key requirement to the PAT method. Compliance is achieved through testing and documentation of the method in correspondence with the relevant regulatory guidelines. In this paper, the development and implementation of a PAT method as part of the control strategy for a QbD filing is discussed. The described case study is extragranular blend of a recently filed and approved solid dose compound. The blend is performed in a manufacturing scale drum blender where granules and excipients are blended to achieve homogeneity. The control strategy consideration is whether PAT can be used to determine the end point of the blending. Some aspects of the method development relevant to compliance and validation of the PAT method are also highlighted, and the benefits achieved from using the method are shared.

Methods and Theory The presented project is executed on standard rotational blending equipment where a drum is hoisted up and rotated continuously around the vertical axis at constant speed (20 rpm) until blend completion. There is a Sapphire window in the drum lid onto which a purpose built near infrared (NIR) probe (Expo ePAT) can be fitted, and this probe is connected via an optical cable to a measurement unit mounted on the blender. Spectra are collected for each rotation and the measurement unit sends the new spectra via wireless communication to a PC in the suite where the spectra are recorded and displayed in [near] real time. Prior to any analysis SNV is applied to the spectra. The blend is started and stopped manually by the operator. The batch size is approximately 55 kg, and the blend is extragranular mixing of active granules, disintegrant, lubricant, and filler. NIR System The NIR system used in this project was the EXPO ePAT601 with the 10 mm spot size optical head. The acquisition range was 1,3501,800 nm with 2 nm resolution. Each spectrum was a single acquisition of approximately 40 ms and the 10 mm spot size at focal length was fixed by window design. Caterpillar The Caterpillar is a moving window algorithm developed from first-principle understanding of the powder blending process to identify the end-point of blending [2]. It is observed that for powder mixing, the end point is defined by a homogeneous distribution of the particles in the blend. Prior to reaching the end-point, the distribution of particles is heterogeneous. This can be utilised in a process measurement technology setup where measurements are made at one point of a rotating blender. The moving powder results in an initial signal with variation and gradient, changing into a stationary signal which is recognised as the end-point. An underlying assumption is that the particle sizes are not significantly influenced by the blending process, so the only change gradient is the transition of the powder blend from the heterogeneous to the homogeneous state. The Caterpillar algorithm works by assessing changes in spectral data variation along time. Two fixed-size windows of data separated by a fixed distance in time are moved through the data at the sampling frequency. The end-point is identified by an F test comparing the signal variation within the two windows: F s2 1 s2 2 1

J Pharm Innov

where s2 and s2 are the variances in the two windows 1 2 determined from the scores after a principal component analysis (PCA) decomposition. The test criteria is F > Fa;df 1;df 2 2 where, is the significance level and df1 and df2 are the relevant degrees of freedom. Before using the algorithm the analyst must set window sizes, interwindow distance, number of components to be used in the PCA representation of the data, and the significance level of the F test [2]. Software and Infrastructure The Caterpillar is not available in any commercial software package so a bespoke implementation was required. The most convenient implementation environment for the development team was Matlab (MathWorks, MA, USA). In addition to the Matlab core package, the OPC toolbox (MathWorks), and the PLS Toolbox (Eigenvector, WA, USA) were used. The code used in the manufacturing environment was compiled to remove the risk of uncontrolled change to the scripts. A graphical user interface was developed to facilitate usage of the method by the manufacturing operators. However, a closed-loop control was not developed so the system is relying on the operator to manually stop the blending when the software flags that the end-point has been reached. The NIR signal is collected directly from the instrument via the OPC interface using the OPC server provided by the instrument vendor. The system is not linked to the manufacturing execution system infrastructure in the plant but developed as a local stand-alone system in the manufacturing suite.

Results and Discussion The PAT method presented in this paper was developed and first implemented at full scale in the manufacturing environment. This is not an ideal approach and it was only made possible since relevant competency and experience was available in the project team at the manufacturing site. The proof of concept for monitoring powder blends by spectroscopy is well documented in the literature [39], and this minimised the requirement for specific development batches as the challenge is reduced to verify that the current application is viable. The drawback of development at large scale in the manufacturing environment is not technical but relating to the cost of development batches and the rigor for any modifications to equipment and process. Requirement 1: Understanding Understanding of the process dynamics is essential in the development of the PAT method, and the concept for this

case study is demonstrated in earlier works [29]. If the proof of concept does not exist, the understanding must be verified in order to be regarded as more than a hypothesis. Blending of powders is normally a straightforward physical process starting from an unblended mix of powders moving to a blended mix. The dynamics of the blend process is to some extent system dependent but the start state is a heterogeneous blend and the end state is defined as a homogeneous blend. Homogeneous means that a random sample from the bulk of the blend will be equal to any other similar sized sample from the blend, both with respect to constituents and particle distribution. The implication is that if it is possible to identify when the blend is homogeneous it is possible to identify the end-point. Some powders can deviate from this simple transition due to agglomeration, breakage, and other particle interactions, but this is not further discussed here as it was not relevant for the case study being reported. Based on the understanding of the blend process, a sampling strategy can be derived. It is realised that if a point measurement is made in a vessel in which the unblended powder is continuously moving, the measurement signal is changing and it is nonstationary since the bulk of powder is heterogeneous. Performing the same point measurement on a bulk of blended powder gives a stationary signal since the bulk of the powder is homogeneous. The underlying assumption is of course that the chosen measurement method is able to distinguish the powders in the blend and represent the relative abundance for all of them simultaneously. The CPP of the blend operation is blend time, which is the required time for the blend to reach homogeneity. The CQA is the amount of drug in the tablets and the content uniformity. To ensure patient safety and drug benefit, the amount of powder, particularly active ingredient, must be the same in every tablet. In order to secure the right quality of the end product, the blending process must provide fully mixed batches. Requirement 2: Ability to Measure From the understanding of the process, it is clear that if it is possible to make a point measurement in the rotating vessel and identify the relative abundances of the constituents from the signal, it is possible to develop a PAT method to identify the blend end-point. The underlying CQA is the amount of API in the tablets so strictly it is sufficient to prove that the distribution of API is homogeneous across the blend. In the available setup, the verification of the ability to measure comes down to confirming that the blend constituents are optically active in the NIR frequency range and that the mixing gradient is captured in the NIR signal. It is important to emphasize that the discussion about ability to measure does not need to be limited to NIR probes. NIR is widely used in

J Pharm Innov

PAT applications but other analytical techniques could be explored if NIR does not meet the requirements. Here, exploring is not to start from scratch as specialists can predict the likelihood of success for various technologies in advance. Figure 1 shows the NIR spectra of the product blend as well as the pure spectra of all the constituents. The most abundant excipient, sthe filler, has a clear peak at about 1,482 nm while the API has well-defined peaks at 1,630 and 1,680 nm. As can be seen in Fig. 1, the disintegrant and lubricant are also giving signal within the spectral region. All the main constituents of the blend are thus optically active in the NIR region, so the next question is whether the NIR signal is able to reflect the dynamics of the blend process. In Fig. 2, the spectra from the blend operation through a typical blend is shown, and one can see that the signal around the filler peak (1,482 nm) is decreasing until stationary while the signals for both the API peaks (1,630 and 1,680 nm) are increasing until stationary. This is consistent with the loading of the blend vessel where granules are first in and filler (and the other excipients) are added on the top as all the spectra are measured through the drum lid when the drum is upside down. The stationary signals for both filler and API have less variation than the initial spectra, so the trend is towards a stationary and smaller variation in the signal. Moving block standard deviation plots of the signal measured at each relevant peak with, e.g., window size 6 starts high and levels out after 2025 rotations for both filler and API but this is not shown here. The same trend of initially large variation from spectrum to spectrum is decreasing until stationary variation between successive spectra is also seen for the whole spectral range. This is illustrated in Fig. 3 where the first spectra collected for each batch is shown with 3 SD variation limits (Fig. 3a) next to the average of the five last spectra collected for each batch after blending for 10 min with 3 SD variation limits (Fig. 3b). Overall, it seems possible to monitor the blend progress using NIR as the constituents of the blend are optically active in the NIR range and the physical transition from a heterogeneous to a homogeneous mix is reflected.
Fig. 1 The pure NIR spectra for the compounds in the extra granular blend and the API are shown together with the total spectrum of the blend

Requirement 3: Ability to Translate Measured Signal into Relevant Information Having established that the NIR signal reflects the progress of the blending, the next question is whether it is possible to quantify the blend progress and clearly identify the end point. This can be approached in several ways including moving block standard deviation [4, 7] or PLS models [10] and other multivariate approaches [11, 12] to describe the end state. For the method being described, the Caterpillar approach has been applied [2]. The Caterpillar is a welldefined approach with a sound statistical basis and the use of a batch-specific reference makes it robust to interbatch variation. Importantly, the end-point determination is a straight forward statistical test which simplifies the communication and provides a clearly defined end-point. The Caterpillar algorithm uses, as mentioned, a moving window approach where two windows of a defined width are moved through the data at a fixed interval. The parameters to be set in order to use the Caterpillar approach include window size, interwindow distance, number of components, and significance level. The setting of the parameters relates to the system properties and end-point requirements, and the settings in this application were six sampling points, six sampling points, three components, and 0.05, respectively. The CPP of the blend is blend time required to achieve homogeneity at the unit dose level to ensure that every tablet (dose unit) has the right composition (CQA). This means that the variation represented by the windows in the Caterpillar approach needs to be equal to or smaller than the unit dose which is equivalent to the tablet size. To verify this, it is necessary to assess the sampling volume measured by the NIR which can be derived from the sampling area and penetration depth. The sampling volume for each NIR measurement was found to be approximately 1/6 of the smallest unit dose size which means that six spectra represent a sample volume similar to one unit dose for the smallest tablet. Accordingly, the PAT method consisting of NIR measurements and the Caterpillar algorithm does have

Absorbance SNV

API Filler Disintegrant Binder Blended granule Extragranular blend

1400

1500

1600

1700

1800

Wavelength (nm)

J Pharm Innov
2 1.5 1

Mean initial Additives blend spectra + 3 SD 3 2

Absorbance SNV

Absorbance SNV
1400 1450 1500 1550 1600 1650 1700 1750 1800

0.5 0 0.5 1 1.5 2 2.5 3

1350

1 0 1 2 3 4

Wavelength (nm)

1350

1400

1450

1500

1550

1600

1650

1700

1750

1800

Fig. 2 The spectra collected through a typical blend. The spectra which are most different from the others correspond to spectra collected early in the blend, while the tightly overlapping majority of the spectra are collected towards the end of the batch subsequent to reaching the end-point of the blend. The arrows indicate the direction of change with time

Wavelength (nm)

Mean final Additives blend spectra + 3 SD 2 1.5 1 0.5

the capability to identify the true end-point in the blend process. The PAT method was executed for a number of blend batches in order to verify the applicability and the results are shown in Fig. 4. The end-point is determined as the time point where the number of data points under the statistical limit corresponds to the number of spectra between the windows plus one. This is a conservative approach to eliminate any doubt about the data between the two windows being nonstationary. As seen in Fig. 4, the end point of the blend is consistently found after approximately 30 rotations. This consistency is an indication that the method is fit for purpose and that implementation can be considered. The decisive test in the case study was however to run three batches and stop the blend at the indicated end point and analyse the manufactured tablets by high-performance liquid chromatography, the current reference method. This was again a conservative approach chosen because of the novelty of using a PAT method as control strategy, but one can argue that the decisive test was redundant considering the first principles understanding and strength of the proof of concept study. The comparative test was however required to support method validation and provide the required assurance of equivalence to quality and regulators. It is concluded that by using the Caterpillar algorithm, it is possible to quantify the blend CPP (blend time required to reach homogeneity) to ensure that the CQA (uniform API distribution in tablets) is satisfied. Albeit, there is no control loop implemented the application can be seen as an active process control as the operator is acting on a signal or control flag.

0 0.5 1 1.5 2 2.5

1350 1400 1450 1500 1550 1600 1650 1700 1750 1800

Wavelength (nm)

Fig. 3 The gradient from large to smaller interspectra variation during the blend is shown. a The mean spectrum for the five first spectra collected during a blend shown together with the variation illustrated by 3 SD. b The mean spectrum for the five last spectra collected during a blend shown together with the variation illustrated by 3 SD

Requirement 4: Feasibility of Implementation Having established that the understanding of the system is valid, and that the CPP of interest is possible to measure and quantify, the PAT method is ready for implementation in the manufacturing process. However, prior to implementation and use in manufacturing, the feasibility of the implementation must be assessed. Technical feasibility, i.e., whether it can be done and business feasibility, i.e., whether the return on investment is satisfying, are the two aspects which must be considered. These are clearly interlinked and if the manufacturing challenge is big enough, e.g., the method mitigates a significant risk for the unit operation or process, the business requirement can provide the incentive to identify or develop a suitable technical solution.

J Pharm Innov
100 90 80 70
Batch 1 Batch 2 Batch 3 Batch 4 Batch 5 Batch 6 Batch 7 Batch 8 Batch 9 Batch 10 Batch 11

Requirement 5: Compliance The method and all related systems must be compliant since the purpose of the method is to identify the end point of a blend operation for a commercial drug product. An itemised list of requirements to compliance can be summarised in four steps: 1. Scope and solution. Define what the intention of the method is and purpose for application, as well as clear description of any technical solutions. 2. Risk assessment. Within the scope of the solution, perform a risk assessment to identify any connected risk and identify areas for possible testing, and even improvement. 3. Verification. Verify that the method or approach is doing what it is supposed to do within the limitations of the scope. Testing of performance outside the defined scope of the method is not necessary although it is essential to ensure that the scope is clearly defined and documented for the method. 4. Traceability. Document purpose of the approach, perceived risks and extent and outcome of testing as well as application areas and validation ranges. The compliance requirements are analogous to the steps in any method development as described in PAT requirements 14, where scope and solution encompass the development activity. Verification is part of a robust method development and corresponds to the testing during development. Risk assessment is not always formalised although it is inherently done during a robust method development and feasibility study. One will for instance consider the likelihood of window fouling as part of the method development and feasibility considerations. The biggest difference between a regular PAT method development and a compliant PAT method is the traceability, i.e., the documentation requirements. This means that with good routines the work load for developing a compliant PAT method and a noncompliant method should be similar, assuming that the equipment is in compliance. The validation of the reported blend end-point detection method to ensure compliance was complex due to the mix of noncompliant systems, e.g., software and method development. In addition to validating the PAT method, the bespoke software scripts and the model needed validation. This required a fairly extensive number of documents as listed in Table 1 and since there is a strong overlap among method,
User/ Manufacturing process documentation

Fvalue

60 50 40 30 20 10 0 20 40 60 80 100 120 140

160

180

200

Rotation

Fig. 4 The output from the Caterpillar algorithm is shown for 11 development batches. The calculated F value is plotted against the number of rotations. The dotted horizontal line indicates the critical F value for the relevant degrees of the freedom. The end-point is reached when the batch-line is going below the critical value. In the implementation presented here the end-point is conservatively only accepted after six values below the line to ensure that the end point is not transient

The case study discussed herein was part of a strategic GlaxoSmithKline (GSK) project for a QbD filing and the focus was on the technical feasibility and this is also central in this discussion. An overview of the building blocks of the implementation is shown in Fig. 5, and although not shown interfaces between the building blocks are also required. The sampling interface and measurement system are both part of the e-PAT system and thus in place at manufacturing scale. The e-PAT system has a built-in OPC server providing data for further use in other systems in near real time. The challenge is the model implementation since Caterpillar is not currently available in off the shelf software. However, good understanding of the Caterpillar approach and MATLAB competency was available to the project, so bespoke software capable of receiving spectra from the OPC server, performing the Caterpillar calculations and providing the algorithm results as an output to the operator was developed in house. Once the end-point is flagged, the operator will stop the process and record relevant output from the software in the process documentation. This completes the system described in Fig. 5 and the method is deemed feasible to implement.

Manufacturing Process (Blending)

Instrument (NIR)

OPC server

Software

Output

Fig. 5 The main elements of the Caterpillar implementation are shown

J Pharm Innov Table 1 The documents required to achieve compliance for the PAT method are listed

Document

Method

Model

Script

PAT Requirement 1, 3 3 3, 4 3 4 3, 4 3, 4 3 1 4 4 3, 4 3, 4 3, 4 4 3 2, 3 2, 3, 4 4 3, 4

Business requirements Compliance determination: Matlab and PLS Toolbox Design review report for the Caterpillar implementation Design specification for the Caterpillar implementation (CI) Method application protocol Risk assessment Software categories assessmentMatlab and PLS Toolbox Source code review Systems overview for Caterpillar implementation User system requirements Technical installation plan for Caterpillar algorithm Technical installation report for the Caterpillar algorithm The number of documents is large because the validation includes method, model, and script. Column 35 indicates which element the corresponding document is covering. The last column indicates which of the PAT requirements the document is related to, i.e., (1) understanding, (2) ability to measure, (3) ability to translate measured signal into relevant information, (4) method feasibility Unit specification for the Caterpillar implementation Validation plan Validation summary report Blend homogeneity method Model development report Proof of concept Method development protocol EMPACT method risk assessment Model validation report Method development and validation report AM QbD summary report X X

X X X X X X X X

X X

X X X X X X X X

X X X X X X X X X X

X X

2, 3, 4 4

software, and model the document links are also quite complex. In addition to listing the documents, Table 1 indicates which element, method, model, or script, is covered by each document, and it also tentatively indicates which of the PAT requirements each document relates to. Some of the complexity is due to the novelty of this project, the Caterpillar algorithm is for instance calibration free which means that there is no direct reference against which to test the algorithm output. It is expected that the document trail can be simplified without loss of rigour for future methods. One obvious area for improvement is to clearly establish the hierarchy between method, model and the software scripts. In the current validation approach, these concepts were not clearly defined up front resulting in overlapping documents. The Caterpillar model is for instance validated both as an entity and again as part of the script validation. Another potential improvement is to align the documentation with the PAT requirements as these are following the flow of PAT method development. An immediate benefit would be that the documentation of analogous methods for different compounds would be much simpler. The first requirement relating to understanding could be a generic proof of concept report, and the documentation of measurement (requirement 2) and quantification (requirement 3) could be short add-ons to a primary report.

Benefits As mentioned earlier, the submission of the PAT end-point detection for blending was part of a strategic initiative for GSK to prepare a QbD submission for a solid dose product. The submission was filed and after a joint audit by the United States Food and Drug Administration and European Medicines Agency, the method was approved for use as part of the control strategy for the product in both the North American and the European markets. A major benefit of the developed method was the contribution to the QbD submission and thus supporting the strategic ambition for the business. Executing the project also provided the project team with valuable experience and understanding of the requirements to develop a PAT method in a compliant environment. The project team had experience in all required technical aspects of method development and implementation prior to starting the project but the challenges of the development and implementation in a compliant environment as part of a submission gave valuable insight which will benefit GSK in future QbD submissions. The business value of a defined end-point in the blend process is manifold. The PAT method has improved the understanding of the blend operation and been valuable in the process development providing a control strategy for

J Pharm Innov

blending. The method gives a statistically valid confirmation of successful extragranular blend for every batch. Looking at all the blend batches for this product manufactured since implementing the method, it turns out that for most of the batches the end-point is reached after 38 rotations or less. There are however three instances where more rotations are needed, the longest one requiring 57 rotations. In an imagined development where the blend time is fixed, the development campaigns could have justified a blend time of 38 introducing a risk for not meeting the CQA requirements for all batches. This is a clear demonstration of the inherent benefits of monitoring process performance during manufacture and to incorporate the results in the control strategy where possible as any deviations can then be handled as they occur. Business value from the PAT method was also realised during the product development. The initial 10 min blend time was reduced to about 2 min based on the information and understanding gained from the PAT method. This equals an 80% reduction in blend time, and if it is transferable to other products a major opportunity in energy savings across the GSK manufacturing network is available.

Acknowledgments This paper was presented at IFPAC10. Colleagues from the validation and automation departments in GSK are acknowledged for their support.

References
1. US Department of Health and Human Services. Food and Drug Administration, Guidance: PATA Framework for Innovative Pharmaceutical Development, Manufacturing and Quality Assurance, 2004. 2. Flten GR, Belchamber R, Collins M, Walmsley AD. Caterpillar an adaptive algorithm for detecting process changes from acoustic emission signals. Anal Chim Acta. 2005;544:28091. 3. Sanchez FC, Toft J, van den Bogaert B, Massart DL, Dive SS, Hailey P. Monitoring powder blending by NIR spectroscopy. J Anal Chem. 1995;352:7718. 4. Sekulic SS, Ward HW, Brannegan DR, Stanley ED, Evans CL, Sciavolino ST, et al. On-line monitoring of powder blend homogeneity by near-infrared spectroscopy. Anal Chem. 1996;68:50913. 5. De Maesschalck R, Sanchez FC, Massart DL, Doherty P, Hailey P. On-line monitoring of powder blending with near-infrared spectroscopy. Appl Spectrosc. 1998;52:72531. 6. Berntsson O, Danielsson L-G, Johansson MO, Folestad S. Quantitative in-line monitoring of powder blending by near infrared reflection spectroscopy. Powder Technol. 2002;185:18593. 7. Blanco M, Gozlez Ba R, Bertran E. Monitoring powder blending in pharmaceutical processes by use of near infrared spectroscopy. Talanta. 2002;56:20312. 8. Wargo DJ, Drennen JK. Near-infrared spectroscopic characterization of pharmaceutical powder blends. J Pharm Biomed Anal. 1996;14:141523. 9. Bellamy LJ, Nordon A, Littlejohn D. Real-time monitoring of powder mixing in a convective blender using non-invasive reflectance NIR spectrometry. Analyst. 2008;133:5864. 10. Sulub Y, Wabuyele B, Gargiulo P, Pazdan J, Cheney J, Berry J, et al. Real-time on-line blend uniformity monitoring using near-infrared reflectance spectrometry: A non-invasive off-line calibration approach. J Pharm Biomed Anal. 2009;49:4854. 11. Skibsted ETS, Boelens HFM, Westerhuis JA, Witte DT, Smilde AK. Simple assessment of homogeneity in pharmaceutical mixing processes using a near-infrared reflectance probe and control charts. J Pharm Biomed Anal. 2006;41:2635. 12. Puchert T, Holzhauer CV, Menezes JC, Lochman D, Reich G. A new PAT/QbD approach for the determination of blend homogeneity: combination of on-line NIRS analysis with PC Scores Distance Analysis (PC-SDA). Eur J Pharm Biopharm. 2011;78:17382.

Conclusions A compliant PAT method for detection of the end-point in extragranular blends has been developed, validated, filed, and approved, and is currently in use in the manufacturing of a pharmaceutical solid dose product at the GSK Ware manufacturing site. In this paper, the fundamental requirements to any PAT method are shared and it is demonstrated how the development of the end-point method fits within the requirements framework. The activities required to make the method compliant are also shared. The project has provided an immediate saving in blend time and continuous verification of successful blending in the manufacturing of the product. The project has also been of strategic value to GSK and it is used as a stepping stone towards the new QbD paradigm within pharmaceutical manufacturing.