A Practical & Structured Approach To Deploying Process Analytical Technology In a Manufacturing Environment

July 31, 2012
Let’s redefine PAT as the Practical Application of Technology

Despite the many operational and cost saving benefits provided by Process Analytical Technology (PAT), its adoption in the Life Sciences sector has been slow. There are many reasons for this: natural resistance to change, perceived cost, complexity, risk,  regulatory concerns, depth and length of commitment required, and deployment methodology, which is not clear to all. This has meant that PAT deployments may take too long. The purpose of this paper is to address these issues and propose a practical structure for the implementation of PAT. We would ‘coin’ our structured approach as the ‘Practical Application of Technology’.

PAT has been defined, by the USA’s Food and Drug Administration (FDA), ‘as a mechanism to design analyse and control pharmaceutical manufacturing processes through the measurement of Critical Process Parameters (CPP) that affect Critical Quality Attributes (CQA)’. The implementation of PAT requires a fundamental change in pharmaceutical manufacturing methodology to a more dynamic approach delivering substantial reductions in WIP, cost and time to market together with improvements in quality for manufacturers.

The long term goals for PAT can be quantified as follows:

  • increase process understanding
  • reduce development and scale up time
  • reduce production cycle times
  • enable use of alternative (less expensive) raw materials
  • reduce wastage during production
  • prevent rejection of batches
  • improvement in quality and consistency of quality
  • enable real time release of  product
  • improve energy and material use                                             
  • facilitate continuous processing
  • release the capability of automation to improve the production process                                           

PAT Challenges
The challenges facing PAT implementations are many. First, legacy practices: people generally do not like change. Second, human resources: the perception that the implementation of PAT may adversely affect roles and responsibilities. Third, regulatory: regulatory requirements still have to be met at a time of great change; but the FDA has created opportunities for streamlined registration of such changes. Fourth, skilled resources: PAT deployment calls for a wide range of skill sets.

Fifth, pragmatism: a detailed yet pragmatic & risk based approach is required. Sixth, technology for deployment: correct & appropriate PAT technologies are essential for project success. This paper concentrates on these last 2 challenges.

Necessary Technologies for PAT Deployment
The Necessary Technologies for PAT deployment is a two phase process. Phase 1 is concerned with process model & knowledge building – control is seldom of concern at this stage. For this phase the requirements are typically a univariate data acquisition system, spectral instruments, an MVA package, and a PAT data management system. Phase 2 deals with control model building and control with PAT. For this phase all the items listed in Phase 1 are required, plus a control system.

Step 1: Target Product Identification
PAT Implementation Strategy is all about knowing where to start, and what questions to ask. If the project is going to be related to a new product application, typical questions would be: Can I produce without PAT? What are the benefits of using PAT, and how do I quantify them?  Do I apply PAT at the R&D, PD or production phase?
For an existing product application the questions are fundamentally different. Typically, they include: How do I work with the regulator on an existing process? What are the potential financial, quality and timing gains?

What are the risks and how do I mitigate them? An often unspoken concern also relates to answering the question - what happens if the PAT process reveals that my existing quality isn’t to the standard that I always thought that it was?
Step 2: Unit Operation Identification
Of even more importance is the question of which process to start with. Should the easiest or most complex be targeted? Or should it be the one that is the most stable or most problematic? Alternatively, what about the process with the highest potential financial or quality gains? Or should it be the process that is likely to be the fastest to deploy? The outcome of this targeting strategy is critical because of the high level of interest that is likely to be within the business in the outcome of the PAT trials on the 1st unit operation.

Summary of Steps 1 and 2
As a general rule, no matter whether the PAT implementation is going to be on a new or existing product, for your 1st PAT implementation choose a unit operation that is not too complex and where the PAT deployment can be carried out in a timely fashion. Even if only modest financial or quality benefits of using PAT are gained, provided that the project can be delivered in a timely fashion then it is better to follow this route than one where the financial or quality gains are much greater but will take much longer to deploy. If the 1st project becomes protracted then the sponsors can lose heart and interest with the result that PAT gets a bad name and its further adoption gets delayed.

By taking this approach the 1st project will be invaluable in terms of learning and will make further projects much more straightforward.  You can then move on with confidence and experience to the more complex projects that perhaps deliver more financial and quality gains. Further key recommendations are that PAT should be employed as soon as possible in the process development cycle; because PAT models developed early on in the development cycle can be hugely beneficial to the development cycle as a whole.

Step 3: Regulatory and Human Resources Acceptance
The importance of people in the success of a PAT Implementation Strategy is self-evident, and so engagement, both with the regulatory authorities, and with the staff concerned is vital.

Staff need to be advised of the strategy and objectives of the project to ensure that, as far as possible, everyone is ‘on side’. By engaging with the regulator early on you are able to set up a dialog in relation to your intentions and agree upon the way forward.  It may pay to engage a consultant in this step, as to approach this incorrectly may be to the detriment of the project, however if the project is based on sound science and your approach is correct then the regulators are invariably positive.

Step4: Put Team in Place
An in house audit is necessary, to identify all the available in-house skills and engage them on the project; and to identify skill sets that are missing. Any skill gaps can be filled either via recruitment or by connecting with an appropriate PAT services provider.  Again in the early stages it may pay to engage with a consultant – they can work with you on the project and advise you as to what skill sets are going to be required.

Step 5 Data Acquisition Technology
The next step is to identify the Critical to Quality Attributes (CQA) for the chosen target unit operation, and also the CQAs for the up and downstream processes. What also must be considered at this stage is that future CQAs may affect the choice of instruments made at the present.  For example it may in the long term be more cost effective to purchase a multi head instrument that can also in due course be used on an up or downstream process.                                                          
From identification of CQAs the next step is to determine how the raw data for determining the CQAs is to be measured.  This invariably requires the selection of one or more instruments that output multivariate (spectral) data and this may need to be combined with univariate data.

Instrument vendors or consultants can be of great assistance in determining the best instruments and the finer details relating to sampling position, probe types, sample sizing and the like.

Step 6: Data Management and MVA Technology
At this halfway stage of the implementation strategy, the next action is to identify the optimum PAT Data Management product. If a PAT Data Management product is not employed then the whole PAT model building exercise will take much longer, and unless bespoke systems are employed then there will be huge problems in maintaining control of all the data that is to be processed in a way that is acceptable to a GMP process and indeed the regulator. The PAT Data Manager should be: instrument neutral – use the most suitable for the process; MVA package neutral -use the best of breed; and control system neutral – so that there is no need to upgrade your control system. Other key requirements are: cost effectiveness – a low cost of purchase & ownership; scalability – a system that can be scaled to suit your system as it grows; ease of configuration – to enable use by a range of users; and flexibility – allowing the product to be deployed in many ways from laboratory to production.

Next, identify the optimum MVA package. This should have specific functionality for the process concerned and be a good fit with staff. Many clients will already have a site standard due to other pre-existing needs, however PAT may place different demands on such a type of package. MVA vendors and consultants can assist with the selection of the best MVA package to suit your project.
Step 7: Design Experiments for Model Building
Once the data management and MVA packages are decided upon, the stage is set to design the experiments required for model building. Before embarking, it is important to bear in mind that as the understanding of a process grows, the choice and number of CQAs may change. The experiments themselves should be optimised using DoE, and all data should be saved. Again a PAT consultant may initially be useful in assisting.

Data acquisition and data association are key activities at this stage. As regards data acquisition, both time- synchronised univariate and multivariate data need to be collected using the PAT Data Management product, and physical samples for off-line analysis in the laboratory have to be taken during the experiments at known points in time.                                    

It is very important for all the asynchronous data sources to be synchronised by your PAT Data Manager in order to ensure that all the data gathered at any one point in time is valid.  The time synchronisation of physical sample taking is also critical to ensuring that the data for model building is correct.  The samples that have been gathered should then be analysed retrospectively in a laboratory and the results added back into your PAT Data Manager and associated with the real time product data gathered at the precise moment that the sample was taken.

The multiple data sets of raw data and laboratory results are then collated together as a group within your PAT Data Management product and then exported to your MVA package.  All model building is executed within the MVA package, and the model is then imported into the PAT Data Management product.  By controlling the export of data and import of model directly between your PAT Data Management product and the MVA package then the provenance of the model is assured as all of the audit data can be stored together with the model.  Without a full creation history then the value of a model is significantly reduced – potentially it will only be suitable for non GMP operations.
Step 8 Process Model Testing
The model, or models, that are produced by the processes in the previous step, are loaded into instances of a real time prediction engine - these engines usually being supplied from the MVA vendor but being managed for real time use by the PAT Data Management package. Here time can be saved by testing models initially against historic raw data and comparing the ‘live’ results against laboratory results that had not been used in the model building, i.e. a ‘virtual’ process can be run.

Also time can be saved by testing multiple models concurrently – for example you may have several versions of a model that you would like to test and these could all be executed at the same time either virtually else with a live plant.

When running the actual process the operations are very similar to those executed during the model building phase, i.e. at known and recorded points in the process samples are taken for retrospective analysis.  The results are then compared with the historical CQA output value that the real time engine generated at the exact time that the sample was taken.

If the model is not generating the correct results then it needs to be refined and this can be done by gathering more data and laboratory results to optimise the model design and then repeating the testing regime. However it must be pointed out that at this stage the model will not be perfect, you should apply the 80:20 rule – i.e. if the model is capable of being used to produce product that is of acceptable quality (not optimised but adequate), then use the model and move on.

PAT projects can be lost at this stage by users trying to optimise a model and by not being pragmatic. PAT embraces continuous improvement and so by definition the models will not be perfect from day one, however after gathering the data from many batches over many months or years the models can be continuously improved to optimise the financial and quality gains. If too long is taken at this stage then project sponsors will tire of waiting and there is a real risk that the project will be cancelled and all future PAT projects will be vetoed for years to come.

Step 9: Develop Process Understanding
For a true PAT system, developing Process Understanding is essential. To achieve this, experiments should be designed to show how in real time the CQAs change with varying control parameters and raw material input quality – the parameters that are critical to the target CQAs – the Critical Control Parameters (CCPs) are therefore derived. The process should be run with all real time data being recorded by using the PAT Data Management system and being displayed in real time by this same package.    
By running the process in this way you will be able to study the effects of input process parameters on your CQAs in real time plus you will be able to retrospectively analyse the results to develop understanding of the mechanistics of the process.
When the understanding has been derived then it will be possible to predict how a CQA will change when an input parameter is changed.           

Step 10: Develop and Test Control Models
The Process Understanding derived from Step 9 is used to design and build the Control Model. At this concept stage, running the control model from within the Data Management product can have advantages. For example: running multiple instances of control model in open loop can speed up the development process, however there is no reason why it shouldn’t at this stage be deployed in your control system of choice – it really depends on the skill sets of the people used in the PAT project.

Initially, the process should be run in open loop. At this stage closed loop control of the process using the control system would not normally be attempted. An initial ‘soft’ option, (if possible and permitted by the process) is to use manual control derived from instructions from the control model.  Provided that the parameter change demands are within the licensed allowable operating envelope of the process then automatically prompted manual control can be undertaken to test the validity of the control model.

Running in open loop at this stage, and with all ‘traditional’ testing in place, offers the major advantage that the model development or improvement exercise can sometimes be undertaken during normal production. In this mode, all salient events are recorded into PAT Data Management system, and the results analysed, retrospectively, to adjust the control model as necessary. Then, with the model refined, an attempt should be made to run the process with full PAT control in place, initially on trial batches.  If the ‘soft’ option is not possible due to the licensed mode of operation then the control model testing will have to be conducted solely on trial batches.                   
The control model is then optimised as necessary; once again applying the 80:20 rule to manage those things that really make a difference to the results. When finally fit for purpose, the model can be deployed onto the production control system. Of course it should not be used until approved by the regulators and the company’s quality unit; and until approved, the production control system should be run with ‘traditional’ QA procedures in place

Step 11: Apply PAT Holistically
At this stage of the implementation process, PAT will have proved its worth on a single unit operation. As a result, the implementer should have management & peer ‘buy in’.

The time is now right to identify the next unit operation (or product) and repeat the process. In certain cases, it may be advantageous to trial PAT on a unit operation with the end goal solely being to gain deployment experience. PAT can then be applied onto a completely different and, perhaps, new product or process that is more complex, but you will be taking on this new project armed with experiences and skills derived from the first deployment.

A single unit operation is not truly PAT, however from this foundation of a quality-approved single unit operation, the development exercise is continually repeated until the whole production process has PAT applied to it. The stage is now set to obtain regulatory approval for real time release as per the PAT guidance. Remember: a science based approach together with detailed and robust Process Understanding is key to the approval process.
Once again, until approval is received, run with ‘traditional’ QA procedures in place. When approved, run the process with full PAT control.

At this stage a continuous improvement programme must be entered in to, where the process and control models are refined and optimised. Here it must be realised that optimisation may never truly cease; this is why models at the early stages must be pragmatic, otherwise PAT may never be deployed. The next step is roll out this approach across all other relevant processes.                                      

Step 12: The Future State
The implementer now has processes running under full PAT control. He has developed rigorous process understanding, and realises that the next objective – the next stage - for many processes is continuous manufacture. Previously impossible, the use of PAT makes controlled continuous manufacture a very real possibility. Such manufacture has huge potential benefits to the industry, and its potential impact cannot be overstated. However, to get to this stage, the implementer has to work with a wide range of skill sets and employ a wide range of technologies in order to develop a proven PAT system that produces product of improved quality faster and with less cost. This means working within a defined structure, using a risk based ethos combined with practical pragmatism.  As mentioned before, the approach is therefore the Practical Application of Technology.

About Optimal Automation
Optimal Automation is now one of the leading developers and integrators of Process Analytical Technology (PAT) solutions for the Life Sciences and Chemical sectors using its own in-house developed synTQ® data management software. The software is now used all around the world, following the signing of a Global Marketing Alliance with Emerson Process Management in Feb 2009.

Optimal is also a leader in vision systems - manufacturing its own 21 CFR Part 11 compliant packages – and is in the first rank of UK systems integrators, evidenced by its appointments as a Siemens Solution Partner, an Approved Systems Integrator for ABB, and a Software Solutions Provider for Rockwell Automation. In addition, Optimal has for some years been an approved Siemens WinCC Professional and PCS7 Integrator.   

About the Author

Martin Gadsby | Director