In 1986 Motorola introduced the concept of Six Sigma as a means of standardising and then reducing defects in electronics manufacture. Six Sigma didn’t just pop up out of the blue; the statistical science behind it has evolved and enjoyed various eureka! moments over the past four centuries. The work of Leibnitz, DeMoivre, Laplace and Gauss in the eighteenth and nineteenth centuries, and Shewhart and Deming in the last century, established the statistical principles that scientists, engineers and economists now take for granted to explain or predict chance outcomes and probabilities of their occurrence.
The philosophy behind Six Sigma shows that it is possible to statistically control any process as long as you know what your target outcome is and understand what you are able to measure. Unfortunately, there is often a mismatch between what a process can achieve on a consistent basis and what the design requirements demand. This applies in virtually any industry, whether you are making precision instruments, pills, potions or pickled onions! Unless the product originator (designer, engineer, scientist and so on) shares an understanding of the production processes with those who control them (toolmaker, machinist, process engineer, manufacturing chemist …) and vice versa, consistency of quality and acceptable yields will never be guaranteed.
At its simplest level, process capability compares output — assuming an incontrol process — with specification limits. For example, a centred process with a capability (Cpk) of one translates into a yield of 99.73%. This sounds great until you realise it means there will be 2,700 parts per million which fall outside of the specification. However a process with a Cpk of two gives a yield of 99.9999998%, or just two nonconforming parts per billion.
A drug delivery device design has to be ‘capable’ before it is industrialised for manufacture. Yet even the most experienced engineers can sometimes be falsely convinced by the performance of a single prototype and believe that exactly the same reliability can be replicated in production when this is very rarely the case.
The challenge of high volume manufacture
Designing drug delivery devices for manufacture can be challenging especially considering that production volumes can be very high; for example, GSK have manufactured nearly one billion Accuhaler/Diskus dry powder inhaler devices since product launch in 1995, roughly 50 million units per year.
Each device that leaves the production line must work first time, every time, highlighting the fact that design and production capability is not just important, it is absolutely essential.
An engineer can do for a nickel what any damn fool can do for a dollar. – Henry Ford
Functionality, quality and manufacturability have to be ‘designed in’ to such essential high volume products, not ‘added later’. This demands that the design is informed by a good understanding of the production processes, which must include early dialogue with the manufacturer. A balance is essential between the theoretical understanding (the use of design tools such as mathematical modelling for example) and practical understanding of production processes and control techniques. Specifying a design that demands excessively tight tolerances from a manufacturing process is usually a bad idea; if the design for a new injector pen or inhaler is so complex that it can only be produced by Rolex, the cost of goods is going to be a problem! The obvious deduction is to engineer capable device designs, which meet all performance, safety and cost objectives, with components designed to achieve agreed and rational Cpk indices in manufacture.
Ultimately, if the aim is for a ‘near zero’ defect level, it must be acknowledged that it is unrealistic to set up in production, build a million units and then see how many fail in order to validate the design. A thorough, statistically based predictive approach, grounded on a sound understanding of the product performance limits and of the production processes involved, renders this sort of ‘suck it and see’ method obsolete.
For the pharmaceutical industry, acceptability based on testing completed batches may have been adequate when profit margins were high, formulations relatively simple and regulatory requirements less onerous. This simply won’t do today, especially where biologics and complex products are concerned. Process Analytical Technology (PAT) represents a major departure from the traditional approach of testing the quality of the finished product (‘Is this batch of white powder OK? Does it meet the specification limits? Yes = good, No = bad).
Henry Ford is quoted as saying: “An engineer can do for a nickel what any damn fool can do for a dollar”. It is worth observing that a family saloon car comprises around 30,000 parts down to nut and bolt level, yet we take car reliability, exemplified by 100,000 mile or seven year warranties, for granted. The only way car manufacturers can do this is by adopting designs based on a sound understanding of the production processes required to hit capability targets. If this approach works for a volume manufactured product with 30,000 parts, it should also apply to a drug delivery device with just a handful of components.
Theoretical mathematical modelling techniques, such as statisticallybased tolerance analyses, are powerful tools to help direct engineering design. A realistic tolerance model should start with the identification of the performance objectives of the ‘critical-to-function’ and safety interactions. These should then be translated into critical dimensions (on the 2D component drawings) with tolerance limits grounded in a realistic representation of the intended production process. For injection moulded plastic components, standards such as DIN 16742 provide a good starting point for the identification of achievable tolerance limits. For complex, multi-dimensional assemblies, a deterministic arithmetic method of finding the variance may well be unfeasible. In such cases, a ‘Monte Carlo’ simulation, using a random sampling-based computational algorithm, has proved useful for tolerance investigations when complex trigonometry or other calculations render a deterministic arithmetic method of establishing variance unfeasible. This is generally the case when it is not possible to describe a function using onedimensional tolerance stacks.
Avoiding design paralysis
Having stated the importance of effort up-front to ensure that a design is suitable and capable for industrialisation, it should also be emphasised that too much analysis, too early, risks ‘design paralysis’. As the design development progresses, the emphasis on manufacturing considerations and capability should be increased. Early activities may include consideration of the component construction for tooling, material options and identifying the critical design interactions, eventually leading on to more detailed considerations such as tolerance analysis modelling and component detailing to accommodate specific production processes.
Even with all of the analysis in place, analytical models cannot provide definitive proof of design validity.
Characteristics such as component flexing or friction can be difficult to predict precisely and moulding tolerances do not allow for warping in the components. Therefore empirical test methods must also be used to determine whether the design can be verified.
In pharmaceutical development, PAT shifts the concept of product quality to a more dynamic approach than has been ‘traditional’. As variability in input materials, personnel, manufacturing equipment and conditions will always be present, monitoring and compensating for such variability to maintain predictable and acceptable output is therefore required. PAT requires the definition of the Material Attributes (MA) and Critical Process Parameters (CPP) which affect the Critical Quality Attributes (CQA), and this finds parallels in other industrial sectors. The metrics and attributes to be monitored and controlled are different from those key to the production of drug delivery devices, however the principles are very similar.
Learning lessons from food and brewing
Interestingly, the food and brewing sectors have been dealing with similar process requirements and constraints to those faced by pharma for a long time — and these sectors have generally embraced capability in manufacture and key aspects of Six Sigma philosophy. The food sector has sometimes been regarded as the ‘poor relation’ by pharma, yet a great many of the issues faced by food and pharma are the same. One of the big differences is financial; food margins tend not to allow the wholesale discarding of non-conforming batches, so process control methods, together with an understanding of CPP, CQA and Cpk, have (for most successful organisations in the food sector) been essential to a profitable business.
Some aspects of implementing QbD, PAT and the overall principles of ‘capable manufacture’ are pretty challenging, especially if the objective is the production of high molecular weight proteins or complex dosage forms. But it has to be said that the pharma industry has previously got away with what were, in effect, large scale laboratory setups, rather than ‘modern’ production techniques because it was acceptable to rely solely upon batch release testing, with the probability that some complete batches will be rejected.
The food and brewing sectors have been dealing with similar process requirements and constraints to those faced by pharma for a long time.
Change happens slowly in pharma, in part because the industry is highly regulated (with different regulatory expectations in different territories), but also because historically, profit margins were large and lessons from outside the sector were absorbed very slowly, if at all.
Nevertheless, it appears that all of us involved in the pharma sector are arriving at an accommodation or understanding of the same fundamental principles (some a bit earlier than others). It’s fair to say we are all beginning to speak dialects of a common language, and starting to think in terms of Cpk, CPP, CQA, MA, PAT and QbD, generally understanding not only what those in other areas are talking about but also why, whether they are from clinical, formulation, device development, manufacturing, fill/finish or wherever.
So, if we are all converging on the same standards, could this be a good time to step outside a bit more often and see what else the smart folks are doing in other industries? And also to see where else in our own sector these outside principles are not yet appreciated? OK, this may not provide easy solutions to all aspects of pharma, but if these principles could help sort out some of the more controllable problems — and thus free up intellect, resources and capital to focus on more difficult issues truly unique to the sector — surely that can only be a good thing?
About the authors
– Andy Fry founded Team Consulting in 1986. He has been involved in development of nasal delivery devices since the early 1990s and is a named inventor on a number of patents.
– Brennan Miles is a senior consultant and has worked on a range of surgical and drug delivery products including dry powder inhalers, injectors and ophthalmic devices.