Application of Design of Experiment (DOE)Techniques to Process Validation in Medical DeviceManufactureTuesday, August 29, 2006 - Journal of Validation Technology, February 2006 Volume. 12, Number2This article has been nominated for the Journal of Validation TechnologyArticle of the Year Please give us your opinion and vote here.By Dr. D. Dixon, Dr. J. Eatock, Prof. B.J. Meenan, and M. MorganABSTRACTProcess validation is a requirement in heavily regulated industries such as the automotive and aerospaceindustries. Both the International Organization for Standardization (ISO) 9000/13485 and the U.S. Food andDrug Administration (FDA) Quality System Regulations (QSR) 820 Part 21 Code of Federal Regulations (CFR)“Quality Systems for Medical Devices,” require process validation as a regulatory requirement. Design ofexperiment (DOE) statistical methods will allow these requirements to be met in the most resource-efficientmanner possible, whilst providing a greater understanding of the process and highlighting opportunities forquality improvement.This paper provides an overview of the application of DOE techniques during key aspects of processvalidation and discusses two practical examples from medical device manufacturing. The attendant benefitsof this type of procedural approach to the medical device sector are considered. From the results obtained, itis clear that meeting process validation requirements can be taken as an opportunity to better understandand improve manufacturing processes. As such, DOE techniques are pivotal in effectively achieving theseaims in a manner that provides for regulatory compliance.INTRODUCTIONMany medical device manufacturers view validation and other regulatory requirements as a life science sectorburden. Whereas those companies which have developed a mature approach to regulatory compliance havean integrated approach, many others in this sector only consider this aspect once the main elements ofproduct and process development have been completed. When viewed from this perspective, meetingregulations can become, for the most part, both a paper generating exercise and a drain on resources.It has been considered that „The blind, bureaucratic approach [to validation] being followed by somecompanies is a needlessly expensive process that achieves nothing more than a temporary reprieve from theregulatory authorities.‟1However, these requirements can be taken as an opportunity to increase process understanding, ensure thatprocesses are operated under optimum conditions, improve quality, and reduce costs. Management may lackan understanding of what is required during a process validation (i.e.: to ensure that an acceptable productis produced under all foreseeable conditions). This often produces a mindset that simply aspires to have allthe relevant documentation completed as soon as possible.Ideally, process validation considerations should be integrated into the new product development processfrom an early stage, for example, when writing a product specification or user requirements document.Quality functional development (QFD) can be seen as a useful method for determining what the customerreally wants, and then translating such requirements into a definitive product specification.2 This customerfocused specification can then form the basis of the subsequent process validation acceptance criteria.A large number of companies in the medical device sector have adopted the six-sigma approach to qualityimprovement especially those that produce high volume products (e.g.: disposables). Based on total turnoverin 2004, 6 out of the 15 largest medical device companies worldwide promoted the fact that they haveadopted six-sigma on their respective company websites. The primary aim of the six-sigma quality method isquality improvement by reducing the defect rate and by effectively addressing customer needs.Six-sigma refers to six standard deviations between the mean and the specification limit equating to a defect

level of 3.4 defects per million opportunities (DPMO). In six-sigma thinking, process defects are seen asquality improvement opportunities and as such they should be exploited to improve overall quality. 3, 4 Thesix-sigma method uses a number of established statistical tools such as design of experiment methods (DOE)and statistical process control (SPC). The prevalence of six-sigma training has led to a greater industrialawareness of DOE techniques. However, it should be noted that no standards exist regarding the level ofstatistical training needed to obtain six-sigma qualifications such as green or black belt. Hence, this can leadto an inconsistency in its application across the sector.In recent years, DOE has been increasingly recognised as an essential tool for the validation of medicalmanufacturing processes,5 but in the authors‟ experience, the uptake of these methods by the medical devicesector has been limited. An investigation into UK companies across a range of industries reported that only30%, approximately, apply DOE techniques. This is due in part to the inadequate teaching of such statisticalmethods as DOE in university engineering courses.6 The widespread teaching of statistical techniques as partof six-sigma training may improve the situation.The term, design of experiments refers to a set of statistical approaches to experiment design and analysis.DOE is one of the key tools used in six-sigma, and staff (e.g., six-sigma black belts) trained in its use canalso apply the methodology to efficiently fulfil regulatory requirements. DOE is used to model the effect ofone or more process factors on the chosen outputs.In traditional experiments, only one factor at a time is varied while all others are held constant. In DOE allthe factors of interest can be investigated in a single trial, minimizing the size of the experimental schedulerequired and providing information on key process interactions. This latter aspect is of importance sinceoptimal conditions may occur when one factor is high and another low. This type of information oninteractions between factors cannot be easily obtained by investigating the effect of each factor separately.Types of DOE trials include single factor, two-level, factorial designs, Taguchi methods, mixture methods,and response surface models. Trials which only study each factor at a high and low (so called two-level trials)level are useful for determining the important sources of variability for a process and are, therefore, oftenreferred to as „screening trials.‟ Two-level trials are useful for highlighting the critical factors for furtherdetailed study. Taguchi methods are often used to design robust processes by locating the operatingconditions that minimise product variation.7 Response surface experiments are used to map the effect ofvarying two or more factors across a range of values. Other experimental types include mixture trials forinvestigating product formulations.A DOE approach permits efficient use of resources (personnel time, machine time, materials, etc.), providesdetailed analysis, gives information on reproducibility and errors, and provides a predictive capability. 8Applying DOE reduces the size and hence the cost of process validation trials. It is a regulatory requirementto run sufficient trials to demonstrate the statistical significance of results, and DOE can assist in thisproceedural aspect.Some general statistical packages such as Minitab and SPSS have DOE capacity and dedicated software isavailable, including Design-Expert, PQ systems and JMP. The dedicated packages tend to be userfriendly as they lead you through the design and analysis of the experimental outcomes, and often have amore comprehensive DOE capability. First, one decides on the experimental type (e.g., two levels) and thenthe factors and levels are entered. At this point, the software will suggest the number of runs necessary. Onecan increase the accuracy of the trial by adding replicate runs or by reducing the trial size by 1/2 or 1/4factorial design. Unlike in a full design, a fractional design does not require tests at every combination offactor levels. The software then generates the experimental design showing the factor levels for each run.The experiment is then conducted, samples produced and tested, and the data entered into the software.The analysis of the results often consists of determining significant factors, fitting a model to the results, andchecking for outliers. The results can be plotted in various forms including surfaces, box plots, etc. Despitethe ease of use of many software packages, it is still necessary to understand the assumptions made and theunderlying statistical techniques used, such as ANOVA, to avoid drawing incorrect conclusions. For example,a two-level design assumes a linear response between the low and high levels.Process Validation and DOEProcess validation is defined under the FDA quality regulations QSR/Good Manufacturing Practice (GMP) as„establishing documented evidence which provides a high degree of assurance that a specific process willconsistently produce a product that meets predetermined specification and quality attributes.‟9Validation indicates that a process has indeed been subjected to such scrutiny and that the results of theprocess can be practically guaranteed. The complete validation for a product will include both design andprocess validation.The European Union (EU) and FDA regulations operated in the medical device sector do not require all

manufacturing processes to be validated. Under FDA regulations if a process‟ output cannot be fully verifiedduring routine production by inspection or test, the process must be validated according to establishedprocedures [820.75(a)9]. If the output of a manufacturing process can be verified by inspection and it issufficient and cost effective to do so, then validation is not necessary.10 The manufacture and testing ofwiring harnesses is an example of a process that can be satisfactorily covered by verification by testing eachproduct. Validation is vitally important if the predetermined requirements can only be assured by destructivetesting. Examples include testing for mechanical properties, heat treatment processes, sealing and bondingoperations, and sterilization processes.It is important to implement a quality assurance program that is appropriate to the device beingmanufactured.11 The manufacturer of a simple low-risk (e.g., Class 1) device that is not exempt fromQSR/GMP regulations, does not need to implement as extensive and detailed a validation and control plan asthat required for a complex high risk Class III device.Process validation is normally broken into a number of stages:Installation Qualification (IQ) establishes that all key aspects of the equipment installationadhere to the manufacturer’s specification including: equipment description, installation andsupplies, environment, calibration, maintenance, and operator training.Operational Qualification (OQ) demonstrates that the equipment consistently operates tospecification under normal conditions including: testing of alarms, software function,extremes of operating ranges, and machine consistency.Performance Qualification (PQ) produces product within specification when operated underchallenged conditions.Validation is then an umbrella term for the set of qualifications for a particular process. These qualificationstages are not defined in an identical manner in GMP/QSR and ISO standards e.g.: GMP Part 820 refers toprocess performance and product performance stages. However, it is not necessary to divide validation intothese stages. Many manufacturers have a single document to cover both the IO and OQ.12 The ISO 13485and QSR/GMP Part 820 regulations themselves do not provide specific guidance on how to conduct avalidation. Various bodies, including the regulatory authorities, produce guidance documents to offer morepractical and specific help.5, 6, 8 However, unlike the ISO and GMP/QSR Part 820 quality standards, theseguidance documents are optional.13 It should be possible to meet all quality system validation requirementsapplicable to medical device manufacture with one procedure using the GHTF (Global Harmonisation TaskForce) document as a guide.10Regardless of the way in which the validation is subdivided, a protocol is written for each stage stating thetests that will be conducted and the acceptance criteria (based on the product specification). The tests arethen conducted and a report is written based on the protocol. If the acceptance criteria are not met, it maybe necessary to make changes to the process and repeat the qualification. It is often necessary to analysehistorical data or conduct a prequalification study of equipment before writing the validation documentationsince it is only possible to write a qualification protocol for a well-understood and stable process.It is a requirement of the validation process to monitor the long-term stability of a validated process.Statistical process control (SPC), another technique from the six-sigma toolbox, can be used to analyse thelong-term stability of a process and to highlight out of control situations before they become critical.This paper discusses two examples from the authors‟ practical experience of the ways in which DOE can beapplied to process validation. First, the use of a response model type DOE in determining process operationlimits, and second, the application of a two-level design for challenging a process under worst case operatingconditions.Case Study 1.Use of DOE to Establish Process LimitsDevice and process specifications must be established before a piece of equipment can be validated, becausethis information is used to write the protocol.10 Equipment can then be tested to ensure that it is capable ofoperating consistently within these limits. The acceptance criteria for a capability analysis are often set atCpK 1.33. The critical process capability factor (CpK) is defined as (Specification Limit-Mean)/ 3s.The effect of altering process parameters on product quality can be established using DOE, thus defining anacceptable window of operating conditions, i.e.: the process limits. This trial was conducted as part of thepre-qualification study to define the allowable operating conditions.

The heat-sealing of medical packaging is an example of a manufacturing process requiring validation, as it isnot possible to test seal integrity non-destructively. In the process, a heated platen tool is used to thermallyseal together the two sides of a medical package. Failure of a seal during sterilisation or handling wouldresult in loss of sterility and thus affect patient safety. The process variables during the sealing operation areplaten temperature, pressure applied, and seal time. Each of these variables must be controlled within thecorrect limits to achieve acceptable seal quality. DOE is the most efficient method to determine these limits.The table in Figure 1 illustrates the high and low levels for each of the three variables, the influence of whichcan be investigated using one compact DOE trial.14 The range investigated for each factor is chosen based onknowledge of the system. Traditional testing methods involve using a matrix approach to assess theacceptable processing window on form, fill, and seal equipment.14 Using this method, ten temperatures, threepressure levels, and five dwell times were investigated resulting in 150 separate runs. Using a DOE approach,it was possible to complete the trial with only 20 runs. This approach is, therefore, very efficient in terms ofresources required and is also more robust statistically.14Figure 1Heat-sealing Process Factor LevelsSince in DOE trials multiple variables are varied in a single run, this makes it possible to fully comprehendthe effects of all treatments and the interactions between them while limiting the size of the study. For thistrial, a central composite type of response surface model (RSM) was chosen. This method studies eachvariable at 5 levels over the 20 runs. The required number of samples produced and tested for each rundepends upon the number of factors involved, the process variability, and by what is deemed to be asignificant change in the output.A software package (Design Expert 6.0) was used to design the experiment, i.e.: generate the conditions foreach of the twenty runs and analyse the results. The software randomises the run order to reduce theinfluence of external factors. If, for example, one was to start with all the low temperature settings and thenthe high temperature runs, a noise factor such as material variability may be wrongly attributed to thetemperature change. A randomised run order is a critical aspect of DOE. Although in certain circumstancesthis may lead to prolonged change over times between runs. For example, it may be necessary to allow theequipment to cool when changing the temperature between runs.In this study, the equipment was set up under the conditions needed for each run and a number of sampleswere then produced and tested. The results were then entered into the software. In this case, acceptancewas based on a specification being met in terms of peel strength of the bond formed during the sealingprocess. This is normally the specification from the data sheet that is used to sell the product and, therefore,its correctness is paramount.Figure 2 illustrates the relationship between dwell time, temperature, and seal strength on a contour plot.The values on the contours correspond to the measured bond strength in N/mm2. Of the three processvariables, pressure was found to have only a minimal effect on seal strength and it was, therefore, decided tofix this at the normal operating level of 65 psi. It should be noted that longer dwell times and highertemperatures result in greater bond strength. This is due to the higher heat input and, therefore, greatermelting at the interface, which consequently increases the bond strength. Short dwell times are desirable forhigh machine throughput. However, higher temperatures are required at short dwell times in order toachieve the required amount of heat transfer. The maximum temperature is limited to 150 C, as highertemperatures cause thermal damage to the packaging materials.Figure 2Effect of Temperature and Time on Seal Strength

The design specification for the packaging calls for a peel strength of 15.5N/mm2. Therefore, the acceptableoperating window can be defined as being between 1.5-2 seconds duration and 130 -145 C, correspondingto the box in the upper right hand corner of Figure 2. This operating window provides for a safety margin inthe product specification. Hence, the DOE method of investigating the effect of process variables allows thelevels of the operating limits to be evidence-based and ensures that the process is being operated underoptimal conditions.Capability analysis is often used as part of the OQ stage of process validation. This is conducted to prove thatthe process is capable of consistently operating within the set limits (e.g.: temperature between 130 145 C). As mentioned earlier a CpK value of 1.33 is often defined as the acceptance criteria. The normaloperating conditions should be set at the centre point of the operating window, in this case 1.75s and137.5 C. SPC could then be used to monitor temperature and time on a continuous basis to ensure that theprocess remains in control.Case Study 2.DOE for Determination of the Worst Case Operating Conditions during Process Qualification (PQ)The IQ and OQ phases of a validation demonstrate that equipment is installed correctly, functions asdesigned, and can operate consistently within the required limits. It is often necessary, however, to challengethe process under worst-case conditions during validation; this is commonly conducted as part of the PQstage. The GHTF states that:“the challenge should include the range of conditions as defined by operators‟ practices and procedures asestablished in the OQ phase.”10Challenges should include the range of conditions allowed in the written standard operating procedures(SOPs) and should be repeated enough times to ensure that the results are statistically meaningful.Challenges may need to include forcing the preceding process to operate at its allowed upper and lowerlimits. The PQ phase proves that an acceptable product is produced under all foreseeable operatingconditions.It is not always necessary to manufacture product during the OQ procedure, but it is a prerequisite forprocess qualification (PQ). For example, it would be possible to demonstrate the operation as part of an OQof an autoclave used for sterilisation by recording oven temperature without actually sterilizing equipment.The process qualification study presented here concerns the reflow solder process used in the manufacture ofPCBs (printed circuit boards) for medical device applications. Solder paste is first applied to blank PCBs;components are then robotically placed in the correct locations onto the paste. The board then travelsthrough a reflow solder oven, where the paste melts and then solidifies on exiting the heated zone to holdthe components in place and form the electrical connections. The reflow oven, therefore, consists of an airoven with a number of temperature zones; PCBs are fed though the system on a conveyor belt andexperience a temperature profile.

A trial was not necessary to establish process limits for this process, as industry standards exist regardingthe required temperature profile. The specification states the maximum and minimum, heating rate, peaktemperature, cooling rate, and time above 183 C (solder melting point). During the capability analysis in theOQ phase a DatapaqTM data recording system was used. Thermocouples are connected to the data recorderwhich is placed in a thermally insulated box. This is then sent through the oven in place of a PCB and storesdata from a number of thermocouples; the temperature profiles are then downloaded onto a PC.Figure 3 is an example of a typical temperature profile from the reflow oven used. Data was logged from sixthermocouples placed across the width of the conveyor system corresponding to the six traces provided inFigure 3. This is to ensure that all points on the PCB experience the correct profile. Thirty such profiles wereobtained for this capability study to demonstrate that the oven operated consistently within the requiredlimits. Acceptance was based on a Cpk value of greater than 1.33 for each of the profile specifications (e.g.:heating rate, peak temperature, etc.).Figure 3Reflow Oven Temperature ProfileTo ensure that the process continually produces product to the required specification under all foreseeableconditions, PCBs were produced under worst-case conditions during the PQ stage. Depending upon thenature of the process and its sensitivity, causes of variation may include changes in environmental conditions(ambient temperature, humidity, etc.), drifting in process parameters, human factors (ergonomic factors,training), and material variability. A brain storming session could be used to create a list of possible sourcesof variability. This should be based on knowledge of the system and historical data. The challenge may needto include forcing the preceding process to operate at its allowed upper and lower limits.Brainstorming produced the three possible sources of variation illustrated in Figure 4.The use of a two-level DOE screening experiment was used to demonstrate that an acceptable product isproduced under challenged conditions, determine the important sources of variability and highlight the worstcase conditions. This approach allows the trial to be conducted with the minimum number of runs and givesinformation on interactions that would not be available if each possible cause of variation was investigatedindividually. The worst-case conditions will occur at such interactions, e.g.: a high loading rate combined withprocess start up. A simple two-level factorial experiment can rapidly increase the user‟s knowledge about thebehaviour of the process being studied.16Testing at every combination of the three factors shown in Figure 4 would require an experiment with eightruns. Alternatively, a 1/2 fractional design could be used halving the number of runs required. A general ruleof thumb for two-level experiments is that a minimum of twice as many runs as factors is required, e.g.: tenruns to investigate five factors. A disadvantage of such two-level designs is that a linear response is assumedbetween the low and high points. Conducting an extra run at the centre level of all the variables can be used

to check that the response is linear.16Figure 4Possible Sources of Variation in Reflow Oven OperationThe number of samples required for each run will depend upon the confidence level required (e.g., 95%), thesize of effect one is trying to measure, and the variability inherent in the process.Once boards had been manufactured under each set of conditions, they were tested visually and by flyingprobe. The analysis of these trials revealed which, if any, of the possible sources of variance had astatistically significant effect on product quality. To fulfil the purpose of validation, steps can then be taken toreduce or remove these sources, e.g.: by installing air-conditioning to maintain ambient temperature or byaltering the standard operating procedures (SOP) to change operator behaviour.The acceptance criteria could be based on clinical or user needs. The acceptance criteria for the PQ werebased on the maximum number of minor defects allowable. This limit could have been based on normaldefect rates from historical production data. In this case, the maximum number of defects allowable wasdefined as:UCL A v(3A)Where:A the average defect rate for all trial runsUCL* the maximum allowable defect rate for any run* Upper Control LimitThis commonly used acceptance criteria is based on a Poisson distribution.17 The average defect rate wascalculated for each run, and the defect rate is acceptable if no run averages exceed the upper control limit.The defect rates for all runs were found to be below the upper control limit. This provides evidence that noneof the sources of variance investigated had a large detrimental affect on product quality. If any of thevariables had a significant effect on the defect rate, then the process could be altered to reduce thesesources of variability. If for example, loading rate was found to significantly affect the defect rate, then newSOPs could be written to change the operators‟ loading methods. The acceptance criteria must be set out inthe protocol for the qualification before any tests are conducted.SUMMARYValidation is a regulatory requirement in both FDA and European medical device regulations. Statisticaltechniques such as DOE and SPC, which are taught in six-sigma training, can be usefully applied to processregulation. The use of DOE methods, in particular, ensures that the necessary trials are conducted in themost resource efficient manner possible. It is, however, vital that staff have sufficient understanding of theassumptions and modelling methods used in DOE. This knowledge could, for example, be gained byattending a short course and working through several practical examples.DOE methods allow a large number of variables to be investigated in a compact trial, enable outliers in thedata to be identified, and provide detailed process knowledge.The two case studies presented here show that there are indeed tangible benefits to be gained from suchapproaches. These benefits can result in better production control, i.e.: an increase in product yield whilst atthe same time contributing to regulatory compliance.Article Acronym ListingC CentigradeCFR Code of Federal Regulations

DOE Design of ExperimentDPMO Defects Per Million OpportunitiesEU European UnionFDA Food and Drug AdministrationGHTF Global Harmonization Task ForceGMP Good Manufacturing PracticeIQ Installation QualificationISO International Organization for StandardizationOQ Operational QualificationPCB Printed Circuit BoardPQ Performance QualificationPSI Pressure per Square InchQFD Quality Functional DevelopmentQSR Quality System RegulationsRSM Response Surface ModelSOP Standard Operating ProcedureSPC Statistical Process ControlU.S. United StatesUK United KingdomREFERENCES1. Johnston R., 1995, “Validation Proof or Document Elegance,” Medical Device andDiagnostic Industry, Vol. 17, Part 6, pp. 20-22.2. Booker J.D., 2003, “Industrial Practice in Designing for Quality,” International Journal ofQuality Reliability and Management, Vol. 20, No. 3, pp. 388-203.3. Pande, P.S, Mewman, R.P. and Cavanagh, R.R., 2000, The Six Sigma Way, (McGraw Hill,ISBN 0-07-135806-4).4. Coronado, R.B. and Antony, J, 2002, “Critical Success Factors for the Implementation of SixSigma Projects,” The TQM Magazine, Vol. 14, No. 2, pp. 92-99.5. Mark, J. et al, 1999, “Design of Experiment for Process Validation,” Medical Device andDiagnostic Industry, Vol. 21, pp. 193-199.6. Anthony J, 2000, “Improving the Manufacturing Process Quality and Capability UsingExperimental Design: A Case Study,” International Journal of Production Research, Vol. 38,No 12, pp. 2607-2618.7. Reddy RBS and Babu A.S. 1998, “Taguchi Methodology for Multi Response Optimisation,”International Journal of Quality Reliability and Management, Vol. 15, pp.646-668.8. Montgomery, D.C., 1996, Introduction to Statistical Quality Control, 3rd Edition, (New York:John Wiley & Son), p. 478.9. Good Manufacturing Procedures, U.S. Food and Drug Administration, Process Validation,Section-820, Part 4.10. Quality Management Systems, Process Validation Requirements, Jan. 2004, Edition 2,Global Harmonization Task Force.11. Guidance on the General Principles of Process Validation, May 1987, Guidance Docu

Validation indicates that a process has indeed been subjected to such scrutiny and that the results of the process can be practically guaranteed. The complete validation for a product will include both design and process validation. The European Union (EU) and FDA regulations