02-Peng-Using-Control-Chart-to-Evaluate-Process-Va

1UsingControlChartstoEvaluateProcessVariabilityDanielY.Peng,Ph.D.QualityAssessmentLeadOfficeofProcessandFacility(OPF)OPQ/CDER/FDAPQRI2015AnnualMeetingNorthBethesda,MarylandOctober5,2015WalterAndrewShewhart(1891-1967)Aphysicist,engineerandstatistician“Fatherofstatisticalqualitycontrol”–“Statisticalmethodfromtheviewpointofqualitycontrol”(1939)“CreatorofPDSA(Plan,Do,StudyandAct)cycle”“Creatorofcontrolchart”Originatorofthe“ChanceandAssignablevariation”concept2“Uncontrolledvariationistheenemyofquality”3Dr.W.EdwardsDeming(1900-1993)4SourcesofVariationVariationexistsinallprocesses.Variationcanbecategorizedaseither:–ChanceorCommoncausesofvariation•Inherenttoasystem,random,alwayspresentandhencepredictablewithinstatisticallimits•Eliminateinherentvariability(noise)isdifficult–AssignableorSpecialcausesofvariation•Exteriortoasystem,non-random,notalwayspresent(intermittent)•cancausechangesintheoutputlevel,suchasaspike,shift,drift,ornon-randomdistributionoftheoutput.•Areusuallyeasiertobedetected,controlledoreliminated5ControlChartDefinition:agraphicaldisplayofaproductqualitycharacteristicthathasbeenmeasuredorcomputedperiodicallyfromaprocessatadefinedfrequencyEverycontrolchartconsistsof:–Asetofdata–Acentralline(CL)(mean)–Twostatisticalprocesscontrollimits(UCLandLCL)(IstheprocessStable?)UpperandLowerSpecificationLimits(USLandLSL)–Patient’sneed(SafetyandEfficacy)(IstheprocessCapable?)Qualityattribute(unit)Sample#4.05.06.030405060USLLSLCLUCLLCLPotentialApplicationsToproactivelymonitorandtrendaprocessTodetectthepresenceofspecialcausevariationToidentifycontinualimprovementopportunitiesTomaintaintheprocessinastateofstatisticalcontrol–Usingscienceandrisk-basedapproach–Takeactioninatimelymanner6KeyConsiderationsforConstructingaControlChart7ChoiceofProductQualityCharacteristicsCriticalQualityAttributes(CQA)–Aphysical,chemical,biologicalormicrobiologicalpropertyorcharacteristicofanoutputmaterialincludingfinisheddrugproductthatshouldbewithinanappropriatelimit,range,ordistributiontoensurethedesiredproductquality(ICHQ8)–IdentificationofCQA:primarilybasedupontheseverityofharmtothepatient(safetyandefficacy)Critical(input)materialattributesandcriticalprocessparameters(CMAs/CPPs)Otherrelevantprocesscharacteristicsthatcanassistinprocessmonitoringandcontrolling89TypesofControlChartVariableControlChart–Characteristicsthatcanbemeasured(continuousnumericdata)e.g.Assay,Dissolution,%ofImpurity…–Theaverageandvariabilitychartsareusuallypreparedandanalyzedinpairs•Average–Rangechart(Xbar-Rchart,subgroupsize2-10)•Average–StandardDeviationchart(Xbar-Schart,subgroupsize10)•Individual–MovingRangechart(I-MRchart,n=1)AttributeControlChart–Characteristicsthathavediscretevaluesandcanbecounted,e.g.%defective,#offailedbatchesinamonth–pchart/npChart:forfractionofoccurrenceofanevent-Binominaldistribution•e.g.%ofunsuccessfulbatchatafacilityeverymonth–cchart/ncChart:forcountsofoccurrenceinadefinedtimeorspaceincrement-Poissondistribution•e.g.numberofparticulatematterinaninjectionvialOthertypesofcontrolchart:–cumulativesumcontrolchart(CUSUM)–exponentiallyweightedmovingaveragecontrolcharts(EWMA)10SubgroupSizeandSamplingFrequencySubgroup:theobservationssampledataparticulartimepointSubgroupSizeandSamplingFrequency(NxK)–Thenumberofobservationsineachsubgroup:1n–theobjectiveofthemonitoring(detectlargeorsmallshift)–howquicklytheoutputrespondstoupsets–consequencesofnotreactingpromptlytoaprocessupset–timeandcostofanobservationRationalSubgroup:–Minimizethevariationofobservationswithinasubgroup–MaximizevariationbetweensubgroupsStatisticalProcessControlLimitsUCLandLCL:–thethresholdsatwhichtheprocessoutputisconsideredstatisticallyunlikely–typically,±3SD(Shewhartlimits)Rationale:tobalancethetworisks:–Failingtosignalthepresenceofaspecialcausewhenoneoccurs;–Falsealarmofanout-of-controlsignalwhentheprocessisactuallyinastateofstatisticalcontrol11Howout-of-controlpointsareidentified?RuleNo.1–anypointfallsoutsideUCL/LCLOtherRules–certainnonrandompatternsoftheplotteddata–UseitjudiciouslyRiskof“falsealarm”128WesternElectricRulesOver-Reactionvs.No-Reaction13“Proceduresshoulddescribehowtrendingandcalculationsaretobeperformedandshouldguardagainstoverreactiontoindividualeventsaswellasagainstfailuretodetectunintendedprocessvariability”(2011FDAProcessValidationGuidance)Controlchartandprocesscapabilityanalysisoftengohand-in-handIllustrativeExamples1415WithinBatchVariabilityExampleERcoatedbeads,mixedwithextra-granularcushioningexcipientsandcompressedintotabletsCompression:~5h,samplefrequency:every8-10min(total33subgroups),subgroupsize=6162516021539151514521429135613381309123310544540353025TimeSampleMean__X=29.25UCL=31.75LCL=26.751625160215391515145214291356133813091233105410.07.55.02.50.0TimeSampleRange_R=5.17UCL=10.36LCL=0111111111111Xbar-RChartofDisso@240minNotStable&NotCapable16BetweenBatchVariabilityExample25232119171513119753110210098BatchNo.SubgroupMean__X=100.287UCL=102.108LCL=98.466252321191715131197531420BatchNo.SubgroupRange_R=1.78UCL=4.582LCL=025201510510