Validation in pharmaceutical analysis. Part I An integrated approach

24(2001)755–767

www.elsevier.com/locate/jpba

Validationinpharmaceuticalanalysis.

PartI:Anintegratedapproach

JoachimErmer*

A6entisPharmaAG,GlobalAnalyticalDe6elopment,QOTSS,Poseidonhaus,D-65926FrankfurtamMain,Frankfurt,Germany

Received16May2000;receivedinrevisedform16November2000;accepted17November2000

Abstract

TheICHguidelinesachievedagreatdealinharmonisingthedefinitionsoftherequiredvalidationcharacteristicsandtheirbasicrequirements.However,theyprovideonlyabasisforageneraldiscussionofthevalidationparameters,theircalculationandinterpretation.Itistheresponsibilityoftheanalysttoidentifyparameterswhicharerelevanttotheperformanceofthegivenanalyticalprocedureaswellastodesignpropervalidationprotocolsincludingacceptancecriteriaandtoperformanappropriateevaluation.Inordertofulfilthisresposibilityproperly,thebackgroundofthevalidationparametersandtheirconsequencesmustbeunderstood.Inthispart,thegeneralconceptofanintegratedvalidationisdiscussed.TheinterdependenciestootherICHguidelinesandtopicsduringdrugdevelopment(e.g.impuritiesanddegradants,stabilityandspecificationdesign)mustbetakenintoaccounttodefinetherequiredacceptancecriteria.Evaluationoftheresultsinordertoprovethesuitabilityoftheanalyticalproceduremustbebasedonthespecificationlimits.Importantparametersandaspectsarediscussedfortheindividualvalidationcharacteristics.Inthefollowingparts,theseparameterswillbediscussedindetail.Exampleswillbegivenfortheirinterpretationinordertofacilitatetheselectionofparameterswhicharerelevanttotheperformanceandsuitabilityofthegivenanalyticalprocedure.©2001ElsevierScienceB.V.Allrightsreserved.

Keywords:Pharmaceuticalanalysis;Validation;Qualityassurance;Acceptancelimits;Specificationlimits

1.Introduction

Thevalidationofanalyticalprocedures,i.e.theproofofitssuitabilityfortheintendedpurpose,isanimportantpartoftheregistrationapplicationforanewdrug[1–8].TheInternationalConfer-enceontheHarmonizationoftheTechnicalRe-*Tel.:+49-69-30584890;fax:+49-69-30525538.

E-mailaddress:joachim.ermer@aventis.com(J.Ermer).

quirementsforRegistrationofPharmaceuticalsforHumanUse(ICH)hasharmonisedthere-quirementsintwoguidelines[7,8].Thefirstonesummarizesanddefinesthevalidationcharacteris-ticsneededforvarioustypesoftestprocedures,thesecondoneextendstheprevioustexttoin-cludetheexperimentaldatarequiredandsomestatisticalinterpretation.Theseguidelinesserveasabasisworldwidebothforregulatoryauthoritiesandindustryandbringtheimportanceofa

0731-7085/01/$-seefrontmatter©2001ElsevierScienceB.V.Allrightsreserved.PII:S0731-7085(00)00530-6

756J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767propervalidationtotheattentionofallthoseinvolvedintheprocessofsubmission.

Nowadays,thevalidationcharacteristicsneededforthevarioustestproceduresandtheirgeneralrequirements(seeTable1)arewellunderstood.However,inspiteofrecommendationsnottodoso[8],thereisanincreasingtendencytomisusetheguidelinesasakindofchecklistwhichisautomaticallyapplied.Sucha‘checklistmentality’maybecausedbyanincorrectunderstandingofstandardisationandimprovingefficiency.Duetotheintegrationofanalyticsinallaspectsofdrugdevelopmentandqualitycontrol,inappropriateanalyticalproceduresmay,however,leadto,forexample,wrongdecisions,workwhichhastoberepeated(out-ofspecificationresults)anddelays.Consequently,boththedesignofthevalidationstudiesandtheevaluationoftheresultsmustbeadjustedtotheindividualanalyticalprocedureinordertoachieveanunderstandingofitsrealperformance.Onlyonthisbasiscanproofbeobtainedthattheprocedure‘issuitableforitsintendedpurpose’[14].

2.Integrationandinterdependenciesofanalyticalvalidation

Analyticalproceduresareusedthroughoutdrugdevelopmentandthemanufacturingofdrugsub-stancesanddrugproducts.Importantdecisionssuchastheestablishmentoftheshelf-lifefromstabilitystudies,theneedforadditionaltoxicolog-icaltrialsifnewimpuritiesappearorifknownimpuritiesexceedthequalifiedlevels,therework-ingofbatchesandbatchreleaseorrejectionarebasedonanalyticalresults.Inordertomaketherightdecisionsandtoavoidadditionalwork,anappropriateperformanceoftheanalyticalproce-duresisessential.

Butwhatdoes‘suitabilityforitsintendedpur-pose’mean?Forsomeapplications,therequire-mentsaredefinedintheICHguidelines,forexample,forimpuritytesting[9–11].Ifthereport-inglevelforunknownimpuritiesindrugsub-stancesissetto0.05or0.03%[9],thecorrespondingtestproceduremustbeabletoquantifyimpuritiesatthisconcentrationwithanappropriatelevelofprecisionandaccuracy.

Table1

Validationcharacteristicsnormallyevaluatedforthedifferenttypesoftestprocedures[7]andtheminimumnumberofdeterminationsrequired(ifapplicable)[8]Validationcharacteristics

Minimumnumber

TestprocedureIdentity

ImpuritiesQuantitative

SpecificitybLinearityRangeAccuracyPrecisionRepeatability

Intermediateprecision/reproducibilitycDetectionlimitQuantitationlimit

aAssayaLimitYesNoNoNo

YesYesYesYes

–

5concentrations–

9determinationsover3concentrationlevels(e.g.3×3)

6determinationsat100%or9determinationsover3concentrationlevels(e.g.3×3)2series––

YesNoNoNoYesYesYesYes

NoNoNoNo

YesYesNodYes

NoNoYesNo

YesYesNoNo

Includingdissolution,content/potency.

Lackofspecificityofoneanalyticalprocedurecouldbecompensatedbyothersupportinganalyticalprocedure(s).cIntermediateprecisionsufficientforsubmission.dMaybeneededinsomecases.

bJ.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767757

Fig.1.Suitabilityoftheanalyticalprocedurewithrespecttospecificationlimits.

Withrespecttoassaydeterminations,thevari-abilityoftheanalyticalprocedureisoftenlargerthanthevariabilityofthemanufacturing.Thismustbetakenintoaccountintheestablishmentofspecificationlimits[12,13],i.e.thevariabilityoftheprocedureandtheacceptancelimitsmustbecompatible(Fig.1).Ofcourse,safetyrequire-mentsareofprimaryimportancebutiftheyaresatisfied,specificationlimitscanalsobedefinedbasedontheanalyticalvariability[15,16].

Theanalyticalstateoftheartshouldbetakenintoaccountalthoughitisnottheultimategoaltooptimiseanindividualanalyticalprocedureaswellaspossible.Itisalsoveryimportanttorecognisethatthereleaseofagivenbatchisbasedonawholesetoftestprocedureswhichcomple-mentandsupplementeachother.Theirselectioninthespecificationdesign[12,13]has,therefore,considerableinfluenceontherequiredperfor-manceoftheindividualcontroltestand,hence,onitsvalidation.

Besidesthis‘horizontal’integration,therearealso‘vertical’connections.Thereisastrongfeed-backbetweenmethoddevelopmentandvalidation[17].Duetotheimportantaspectoftime,itisadvisable(ifpossible)toperforma‘progressive’validationstartingfromabasicdatasetwhichissupplemented,forinstance,withrespectto(inter-mediate)precisionandrobustness.Thevalidationresultscanalsohaveafeedbackeffectondetailsoftheanalyticalproceduresuchasonthenumberofreplicatedeterminationsorthemodeofcali-bration(seenextchapter).

Duringmethoddevelopmentandvalidationthecriticalperformanceparametersoftheanalytical

procedureshouldbeidentifiedinordertodesigndiscriminatingsystemsuitabilitytests.

Iftheseinterdependenciesareignoredandifparametersaredeterminedduringthevalidationwhichdonotdescribethecriticalperformanceoftheanalyticalprocedure,severeconsequencescanbeexpected.Forexample,iftheprocedureisnotsufficientlyrobust,problemsarelikelytooccurinamethodtransferandrepeatedadjustmentsinthesystemsuitabilitytestwill,forexample,bere-quired.Iftheacceptancelimitsfor,forexample,assayaretoonarrow,out-ofspecificationresultswillrequireextensiveinvestigations[18].Conse-quently,thetimeandeffort(perhaps)savedina‘checklist’validationapproachwillmostlikelyresultinproblemsatalaterdatewhichcouldbemuchmoreexpensive.

3.AcceptancecriteriaforvalidationparametersUsingtheICHguidelinesasabasis[7,8],itistheresponsibilityoftheanalysttoselectforthegivenindividualtestprocedurerelevantparame-tersandappropriateacceptancecriteriaandtodesigntheexperimentalstudiesaccordingly.Theseacceptancecriteriacanoftenbederivedfromspe-cificationlimits.

Asageneralrule,thestandarddeviationoftheanalyticalprocedureshouldbelowerthan1/6ofthespecificationrange.Adetailedapproachtak-ingthenumberofrepeateddeterminationsintoaccountisbasedonconfidenceintervals[15,16](Fig.2).Basicspecificationlimits(BL)includethevariabilityofthemanufacturingprocessandrep-resentthefinallimits(SL)ifanerror-freeanalyti-calprocedureisused.Describingtheanalyticalvariabilityasanormaldistributionoftheexperi-mentalresults,confidenceintervalscanbecon-structedasarepresentationoftheresultprobabilityforagivennumberofreplicates.Thecombinationofthebasiclimitandthelimit(up-perorlower)ofthe95%confidenceinterval(one-sided)thenrepresentstheoverallspecificationlimit(Eq.(1)).Itmustbetakenintoaccountthatthiscalculationisbasedonthetruestandarddeviationwhereasthestandarddeviationobtainedinavalidationstudyisonlyarandomestimate

758J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767Fig.2.Constructionofspecificationlimitsfrom95%confi-denceintervals.Theprocedureisshownforthelowerlimits.BL,basiclimits,imposedbythevariabilityofthemanufactur-ingprocess;SL,overallspecificationlimit,combinationofBLandthe(lower)limitoftheone-sided95%confidenceintervaloftheanalyticalvariability.

whichalsodisplaysavariability.Areliableexper-imentaldeterminationofthetruevaluerequiresrepeatedintermediateprecisionstudies,butitcanbeestimatedfromthestatistical(󰁸2)distributionofstandarddeviations.Theupper95%confidencelimitofthisdistribution(UL)willrepresentthemaximumvalueforthetruestandarddeviation(Eq.(2)).Forsixexperimentalvalues,itcanbeapproximatedbytwicetheexperimentalstandarddeviation[16].SL=BL9tn−1,0.95×UL(strue)=sexp×

strue󰀊n(1)

RearrangingEqs.(1)and(2)givesthemaxi-mumpermittedexperimentalstandarddeviationforthegivenspecificationlimits,i.e.theaccep-tancelimitforthevalidation.Forexample,adrugsubstanceLC-assay(performedwithfourrepeti-tions)withalowerspecificationlimitof98.0%andalimitforthesumofimpuritiesof0.5%(i.e.alowerbasiclimitof99.5%)wouldrequireanexperimentalstandarddeviationinvalidationbe-low0.64%.ItshouldbenotedthatBLandSLinEq.(3)refertothe‘critical’halfofthespecifica-tionrange.Fordrugsubstances,duetothepres-enceofimpurities,thesearethelowerlimits.Astherequiredstandarddeviationisdependentonthenumberofrepetitionsintheassay,adjust-mentsarepossible(Table2).Thus,thenumberofrepeateddeterminationscanalsobefine-tunedaccordingtotheresultsofthevalidation.Incaseofsufficientlywidespecificationlimitscomparedtotheanalyticalvariability,thisallowsaneffi-ciencyoptimisationoftheanalyticalprocedure.smax=

󰀌(BL−SL)󰀌×󰀊n2×tn−1,0.95(3)

'

(n−1)󰁸2n−1,0.95(2)

Whileinthecaseofadrugsubstanceassaythe

basiclimitsaredefinedmainlybythesumofimpurities,inthecaseofdrugproducts,oftenonlyanestimationispossible.Forsimpledosageforms,thesamevariabilitymaybeexpectedforanalyticsandmanufacturing.

Itshouldbenotedthattheestimationofthetruestandarddeviationwillresultinamaximumvalueand,therefore,reducetherequiredaccep-tancelimit.Consequently,incriticalcasesanex-

Table2

Maximumpermittedstandarddeviationforassaydeterminationsindependenceonthenumberofrepetitionsandthespecificationrange

Drugproduct(%)

SpecificationrangeBasiclimit(lower)Numberofrepetitions2346

95–105

97.5(estimated)

Drugproduct(%)95–105

99.0(estimated)

Drugsubstance(%)98–102

99.5(sumofimpurities)

Acceptancelimitforexperimentalstandarddeviationinvalidation(n=6)0.280.741.061.44

0.451.191.72.3

0.170.450.640.86

J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767759

perimentaldetermination(usingatleastfourtosixseries)mightbereasonabletoobtainanesti-mateforthetruestandarddeviation.Then,Eq.(1)canbedirectlyrearrangedandthefactoroftwoinEq.(3)canbeomitted.

Ifspecificationlimitsarenotyetdefinedorimpliedduetosafetyrequirements,Eq.(1)canbeuseddirectlytocalculatethelimits.Forthetruestandarddeviation,theupperlimitoftheexperi-mentaldetermination(Eq.(2))canbecalculatedortwicethisvalueasanapproximation.Alterna-tively,asoundlybasedintermediateprecisionmaybeusedasanestimationforthetruestandarddeviation.

Thevalidationacceptancelimitshouldthenbederivedfrompreviousexperiencesofcomparableanalyticalprocedures(analyticalstateoftheart).Forexample,fromawellbasedaswellasaccept-ableintermediateprecisionof1.0%andabasiclimitof99.0%,thespecificationrangeforadrugsubstancetobedeterminedwithfourrepetitionscanbecalculatedto97.8–101.2%.Iftheexperi-mentalrepeatabilitywasdeterminedwith1.0%,limitsfrom96.4–102.4%wouldresult.

Withrespecttoimpuritydeterminations,theICHreportingthresholdof0.05%[9]canberegardedastherequiredquantitationlimitforunknownimpuritieswhichwouldguaranteeareli-ablequantitationatthespecificationlimitof0.1%.Ifotherlimitsarerequireddue,forexam-ple,tosafetyconsiderations,theabove-mentionedapproachcanbeapplied.

StatisticaltestssuchastheStudent’st-testortheevaluationof95%confidenceintervalsshouldonlybecarefully(directly)appliedasacceptancecriteriabecausetheytestforstatisticaldifferences.Duetosometimesabnormallysmallvariabilitiesintheanalyticalseries,differencesareidentifiedassignificantwhichareofnopracticalrelevance[19].Inaddition,whencomparingindependentmethodsfortheproofofaccuracy,differentspe-cificitiescanbeexpectedwhichaddasystematicbias,thusincreasingtheriskoftheaforemen-tioneddanger.

Theanalystmustdecide,ifdetectedstatisticaldifferencesareofpracticalrelevance.Forexam-ple,if(e.g.duetoabnormalsmallvariabilityinoneseries)forameant-testorinlinearitya

Fig.3.Dependenceofthetestpoweronthenumberofdeterminations.Meant-testat95%levelofstatisticalsignifi-cance,assumingatruestandarddeviationof0.5andthesamenumberofdeterminationsforeachmean.

(statistical)significantdifferencetoanotherseriesoroftheintercepttozeroisdetected,theabsolutemagnitudeofthisdifferenceshouldbetakenintoaccount.Ontheotherhand,alargevariabilitycanalsoobscuredifferenceswhicharenotacceptable.

Inaddition,thepowerofthesetestsincreaseswiththenumberofdeterminations.AsshowninFig.3,withsixdeterminations,adifferenceof0.64betweentwomeanscanbedetected,whereasadifferenceof0.25canbedistinguishedwith32repetitions.Althoughthelatterdifferenceisstatis-ticallysignificant,inmostcases(suchasLC-assaydetermination)ithasnopracticalrelevance.

Forpracticalpurposes,asufficientagreementbetweentworesults(twomeansorameanandanominalvalue,e.g.recovery)iscompletelyade-quate.Theacceptancecriteriacanbederivedfrompreviousexperienceorcalculatedbasedonspecificationlimitsandstatisticalconsiderations(seenextpart).Forexample,assumingaLC-as-saywithspecificationlimitsfrom95to105%,arecoveryrangefrom98to102%wouldbeacceptable.

Theacceptancecriteriaandlimitsshouldbedefinedbeforestartingthevalidationandincludedinaprotocol.Afterthevalidationstudies,theywillserveasabasisfortheevaluation.

760J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–7674.Validationcharacteristics

Whenperformingvalidationstudies,thewholeanalyticalprocedureincludingallthestepsofthesamplepreparationshouldbeapplied,asfaraspossible.Incontrast,theterm‘method’shouldberestrictedtothemodeofanalyticaldeterminationalone(e.g.capillaryelectrophoresis,(reversed-phase)chromatographyandspectrometry).

Allowedexceptionstothewrittenprocedureconcernthenumberofrepetitionsasthenumberofdeterminationsforthevariousvalidationchar-acteristicsisdescribedintheICHguideline(Table2)andtherepetitionsintheproceduremaybefine-tunedbasedonthevalidationresults(seepreviouschapter).Suchanadjustmentmayalsobeusedforthefinalcalibrationmode(Section4.2).

Asfaraspossible,theanalyticalprocedureshouldbeindependentoftheactualequipmentusedprovidedthattheequipmenthasbeenappro-priatelyqualified.Thismustbetakenintoaccountforthevalidationstudies.

4.1.Specificity

Therehasbeensomecontroversyregardingthetechnicaltermforthisvalidationcharacteristic,i.e.specificityvs.selectivity[20].Incontrasttoanisolatedtestprocedure,inpharmaceuticalanalysisthesumofvariouscontroltestsandhencetheircombinedspecificityisusedfortheoverallbatchevaluation.Averypragmaticdefinitiondescribesselectivityasthe(physical)separationofsub-stancemixtureswith,forexample,chromatogra-phyandelectrophoresis,i.e.thedeterminationoftheanalyteinadditiontoothersubstances.Theindividualdeterminationofananalyteinthepres-enceofothersubstances,(i.e.withoutsignificantinfluenceofothersubstancesorclassesofsub-stances)isdefinedasspecificby,forinstance,massspectrometry,NMR,infrared,fluorescenceorUVspectrometry,electrochemicaldetectionandtitration[21].

Inspiteofthisdiscussion,thereisabroadagreementthatthisvalidationcharacteristicisofcrucialimportanceandisthecriticalbasisforeachanalyticalprocedure.Asnoabsoluteand

quantitativemeasureexists(atleastfortheoverallspecificity),therequirementsdependontheindi-vidualanalyticalprocedureaswellasonitscom-binationwithothers.Forexample,theoverallspecificityofaqualitycontrolcanbeobtainedbysecuring(orcorrecting)apreciseandefficientassaytitrationwithaselectivechromatographicimpuritydetermination.Assumingatitrationfortheassayofabasicdrugsubstancewithamolec-ularweightof300.Threeimpuritiesarespecified,twoofwhicharealsobasiccompounds(A:200andB:100MW).Thethirdimpurity,aneutralmoleculewillhavenotitrationresponse,fortheotherimpuritiestitrationresponsefactorsof300/200=1.5and300/100=3.0canbecalculated.Therefore,theresultofthebatchtitrationmustbecorrectedwiththeamountsofthebasicimpuritiesobtainedbyaselectiveLC-procedure.Ofcourse,theamountofunknownimpuritiesmustbelim-itedinordertopreventnonpredictabletitrationresponses.

drug[%asis]=titration[%]×

100−1.5×A[%,LC]−3×B[%,LC]

100

(4)

Withrespecttochromatographictechniques,specificitycanbedemonstratedbyasufficientseparationofthesubstancespresent.Fortheas-say,appropriateseparationmeansanadequateresolutionbetweenthemainpeakandtheimpu-rityandplacebopeakswhichneednottobeseparatedfromeachother.Thesamecanbeappliedtoindividualimpurityordegradantdeter-mination.Incontrast,universalproceduresforthedeterminationofimpuritiesrequireasufficientseparationofallrelevantimpuritypeaks.Therequiredresolutionisstronglydependentonthedifferenceinthesizeofthecorrespondingpeaksaswellasontheirelutionorder[22].Inordertobeabletodetectthecoelutionofunknownsub-stances,peakhomogeneityinvestigationssuchasrechromatography,diodearraydetectionorLC-MScouplingshouldbeperformed[nextpart].Ifsamplesfromstresstestingareusedtodemonstrateappropriateseparationpower,care

J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767761

shouldbetakentoavoidoverdegradationasthiswouldresultinsecondary(orevenhigherorder)degradantswhichareofnopracticalrelevance.Therefore,degradationshouldberestrictedtoabout10%.Alternatively,samplesfromregularstabilitystudies(acceleratedstorageconditions)maybeused.

4.2.Linearity/range

Alineardependenceofthesignalandtheana-lyteconcentrationiscertainlythemostconvenientcaseandwidelyusedinpharmaceuticalanalysis.However,thereareanalyticalprocedureswitha

nonlinearresponsesuchasTLC,fluorescencede-tectionandatomabsorptionspectrometry.There-fore,theterm‘analyticalresponse’wouldhavebeenmoreappropriateforthisvalidationcharacteristic.

Theessentialquestiontobeansweredhereisonthesuitabilityofthecalibrationmodetobeusedinthetestprocedure.Therequirementsandrele-vantparametersforthevariouscalibrationsaregiveninTable3.Adetaileddiscussionwillfollowinthenextpart.

Itshouldbenotedthatinmostcasesonlyaqualitativestatementisneeded.Forexample,ifasingle-pointcalibration(externalstandard)is

Table3

RequirementsfordifferentcalibrationmodeswithrelevantparametersQuantitation

Single-pointcalibrationExternalstandard

Requirements

Relevantparameters

Linearfunction

Non-significantordinateinterceptHomogeneityofvariancesaMultiple-pointcalibrationLinear,unweighted

Linearfunction

Standarderrorofslope(residualstandarddeviation),sensitivities(relativestandarddeviation,graph),residualanalysis,statisticaltests(vs.quadraticregression)

Inclusionofzeroinconfidenceintervaloftheordinateintercept,magnitudeoftheintercept(aspercentofthesignalat100%testconcentration)

F-testofthevariancesatthelowerandupperlimitoftherange

Linear,weighted

Non-linear

100%-method(areanormalisationforimpurities):

Standarderrorofslope(residualstandarddeviation),sensitivities(relativestandarddeviation,graph),residualanalysis,statisticaltests(vs.quadraticregression)

HomogeneityofF-testofthevariancesatthelowerandupperlimitofthe

avariancesrangeLinearfunctionStandarderrorofslope(residualstandarddeviation),

sensitivities(relativestandarddeviation,graph),residualanalysis,statisticaltests(vs.quadraticregression)

ContinuousfunctionAppropriateequationFormainpeak:linearfunctionNon-significantordinateinterceptHomogeneityofvariancesaForimpurities:linearfunction

Standarderrorofslope(residualstandarddeviation),sensitivities(relativestandarddeviation,graph),residualanalysis,statisticaltests(vs.quadraticregression)

Inclusionofzeroinconfidenceintervaloftheordinateintercept,magnitudeoftheintercept(aspercentofthesignalat100%testconcentration)

F-testofthevariancesatthelowerandupperlimitoftherange

Standarderrorofslope(residualstandarddeviation),sensitivities(relativestandarddeviation,graph),residualanalysis,statisticaltests(vs.quadraticregression)

aMaybepresumedforalimitedrange(factor10–20).

762J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767Table4

QuantitativeapproachestodemonstrateaccuracyaccordingtoICH[8]

DrugsubstanceApplicationoftheanalyticalprocedureto

areferencematerial

Comparisonoftheresultswiththoseofasecond,wellcharacterisedprocedureDrugproduct

Applicationoftheanalyticalproceduretosyntheticmixturesofdrugproductcomponents

Spikingofanalyttodrugproduct

Comparisonoftheresultswiththoseofasecond,wellcharacterisedprocedure

ImpuritiesSpikingoftheimpuritytodrugsubstance(quantitative)orproduct

Comparisonoftheresultswiththoseofasecond,wellcharacterisedprocedure

aimedat,therequirementsarealinearresponsefunctionandthezerointercept.Iftheseprerequi-sitesarefulfilled,theactualfiguresforthestan-darderrorofslopeortheconfidenceintervaloftheintercept,forexample,arenotusedfurtherorreferredto.Therefore,itmakesnosensetorepeatlinearityinvestigationsonotherdaysorwithotheroperators[23].However,careshouldbetakentoremainwithinthelinearrangeoftheindividualdetectorusedbutthisinformationcanbeobtainedfromtheequipmentqualification.

4.3.Accuracy

TheICHguidelinerecommendsthedemonstra-tionofaccuracyoverthewholeworkingrange(seeTable1).However,ifonlyanarrowrangeisrequired(e.g.assayorimpuritieswithalowspe-cificationlimit),asix-folddeterminationata100%testconcentrationasdescribedforthepreci-sionstudiesmayalsobeused.

SeveralapproachesdiscussedintheICHguide-linearegiveninTable4.

Iftheanalyticaltesttobevalidatediscom-paredwithanotherprocedureorappliedtoareferencesubstance,theprobablydifferentspecifi-citiesmustbetakenintoaccount.Therefore,statisticaltestsshouldbeperformedonlyifthesystematicbiasbasedonthesedifferencescanbequantifiedandthuscorrectedorarenegligible.

Otherwise,thecomparisonshouldbeperformedasaqualitativeverificationofplausibilityoranacceptablemaximumdifferenceshouldbedefined(e.g.2.0%foranLCassay).

Spikingexperimentsforrecoveryinvestigationsshouldbeperformedascloselytotheauthenticconditionsaspossiblesothatpossibleinterfer-encesbetweentheanalyteandmatrixcanberecognised.Thisranges,forexample,fromthedirectpreparationofadrugproductwithvariouscontentsofactiveingredienttowhichthewholeanalyticalprocedureisappliedtotheadditionofadrugsubstancestocksolutiontoaplacebosolution.

Forthequantitationoftheanalyte,thesamecalibrationmodeasdescribedinthefinaltestproceduremustbeused.Again,statisticaltestsshouldbeusedcarefully,especiallywithcomplexmatricesandlowconcentrationsofimpurities.Alternatively,acceptabledeviationsfromthethe-oreticalrecoveryof100%canbedefinedbasedontheapplication,experiencesorgeneralstatisticalconsiderations(nextpart).

UsingUVdetection,responsefactorsforknownandavailableimpuritiesanddegradantscanbeobtainedfromlinearityorrecoverystudies(ratiooftheslopesorofspecificpeakareaofanalyteandactiveingredient).However,thisisnotpossibleforunknownimpuritiesorimpuritieswhicharenotavailable.Inthesecases,safetychecksshouldbeperformedtoidentifypossibleproblems.Asafirststep,thepeakareapercentcanbeobtainedadditionallyata‘check’wave-lengthinthelowerUVrange(e.g.210nm)[24]whereabsorbancecoefficientsoftendifferless.Comparableresultsforthetwowavelengthsindi-catesimilarresponsefactors,wheraslargediffer-encesinthepeakareamayindicateresponsefactorsdifferenttounity(Fig.4).However,thelatterbelongsrathertotheanalyticaldevelopmentandthedesignoftheanalyticalprocedure.

4.4.Precision

InadditiontotheICHprecisionlevels(Table1),itisadvisabletodeterminethesystempreci-sion(injectionrepeatability)eitherbyrepeated

J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767763

determinationsofthesametestsolutionorfromdoubledeterminationsofeachtestsolutionusedforrepeatability(Eq.(5)).s1md=

'

%(xi,1−xi,2)22m(5)

i=1Repeatability,alsotermedintra-assayprecision,referstotheprecisionobtainedunderthesameoperatingconditionsoverashortintervaloftimebyapplyingthewholeanalyticalproceduretothesample.Intermediateprecisionreferstowithin-laboratoryvariations.Theextentofinvestigationswilldependontheintendeduseoftheanalyticalprocedure.Atypicalinvestigationmightincludeanalysts,days,equipment,reagents,columns,etc.Preferably,theintermediateprecisionstudiesshouldbeextendedoveralongerperiodoftime,inordertoobtainameasureoftheanalyticalvariabilitywhichisrepresentativeforthelong-termroutineuse.Abasicvalidationstudycanalsobesupplementedbyincorporatinginvestiga-tionsintotheroutineapplicationoftheanalyticalprocedure.

Repeatabilityandintermediateprecisioncanbecalculatedbyananalysisofvariances[25,26].Theformerfigurerepresentstheoverallvariabilitywithintheperformedseries,thelatteralsoin-cludesthevariabilitybetweentheseries(Table5).Thedifferencebetweentheprecisionlevelsaswell

Fig.4.SafetycheckfordeviatingresponsefactorsofunknownimpuritiesinUVdetection.Inadditiontothetestprocedurewavelengthof240nm,relativepeakareaaredisplayedat220nm.Theratioforthelabelledpeaksisaboutseven.Infurtherinvestigations,aresponsefactorof14wasdetermined.

Table5

AnalysisofvariancesfortheinvestigationofintermediateprecisionaSeries1

Series2Numberofdeterminations6

7

Relativestandarddeviation0.32%0.63%Mean

10.0710.1495%confidenceinterval0.0340.060

Overallmean10.11Repeatability

0.52%Intermediateprecision

0.65%

aTwooperatorsperformedanassayofatabletformulation(10mg)withdifferentLC-systems,mobilephases,andcolumns.

astheirabsolutemagnitudeindicatetherobust-nessoftheanalyticalprocedure.Fortheevalua-tionofthesuitability(compatibilitywithspecificationlimits,seechapter‘AcceptanceCrite-riaforValidationParameters’),theintermediateprecisioncanberegardedastherelevantparame-ter,especiallyiftheanalyticalinvestigationsex-tendoverseveralyearssuchasduringstabilitystudies.

Fromthestandarddeviation,therepeatabilitylimitcanbecalculated(Eq.(6))whichrepresentsthemaximumpermitteddifferencebetweentworepeatedmeasurements.Inthisway,astraightfor-wardverificationispossibleifthedegreeofscat-teringundertheactualconditionsiscomparabletothevalidation.Bythisparameter,theprecisionobtainedduringthevalidationstudiesusingalargernumberofdata(thusincreasingthereliabil-ity)islinkedtotheroutineanalyseswithoutre-quiringalargernumberofexperimentaldata.r=tn−1,0.95×󰀊2×s:2.8×s

(6)

Ofcourse,areliableestimateofthestandarddeviationisrequiredtocalculateappropriatere-peatabilitylimits.

Moststatisticaltestsandcalculationsarebasedontheassumptionthattheexperimentalvaluesareonlyinfluencedbyrandomvariability(i.e.thattheyarenormallydistributed).Data,whichdonotfulfilltheseassumptions(e.g.duetosocalled‘grosserrors’,weighing,dilution,orbyproblemswiththeinstrument,etc.)willaffecttheresults.

764J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767Suchvaluescanbeidentifiedbystatisticaloutliertests(e.g.accordingtoDixonorGrubbs[25])inordertoeliminatethembeforeperformingfurthercalculations.However,theproblemis—espe-ciallywithasmallnumberofdatawheregroup-ingscouldeasilyoccur—avoidingtheincorrectrejectionofvaluesbelongingtothesamedistribu-tion.Whenanoutlierisidentified,theabsolutemagnitudeofthestandarddeviationmustalsobeconsideredforevaluation.Ifthisparameter(cal-culatedincludingthesuspected‘outlier’)liesinanormallyexpectedrange,preferablyallvaluesshouldberetained.Outliertestsshouldbeappliedcarefullyandonlyobviouslydeviatingvalues(‘grosserrors’)shouldbeeliminated.Itisprefer-abletoincreasethereliabilityoftheobtainedstandarddeviationbyrepeatedinvestigations,e.g.usingtheoverallrepeatabilityobtainedwithananalysisofvariancesofanintermediateprecisionstudy.

4.5.Detectionandquantitationlimit

SeveralapproachesaregivenintheICHguide-linetodeterminethedetectionandquantitationlimits(Table6).Generally,theyarebasedeitherontheanalysisofblanksoronthescattering(variability)oftheanalyticalsignalsinthelowconcentrationrange.

Usingtheblankprocedures,thecorrespondingcalculationvalueismultipliedbythefactorsof

Table6

Approachesfordeterminingthedetectionandquantitationlimit[8]aApproach

Detectionlimit

Quantitationlimit

VisualevaluationMinimumlevelMinimumleveldetectablequantifiableSignal-to-noise

3:1or2:110:1Standarddeviationof3.3×|/S

10×|/S

theresponse(|)bandtheslope(S)

aVerificationwithasuitablenumberofsamples.

bStandarddeviationoftheblank,residualstandarddevia-tionofthecalibrationline,orstandarddeviationoftheintercept.

3.3and10forthedetectionandquantitationlimits,respectively.Thecalculationvaluemayrepresentthesignaloftheblank,thestandarddeviationoftheblankoroftheinterceptofacalibrationline(correspondingtoanextrapolatedblank).Inthelattertwocases,theanalyticalsignalistransformedbytheslopeofthecalibra-tionlineintoaconcentration[8].Usingthecali-brationlinedirectly,theaforementionedfactors(3.3and10)canbemultipliedbytheratiofromtheresidualstandarddeviationandtheslope(cor-respondingtothestandarderrorofslope)[8].Limitscalculatedorextrapolatedbythesepro-ceduresshouldbeverifiedbytheanalysisofsam-plesinthecorrespondingconcentrationrange[8].Thisadditionalverificationisalreadyincludedinotherprocedureswhichmakedirectuseofthescatteringaroundthecalibrationlinebymeansofthe95%predictionintervalaroundtheregressionline[27,28].

Thequantitationlimitcanalsobeobtaineddirectlyfromprecisionstudies.Forthisapproach,decreasinganalyteconcentrationsareanalysedre-peatedly.Thecoefficientofvariation(relativestandarddeviation)isplottedagainstthecorre-spondingconcentration.Ifapredefinedlimitforthecoefficientofvariation(relativestandarddevi-ation)isexceeded(e.g.10or20%),thecorre-spondingconcentrationisestablishedasthequantitationlimit[6,29].However,asufficientlylargenumberofanalyteconcentrationsmustbeanalysedbecausealargescatteringofstandarddeviationsoccurinthelowconcentrationrange.Morethantheothervalidationcharacteristics,detectionandquantitationlimitsaredependentontheequipmentusedandtheactualconditionsaswellasonthemethodofcalculation.InFig.5,theresultsofarepeateddeterminationusingfiveLC-systemsoveraperiodofabout9monthsareshown.Acalibrationusingamodelcompoundwasperformedwithsixconcentrationsintherangefrom0.05–1mg/ml.Threeapproacheswereappliedtoobtainthequantitationlimit:thecalcu-lationsfromtheresidualstandarddeviationandfromthe95%predictionintervalwereperformedusingthelinearregressionparameters,thecon-centrationscorrespondingtoasignal-to-noisera-tioof10wereinterpolatedfromtheexperimental

J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767765

Fig.5.Reproducibilityofthequantitationlimit(‘intermediateQL’).Threedifferentprocedureswereusedtocalculatethequantitationlimit.1.1–1.6,sixdeterminationswereperformedonthesameLCsystem;2.1–2.3,threeserieswereperformedonasecondLC-system;3.1–5.1,threeotherLC-systemswereused.Theserieswereanalysedoveratimeperiodofabout9months.

signal-to-noiseratiooftheconcentrationsbelow0.5mg/ml.

Largedifferencescanbeseenforthevariouscalculationprocedures(factor2–3),butalsointhecaseofrepeatedanalysisforthesamecalcula-tion(factor2).

However,theselimitsareofspecialimportanceinthetransferofanalyticalproceduresandforthereportingofimpurities.Therefore,the(gen-eral)quantitationlimitoftheanalyticalprocedureshouldbedefinedtakingtherequirementsoftheanalyticalprocedureintoaccount,ratherthandeterminedexperimentally.Inordertoguaranteeareliablequantitation,itmustbeatleastthreestandarddeviationsawayfromthespecificationlimit.Fororientationpurposes—takingthehighervariabilityinthisconcentrationrangeintoaccount—thequantitationlimitmaybedefinedas50%ofthespecificationlimit.Forexample,withregardtoanimpurityspecifiedas0.1%,thequantitationlimitcanbeestablishedat0.05%.ThiscorrespondstothereportingthresholdsoftheICHimpurityguideline[9].Duringvalidation,oneofthedescribedmethodsisusedtoverify(generally)whetherthisquantitationlimitcanbereliablyachieved.Takingtherandomvariability

ofthedeterminationintoaccount,itdoesnotmatterwhetherthe‘actual’QLisdeterminedto0.02or0.04%,providedthattheupperlimitofthe‘general’QLof0.05%canbeachieved.Ifimpuritiesarepresentorcanbespikedatthequantitationlimitinbatchesusedforassaypreci-sionstudies,thestandarddeviationwithrespecttotheimpuritiescanbecalculatedfromthesameexperimentaldata.TheQLcanberegardedasverifiedifanacceptableprecision(e.g.below10–20%)isachieved.

5.Conclusions

Beyondtheregulatoryrequirements,theperfor-manceandreliabilityofthecontroltestprocedureareessentialtothequalitycontrolofdrugs.Re-sultswhichreflectthequalityoftheanalyticalproceduremorethanthequalityofthepharma-ceuticaltobetestedmayeasilyleadtodrasticfinancialconsequences.Therefore,validationshouldberegardedaspartofanintegratedcon-cepttoensurethequality,safety,andefficacyofpharmaceuticals.

766J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767Basedonthevalidationcharacteristicsandre-quirementsoftheICHguidelines,eachanalyticalproceduremustbevalidatedwithrespecttoparameterswhicharerelevanttoitsperformance.Itistheresponsibilityoftheanalysttoidentifythecriticalperformanceparametersanddesignthevalidationstudyaccordingly.Acceptancecriteriashouldbedefinedinthevalidationproto-col.Theycanbeestablishedfrompreviousexperi-ences(analyticalstateoftheart)orcalculatedfromspecificationlimits.Fortheintendeduseofthetestprocedureacceptableabsoluteacceptancelimitsarepreferred.Statisticaltestsshouldbeusedcarefullyandpreferablyfororientationpur-poses.Theevaluationofthevalidationresultsistheresponsibilityoftheanalystandmustnotbeleftorreducedtotheoutcomeofastatisticaltest!However,statisticalanalysisandconsiderationsareveryhelpfulinverifyingthecompatibilityofspecificationlimitsandanalyticalvariability,cal-culatingacceptancelimitsandperformingsimula-tionsinordertopredictfuturerisks.

Relevantparametersfortheevaluationoflin-earityaredependentontheintendedcalibrationmodeoftheanalyticalprocedure.Fortheevalua-tionandfurthercalculations,thedifferentlevelsofprecisionmustbetakenintoaccount.Duetothelargevariabilityofexperimentallyobtainedresults,the‘general’quantitationlimitshouldbedefinedaccordingtotherequirements,forexam-plethereportingthresholdforunknownimpuri-tiesof0.05%.

References

[1]Erla¨uterungendesBGAzumAntragaufZulassungeines

Arzneimittels(Febr.1988).

[2]TheRulesGoverningMedicinalProductsintheEu-ropeanCommunity,volume3Addendum(1990).

[3]CDERGuidelineonValidationofChromatographic

Methods,ReviewerGuidanceofChromatographicMeth-ods,U.S.FoodandDrugAdministration,CenterforDrugsandBiologics,DepartmentofHealthandHumanServices(1994).

[4]GuidelinesforSubmittingSamplesandAnalyticalData

forMethodsValidation,U.S.FoodandDrugAdminis-tration,CenterforDrugsandBiologics,DepartmentofHealthandHumanServices(1987).

.

[5]UnitedStatesPharmacopeia,Section󰂎1225󰂏‘Validation

ofCompendialMethods’,UnitedStatesPharmacopeialConvention,Rockville(1995).

[6]A‘cceptableMethods’,DrugDirectorateGuidelines,Na-tionalHealthandWelfare,HealthProtectionBranch,HealthandWelfareCanada(1992).

[7]InternationalConferenceontheHarmonizationofTech-nicalRequirementsfortheRegistrationofPharmaceuti-calsforHumanUse(ICH)Q2A:ValidationofAnalyticalMethods(DefinitionsandTerminology)(October1994).[8]ICH:Q2BAnalyticalValidation-Methodology(Novem-ber1996).

[9]ICH:Q3A,ImpuritiesinNewDrugSubstances(March

1995,revisedOctober1999,Step2).

[10]ICH:Q3B,ImpuritiesinNewDrugProducts(November

1996,revisedOctober1999,Step2).

[11]ICH:Q3C,ResidualSolvents(July1997).

[12]ICH:Q6A,Specifications:testproceduresandacceptance

criteriafornewdrugsubstancesandnewdrugproducts,ChemicalSubstances(October1999).

[13]ICH:Q6B,Specifications:testproceduresandacceptance

criteriaforbiotechnological/biologicalproducts(March1999).

[14]L.Huber,LC-GCInternational,Febr.(1998)96–105.[15]A.G.J.Daas,J.H.McB.Miller,Pharmeuropa9.1(1997)

148–156.

[16]F.J.VandeVaart,Pharmeuropa9.1(1997)139–143.[17]B.Renger,Analyticalvalidation:formalrequirements

andpracticalapproaches,solutionforscientistssympo-sium,29–30Nov.1999,London.

[18]Draftguidanceforindustryoninvestigatingoutofspe-cification(OOS)testresultsinpharmaceuticalproduc-tion.USFoodandDrugAdministration,CenterforDrugEvaluationandResearch(CDER)(1998).

[19]R.G.Whitfield,D.W.Hughes,T.P.Layloff,R.R.Cox,

G.E.Gressett,P.J.Jimenez,J.Andersen,R.R.Reck,S.Schniepp,PharmacopeialForum24(1998)7051–7056.[20]B.-A.Persson,J.Vessman,R.D.McDowall.LC-GCIn-ternational,March1998,160-164.

[21]C.M.Riley,Statisticalparametersandanalyticalfigures

ofmerit,in:C.M.Riley,T.W.Rosanske(Eds.),Develop-mentandValidationofAnalyticalMethods,Elsevier,Oxford,1996,p.15.

[22]V.R.Meyer,Chromatographia40(1995)15–22.

[23]J.Caporal-Gautier,J.M.Nivet,P.Algranti,M.Guil-loteau,M.Histe,M.Lallier,J.J.N’Guyen-Huu,R.Rus-sotto,S.T.P.PharmaPratiques2(4)(1992)205–226.[24]Guidanceforindustry:analyticalproceduresandmethods

validation,chemistry,manufacturing,andcontrolsdocu-mentation,DraftGuidance(FoodandDrugAdministra-tion),August2000.

[25]DINISO5725-2:Accuracy(truenessandprecision)of

measurementmethodsandresults;abasicmethodforthedeterminationofrepeatabilityandreproducibilityofastandardmeasurementmethod(1990).

J.Ermer/J.Pharm.Biomed.Anal.24(2001)755–767[26]MVA-Methodvalidationinanalytics(PC-software,Win-dowsNT),NOVIAGmbH,Saarbru¨cken,Germany(http://www.novia.de).

[27]W.Funk,V.Dammann,C.Vonderheid,G.Oehlmann,

StatistischeMethodeninderWasseranalytik,VerlagChemie,Weinheim,1985.

[28]DIN32645:ChemischeAnalytik:Nachweis-,Erfassungs-

767

undBestimmungsgrenze,ErmittlungunterWiederholbe-dingungen.Begriff,Verfahren,Auswertung.BeuthVerlagGmbH,Berlin.

[29]EURACHEMGuidanceDocumentNo.WDG2:Ac-creditationforchemicallaboratories:GuidanceontheinterpretationoftheEN45000seriesofstandardsandISO/IECGuide25,1993.

.