INTERNATIONALCONFERENCEONHARMONISATIONOFTECHNICAL
REQUIREMENTSFORREGISTRATIONOFPHARMACEUTICALSFORHUMAN
USE
ICHHARMONISEDTRIPARTITEGUIDELINE
EVALUATIONFORSTABILITYDATA
Q1E
CurrentStep4version
dated6February2003
ThisGuidelinehasbeendevelopedbytheappropriateICHExpertWorkingGroup
andhasbeensubjecttoconsultationbytheregulatoryparties,inaccordancewiththe
ICHProcess.AtStep4oftheProcessthefinaldraftisrecommendedforadoptionto
theregulatorybodiesoftheEuropeanUnion,JapanandUSA.
Q1E
DocumentHistory
First
Codification
HistoryDate
New
Codification
November
2005
Q1EApprovalbytheSteeringCommitteeunderStep2and
releaseforpublicconsultation.
6February
2002
Q1E
CurrentStep4version
Q1EApprovalbytheSteeringCommitteeunderStep4and
recommendationforadoptiontothethreeICHregulatory
bodies.
6February
2003
Q1E
ii
EVALUATIONFORSTABILITYDATA
ICHHarmonisedTripartiteGuideline
HavingreachedStep4oftheICHProcessattheICHSteeringCommitteemeeting
on6February2003,thisguidelineisrecommendedfor
adoptiontothethreeregulatorypartiestoICH
TABLEOFCONTENTS
1.INTRODUCTION....................................................................................................1
1.1ObjectivesoftheGuideline........................................................................................1
1.2Background.................................................................................................................1
1.3ScopeoftheGuideline................................................................................................1
2.GUIDELINES..........................................................................................................1
2.1GeneralPrinciples......................................................................................................1
2.2Datapresentation......................................................................................................3
2.3Extrapolation..............................................................................................................3
2.4DataEvaluationforRetestPeriodorShelfLifeEstimationforDrug
SubstancesorProductsIntendedforRoomTemperatureStorage.........................3
2.4.1Nosignificantchangeatacceleratedcondition........................................................3
2.4.2Significantchangeatacceleratedcondition.............................................................5
2.5DataEvaluationforRetestPeriodorShelfLifeEstimationforDrug
SubstancesorProductsIntendedforStorageBelowRoomTemperature..............5
2.5.1Drugsubstancesorproductsintendedforstorageinarefrigerator.......................5
2.5.2Drugsubstancesorproductsintendedforstorageinafreezer...............................6
2.5.3Drugsubstancesorproductsintendedforstoragebelow-20°C..............................7
2.6GeneralStatisticalApproaches.................................................................................7
3.APPENDICES.........................................................................................................8
AppendixA:DecisionTreeforDataEvaluationforRetestPeriodorShelfLife
EstimationforDrugSubstancesorProducts(excludingFrozenProducts).........................8
AppendixB:ExamplesofStatisticalApproachestoStabilityDataAnalysis.....................8
i
EVALUATIONOFSTABILITYDATA
1.INTRODUCTION
1.1ObjectivesoftheGuideline
Thisguidelineisintendedtoproviderecommendationsonhowtousestabilitydata
generatedinaccordancewiththeprinciplesdetailedintheICHguideline“Q1A(R)
StabilityTestingofNewDrugSubstancesandProducts”(hereafterreferredtoasthe
parentguideline)toproposearetestperiodorshelflifeinaregistrationapplication.
Thisguidelinedescribeswhenandhowextrapolationcanbeconsideredwhen
proposingaretestperiodforadrugsubstanceorashelflifeforadrugproductthat
extendsbeyondtheperiodcoveredby“availabledatafromthestabilitystudyunder
thelong-termstoragecondition”(hereafterreferredtoaslong-termdata).
1.2Background
Theguidanceontheevaluationandstatisticalanalysisofstabilitydataprovidedin
theparentguidelineisbriefinnatureandlimitedinscope.Theparentguideline
statesthatregressionanalysisisanappropriateapproachtoanalyzingquantitative
stabilitydataforretestperiodorshelflifeestimationandrecommendsthata
statisticaltestforbatchpoolabilitybeperformedusingalevelofsignificanceof0.25.
However,theparentguidelineincludesfewdetailsanddoesnotcoversituations
wheremultiplefactorsareinvolvedinafull-orreduced-designstudy.
ThisguidelineisanexpansionoftheguidancepresentedintheEvaluationsectionsof
theparentguideline.
1.3ScopeoftheGuideline
Thisguidelineaddressestheevaluationofstabilitydatathatshouldbesubmittedin
registrationapplicationsfornewmolecularentitiesandassociateddrugproducts.The
guidelineprovidesrecommendationsonestablishingretestperiodsandshelflivesfor
drugsubstancesanddrugproductsintendedforstorageatorbelow“room
temperature”.Itcoversstabilitystudiesusingsingle-ormulti-factordesignsandfull
orreduceddesigns.
Note:Theterm“roomtemperature”referstothegeneralcustomaryenvironment
andshouldnotbeinferredtobethestoragestatementforlabeling.
ICHQ6AandQ6Bshouldbeconsultedforrecommendationsonthesettingand
justificationofacceptancecriteria,andICHQ1Dshouldbereferencedfor
recommendationsontheuseoffull-versusreduced-designstudies.
2.GUIDELINES
2.1GeneralPrinciples
Thedesignandexecutionofformalstabilitystudiesshouldfollowtheprinciples
outlinedintheparentguideline.Thepurposeofastabilitystudyistoestablish,based
ontestingaminimumofthreebatchesofthedrugsubstanceorproduct,aretest
periodorshelflifeandlabelstorageinstructionsapplicabletoallfuturebatches
manufacturedandpackagedundersimilarcircumstances.Thedegreeofvariabilityof
individualbatchesaffectstheconfidencethatafutureproductionbatchwillremain
withinacceptancecriteriathroughoutitsretestperiodorshelflife.
1
EvaluationofStabilityData
2
Althoughnormalmanufacturingandanalyticalvariationsaretobeexpected,itis
importantthatthedrugproductbeformulatedwiththeintenttoprovide100percent
ofthelabeledamountofthedrugsubstanceatthetimeofbatchrelease.Iftheassay
valuesofthebatchesusedtosupporttheregistrationapplicationarehigherthan100
percentoflabelclaimatthetimeofbatchrelease,aftertakingintoaccount
manufacturingandanalyticalvariations,theshelflifeproposedintheapplicationcan
beoverestimated.Ontheotherhand,iftheassayvalueofabatchislowerthan100
percentoflabelclaimatthetimeofbatchrelease,itmightfallbelowthelower
acceptancecriterionbeforetheendoftheproposedshelflife.
Asystematicapproachshouldbeadoptedinthepresentationandevaluationofthe
stabilityinformation.Thestabilityinformationshouldinclude,asappropriate,results
fromthephysical,chemical,biological,andmicrobiologicaltests,includingthose
relatedtoparticularattributesofthedosageform(forexample,dissolutionratefor
solidoraldosageforms).Theadequacyofthemassbalanceshouldbeassessed.
Factorsthatcancauseanapparentlackofmassbalanceshouldbeconsidered,
including,forexample,themechanismsofdegradationandthestability-indicating
capabilityandinherentvariabilityoftheanalyticalprocedures.
Thebasicconceptsofstabilitydataevaluationarethesameforsingle-versusmulti-
factorstudiesandforfull-versusreduced-designstudies.Datafromformalstability
studiesand,asappropriate,supportingdatashouldbeevaluatedtodeterminethe
criticalqualityattributeslikelytoinfluencethequalityandperformanceofthedrug
substanceorproduct.Eachattributeshouldbeassessedseparately,andanoverall
assessmentshouldbemadeofthefindingsforthepurposeofproposingaretestperiod
orshelflife.Theretestperiodorshelflifeproposedshouldnotexceedthatpredicted
foranysingleattribute.
ThedecisiontreeinAppendixAoutlinesastepwiseapproachtostabilitydata
evaluationandwhenandhowmuchextrapolationcanbeconsideredforaproposed
retestperiodorshelflife.AppendixBprovides(1)informationonhowtoanalyzelong-
termdataforappropriatequantitativetestattributesfromastudywithamulti-
factor,fullorreduceddesign,(2)informationonhowtouseregressionanalysisfor
retestperiodorshelflifeestimation,and(3)examplesofstatisticalproceduresto
determinepoolabilityofdatafromdifferentbatchesorotherfactors.Additional
guidancecanbefoundinthereferenceslisted;however,theexamplesandreferences
donotcoverallapplicablestatisticalapproaches.
Ingeneral,certainquantitativechemicalattributes(e.g.,assay,degradationproducts,
preservativecontent)foradrugsubstanceorproductcanbeassumedtofollowzero-
orderkineticsduringlong-termstorage
1
.Datafortheseattributesaretherefore
amenabletothetypeofstatisticalanalysisdescribedinAppendixB,includinglinear
regressionandpoolabilitytesting.Althoughthekineticsofotherquantitative
attributes(e.g.,pH,dissolution)isgenerallynotknown,thesamestatisticalanalysis
canbeapplied,ifappropriate.Qualitativeattributesandmicrobiologicalattributes
arenotamenabletothiskindofstatisticalanalysis.
Therecommendationsonstatisticalapproachesinthisguidelinearenotintendedto
implythatuseofstatisticalevaluationispreferredwhenitcanbejustifiedtobe
unnecessary.However,statisticalanalysiscanbeusefulinsupportingthe
extrapolationofretestperiodsorshelflivesincertainsituationsandcanbecalledfor
toverifytheproposedretestperiodsorshelflivesinothercases.
EvaluationofStabilityData
3
2.2Datapresentation
Dataforallattributesshouldbepresentedinanappropriateformat(e.g.,tabular,
graphical,narrative)andanevaluationofsuchdatashouldbeincludedinthe
application.Thevaluesofquantitativeattributesatalltimepointsshouldbereported
asmeasured(e.g.,assayaspercentoflabelclaim).Ifastatisticalanalysisis
performed,theprocedureusedandtheassumptionsunderlyingthemodelshouldbe
statedandjustified.Atabulatedsummaryoftheoutcomeofstatisticalanalysis
and/orgraphicalpresentationofthelong-termdatashouldbeincluded.
2.3Extrapolation
Extrapolationisthepracticeofusingaknowndatasettoinferinformationabout
futuredata.Extrapolationtoextendtheretestperiodorshelflifebeyondtheperiod
coveredbylong-termdatacanbeproposedintheapplication,particularlyifno
significantchangeisobservedattheacceleratedcondition.Whetherextrapolationof
stabilitydataisappropriatedependsontheextentofknowledgeaboutthechange
pattern,thegoodnessoffitofanymathematicalmodel,andtheexistenceofrelevant
supportingdata.Anyextrapolationshouldbeperformedsuchthattheextendedretest
periodorshelflifewillbevalidforafuturebatchreleasedwithtestresultscloseto
thereleaseacceptancecriteria.
Anextrapolationofstabilitydataassumesthatthesamechangepatternwillcontinue
toapplybeyondtheperiodcoveredbylong-termdata.Thecorrectnessoftheassumed
changepatterniscriticalwhenextrapolationisconsidered.Whenestimatinga
regressionlineorcurvetofitthelong-termdata,thedatathemselvesprovideacheck
onthecorrectnessoftheassumedchangepattern,andstatisticalmethodscanbe
appliedtotestthegoodnessoffitofthedatatotheassumedlineorcurve.Nosuch
internalcheckispossiblebeyondtheperiodcoveredbylong-termdata.Thus,aretest
periodorshelflifegrantedonthebasisofextrapolationshouldalwaysbeverifiedby
additionallong-termstabilitydataassoonasthesedatabecomeavailable.Care
shouldbetakentoincludeintheprotocolforcommitmentbatchesatimepointthat
correspondstotheendoftheextrapolatedretestperiodorshelflife.
2.4DataEvaluationforRetestPeriodorShelfLifeEstimationforDrug
SubstancesorProductsIntendedforRoomTemperatureStorage
Asystematicevaluationofthedatafromformalstabilitystudiesshouldbeperformed
asillustratedinthissection.Stabilitydataforeachattributeshouldbeassessed
sequentially.Fordrugsubstancesorproductsintendedforstorageatroom
temperature,theassessmentshouldbeginwithanysignificantchangeatthe
acceleratedconditionand,ifappropriate,attheintermediatecondition,andprogress
throughthetrendsandvariabilityofthelong-termdata.Thecircumstancesare
delineatedunderwhichextrapolationofretestperiodorshelflifebeyondtheperiod
coveredbylong-termdatacanbeappropriate.AdecisiontreeisprovidedinAppendix
Aasanaid.
2.4.1Nosignificantchangeatacceleratedcondition
Wherenosignificantchangeoccursattheacceleratedcondition,theretestperiodor
shelflifewoulddependonthenatureofthelong-termandaccelerateddata.
EvaluationofStabilityData
4
2.4.1.1Long-termandaccelerateddatashowinglittleornochangeovertime
andlittleornovariability
Wherethelong-termdataandaccelerateddataforanattributeshowlittleorno
changeovertimeandlittleornovariability,itmightbeapparentthatthedrug
substanceorproductwillremainwellwithintheacceptancecriteriaforthatattribute
duringtheproposedretestperiodorshelflife.Inthesecircumstances,astatistical
analysisisnormallyconsideredunnecessarybutjustificationfortheomissionshould
beprovided.Justificationcanincludeadiscussionofthechangepatternorlackof
change,relevanceoftheaccelerateddata,massbalance,and/orothersupportingdata
asdescribedintheparentguideline.Extrapolationoftheretestperiodorshelflife
beyondtheperiodcoveredbylong-termdatacanbeproposed.Theproposedretest
periodorshelflifecanbeuptotwice,butshouldnotbemorethan12monthsbeyond,
theperiodcoveredbylong-termdata.
2.4.1.2Long-termoraccelerateddatashowingchangeovertimeand/or
variability
Ifthelong-termoraccelerateddataforanattributeshowchangeovertimeand/or
variabilitywithinafactororamongfactors,statisticalanalysisofthelong-termdata
canbeusefulinestablishingaretestperiodorshelflife.Wheretherearedifferences
instabilityobservedamongbatchesoramongotherfactors(e.g.,strength,container
sizeand/orfill)orfactorcombinations(e.g.,strength-by-containersizeand/orfill)that
precludethecombiningofdata,theproposedretestperiodorshelflifeshouldnot
exceedtheshortestperiodsupportedbyanybatch,otherfactor,orfactorcombination.
Alternatively,wherethedifferencesarereadilyattributedtoaparticularfactor(e.g.,
strength),differentshelflivescanbeassignedtodifferentlevelswithinthefactor
(e.g.,differentstrengths).Adiscussionshouldbeprovidedtoaddressthecauseforthe
differencesandtheoverallsignificanceofsuchdifferencesontheproduct.
Extrapolationbeyondtheperiodcoveredbylong-termdatacanbeproposed;however,
theextentofextrapolationwoulddependonwhetherlong-termdatafortheattribute
areamenabletostatisticalanalysis.
?Datanotamenabletostatisticalanalysis
Wherelong-termdataarenotamenabletostatisticalanalysis,butrelevant
supportingdataareprovided,theproposedretestperiodorshelflifecanbeuptoone-
and-a-halftimes,butshouldnotbemorethan6monthsbeyond,theperiodcoveredby
long-termdata.Relevantsupportingdataincludesatisfactorylong-termdatafrom
developmentbatchesthatare(1)madewithacloselyrelatedformulationto,(2)
manufacturedonasmallerscalethan,or(3)packagedinacontainerclosuresystem
similarto,thatoftheprimarystabilitybatches.
?Dataamenabletostatisticalanalysis
Iflong-termdataareamenabletostatisticalanalysisbutnoanalysisisperformed,
theextentofextrapolationshouldbethesameaswhendataarenotamenableto
statisticalanalysis.However,ifastatisticalanalysisisperformed,itcanbe
appropriatetoproposearetestperiodorshelflifeofuptotwice,butnotmorethan12
monthsbeyond,theperiodcoveredbylong-termdata,whentheproposalisbackedby
theresultoftheanalysisandrelevantsupportingdata.
EvaluationofStabilityData
5
2.4.2Significantchangeatacceleratedcondition
Wheresignificantchangeoccursattheacceleratedcondition,theretestperiodor
shelflifewoulddependontheoutcomeofstabilitytestingattheintermediate
condition,aswellasatthelong-termcondition.
Note:Thefollowingphysicalchangescanbeexpectedtooccurattheaccelerated
conditionandwouldnotbeconsideredsignificantchangethatcallsforintermediate
testingifthereisnoothersignificantchange:
?softeningofasuppositorythatisdesignedtomeltat37oC,ifthemeltingpointis
clearlydemonstrated,
?failuretomeetacceptancecriteriafordissolutionfor12unitsofagelatincapsule
orgel-coatedtabletifthefailurecanbeunequivocallyattributedtocross-linking.
However,ifphaseseparationofasemi-soliddosageformoccursattheaccelerated
condition,testingattheintermediateconditionshouldbeperformed.Potential
interactioneffectsshouldalsobeconsideredinestablishingthatthereisnoother
significantchange.
2.4.2.1Nosignificantchangeatintermediatecondition
Ifthereisnosignificantchangeattheintermediatecondition,extrapolationbeyond
theperiodcoveredbylong-termdatacanbeproposed;however,theextentof
extrapolationwoulddependonwhetherlong-termdatafortheattributeareamenable
tostatisticalanalysis.
?Datanotamenabletostatisticalanalysis
Whenthelong-termdataforanattributearenotamenabletostatisticalanalysis,the
proposedretestperiodorshelflifecanbeupto3monthsbeyondtheperiodcoveredby
long-termdata,ifbackedbyrelevantsupportingdata.
?Dataamenabletostatisticalanalysis
Whenthelong-termdataforanattributeareamenabletostatisticalanalysisbutno
analysisisperformed,theextentofextrapolationshouldbethesameaswhendata
arenotamenabletostatisticalanalysis.However,ifastatisticalanalysisis
performed,theproposedretestperiodorshelflifecanbeuptoone-and-halftimes,but
shouldnotbemorethan6monthsbeyond,theperiodcoveredbylong-termdata,
whenbackedbystatisticalanalysisandrelevantsupportingdata.
2.4.2.2Significantchangeatintermediatecondition
Wheresignificantchangeoccursattheintermediatecondition,theproposedretest
periodorshelflifeshouldnotexceedtheperiodcoveredbylong-termdata.In
addition,aretestperiodorshelflifeshorterthantheperiodcoveredbylong-termdata
couldbecalledfor.
2.5DataEvaluationforRetestPeriodorShelfLifeEstimationforDrug
SubstancesorProductsIntendedforStorageBelowRoom
Temperature
2.5.1Drugsubstancesorproductsintendedforstorageinarefrigerator
Datafromdrugsubstancesorproductsintendedtobestoredinarefrigeratorshould
beassessedaccordingtothesameprinciplesasdescribedinSection2.4fordrug
EvaluationofStabilityData
6
substancesorproductsintendedforroomtemperaturestorage,exceptwhere
explicitlynotedinthesectionbelow.ThedecisiontreeinAppendixAcanbeusedas
anaid.
2.5.1.1Nosignificantchangeatacceleratedcondition
Wherenosignificantchangeoccursattheacceleratedcondition,extrapolationof
retestperiodorshelflifebeyondtheperiodcoveredbylong-termdatacanbeproposed
basedontheprinciplesoutlinedinSection2.4.1,exceptthattheextentof
extrapolationshouldbemorelimited.
Ifthelong-termandaccelerateddatashowlittlechangeovertimeandlittle
variability,theproposedretestperiodorshelflifecanbeuptoone-and-a-halftimes,
butshouldnotbemorethan6monthsbeyond,theperiodcoveredbylong-termdata
normallywithoutthesupportofstatisticalanalysis.
Wherethelong-termoraccelerateddatashowchangeovertimeand/orvariability,the
proposedretestperiodorshelflifecanbeupto3monthsbeyondtheperiodcoveredby
long-termdataif(1)thelong-termdataareamenabletostatisticalanalysisbuta
statisticalanalysisisnotperformed,or(2)thelong-termdataarenotamenableto
statisticalanalysisbutrelevantsupportingdataareprovided.
Wherethelong-termoraccelerateddatashowchangeovertimeand/orvariability,the
proposedretestperiodorshelflifecanbeuptoone-and-a-halftimes,butshouldnot
bemorethan6monthsbeyond,theperiodcoveredbylong-termdataif(1)thelong-
termdataareamenabletostatisticalanalysisandastatisticalanalysisisperformed,
and(2)theproposalisbackedbytheresultoftheanalysisandrelevantsupporting
data.
2.5.1.2Significantchangeatacceleratedcondition
Ifsignificantchangeoccursbetween3and6months’testingattheaccelerated
storagecondition,theproposedretestperiodorshelflifeshouldbebasedonthelong-
termdata.Extrapolationisnotconsideredappropriate.Inaddition,aretestperiodor
shelflifeshorterthantheperiodcoveredbylong-termdatacouldbecalledfor.Ifthe
long-termdatashowvariability,verificationoftheproposedretestperiodorshelflife
bystatisticalanalysiscanbeappropriate.
Ifsignificantchangeoccurswithinthefirst3months’testingattheaccelerated
storagecondition,theproposedretestperiodorshelflifeshouldbebasedonlong-term
data.Extrapolationisnotconsideredappropriate.Aretestperiodorshelflifeshorter
thantheperiodcoveredbylong-termdatacouldbecalledfor.Ifthelong-termdata
showvariability,verificationoftheproposedretestperiodorshelflifebystatistical
analysiscanbeappropriate.Inaddition,adiscussionshouldbeprovidedtoaddress
theeffectofshort-termexcursionsoutsidethelabelstoragecondition(e.g.,during
shippingorhandling).Thisdiscussioncanbesupported,ifappropriate,byfurther
testingonasinglebatchofthedrugsubstanceorproductattheacceleratedcondition
foraperiodshorterthan3months.
2.5.2Drugsubstancesorproductsintendedforstorageinafreezer
Fordrugsubstancesorproductsintendedforstorageinafreezer,theretestperiodor
shelflifeshouldbebasedonlong-termdata.Intheabsenceofanacceleratedstorage
conditionfordrugsubstancesorproductsintendedtobestoredinafreezer,testingon
asinglebatchatanelevatedtemperature(e.g.,5°C±3°Cor25°C±2°C)foran
EvaluationofStabilityData
7
appropriatetimeperiodshouldbeconductedtoaddresstheeffectofshort-term
excursionsoutsidetheproposedlabelstoragecondition(e.g.,duringshippingor
handling).
2.5.3Drugsubstancesorproductsintendedforstoragebelow-20°C
Fordrugsubstancesorproductsintendedforstoragebelow-20°C,theretestperiodor
shelflifeshouldbebasedonlong-termdataandshouldbeassessedonacase-by-case
basis.
2.6GeneralStatisticalApproaches
Whereapplicable,anappropriatestatisticalmethodshouldbeemployedtoanalyze
thelong-termprimarystabilitydatainanoriginalapplication.Thepurposeofthis
analysisistoestablish,withahighdegreeofconfidence,aretestperiodorshelflife
duringwhichaquantitativeattributewillremainwithinacceptancecriteriaforall
futurebatchesmanufactured,packaged,andstoredundersimilarcircumstances.
Incaseswhereastatisticalanalysiswasemployedtoevaluatelong-termdataduetoa
changeovertimeand/orvariability,thesamestatisticalmethodshouldalsobeused
toanalysedatafromcommitmentbatchestoverifyorextendtheoriginallyapproved
retestperiodorshelflife.
Regressionanalysisisconsideredanappropriateapproachtoevaluatingthestability
dataforaquantitativeattributeandestablishingaretestperiodorshelflife.The
natureoftherelationshipbetweenanattributeandtimewilldeterminewhetherdata
shouldbetransformedforlinearregressionanalysis.Therelationshipcanbe
representedbyalinearornon-linearfunctiononanarithmeticorlogarithmicscale.
Insomecases,anon-linearregressioncanbetterreflectthetruerelationship.
Anappropriateapproachtoretestperiodorshelflifeestimationistoanalyzea
quantitativeattribute(e.g.,assay,degradationproducts)bydeterminingtheearliest
timeatwhichthe95percentconfidencelimitforthemeanintersectstheproposed
acceptancecriterion.
Foranattributeknowntodecreasewithtime,thelowerone-sided95percent
confidencelimitshouldbecomparedtotheacceptancecriterion.Foranattribute
knowntoincreasewithtime,theupperone-sided95percentconfidencelimitshould
becomparedtotheacceptancecriterion.Foranattributethatcaneitherincreaseor
decrease,orwhosedirectionofchangeisnotknown,two-sided95percentconfidence
limitsshouldbecalculatedandcomparedtotheupperandloweracceptancecriteria.
Thestatisticalmethodusedfordataanalysisshouldtakeintoaccountthestability
studydesigntoprovideavalidstatisticalinferencefortheestimatedretestperiodor
shelflife.Theapproachdescribedabovecanbeusedtoestimatetheretestperiodor
shelflifeforasinglebatchorformultiplebatcheswhenthedataarecombinedafter
anappropriatestatisticaltest.Examplesofstatisticalapproachestotheanalysisof
stabilitydatafromsingleormulti-factor,full-orreduced-designstudiesareincluded
inAppendixB.ReferencestocurrentliteraturesourcescanbefoundinAppendixB.6.
EvaluationofStabilityData
3.APPENDICES
AppendixA:DecisionTreeforDataEvaluationforRetestPeriodorShelf
LifeEstimationforDrugSubstancesorProducts(excludingFrozen
Products)
No
Significant
changeataccelerated
conditionwithin
3months?
Yes
Noextrapolation;shorter
retestperiodorshelflifeand
datacoveringexcursions
canbecalledfor;statistical
analysisiflong-termdata
showvariability
Tabulateand/orplot
stabilitydataonall
attributesatallstorage
conditionsandevaluate
eachattributeseparately
8
Significant
changeataccelerated
conditionwithin
6months?
Yes
Intended
tobestoredina
refrigerator?
No
No
Yestoboth
Y=upto2X,butnot
exceedingX+12months;
orifrefrigerated,
Y=upto1.5X,butnot
exceedingX+6months
Yes
Noextrapolation;shorter
retestperiodorshelflife
canbecalledfor;statistical
analysisiflong-termdata
showvariability
Noto(1)
or(2)
Ifbackedbystatistical
analysisandrelevant
supportingdata:Y=upto
2X,butnotexceedingX+
12months;orifrefrigerated,
Y=upto1.5X,butnot
exceedingX+6months
Noto(1)
or(2)
Ifbackedbyrelevant
supportingdata:
Y=uptoX+3months
(1)Long-term
dataamenableto
statisticalanalysisand
(2)statisticalanalysis
performed?
Noto(1)or
(2)orboth
Noto(1)or
(2)orboth
Long-term
datashow:(1)littleor
nochangeovertime
and(2)littleorno
variability?
Accelerated
datashow:(1)littleor
nochangeovertime
and(2)littleorno
variability?
Statisticalanalysis
isnormally
unnecessary
Yestoboth
(1)Long-term
dataamenableto
statisticalanalysisand
(2)statisticalanalysis
performed?
Significant
change
atintermediate
condition?
Yes
No
Yestoboth
Ifbackedbystatistical
analysisandrelevant
supportingdata:Y=up
to1.5X,butnot
exceedingX+6months
Yestoboth
Y=Proposedretestperiodorshelflife
X=Periodcoveredbylong-termdata
Ifbackedbyrelevant
supportingdata:Y=up
to1.5X,butnot
exceedingX+6months;
orifrefrigerated,Y=up
toX+3months
EvaluationofStabilityData
9
AppendixB:ExamplesofStatisticalApproachestoStabilityDataAnalysis
Linearregression,poolabilitytests,andstatisticalmodeling,describedbelow,are
examplesofstatisticalmethodsandproceduresthatcanbeusedintheanalysisof
stabilitydatathatareamenabletostatisticalanalysisforaquantitativeattributefor
whichthereisaproposedacceptancecriterion.
B.1DataAnalysisforaSingleBatch
Ingeneral,therelationshipbetweencertainquantitativeattributesandtimeis
assumedtobelinear
1
.Figure1showstheregressionlineforassayofadrugproduct
withupperandloweracceptancecriteriaof105percentand95percentoflabelclaim,
respectively,with12monthsoflong-termdataandaproposedshelflifeof24months.
Inthisexample,two-sided95percentconfidencelimitsforthemeanareapplied
becauseitisnotknownaheadoftimewhethertheassaywouldincreaseordecrease
withtime(e.g.,inthecaseofanaqueous-basedproductpackagedinasemi-permeable
container).Thelowerconfidencelimitintersectstheloweracceptancecriterionat30
months,whiletheupperconfidencelimitdoesnotintersectwiththeupperacceptance
criterionuntillater.Therefore,theproposedshelflifeof24monthscanbesupported
bythestatisticalanalysisoftheassay,providedtherecommendationsinSections2.4
and2.5arefollowed.
Whendataforanattributewithonlyanupperoraloweracceptancecriterionare
analyzed,thecorrespondingone-sided95percentconfidencelimitforthemeanis
recommended.Figure2showstheregressionlineforadegradationproductinadrug
productwith12monthsoflong-termdataandaproposedshelflifeof24months,
wheretheacceptancecriterionisnotmorethan1.4percent.Theupperone-sided95
percentconfidencelimitforthemeanintersectstheacceptancecriterionat31
months.Therefore,theproposedshelflifeof24monthscanbesupportedby
statisticalanalysisofthedegradationproductdata,providedtherecommendationsin
Sections2.4and2.5arefollowed.
Iftheaboveapproachisused,themeanvalueofthequantitativeattribute(e.g.,
assay,degradationproducts)canbeexpectedtoremainwithintheacceptancecriteria
throughtheendoftheretestperiodorshelflifeataconfidencelevelof95percent.
Theapproachdescribedabovecanbeusedtoestimatetheretestperiodorshelflifefor
asinglebatch,individualbatches,ormultiplebatcheswhencombinedafter
appropriatestatisticaltestsdescribedinSectionsB.2throughB.5.
B.2DataAnalysisforOne-Factor,Full-DesignStudies
Foradrugsubstanceorforadrugproductavailableinasinglestrengthandasingle
containersizeand/orfill,theretestperiodorshelflifeisgenerallyestimatedbasedon
thestabilitydatafromaminimumofthreebatches.Whenanalyzingdatafromsuch
one-factor,batch-only,full-designstudies,twostatisticalapproachescanbe
considered.
?Theobjectiveofthefirstapproachistodeterminewhetherthedatafromall
batchessupporttheproposedretestperiodorshelflife.
?Theobjectiveofthesecondapproach,testingforpoolability,istodetermine
whetherthedatafromdifferentbatchescanbecombinedforanoverallestimateof
asingleretestperiodorshelflife.
EvaluationofStabilityData
10
B.2.1Evaluatingwhetherallbatchessupporttheproposedretestperiodor
shelflife
Theobjectiveofthisapproachistoevaluatewhethertheestimatedretestperiodsor
shelflivesfromallbatchesarelongerthantheoneproposed.Retestperiodsorshelf
livesforindividualbatchesshouldfirstbeestimatedusingtheproceduredescribedin
SectionB.1withindividualintercepts,individualslopes,andthepooledmeansquare
errorcalculatedfromallbatches.Ifeachbatchhasanestimatedretestperiodorshelf
lifelongerthanthatproposed,theproposedretestperiodorshelflifewillgenerallybe
consideredappropriate,aslongastheguidanceforextrapolationinSections2.4and
2.5isfollowed.Thereisgenerallynoneedtoperformpoolabilitytestsoridentifythe
mostreducedmodel.If,however,oneormoreoftheestimatedretestperiodsorshelf
livesareshorterthanthatproposed,poolabilitytestscanbeperformedtodetermine
whetherthebatchescanbecombinedtoestimatealongerretestperiodorshelflife.
Alternatively,theaboveapproachcanbetakenduringthepoolingprocessdescribed
inSectionB.2.2.Iftheregressionlinesforthebatchesarefoundtohaveacommon
slopeandtheestimatedretestperiodsorshelflivesbasedonthecommonslopeand
individualinterceptsarealllongerthantheproposedretestperiodorshelflife,there
isgenerallynoneedtocontinuetotesttheinterceptsforpoolability.
B.2.2Testingforpoolabilityofbatches
B.2.2.1Analysisofcovariance
Beforepoolingthedatafromseveralbatchestoestimatearetestperiodorshelflife,a
preliminarystatisticaltestshouldbeperformedtodeterminewhethertheregression
linesfromdifferentbatcheshaveacommonslopeandacommontime-zerointercept.
Analysisofcovariance(ANCOVA)canbeemployed,wheretimeisconsideredthe
covariate,totestthedifferencesinslopesandinterceptsoftheregressionlinesamong
batches.Eachofthesetestsshouldbeconductedusingasignificancelevelof0.25to
compensatefortheexpectedlowpowerofthedesignduetotherelativelylimited
samplesizeinatypicalformalstabilitystudy.
Ifthetestrejectsthehypothesisofequalityofslopes(i.e.,ifthereisasignificant
differenceinslopesamongbatches),itisnotconsideredappropriatetocombinethe
datafromallbatches.Theretestperiodsorshelflivesforindividualbatchesinthe
stabilitystudycanbeestimatedbyapplyingtheapproachdescribedinSectionB.1
usingindividualinterceptsandindividualslopesandthepooledmeansquareerror
calculatedfromallbatches.Theshortestestimateamongthebatchesshouldbe
chosenastheretestperiodorshelflifeforallbatches.
Ifthetestrejectsthehypothesisofequalityofinterceptsbutfailstorejectthatthe
slopesareequal(i.e.,ifthereisasignificantdifferenceininterceptsbutnosignificant
differenceinslopesamongthebatches),thedatacanbecombinedforthepurposeof
estimatingthecommonslope.Theretestperiodsorshelflivesforindividualbatches
inthestabilitystudyshouldbeestimatedbyapplyingtheapproachdescribedin
SectionB.1,usingthecommonslopeandindividualintercepts.Theshortestestimate
amongthebatchesshouldbechosenastheretestperiodorshelflifeforallbatches.
Ifthetestsforequalityofslopesandequalityofinterceptsdonotresultinrejectionat
alevelofsignificanceof0.25(i.e.,ifthereisnosignificantdifferenceinslopeand
interceptsamongthebatches),thedatafromallbatchescanbecombined.Asingle
retestperiodorshelflifecanbeestimatedfromthecombineddatabyusingthe
EvaluationofStabilityData
11
approachdescribedinSectionB.1andappliedtoallbatches.Theestimatedretest
periodorshelflifefromthecombineddataisusuallylongerthanthatfromindividual
batchesbecausethewidthoftheconfidencelimit(s)forthemeanwillbecome
narrowerastheamountofdataincreaseswhenbatchesarecombined.
Thepoolingtestsdescribedaboveshouldbeperformedinaproperordersuchthatthe
slopetermsaretestedbeforetheinterceptterms.Themostreducedmodel(i.e.,
individualslopes,commonslopewithindividualintercepts,orcommonslopewith
commonintercept,asappropriate)canbeselectedforretestperiodorshelflife
estimation.
B.2.2.2Othermethods
Statisticalprocedures
2-6
otherthanthosedescribedabovecanbeusedinretestperiod
orshelflifeestimation.Forexample,ifitispossibletodecideinadvancethe
acceptabledifferenceinslopeorinmeanretestperiodorshelflifeamongbatches,an
appropriateprocedureforassessingtheequivalenceinslopeorinmeanretestperiod
orshelflifecanbeusedtodeterminethedatapoolability.However,suchaprocedure
shouldbeprospectivelydefined,evaluated,andjustifiedand,whereappropriate,
discussedwiththeregulatoryauthority.Asimulationstudycanbeuseful,if
applicable,todemonstratethatthestatisticalpropertiesofthealternativeprocedure
selectedareappropriate
7
.
B.3DataAnalysisforMulti-Factor,Full-DesignStudies
Thestabilityofthedrugproductcoulddiffertoacertaindegreeamongdifferent
factorcombinationsinamulti-factor,full-designstudy.Twoapproachescanbe
consideredwhenanalyzingsuchdata.
?Theobjectiveofthefirstapproachistodeterminewhetherthedatafromallfactor
combinationssupporttheproposedshelflife.
?Theobjectiveofthesecondapproach,testingforpoolability,istodetermine
whetherthedatafromdifferentfactorcombinationscanbecombinedforan
overallestimateofasingleshelflife.
B.3.1Evaluatingwhetherallfactorcombinationssupporttheproposed
shelflife
Theobjectiveofthisapproachistoevaluatewhethertheestimatedshelflivesfromall
factorcombinationsarelongerthantheoneproposed.Astatisticalmodelthat
includesallappropriatefactorsandfactorcombinationsshouldbeconstructedas
describedinSectionB.3.2.2.1,andtheshelflifeshouldbeestimatedforeachlevelof
eachfactorandfactorcombination.
Ifallshelflivesestimatedbytheoriginalmodelarelongerthantheproposedshelf
life,furthermodelbuildingisconsideredunnecessaryandtheproposedshelflifewill
generallybeappropriateaslongastheguidanceinSections2.4and2.5isfollowed.If
oneormoreoftheestimatedshelflivesfallshortoftheproposedshelflife,model
buildingasdescribedinSectionB.3.2.2.1canbeemployed.However,itisconsidered
unnecessarytoidentifythefinalmodelbeforeevaluatingwhetherthedatasupport
theproposedshelflife.Shelflivescanbeestimatedateachstageofthemodel
buildingprocess,andifallshelflivesatanystagearelongerthantheoneproposed,
furtherattemptstoreducethemodelareconsideredunnecessary.
EvaluationofStabilityData
12
Thisapproachcansimplifythedataanalysisofacomplicatedmulti-factorstability
studycomparedtothedataanalysisdescribedinSectionB.3.2.2.1.
B.3.2Testingforpoolability
Thestabilitydatafromdifferentcombinationsoffactorsshouldnotbecombined
unlesssupportedbystatisticaltestsforpoolability.
B.3.2.1Testingforpoolabilityofbatchfactoronly
Ifeachfactorcombinationisconsideredseparately,thestabilitydatacanbetestedfor
poolabilityofbatchesonly,andtheshelflifeforeachnon-batchfactorcombination
canbeestimatedseparatelybyapplyingtheproceduredescribedinSectionB.2.For
example,foradrugproductavailableintwostrengthsandfourcontainersizes,eight
setsofdatafromthe2x4strength-sizecombinationscanbeanalyzedandeight
separateshelflivesshouldbeestimatedaccordingly.Ifasingleshelflifeisdesired,
theshortestestimatedshelflifeamongallfactorcombinationsshouldbecomethe
shelflifefortheproduct.However,thisapproachdoesnottakeadvantageofthe
availabledatafromallfactorcombinations,thusgenerallyresultinginshortershelf
livesthandoestheapproachinSectionB.3.2.2.
B.3.2.2Testingforpoolabilityofallfactorsandfactorcombinations
Ifthestabilitydataaretestedforpoolabilityofallfactorsandfactorcombinations
andtheresultsshowthatthedatacanbecombined,asingleshelflifelongerthan
thatestimatedbasedonindividualfactorcombinationsisgenerallyobtainable.The
shelflifeislongerbecausethewidthoftheconfidencelimit(s)forthemeanwill
becomenarrowerastheamountofdataincreaseswhenbatches,strengths,container
sizesand/orfills,etc.arecombined.
B.3.2.2.1Analysisofcovariance
Analysisofcovariancecanbeemployedtotestthedifferenceinslopesandintercepts
oftheregressionlinesamongfactorsandfactorcombinations
7,8
.Thepurposeofthe
procedureistodeterminewhetherdatafrommultiplefactorcombinationscanbe
combinedfortheestimationofasingleshelflife.
Thefullstatisticalmodelshouldincludetheinterceptandslopetermsofallmain
effectsandinteractioneffectsandatermreflectingtherandomerrorofmeasurement.
Ifitcanbejustifiedthatthehigherorderinteractionsareverysmall,thereis
generallynoneedtoincludethesetermsinthemodel.Incaseswheretheanalytical
resultsattheinitialtimepointareobtainedfromthefinisheddosageformpriortoits
packaging,thecontainerintercepttermcanbeexcludedfromthefullmodelbecause
theresultsarecommonamongthedifferentcontainersizesand/orfills.
Thetestsforpoolabilityshouldbespecifiedtodeterminewhetherthereare
statisticallysignificantdifferencesamongfactorsandfactorcombinations.Generally,
thepoolingtestsshouldbeperformedinaproperordersuchthattheslopetermsare
testedbeforetheintercepttermsandtheinteractioneffectsaretestedbeforethemain
effects.Forexample,thetestscanstartwiththeslopeandthentheintercepttermsof
thehighestorderinteraction,andproceedtotheslopeandthentheintercepttermsof
thesimplemaineffects.Themostreducedmodel,obtainedwhenallremainingterms
arefoundtobestatisticallysignificant,canbeusedtoestimatetheshelflives.
EvaluationofStabilityData
13
Alltestsshouldbeconductedusingappropriatelevelsofsignificance.Itis
recommendedthatasignificancelevelof0.25beusedforbatch-relatedterms,anda
significancelevelof0.05beusedfornon-batch-relatedterms.Ifthetestsfor
poolabilityshowthatthedatafromdifferentfactorcombinationscanbecombined,the
shelflifecanbeestimatedaccordingtotheproceduredescribedinSectionB.1using
thecombineddata.
Ifthetestsforpoolabilityshowthatthedatafromcertainfactorsorfactor
combinationsshouldnotbecombined,eitheroftwoalternativescanbeapplied:(1)a
separateshelflifecanbeestimatedforeachlevelofthefactorsandofthefactor
combinationsremaininginthemodel;or(2)asingleshelflifecanbeestimatedbased
ontheshortestestimatedshelflifeamongalllevelsoffactorsandfactorcombinations
remaininginthemodel.
B.3.2.2.2Othermethods
Alternativestatisticalprocedures
2-6
tothosedescribedabovecanbeapplied.For
example,anappropriateprocedureforassessingtheequivalenceinslopeorinmean
shelflifecanbeusedtodeterminethedatapoolability.However,suchaprocedure
shouldbeprospectivelydefined,evaluated,properlyjustified,and,whereappropriate,
discussedwiththeregulatoryauthority.Asimulationstudycanbeuseful,if
applicable,todemonstratethatthestatisticalpropertiesofthealternativeprocedure
selectedareappropriate
7
.
B.4DataAnalysisForBracketingDesignStudies
ThestatisticalproceduresdescribedinSectionB.3canbeappliedtotheanalysisof
stabilitydataobtainedfromabracketingdesignstudy.Forexample,foradrug
productavailableinthreestrengths(S1,S2,andS3)andthreecontainersizes(P1,
P2,andP3)andstudiedaccordingtoabracketingdesignwhereonlythetwoextremes
ofthecontainersizes(P1andP3)aretested,sixsetsofdatafromthe3x2strength-
sizecombinationswillbeobtained.Thedatacanbeanalyzedseparatelyforeachof
thesixcombinationsforshelflifeestimationaccordingtoSectionB.3.2.1,ortestedfor
poolabilitypriortoshelflifeestimationaccordingtoSectionB.3.2.2.
Thebracketingdesignassumesthatthestabilityoftheintermediatestrengthsor
sizesisrepresentedbythestabilityattheextremes.Ifthestatisticalanalysis
indicatesthatthestabilityoftheextremestrengthsorsizesisdifferent,the
intermediatestrengthsorsizesshouldbeconsiderednomorestablethantheleast
stableextreme.Forexample,ifP1fromtheabovebracketingdesignisfoundtobe
lessstablethanP3,theshelflifeforP2shouldnotexceedthatforP1.No
interpolationbetweenP1andP3shouldbeconsidered.
B.5DataAnalysisForMatrixingDesignStudies
Amatrixingdesignhasonlyafractionofthetotalnumberofsamplestestedatany
specifiedtimepoint.Therefore,itisimportanttoascertainthatallfactorsandfactor
combinationsthatcanhaveanimpactonshelflifeestimationhavebeen
appropriatelytested.Forameaningfulinterpretationofthestudyresultsandshelf
lifeestimation,certainassumptionsshouldbemadeandjustified.Forinstance,the
assumptionthatthestabilityofthesamplestestedrepresentsthestabilityofall
samplesshouldbevalid.Inaddition,ifthedesignisnotbalanced,somefactorsor
factorinteractionsmightnotbeestimable.Furthermore,fordifferentlevelsoffactor
combinationstobepoolable,itmighthavetobeassumedthatthehigherorderfactor
EvaluationofStabilityData
14
interactionsarenegligible.Becauseitisusuallyimpossibletostatisticallytestthe
assumptionthatthehigherordertermsarenegligible,amatrixingdesignshouldbe
usedonlywhenitisreasonabletoassumethattheseinteractionsareindeedvery
small,basedonsupportingdata.
ThestatisticalproceduredescribedinSectionB.3canbeappliedtotheanalysisof
stabilitydataobtainedfromamatrixingdesignstudy.Thestatisticalanalysisshould
clearlyidentifytheprocedureandassumptionsused.Forinstance,theassumptions
underlyingthemodelinwhichinteractiontermsarenegligibleshouldbestated.Ifa
preliminarytestisperformedforthepurposeofeliminatingfactorinteractionsfrom
themodel,theprocedureusedshouldbeprovidedandjustified.Thefinalmodelon
whichtheestimationofshelflifewillbebasedshouldbestated.Theestimationof
shelflifeshouldbeperformedforeachofthetermsremaininginthemodel.Theuseof
amatrixingdesigncanresultinanestimatedshelflifeshorterthanthatresulting
fromafulldesign.
Wherebracketingandmatrixingarecombinedinonedesign,thestatisticalprocedure
describedinSectionB.3canbeapplied.
B.6References
1.Carstensen,J.T.,“StabilityandDatingofSolidDosageForms”
PharmaceuticsofSolidsandSolidDosageForms,Wiley-Interscience,182-185,
1977
2.Ruberg,S.J.andStegeman,J.W.,“PoolingDataforStabilityStudies:Testing
theEqualityofBatchDegradationSlopes”
Biometrics,47:1059-1069,1991
3.Ruberg,S.J.andHsu,J.C.,“MultipleComparisonProceduresforPooling
BatchesinStabilityStudies”
Technometrics,34:465-472,1992
4.Shao,J.andChow,S.C.,“StatisticalInferenceinStabilityAnalysis”
Biometrics,50:753-763,1994
5.Murphy,J.R.andWeisman,D.,“UsingRandomSlopesforEstimatingShelf-
life”
ProceedingsofAmericanStatisticalAssociationoftheBiopharmaceutical
Section,196-200,1990
6.Yoshioka,S.,Aso,Y,andKojima,S.,“AssessmentofShelf-lifeEquivalenceof
PharmaceuticalProducts”
Chem.Pharm.Bull.,45:1482-1484,1997
7.Chen,J.J.,Ahn,H.,andTsong,Y.,“Shelf-lifeEstimationforMultifactor
StabilityStudies”
DrugInf.Journal,31:573-587,1997
8.Fairweather,W.,Lin,T.D.,andKelly,R.,“Regulatory,Design,andAnalysis
AspectsofComplexStabilityStudies”
J.Pharm.Sci.,84:1322-1326,1995
EvaluationofStabilityData
B.7Figures
Figure1
ShelflifeEstimationwithUpperandLowerAcceptanceCriteriaBasedonAssayat
25C/60%RH
80
85
90
95
100
105
110
115
120
036912151821242730333639424548
TimePoint(Months)
A
ssa
y
(
%
o
f
La
bel
C
lai
m
)
RawData
Upperconfidencelimit
Lowerconfidencelimit
Regressionline
Upperacceptance
criterion:105
Loweracceptance
criterion:95
Figure2
ShelflifeEstimationwithUpperAcceptanceCriterionBasedonaDegradation
Productat25C/60%RH
0.0
0.5
1.0
1.5
2.0
2.5
3.0
036912151821242730333639424548
TimePoint(Months)
D
e
g
r
a
d
at
i
o
nPro
duc
t
(
%
)
RawData
Upperconfidencelimit
Regressionline
Upperacceptance
criterion:1.4
15
|
|
