getting to access: squeezing value kristian thorlund, phd
TRANSCRIPT
OUTLINE
• HOWTEACUPSANDCROPFIELDSSHAPEDCLINICALRESEARCH
• BAYESIANTHINKINGINRAREDISORDERSRESEARCH
• 4SLIDESWITHBULLETPOINTS(TIMETOCHECKYOUREMAIL)
• THEGRANDFINALE
UTILITY OF STATISTICS INRAREDISORDERS
Thepremiseofconven/onalsta/s/csisreplicabilityofexperiments!
InthehistoryofHTAdecision-making,howo@endidap-value>0.10s/mulateaposi/vedecisiontofund?
FROM BIG TO SMALL IN HEALTH RESEARCH
Replicable&adequately
powereddata
DatasizevsstaVsVcal/analyVcmethod
BIGDATA
Raredisorderdata
P-values95%CIsANOVA
Regression…
NeuralnetworksMachinelearningBayesianmethodsComb.Algorithms
…
BayesianthinkingBayesianmethodsNovelstudydesignsOpVmizedmonitoring
…
BEING A BAYESIAN THINKER Priorbelief+newinformaVon(data)=updatedbeliefThelesscertainweareinourpriorbeliefs,themorenewdataweightinourupdatedbelief,andviceversa
Priorbelief
Newdata
Newbelief
Txeffectcertainty
Priorbelief
Newdata
Newbelief
Txeffectcertainty
BAYESIAN THINKING FOR RARE DISORDERS
!me
Bayesianapproachestohealthtechnologyassessmentcanbedefinedasan“explicitandqualitaVveuseofexternalevidenceinthedesign,monitoring,analysis,interpretaVonandreporVngoftheresults”1
1.Abrahayanetal.JPTCP2014:21(1):e66-e78.
LEARNING FROM UK … AND BEYOND
• TheUKhasledthewaywithmanagedaccessagreements
• ConVnualmonitoringanddatacollecVonisfacilitated
• ForMPStypeVIa,paVentsaregrantedaccesstoelosulfasealfaifthey(agreementissuedNov2015)• Akend3clinicalassessmentsperyear• Respondtotreatmentwithinyear1thebymeeVngat
least4of5domainresponsecriteria
ONMONITORING
LEARNING FROM UK … AND BEYOND
4/5domainresponsecriteria:• 10%improvementon6MWT• 5%improvementonFVC/FEV-1• StabilizaVonin2/3ofQoL,Depression,Pain• 20%reducVoninuKSs• 10%declineinejecVonfracVon
ONMONITORING
Sincethenonenaturalhistorystudyandfourclinicaltrialsinvariouspopula/onsandofvariousdesignshavebeenpublished
NOVEL CLINICAL TRIAL DESIGNS • Randomizedplacebophasedesigns,randomlyassignspaVents
tovariousduraVonsofplacebobeforeswitchingtoacVvetreatment.SimulaVonshaveshowclosetosimilarpowertoconvenVonalRCTs.PastevidencecaninformresponseVme.
• Simpleadap!vetrials(e.g.,playthewinner)canprovidequickanswers,parVcularlywithBayesianpriors.
• Cross-overdesignsdramaVcallyreducessamplesizeandavoidsimbalanceinprognosVcfactors
MORE EXTERNAL EVIDENCE • EXAMPLE(NCT02437253):PhaseII/III2x16wkcross-overtrial
ofAdalimumabvsplaceboforMSPtypeI,IIandIV.Opportunitytolearnabout(dis)similariVesacrosstypes.
• Lymphangioleiomyomatosis(LAM)(condiVonofthelungs)andRenalAngiomyolipoma(AML)associatedwithTSCarebothcharacterisedbyadiffuseinfiltraVonbybioplasVcsmoothmuscle-likecellsthatinvadetheorgan’scells.SirolimusisusedforLAM,everolimusforAML.
TSC:TUBEROUSSCLEROSISCOMPLEX
SUMMARY • ConvenVonalstaVsVcsbreakdownintheraredisease
scenario,sowhyarewesVllinsisVngonp-valuesandrandomizaVon?
• DynamicandBayesianthinkingoffergreatawayofgerng(much)moreoutofthedata,butonlywithproperplanning
• Well-designedinnovaVvetrialscoupledwithhistoricdataandexternalevidencearemoreefficientatextracVngvaluefromthedata