生物统计学英文课件1.ppt

Basic Design ConsiderationsSTAT 585EpidemiologyReview from STAT 584Types of studiesCross-sectional studyData are obtained from a random sample at one point in time which gives a snapshot of a populationProspective studyA cohort of individuals are identified who are free of a particular disease under study and data are collected on certain risk factorsThese individuals are then followed over some specified period of time to determine whether they get disease or notEpidemiologyRetrospective or case-control studyIndividuals with disease(called cases)and individuals without disease(called controls)are identifiedUsing records or questionnaires the investigators go back in time and ascertain exposure status and risk factors from their past.Such data are used to estimate the relative risk of developing disease between exposed and un-exposed.Clinical trialA prospective study(or an intervention study)comparing the effect and value of intervention against a control in human beingsA clinical trial is an experiment which involves patients and is designed to elucidate the most appropriate treatment of future patientsEssential feature is that the allocation of subject to treatment is planned.Contrast this with cross-sectional surveys,cohort and case-control studies wherein,for example,there is no control over who is and who is not a cigarette smoker.Clinical TrialPrerequisitesNeed for the trialHigher incidence or poor prognosis for the diseaseNo existing treatment or lack thereofThe intervention must have promise of efficacyTrial question must be appropriate or unambiguousValid trial architecture(via,random allocation of patients,blinded trial,use of placebo etc)Appropriate inclusion/exclusion criteriaFeasible trial protocolEffective trial administrationObjectives and outcomes of a clinical trialPrimary objective:What is the primary question to be answered?ideally just oneimportant,relevant to care of future patientscapable of being answeredPrimary outcome(endpoint)ideally just onerelatively simple to analyze and reportshould be well defined;objective measurement is preferred to a subjective one.For example,clinical and laboratory measurements are more objective than say clinical and patient impressionSecondary objectivesSecondary Questionsother outcomes or endpoints of interestsubgroup analysessecondary questions should be viewed as exploratorytrial may lack power to address themmultiple comparisons will increase the chance of finding“statistically significant”differences even if there is no effectavoid excessive evaluations;as well as problem with multiplecomparisons,this may effect data quality and patient supportExample(1)Physicians Health Study(PHS)Ref:N Engl J Med 2005;352:1293-304.Why do the trial?Randomized trials have shown that low-dose aspirin decreases the risk of a first myocardial infarction in men,with little effect on the risk of ischemic stroke.There are few similar data in women.MethodRandomly assigned 39,876 initially healthy women 45 years of age or older to receive 100 mg of aspirin on alternate days or placebo and then monitored them for 10 years for a first major cardiovascular event(i.e.,nonfatal myocardial infarction,nonfatal stroke,or death from cardiovascular causes).Example(1)Primary EndpointThe primary end point was a combination of major cardiovascular events,including nonfatal myocardial infarction,nonfatal stroke,and death from cardiovascular causesThe trial was initially designed to have a statistical power of 86 percent to detect a 25 percent reduction in this end point.Example(1)Secondary endpointSecondary end points included the individual end points of fatal or nonfatal myocardial infarction,fatal or nonfatal stroke,ischemic stroke,hemorrhagic stroke,and death from cardiovascular causes.Example(2)Eastern Cooperative Oncology group(ECOG 1178)Tamoxifen(Soltamox)blocks actions of estrogen and is used to treat and prevent some types of breast cancerTamoxifen vs placeboPrimary:tumor recurrence/relapse,disease-free survivalSecondary:total mortalityExample(3)Multicenter Investigation of Limitation of Infarction Size(MILIS)Propranolol is a non-selective beta blocker used in the treatment of hypertensionPropranolol vs placeboPrimary:ultimate size of an acute myocardial infarctionSecondary:left ventricular ejection fractionDealing with primary and secondary outcomesMany advocate having a single primary endpointdrives sample size calculationstest based on this endpoint has a 5%type I error rateAll other endpoints are“secondary”Delineate primary and secondary outcomesCan be hard to adhere to in practiceFor example,what if primary outcome is not different among groups,but all secondary outcomes are?What to do with primary and secondary endpoints?ONeill,R.(1997)“Secondary endpoints cannot be validly analyzed if the primary endpoint does not demonstrate clear statistical significance”Controlled Clinical Trials,550 556Davis,C.E.(1997)“Secondary endpoints can be validly analyzed,even if the primary endpoint does not provide clear statistical significance”Controlled Clinical Trials,557-560ONeill(1997)Primary endpoint definition:“clinical endpoint that provides evidence sufficient to fully categorize clinically the effect of a treatment that would support a regulatory claim for the treatment”Secondary endpoint:“additional clinical characterization of a treatment but could not,by itself,be convincing of a clinically significant treatment effect”ONeill(1997)Argues that primary and secondary outcomes are generally relatedAnalysis of secondary should be conditional on the primary outcome analysis resultespecially true when“secondary”outcomes depend directly on primary(survival)Cant quantify the uncertainty in analyses done after looking at resultsDavis(1997)Strict adherence could miss important and unexpected results Argues that the major problem is multiple comparison issueits a statistical problem,so use a statistical solutionOne such“solution”is the Bonferroni adjustment(will be discussed later in the course)Choice of primary endpointExample:new treatment for HIV patients;attacks HIV virus.1.Increase in CD4 count.2.Viral RNA reduction.Measures the amount of virus in the body3.Time to the first opportunistic infection4.Time to death from any cause5.Time to death or first opportunistic infection,whichever comes firstOutcomes 1&2 are good for a phase II trial(they are objectively measurable).Outcome 4 is the ultimate endpoint;may lead to obsolescence.Outcome 3 is not complete.Outcome 5 is good one in a phase III trial.Opportunistic infectionsPeople with HIV can get many infections(called opportunistic infections,or OIs).Many of these illnesses are very serious,and they need to be treated.Some can be prevented and might take long to develop.Example Michalowicz et al(NEJM,Nov 2,2006)Study of periodontal therapy and birth outcomeSeveral outcomes of interest:preterm birth(before 37 weeks),birth weight,proportion of infants who are small for gestational age,Apgar scores,admissions to NICU.Whats the problem?If test each endpoint at the 5%level:overall chance of finding at least one endpoint where there is a significant difference is larger than 5%,even if the treatments are identicalProne to distorted reporting(i.e.pick most significant)Good reference:Pocock(1997)Controlled Clinical Trials,p 530-545Multiplicity issue will be discussed laterDealing with problems with multiple endpointsHave a pre-defined strategySome advocate:all results pre-written,with results filled in as trial concludesAlternative view:need to be flexibleneed to allow for unexpected findingsbut recognize potential for problems:type I error rate is not 5%Surrogate endpointsWhy do we use surrogate endpoint?Can be measured earlierConvenient or less invasiveCan be measured more frequentlyCan accelerate the approval processAdvantages:May reduce the size of clinical trialsMay shorten the duration of clinical trialsMay reduce the cost of clinical trialsEndpoints in clinical trialsA surrogate endpoint does not directly measure any clinical benefit to patient,it only predicts the outcomeA mixed surrogate/clinical benefit endpoint directly measures significant benefit to patient and predicts an additional,more substantial benefit to patientA clinical outcome directly measures substantial clinical benefit to patientEndpoints inappropriately characterized as surrogatesQuality of life It is an outcome measure(not a surrogate endpoint)Morbidity scaleIt is a clinical benefit endpoint(not a surrogate endpoint)When is the use of surrogate endpoints justified?Screening For promising new therapiesEvaluation of biological activity in phase I/II trialsCaution in using surrogate endpoints:Using biological markers as a surrogate endpoint,one may obtain misleading false positive or false negative conclusion when assessing treatment effects of longer term clinical outcomeRequirements:Before a surrogate endpoint can replace a primary endpoint,it must be formally validated Examples of FDA approval of Drugs using SurrogatesLower cholesterol without evidence of survival benefitLower blood pressure without evidence of benefit for stroke,MI,congestive heart failure or survivalIncreased bone density without evidence of decreased fractures in osteoporosisExamples of FDA Approval of Drugs using SurrogatesIncreased cardiac function in congestive heart failure without evidence of survival benefitDecreased rate of arrhythmias without evidence of survival benefitLower blood glucose and glycosylated hemoglobin without evidence about diabetic complications or survival benefit.Prentice(1989)definition of surrogate endpointSuppose f(T|Z)is the conditional distribution of the true endpoint T given the treatment assignment Z,and S is the surrogate endpoint.All four of the following conditions must be met:ExampleOsteoporosis(Riggs et al,NEJM,1990)Bone loss in post-menopausal women leads to increase risk of fractureSodium Fluoride stimulates bone formation and increased bone mass(double)HypothesisWill Fluoride treatment decrease rate of vertebral fractures?DesignRandomized,double blind,placebo-controlled202 postmenopausal women randomizedAll received calcium supplementOsteoporosis Fluoride trial resultsFluoride increased bone density by 35%35%(p=0.0001)in spine12%(p=0.0001)in femoral neckFluoride decreased bone density by 4%in wrist(p=0.02)Vertebral fractures higher on Fluoride(F 163,P 136,p0.05)Non-vertebral fractures higher on Fluoride(72 vs 24;p=0.01)Fluoride was concluded to be not effective as a treatment for post-menopausal osteoporosisThough increase in bone density was noticed,vertebral fracture risk remained.CommentsAll four conditions imply that the surrogate captures the full of effect of treatment on the primary endpoint,very stringent and difficult to verify.Can fit a modeland test the null H0:b1=0 to verify the last condition P4A significant treatment effect on the primary endpoint after adjustment for S leads to the conclusion that S is a poor surrogate.On the other hand,failing to reject is inadequate to validate S as a good surrogatePurpose of control groupTo allow discrimination of patient outcomes caused by experimental intervention from those caused by other factorsNatural progression of diseaseObserver/patient expectationsOther treatmentFair comparisonsNecessary to be informativeChoice of control groupGoals of Controlled Clinical TrialsTypes of Control GroupsSignificance of Control GroupAssay SensitivityICH E-10,Choice of control groups in clinical trialsConsiderations in Choice of Control GroupAvailable standard therapiesAdequacy of the control evidence for the chosen designEthical considerationsSignificance of control groupInference drawn from the trialEthical acceptability of the trialDegree to which bias is minimizedType of subjectsKind of endpoints that can be studiedCredibility of the resultsAcceptability of the results by regulatory authoritiesOther features of the trial,its conduct,and interpretationTypes of controlsExternalHistoricalConcurrent,not randomizedInternal and concurrentNo treatmentPlaceboDose-responseActive(Positive)controlMultipleBoth an Active and PlaceboMultiple doses of test drug and of an active controlUse of placebo controlThe“placebo effect”is well documentedCould beNo treatment+placeboStandard care+placeboMatched placebos are necessary so patients and investigators cannot decode the treatmentE.g.Vitamin C trial for common coldPlacebo was used,but was distinguishableMany on placebo dropped out of studyThose who knew they were on vitamin C reported fewer cold symptoms and duration than those on vitamin who didnt knowHistorical ControlsA new treatment used in a series of subjectsOutcome compared with previous series of comparable subjectsNon-randomized,non-concurrentRapid,inexpensive,good for initial testing of newtreatmentsTwo sources of historical control data:Literature Subject to publication bias Data baseHistorical controlVulnerable to biasChanges in outcome over time may come from change in:underlying patient populationscriteria for selecting patientspatient care and management peripheral to treatmentdiagnostic or evaluating criteriaquality of data availableChange in definitionTime trendAge-adjusted Death Rates for Selected Causes:United States,1950-76Historical ControlsTend to exaggerate the value of a new treatmentLiterature controls particularly poorEven historical controls from a previous trial in the same institution or organization may still be problematicPocock(1977,Brit Med J)In 19 studies where the same treatment was used in two consecutive trials,differences in survival ranged from 46 to 24,with four differences being statistically significantAdjustment for patient selection may be made,but all other biases will remainConcurrent controlsNot randomizedOne chosen from the same population as the test groupPatients compared,treated by different strategies,same periodAdvantageEliminate time trendData of comparable qualityDisadvantageSelection BiasTreatment groups not comparableCovariance analysis not adequateBias in concurrent control studyTypesMagnitude of effectsFalse positiveSourcesPatient selection Referral patterns Refusals Different eligibility criteriaExperimental environment Diagnosis/staging Supportive care Evaluation methods Data qualityRandomized control studyReference:Byar et al.(1976)New England Journal of MedicinePatients assigned at random to either treatment(s)or controlConsidered to be“Gold Standard”Advantages of Randomized Control Clinical Trial1.Randomization tends to produce comparable groups Design Sources of ImbalanceRandomizedChanceConcurrentChance&Selection Bias(Non-randomized)HistoricalChance,Selection Bias,(Non-randomized)&Time Bias2.Randomization produces valid statistical testsReference:Byar et al(1976)NEJMDisadvantages of Randomized Control Clinical Trial1.Generalizable Results?Subjects may not represent general patient population volunteer effect2.RecruitmentTwice as many new patients3.Acceptability of Rando