9Data set Simulated - standing diastolic BP Eight week study of test drug vs placeboClinic visit every 2 weeksPrimary endpoint – change in standing BP from baseline to week 8Patients completed the study or dropped out at various time pointsMissing completely at random
11Modeling approaches Many proposals to deal with dropouts Mixed model approachRepeated measuresRandom intercept, random slopeSingle imputationMultiple imputationImputation modelAnalysis model
12ANCOVA on LOCF data TREATMENT | 1 2441.0 4.13 0.0444 Source | df MS F p-ValueTREATMENT |CENTER |BASELINE |ERROR |Statistic Test Drug PlaceboRaw MeanAdj MeanStd ErrorN
13Analysis of completed cases Source | df MS F p-ValueTREATMENT |CENTER |BASELINE |ERROR |Statistic Test Drug PlaceboRaw MeanAdj MeanStd ErrorN
14Naive interpretation If LOCF provides statistical significance If completer analysis supports LOCFTrue story may lie between the twoClinical conclusion can be made
15Mixed model analysisFor demonstration purposes, only repeated measure results are presentedproc mixed method=reml ; where week>0 ;class pid trt week ctr ;model y=wk0 trt ctr week trt*week/solution ;repeated week / type=cs subject=pid r rcorr ;estimate 'trt dif at week 8' trt -1 1 trt*week / cl alpha=0.05 ;
16Results from PROC MIXED Num DenEffect DF DF F Value Pr > FBaselineTreatmentCenter <.0001WeekTrt*weekStandardLabel Estimate Error DF t Value Pr > |t|week 8 dif
17Single or multiple imputation Mixed model can be considered as single imputationFor imputation, we can use the same model for imputation and analysisHowever, one model can be used for imputation, but a different one is for analysis
18Should LOCF be used?After the modeling approaches became available, use of LOCF have been discouragedModels are developed with assumptionsMore complicated models require more assumptionsAre these assumptions justified?
19Should LOCF be used?LOCF is a model and there are simple assumptions behind itIn New Drug Applications (NDA), LOCF is still widely usedWhy?
20Different phases in clinical trials Phase I, II, III, IVPhase I – How often?Phase II – How much?Phase III – ConfirmPhase IV – Post-Market
21DOES THE DRUG WORK?Double-blind, placebo controlled, randomized clinical trialTest hypothesis - does the drug work?Null hypothesis (H0) - no difference between test drug and placeboAlternative hypothesis (Ha) - there is a difference
22TYPES OF ERRORSRegulatory agencies focus on the control of Type I errorProbability of making a Type I error is not greater than aIn general, a = 0.05; i.e., 1 in 20Avoid inflation of this errorChanging the method of analysis to fit data will inflate a
23MULTIPLE COMPARISONSFor 20 independent variables (clinical endpoints), one significant at randomFor 20 independent treatment comparisons, one significant at randomSubgroup analyses can also potentially inflate aMultiple comparison adjustment
24Report all data Scientific experiments generate data Outliers may be observedDelete outlier?Clinical trials generate dataA wonder drug cures 9,999 patients of 10,000One died – outlier – delete?
25Statistical Analysis Plan (SAP) Pre-specification of analysisPrior to breaking blindInternal agreement within project teamBinding document to communicate with regulatory authoritiesUse of LOCF or modeling approach need to be pre-specified in SAP
26Modeling approaches Assumptions Can be complicated Difficult to explain to end usersGeorge Box – “All models are wrong, some are useful”
27Why LOCF? Or why not? Easy to understand Easy to communicate between statisticians and clinicians, and between sponsor and regulatorsLots of prior examplesBiased point estimate, biased variance
28Recommendations Understand the disease Understand data to be collected Understand the dropout issuesMake use of Phase II resultsEncourage use of statistical modelsLOCF may still be considered as supportive