5.1 Standard MMRM SAS code for Parkinson's disease example.
5.2 MMRM with spatial power, random subject effect and sandwich estimator.
5.3 GEE with compound symmetry working correlation matrix.
5.4 MMRM SAS code for Parkinson's disease treatment × gender interaction at Visit 8.
5.5 MMRM SAS code for Parkinson's disease cLDA model with baseline as a repeated measurement.
5.6 Logistic GLMM SAS code for the mania study.
5.7 Logistic GEE SAS code for the mania study.
6.1 Using PROC MI to examine patterns of missingness of TST in the insomnia example.
6.2 Imputation using monotone regression with a default model for each imputed variable corresponding to TST assessments in the insomnia example.
6.3 Imputation using monotone regression with a user-specified model for each imputed variable corresponding to a TST assessment in the insomnia example.
6.4 Imputation of TST using MSS assessments as ancillary variables in the insomnia example using monotone regression with a default model.
6.5 Full imputation with the MCMC method for the UPDRS score in the Parkinson's disease example.
6.6 Partial imputation using the MCMC method with the remaining imputations done by monotone regression for the UPDRS score in the Parkinson's disease example.
6.7 An ANCOVA analysis by time point on multiply-imputed data for the insomnia example.
6.8 MMRM analysis on multiply-imputed data for the insomnia example.
6.9 Combining results from ANCOVA analysis of multiply-imputed data for the insomnia example.
6.10 Combining results from ANCOVA analysis of multiply-imputed data using adjusted degrees of freedom for the insomnia example.
6.11 Combining results from an MMRM analysis of multiply-imputed data for the insomnia example.
6.12 Combining results from a logistic regression analysis on multiply-imputed responder status at Visit 5 (study Day 28) for the mania example.
6.13 Combining results from a CMH test on multiply-imputed responder status at Visit 5 (study Day 28) for the mania example.
6.14 Combining Mantel–Haenszel estimates of the common odds ratio for multiply-imputed responder status at Visit 5 (study Day 28) in the mania example.
6.15 Combining estimates of responder proportions in each treatment arm and difference in proportions from multiply-imputed responder status at Visit 5 (study Day 28) in the mania example.
6.16 Tests for no difference in baseline characteristics between completers and dropouts for the insomnia example.
6.17 Longitudinal logistic model of dropout to explore the effects of baseline and post-baseline values on discontinuation in the insomnia example.
6.18 Imputation using monotone regression with a model for each imputed variable corresponding to TST assessments in the insomnia example including treatment by outcome interactions for post-baseline visits used as predictors.
6.19 Diagnostic plots of MCMC convergence for the insomnia example.
6.20 Imputation using monotone regression and predictive mean matching methods for TST and MSS in the insomnia example.
7.1 Simplified MI model to illustrate interaction by visit.
7.2 Simplified direct likelihood MMRM model to illustrate interaction by visit.
7.3 Use MCMC to impute non-monotone missing values.
7.4 Use regression imputation to complete the MAR imputation, to match an MMRM where covariances are estimated across all treatment groups.
7.5 Use regression imputation to complete the MAR imputation; the inclusion of interactions with treatment is consistent with an MMRM where covariances are estimated separately for each treatment group (as with the GROUP=
trt option in PROC MIXED).
7.6 Select observations so as to impute Visit 1 using model based on control arm.
7.7 Impute Visit 1 using model based on control arm.
7.8 Re-assemble the dataset to impute missing values at the next visit.
7.9 Impute Visit 2 using model-based control arm.
7.10 Create dataset with missing values for reference treatment imputed assuming MAR, as preparation for J2R.
7.11 Impute distribution of control to Visit 2 value in experimental arm with baseline as the only covariable.
7.12 Perform the MI assuming MAR for the control arm.
7.13 Prepare to model the distribution of baseline values for BOCF-like imputation.
7.14 Impute baseline-like values for the subject visits with missing outcomes.
7.15 Transpose imputed values to obtain a dataset with one record per subject with all imputed values for a subject on the same record.
7.16 Use baseline-distributed values for imputing the subject visits with missing post-baseline assessments.
7.17 Select subjects with last observation at Visit 1 to impute LOCF values for Visits 2, 3, 4 and 5 for these subjects.
7.18 Create LOCF-like values for subjects whose last observation was at Visit 1.
7.19 Imputing last observation-distributed values, adapting the code from SAS Code Fragment 7.14.
7.20 Imputing baseline-like values for subjects with reason for discontinuation = “AE.”
7.21 Imputing a base imputation before imposing a delta adjustment.
7.22 Adjust imputed value by an amount δ.
7.23 Imposing a delta adjustment after all subjects have been imputed based on MAR assumption via MI with joint modeling as implemented by Roger.
7.24 Imposing an adjustment to imputed binary response with a probability δ.
7.25 Call the
run_tdelta macro repeatedly to implement a tipping point analysis.
8.1 Take a bootstrap sample.
8.2 Simple macro to create missingness indicators.
8.3 Code to calculate the probability of being observed at the first visit.
8.4 Code to calculate the probability of being observed at the first or the second visit.
8.5 Code to calculate the unconditional probability of being observed at each visit.
8.6 Calculate the inverse probability of being observed.
8.7 Convert the dataset to vertical format and run the weighted imputation model.
8.8 Set the predicted outcome from the imputation model as the new response and run the final analysis model.
8.9 Use LSMEANS to get the treatment effect at the final visit for each bootstrap sample and PROC MEANS to get the overall estimate of treatment effect.
8.10 Imputation model for each bootstrap sample.
8.11 Final analysis model to calculate the treatment change from baseline.