Missing Data: Turning Guidance Into Action

Response to Comments by Seaman et al. on “Analysis of Longitudinal Trials With Protocol Deviation: A Framework for Relevant, Accessible Assumptions, and Inference via Multiple Imputation,” Journal of Biopharmaceutical Statistics 23:1352–1371

“…These give a more formal justification for the use of Rubin's rules in this setting, and will be presented in a future article. We conclude by drawing readers' attention to Mallinckrodt et al (2013), who consider the role that methods such as those described in our article-which they term "controlled-imputation methods"-may play in the analysis strategy for a trial with missing outcome data. They comment that they are "specially useful in constructing analyses to assess specific departures from MAR and for assessing effectiveness because the assumptions are transparent."…”

“…Their calculations do not appear to allow for the fact that after withdrawal the distribution is conditional on previous residuals calculated as difference from the reference arm mean rather than using the subject's own previous residuals as in J2R. As a result, the expected value of treatment effect θ should depend on the underlying covariance and is about −2.7, lying between that for CIR (−2.8) and that for J2R (−2.4), as found nearly always in practice by those using this approach (Mallinckrodt et al, 2013).…”

An efficient monotone data augmentation algorithm for multiple imputation in a class of pattern mixture models

“…The model essentially assumes that immediately upon withdrawal from the active group, all benefit from the treatment is gone (Mallinckrodt et al, 2013). …”

“…Two types of PMMs that are commonly used as MNAR sensitivity analyses in confirmatory trials are the control-based PMMs and the delta-adjusted PMMs (Little and Yau, 1996;Carpenter et al, 2013;Ratitch et al, 2013;Mallinckrodt et al, 2013). The control-based methods assume that the statistical behavior of active subjects after dropout is similar to that of control subjects since subjects no longer receive the active treatment after discontinuation (Little and Yau, 1996;Carpenter et al, 2013;Ayele et al, 2014).…”

“…The delta-adjusted PMMs assume that subjects who discontinue from the active treatment will have, on average, their unobserved value worse by some amount ∆ compared with subjects who continue to the next time point . The delta adjusted imputations are often implemented using the tipping point (progressive stress-testing) strategy Mallinckrodt et al, 2013) in the sense that the analyses are conducted with a range of increasing ∆ values in order to find the ∆ value at which the significance of the statistical analysis is lost. It is often challenging to specify in advance some range of clinically plausible values for ∆ without knowledge of the actual data .…”