Missing Data in Neuroscience Clinical Trials: Truth or Consequences

Co-Chairs:  Pilar Lim, PhD  / George Haig, PharmD

When conducting clinical trials, one is searching for the Truth. Unfortunately, when there is substantial missing data, the truth may be unknown and the statistician and clinician must deal with the Consequences (all of the different methods and sensitivity analyses).

It is widely known in neuroscience clinical trials that a fair amount of missing data can be present. Missing data occur when not all patient visits are completed, for whatever reason, resulting in omitted data at scheduled time points. In trials concerning depression and schizophrenia, for example, it is not unusual to have at least a third of the key data missing. This creates a problem for statistical inference in the primary analysis and interpretation of the trial results. As a commonly seen approach, the primary analysis is based on the MAR (missing-at-random) assumption that only observed data explain dropouts, followed by a few sensitivity analyses based on the MNAR (missing-not-at-random) assumption that unobserved data also explain dropouts. This session will present recent biopharmaceutical, academic, and regulatory perspectives concerning approaches for the primary and sensitivity analysis. This topic will also be discussed in light of recent debates concerning the primary estimand (the main clinical quantity of interest to be estimated in a study) and corresponding analysis methods. Varied examples in the neuroscience therapeutic area will be presented. The session will also discuss end-to-end procedures that could be put in place in a clinical development program to prevent or minimize missing data. Examples will also be provided.