Tipping Point Analysis in Handling of Missing Data
*Ruvie Martin, Novartis Pharmaceuticals 

Keywords: tipping point, missing data

Most clinical trials will have at least some missing data. This leads to more discussion on how to handle missing data and different sensitivity analyses. One sensitivity analyses that is gaining popularity even with the Health Authorities is the Tipping Point Analysis (Yan et al., 2009; Campbell et al., 2011). The goal of the tipping point analysis is to identify assumptions about the missing data under which the conclusions change, i.e., under which there is no longer evidence of a treatment effect. Tipping point can look at the difference of means (for continuous data) or the difference of the number of events (for binary data) between the treatment groups in the missing cohort at which the study conclusion is changed. A tipping point analysis replaces the missing value with some values so that the resulting p value of the treatment comparison is equal to (or larger but close to) a pre-specified significance level (Yan et al., 2009). Tipping-point analysis for binary endpoints can analyze all possible cases that may occur in the missing data group. For two-arm comparisons, the results can be displayed by a two-dimensional plot. Robustness and clinical meaningfulness of results goes hand in hand on any sensitivity analyses, hence applications of tipping point for binary data are examined and how the combination of the amount of missing data and clinical relevance provide robust conclusions.