In smoking cessation trials, it is common to assume that all those who fail to provide outcome data have also failed to give up smoking. This widely used, and advocated, “missing=smoking” assumption sets smoking cessation trials apart from those performed in other areas. This is because elsewhere it is common to assume that data are missing at random.
In order to investigate this issue, Dan Jackson, member of Ian White’s group from the BSU’s Design and Analysis of Randomised Trials theme, performed statistical analyses using data from the iQUIT trial, which is an internet based smoking cessation trial. These analyses used information on the repeated attempts (telephone calls and an email) made to try to obtain participants’ smoking statuses. The hypothesis that “missing=smoking” was not supported by these analyses.
A more accessible approach to the analysis of smoking cessation trials with missing outcome data is also currently being developed. This allows trialists to make a range of alternative assumptions, such as “missing=smoking” or instead that data are missing at random. By exploring how sensitive important conclusions are to the assumption made about missing outcome data, trialists can assess the robustness of their conclusions. We call this type of investigation a “sensitivity analysis”.