Sensible analysis in longitudinal study with drop out: An example using weight loss trial

In many randomized weight loss trials, each subject is measured repeatedly throughout the specified time period. It is, unfortunately, inevitable that, for a variety of reasons, some patients do not complete all of their intended follow-up appointments according to the initial protocol. High loss to follow-up has several detrimental implications on the analysis including: loss of statistical power, potential loss of internal validity and possibility of introducing bias. Therefore to achieve valid conclusion, sensible analysis has to be performed. By using The Yale Bright Bodies Weight Management trial as an example, a collection of methods (graphical and analytical) are introduced for assessing missingness patterns and missingness mechanism. To assessing missingness patterns and missingness mechanism, both graphical method and analytical method has been introduced. The assessing of missing patterns has showed that in our trial data there is no significant different missing pattern between treatment group (weight management group) and control group. To assessing the missing mechanism showed we fail to reject the null hypothesis of MCAR/DCAR. Failing to reject the MCAR/DCAR doesn't mean the missing could not be MAR/DAR. Furthermore, since we can never test MAR versus MNAR, I analyzed the data under assumptions of MAR (repeated measure analysis of covariate models) and NMAR (pattern mixture model) to assess sensitivity of results. Under different missing assumptions, the results consistently show that the weight management program does benefit the subjects in long-term weight loss. Furthermore, to compare the performance of the multiple statistical methods, a simulation study was performed.