Analysis of missing mechanism in IVUS imaging clinical trials with missing covariates

Intravascular ultrasound (IVUS) is a clinical imaging procedure used to assess drug effects on the progression of coronary atherosclerosis in clinical trials. It is an invasive medical procedure of measuring coronary artery atheroma (plaque) volume, and leads to high missing rates (often over 30%). This paper uses an IVUS Phase II clinical trial data to explore the missing mechanism of IVUS endpoint, the percent atheroma volume (PAV). We proposed a moving-window method to examine the relationship between continuous covariates such as lipid endpoint and the probability of missing IVUS values, which provides a general approach for missing mechanism exploration. The moving-window method is more intuitive and provides a fuller picture about the relationship. In the example, some covariates such as lipid measures also have high missing rates after 12 months because of compliance issues probably caused by fatigue of blood drawing. We found that if the method of last observation carried forward (LOCF) is used to impute the lipid endpoints, it leads to biologically unexplainable results. Using the multiple imputation approach for the missing covariates results in a more reasonable conclusion about the IVUS missing mechanism. Age, race, and baseline PAV are identified as key potential contributors to the probability of missing IVUS endpoint. This finding can be used to reduce missing values in future IVUS trials by setting up appropriate inclusion and exclusion criteria at the trial design stages.