Introduction We assessed the added value of multiple imputation (MI) of missing values in longitudinal datasets with carotid intima-media thickness (CIMT) as primary endpoint subsequently analysed with linear mixed effects (LME) models.
Methods Analyses were based on a subset of 300 participants from the METEOR (Measuring Effects on Intima-Media Thickness: an Evaluation of Rosuvastatin) trial. CIMT measurements were performed at 12 carotid sites over seven examinations. The “true” difference in rate of change in CIMT between rosuvastatin and placebo was derived from a completed dataset. Scenarios with missing values were defined, both MCAR (Missing Completely At Random) and MAR (Missing at Random), with 10 to 60% missing values, related to, among others, age and treatment allocation. LME analyses were performed with and without preceding MI. The added value of MI was assessed by comparing the LME estimates with the true value in terms of bias and precision.
Results Bias in point estimates for LME analysis with and without preceding MI was similar in scenarios with ≤40% missings. With 60% missing values, LME without MI was superior to LME with MI. Coverage of the 95% CIs was similar for LME with and without MI for all scenarios.
Conclusion Applying MI prior to LME analyses on longitudinal CIMT measurements does not increase precision or reduce bias in the estimated differences in rates of change in CIMT. Hence, MI has no added value in this context, and direct application of LME remains the preferred method in trials using CIMT as primary endpoint.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.