Contemporaneous Outcomes Multiple Imputation (KOMI) can communicate missing Glasgow Outcome Scale - Extended (GOSE) scores: A multi-domain imputation for GOSE in the transforming research and clinical knowledge in traumatic brain injury (TRACK-TBI) data.

Abstract

Traumatic brain injury (TBI) brings a major healthcare burden and is a significant contributor to global morbidity and mortality. TBI clinical research often employs the Glasgow Outcome Scale or Glasgow Outcome Scale - Extended (GOSE) as an endpoint. Especially in long-term follow-ups, imputation of missing GOSE scores is often necessary, using approaches such as last observation carried forward (LOCF) or other model-based methods (e.g., Multivariate Imputation by Chain Equation [MICE]). Imputing a missing GOSE score from previous scores can be problematic due to the risk of statistical circularity in the trajectory analysis. We hypothesised that contemporaneous clinical data would include other indicators and outcome measures, which might constrain the imputation of missing GOSE scores, without making assumptions about trajectory from previous GOSE measurements. Therefore, we compared a new method of Contemporaneous Outcomes Multiple Imputation (KOMI) with the established method of imputing missing GOSE scores from available GOSE scores of the same participant, i.e., Longitudinal GOSE Multiple Imputation (LoGMI). To this end, we created simulated missing GOSE datasets, imputed the missing data with MICE and calculated the imputation errors relative to ground truth GOSE scores with ten-fold cross-validation. The new KOMI approach had superior accuracy to the LoGMI method, and this accuracy was independent of the number of available outcome data contemporaneous to the simulated missing GOSE score. The KOMI method avoids imputation of non-random attrition or missing data and returns valid missing at random (MAR) values.