Multiple imputation in trauma disparity research

Abstract

Background: Missing data has remained a major disparity in trauma outcomes research due to missing race and insurance data. Multiple imputation (M.IMP) has been recommended as a solution to deal with this major drawback.

Study design: Using the National Data Trauma Bank (NTDB) as an example, a complete dataset was developed by deleting cases with missing data across variables of interest. An incomplete dataset was then created from the complete set using random deletion to simulate the original NTDB, followed by five M.IMP rounds to generate a final imputed dataset. Identical multivariate analyses were performed to investigate the effect of race and insurance on mortality in both datasets.

Results: Missing data proportions for known trauma mortality covariates were as follows: age-4%, gender-0.4%, race-8%, insurance-17%, injury severity score-6%, revised trauma score-20%, and trauma type-3%. The M.IMP dataset results were qualitatively similar to the original dataset.

Conclusion: M.IMP is a feasible tool in NTDB for handling missing race and insurance data.