Selection bias found in interpreting analyses with missing data for the prehospital index for trauma

Abstract

Objective To evaluate the effects of missing data on analyses of data from trauma databases, and to verify whether commonly used techniques for handling missing data work well in theses settings. Study design and setting Measures of trauma severity such as the Pre-Hospital Index (PHI) are used for triage and the evaluation of trauma care. As conditions of trauma patients can rapidly change over time, estimating the change in PHI from the arrival at the emergency room to hospital admission is important. We used both simulated and real data to investigate the estimation of PHI data when some data are missing. Techniques compared include complete case analysis, single imputation, and multiple imputation. Results It is well known that complete case analyses and single imputation methods often lead to highly misleading results that can be corrected by multiple imputation, an increasingly popular method for missing data situations. In practice, unverifiable assumptions may not hold, meaning that it may not be possible to draw definitive conclusions from any of the methods. Conclusion Great care is required whenever missing data arises. This is especially true in trauma databases, which often have much missing data and where the data may not missing at random.