Which patients have missing data? An analysis of missingness in a trauma registry

BackgroundTrauma registry data are almost always incomplete. Multiple imputation can reduce bias in registry analyses but the ideal approach would be to improve data capture. The aim of this study was to identify, using multiple imputation, which type of patients were most likely to have incomplete data.MethodsAn analysis of prospectively collected regional trauma registry data over one year was performed. Analyses were conducted following complete data estimation using multiple imputation. Variables necessary for TRISS analysis and with incomplete data were analysed. For each variable, logistic regression analyses were performed to identify predictors of missingness. A p-value of less than 0.05 was considered to be statistically significant.ResultsThere were 2520 cases. The variables with the greatest proportion of missing observations were respiratory rate, GCS, Qualifier (of GCS and respiratory rate) and systolic blood pressure. The Qualifier variable described whether or not the patient was intubated and mechanically ventilated at the time the first hospital GCS and respiratory rate were recorded. GCS and respiratory rate were more likely to be missing (imputed) when abnormal (unadjusted ORs: 8.6 (p < 0.001) and 2.1 (p = 0.02), respectively). The most important determinant of a valid GCS or respiratory rate was the Qualifier. There was no association between whether the systolic blood pressure and Qualifier were missing (imputed) and whether they were estimated to be abnormal. Following multivariable analysis, data for all four variables were more likely to be missing when the patient died in hospital. Additional independent predictors of a missing GCS or respiratory rate were an abnormal pre-hospital GCS and severe chest injury. The Qualifier and systolic blood pressure were more likely to be missing where the patient was transferred from the primary hospital.ConclusionThe major independent predictor of missing primary hospital physiological variables was death in hospital. An abnormal GCS was more likely to be missing from the regional trauma registry dataset. Predictors of a missing GCS or respiratory rate included whether the patient was intubated, an abnormal pre-hospital GCS and severe chest injury. Augmenting resources to record the initial observations of the more severely injured patients would improve data quality. Multiple imputation can be used to inform data capture.