Abstract
Background: Data quality of the Japan Trauma Data Bank (JTDB) has not yet been assessed.
Aim: To assess the data quality of the JTDB by using proportions of missing data.
Material and methods: We used all cases registered in the JTDB between 2004 and 2008 (n=29,562). We investigated the proportion of missing data of the variables required to derive a predictive model for trauma patient outcome, including age, Injury Severity Score (ISS), Revised Trauma Score (RTS) at hospital admission, type of injury (i.e. blunt), mode of admission (i.e. direct admission, referred) and outcome at discharge. We also analysed risk factors associated with missing outcome by logistic regression analysis. We further explored the association between the number of case registrations and the proportion of missing outcomes, and compared age, ISS, RTS, mode of admission and type of injury of outcome-missing cohort to those of outcome-non-missing cohort.
Results: We found 12,484 (42.2%) cases that had at least one missing data in the selected variables. Outcome was the most frequently missed item (28.2%). The risk of missing outcome was high in case of cardiopulmonary arrest on arrival, missing ISS, missing mode of admission, admission in November or December and admission to the hospital that registered 100 or less cases to the JTDB. There was a significant negative linear relationship between the number of case registrations and proportion of missing outcomes. We identified statistically significant difference in all compared variables between outcome-missing and outcome-non-missing cohorts.
Discussion: We found that the proportion of missing outcome of the JTDB was higher than that of the National Trauma Data Bank (0.5%). By providing the participating hospitals in the JTDB with the known risk factors and the proportion of missing data of each hospital, the number of complete data may increase. Researchers should be aware of existing selection bias in research outputs gained from the extracted data from the JTDB by excluding cases with missing outcome.
Conclusion: The proportion of missing data, especially missing outcome, should be reduced to improve data quality of the JTDB.
