A comparison of data sources for motor vehicle crash characteristic accuracy

Objective: To determine the accuracy of police reports (PRs), ambulance rep orts (ARs), and emergency department records (EDRs) in describing motor veh icle crash (MVC) characteristics when compared with an investigation perfor med by an experienced crash investigator trained in impact biomechanics. Me thods: This was a cross-sectional, observational study. Ninety-one patients transported by ambulance to a university emergency department (ED) directl y from the scene of an MVC from August 1997 to April 1998 were enrolled. Po tential patients were identified from the ED log and consent was obtained t o investigate the crash vehicle. Data describing MVC characteristics were a bstracted from the PR, AR, and medical record. Variables of interest includ ed restraint use (RU), air bag deployment (AD), and type of impact (TI). Ag reements between the variables and the independent crash investigation were compared using kappa. Interrater reliability was determined using kappa by comparing a random sample of 20 abstracted reports for each data source wi th the originally abstracted data. Results: Agreement using kappa between t he crash investigation and each data source was 0.588 (95% CI = 0.508 to 0. 667) for the PR, 0.330 (95% CI = 0.252 to 0.807) for the AR, and 0.492 (95% CI = 0.413 to 0.572) for the EDR. Variable agreement was 0.239 (95% CI = 0 .164 to 0.314) for RU, 0.350 (95% CI = 0.268 to 0.432) for AD, and 0.631 (9 5% = 0.563 to 0.698) for TI. Interrater reliability was excellent; (kappa > 0.8) for all data sources. Conclusions: The strength of the agreement betw een the independent crash investigation and the data sources that were meas ured by kappa was fair to moderate, indicating inaccuracies. This presents ramifications for researchers and necessitates consideration of the validit y and accuracy of crash characteristics contained in these data sources.