Objective: To determine the accuracy of ambulance (AR), emergency department (EDR) and police (PR) records in describing motor vehicle crash (MVC) characteristics when compared to a standard - an in-depth motor vehicle crash investigation record (CIR).

Methods: Fifty-six MVC patients transported to a suburban, university hospital emergency department via ambulance and included in a crash investigation were identified. The time period was January 1993 through December 1995. Data sources were the AR, EDR, PR and CIR. The CIR was abstracted to a standard form. The other data sources were abstracted to a standard form using a retrospective, blinded review. Variables included occupant position, restraint use, air bag deployment, type of impact, ejection, and external cause of injury code. Accuracy was measured by determining percent agreement and Kappa for each data source compared to the CIR.

Results: Forty-six cases (82%) had one or more episodes of discordance. The mean percent agreement for the sources studied was AR 0.813, EDR 0.893 and PR 0.932. Mean values for Kappa were AR 0.712, EDR 0.831 and PR 0.885. Among variables, restraint use was determined with the least accuracy with means for percent agreement and Kappa of 0.792 and 0.674, respectively. For the AR, discordance was due to missing information almost 50% of the time.