Background: In France, as in many countries, road casualty statistics are mostly based on police reports. It is generally recognized that these data are incomplete but no measurement has been made of the degree of under-reporting and thus of associated biases. This study aims to demonstrate and quantify these biases.
Methods: The study compares, after data linkage, the 10,202 people reported injured or killed in 1996 in the medical road accident victims Register in the Département du Rhône (France), with the 4,572 victims reported by the police during the same year and in the same area. This Département was chosen, as it is the only region in France where these two independent data sources coexist. Two types of possible biases are studied: injury severity classification bias and selection bias induced by underreporting.
Results: The study shows that the definition of "serious injury" used by the police exaggerates the severity of the victim's condition in over half the cases. This bias depends on road user group. This bias is maximum for pedestrians: compared to a slightly injured car occupant, a pedestrian with the same injury severity level has significantly more chance to be considered as severely injured (RR=1.78; 95% CI: 1.11-2.87). Conversely, significant selection biases are related to data collection by the police. The multivariate analysis shows that the underreporting of victims increases if no third party is involved (i.e. without any other vehicle or pedestrian), and reduces with injury severity. It also varies by road user group (with the largest underreporting for cyclists). Among the most seriously injured in accidents involving third parties, motor cyclists and car users are the most reported category and pedestrians the least (RR=0.80; 95% CI: 0.70-0.92). Biases in Register selection are much more limited and basically concern underreporting of victims of minor accidents who did not require medical care.
Conclusions: This study confirms and quantifies misleading distortions in police statistics used to assess road accidents. These results concern the relevant indicators to be used to define road safety issues.