Introduction: Police crash data, which are the basis for safety research in most countries, are incomplete and biased. We focus here on the extent of under-reporting in France, and how it is related to casualty and crash characteristics.
Methods: Police data are compared with a road trauma registry, on the Rhône county, after record-linkage. The study covers the 1997-2001 period, totalling 59,714 casualties. A multivariate analysis is conducted, modelling the probability of being police-reported among casualties in the registry, as a function of crash and casualty characteristics. Results are expressed as relative risks (RR) and adjusted probabilities.
Results: Police reporting rate is 37.7%. Under-reporting varies mainly according to injury severity (RR=0.35 for slightly injured versus severely injured), to road user type and third party involvement (yes/no): comparing casualties with no third party versus those with one, RR=0.32 among motorcyclists whereas RR=0.78 among car occupants. Under-reporting also varies with road type, road environment (metropole/rural) and type of police force, all of which are structurally dependent.
Conclusion: Any study based on police crash data may be quite misleading. We are therefore working on obtaining unbiased estimates of road casualties figures, by extrapolating the Rhône road trauma registry to the nation-wide level.