Emmanuelle Amoros1, Jean-Louis Martin, Mireille Chiron, Bernard Laumon. 1. Transport, Occupational and Environmental Epidemiology Research and Surveillance Unit (UMRESTTE UMR-T 9405), joint unit of the French National Institute for Transport and Safety Research (INRETS), Bron, F-69500, France. amoros@inrets.fr
Abstract
BACKGROUND: In most countries, epidemiologic knowledge of road crash injury is mainly based on police data, as they very often are the only available data at the nation-wide level. However their validity is of some concern. We focus here on the police severity classification of 'serious' and 'slight' casualties in France. We want to know how the police classification compares with a trauma severity scale, so that we could correctly interpret police based studies. METHOD: The study is based on the Rhône county (population 1.6 million) during the 1997 to 2001 period. Police data have been linked with a road trauma registry, so that both police and New Injury Severity Score (NISS) classifications are available on 14,342 casualties. The police classification of 'slight' and 'serious' casualties is compared with the registry classification grouped into NISS 1-15 and NISS 16-75 categories. We conduct multivariate analyses of the probability of police severity misclassification, over and under-classification, as a function of crash and casualty characteristics. RESULTS: Kappa is estimated at 0.41; the sensitivity of the police classification is 72% and the positive predictive value is 35%. Pedestrian and motorcyclist casualties are the most likely to be over-classified (Relative Risk [RR] = 1.4 and RR = 1.2, respectively compared with car occupants). The 'rural police' are more likely to over-classify than the other police forces (RR = 3.1). Over-classification decreased during the 1997 to 2001 period whereas under- classification increased. CONCLUSION: These misclassification characteristics must be kept in mind when interpreting severity results based on police data. We are working on obtaining unbiased nation-wide estimates of severity figures.
BACKGROUND: In most countries, epidemiologic knowledge of road crash injury is mainly based on police data, as they very often are the only available data at the nation-wide level. However their validity is of some concern. We focus here on the police severity classification of 'serious' and 'slight' casualties in France. We want to know how the police classification compares with a trauma severity scale, so that we could correctly interpret police based studies. METHOD: The study is based on the Rhône county (population 1.6 million) during the 1997 to 2001 period. Police data have been linked with a road trauma registry, so that both police and New Injury Severity Score (NISS) classifications are available on 14,342 casualties. The police classification of 'slight' and 'serious' casualties is compared with the registry classification grouped into NISS 1-15 and NISS 16-75 categories. We conduct multivariate analyses of the probability of police severity misclassification, over and under-classification, as a function of crash and casualty characteristics. RESULTS: Kappa is estimated at 0.41; the sensitivity of the police classification is 72% and the positive predictive value is 35%. Pedestrian and motorcyclist casualties are the most likely to be over-classified (Relative Risk [RR] = 1.4 and RR = 1.2, respectively compared with car occupants). The 'rural police' are more likely to over-classify than the other police forces (RR = 3.1). Over-classification decreased during the 1997 to 2001 period whereas under- classification increased. CONCLUSION: These misclassification characteristics must be kept in mind when interpreting severity results based on police data. We are working on obtaining unbiased nation-wide estimates of severity figures.
Authors: Nicolas Cheynel; Julie Gentil; Marc Freitz; Patrick Rat; Pablo Ortega Deballon; C Bonithon Kopp Journal: World J Surg Date: 2011-07 Impact factor: 3.352
Authors: Jeffrey R Brubacher; Herbert Chan; Penelope Brasher; Shannon Erdelyi; Edi Desapriya; Mark Asbridge; Roy Purssell; Scott Macdonald; Nadine Schuurman; Ian Pike Journal: Am J Public Health Date: 2014-08-14 Impact factor: 9.308