The appropriate use of emergency departments: evidence from a French survey

This paper analyses the determinants of the appropriateness of emergency room visits for adult patients (18+), distinguishing between individual patient characteristics and primary care supply characteristics. There is evidence that emergency care can substitute to some extent to primary care for reasons that have nothing to do with the urgency of care needs. This can lead to avoidable emergency visits that raises concerns about the efficiency of care provision. We take advantage of a unique French survey called "Enquête urgence", implemented in 2013 by the French Ministry of Health, where the characteristics of all ED visits in French hospitals have been exhaustively recorded during one day (24h). While no administrative data collected at the hospital level or from emergency departments (ED) provide direct observation of avoidable ED visits, this survey collected for each visits three direct assessments of appropriateness completed by a physician, of which one is collected ex ante (before the beginning of medical examinations) and two ex post. We first compare our direct assessments of visit appropriateness with classical indicators used in studies on administrative data. We find that 52% of emergency room visits are deemed necessary ex post by the physician, 34% are deemed divertible (conditions that would have been treated more efficiently by a primary care physician the same day) and 15% are judged delayable (there was no emergency and thus the ED visit could have been avoided). Then we examine the determinants of avoidable visits by OLS and multinomial logit (MNL) estimates. We show that (i) financial barriers increase the risk of avoidable ED visits; (ii) accessibility of primary care services is positively associated with appropriateness. Finally we are able to assess the ED visits that were exposed to a "type one error" - i.e. visits that are deemed avoidable at entry but which prove to be appropriate after medical examinations. We find that a non-negligible number of patients, namely 1041 patients (4.27% of the sample) were at risk of a type one error and that this risk is not randomly assigned among patients.