RESUMO
OBJECTIVES: To build a detection algorithm of non-accidental pediatric burns (NAB) using hospital resumes from the French Hospital Discharge Database (HDD) and to describe cases with no judicial or administrative report. MATERIALS AND METHODS: Children aged 0-16 years old hospitalized at the University Hospital of Tours from 2012 to 2017 with a coded burn were included. "Probable" or "possible" HDD cases of NAB were defined based on the International Classification of Diseases 10th version codes during the inclusion stay or the previous year. A chart review was performed on all the HDD cases and HDD non cases matched on sex and age with a 1:2 ratio. Performance parameters were estimated for three clinical definitions of child maltreatment: excluding neglect, including neglect in a restrictive definition, and in a broad definition. For clinical cases, report to the judicial or administrative authorities was searched. RESULTS: Among the 253 included children, 83 "probable" cases and 153 non-cases were analyzed. Sensitivity varied from 48 (95%CI [36-60], excluding neglect) to 90% [55-100] and specificity from 70 [63;77] to 68% [61;74]. The proportion of clinical cases with no report without justification varied from 0 (excluding neglect) to > 85% (with the broadest definition); all corresponded to possible isolated neglect. CONCLUSION: The performances of the algorithm varied tremendously according to the clinical definition of child maltreatment. Neglect is obviously complex and tough to clinically detect. Training for healthcare professionals and qualitative studies on obstacles to report should be added to this work.