[Analysis of off-label prescriptions of medicines in hospital in adult patients and feasibility study of their detection using CIM-10 coding]. / Analyse des prescriptions hospitalières hors AMM des médicaments chez l'adulte et étude de la faisabilité de leur détection par le codage CIM-10 du PMSI.

ABSTRACT

Analysis of off-label prescriptions of medicines in hospital in adult patients and study of feasibility of their detection by use of international disease classification, 10th version (IDC-10 codes). CONTEXT In order to improve the appropriate use of medicines, a method of detection of off label prescriptions, especially in hospitalised patients, should be available. STUDY OBJECTIVES: Evaluate the performance of the detection of off-label prescriptions in hospitalised patients by use of IDC-10 codes. METHODS: Data prescriptions (excluding those directly taken in charge by the national health care system), clinical history and biological results were extracted from Assistance publique des Hôpitaux de Paris (AP-HP) data-warehouse for 108 in-hospital adults patients' journeys. An adjudication committee established the classification reference for the appropriate or off label drug prescriptions status after analysis of medical information for each patient. IDC-10 codification that is performed after every hospitalisation was crossed with those IDC-10 codes that were to be expected corresponding to the marketing authorisation labelling (section 4.1 of specifications of product characteristics [SPC]). Results of IDC-10 coding were compared to the reference for off label use identification. RESULTS: Out of the 1131 analysed prescriptions, 44 (3.9%) were classified as off label by the adjudication committee. Sensitivity of detection by IDC-10 coding was 87 (95% CI [0.73-0.96]) to 92% (95% CI [0.79-0.98]) and specificity 25 (95% CI [0.22-0.27]) to 41% (95% CI [0.38-0.44]) according to the number of characters of ICD-10 that could be used. CONCLUSIONS: Incidence of in-hospital off label use of drugs (restricted to within drug related groups prescriptions) appeared relatively low (3.9%). Its semi-automatic detection by IDC-10 coding appears feasible with a good sensitivity but a low specificity. Such method could be further assessed as a first step detection focusing on one pharmacological class or on one pathologic condition.