Abstract Number: PB1203
Meeting: ISTH 2021 Congress
Background: New oral anticoagulants (DOAC) are indicated in atrial fibrillation (AF), thromboembolic events (VTE), and prevention of VTE after hip or knee prosthesis (PVTE).
Aims: The objective was to differentiate these indications using medico-administrative databases.
Methods: Two sources of data were used, LPD and LRx, including data of near 2.5 and 40 million patients, respectively. LPD, a medicalized database, included 56,665 patients treated by DOAC in 2019 followed-up by 1,800 general practitioners and/or specialists who participating in a permanent longitudinal observatory of prescription in ambulatory medicine. LRx, contained all anonymized medication dispenses prescribed in outpatient care from a representative panel of 45% of all French retail pharmacies. After derivation on LPD, the best gradient boosting model was selected in order to identify AF and/or VTE (accuracy of 91.5% and 90.5% for AF, and VTE, respectively). The model was then implemented in LRx in order to obtain AF and VTE patients in 2019. In order to identify PVTE patients on LRx, rules-based algorithm was defined. We obtained a raw number of DOAC patients and performed demographic characteristics. We calculated the extrapolated number of DOAC patients.
Results: Over the 944,892 DOAC patients identified in LRx, 73% were classified as having AF, 20% VTE, 4% AF/VTE, 3% PVTE, and 0.3% unclassified. The percentage of female and mean age (SF) were 45% and 79 (10), 54% and 65 (15), 57% and 76 (9), 51% and 77 (11), 93% and 88 (9), for AF, VTE, AF/VTE, PVTE, unclassified, respectively. The extrapolated number of DOAC patients in France in 2019 was 1,8 million.
Conclusions: A combined approach using machine learning and rules-based algorithm could be used in order to distinguish the different indications for which DOAC could be prescribed. According to demographics characteristics in France of each medical condition our results correspond to patients seen in clinical practice and literature.
To cite this abstract in AMA style:Emmerich J, Chekroun-Martinot A, Petri C, Sigogne R, Perray L, Maravic M. Machine Learning and Algorithmic Diagnosis Identification of PatientsTreated by Direct Oral Anticoagulants Using Medico-administrative Databases [abstract]. Res Pract Thromb Haemost. 2021; 5 (Suppl 2). https://abstracts.isth.org/abstract/machine-learning-and-algorithmic-diagnosis-identification-of-patientstreated-by-direct-oral-anticoagulants-using-medico-administrative-databases/. Accessed November 27, 2021.
« Back to ISTH 2021 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/machine-learning-and-algorithmic-diagnosis-identification-of-patientstreated-by-direct-oral-anticoagulants-using-medico-administrative-databases/