Predicting major bleeding events in anticoagulated cancer patients with venous thromboembolism using real-world data and machine learning.

• DOI: 10.1200/jco.2022.40.16_suppl.e18744
• Publication Date: 2022-06-06T16:51:41Z
• Authors: Andrés J. Muñoz M...; María Luisa Palacios; Juan Carlos Souto; Berta Obispo; Jorge Aparicio; Andres Garcia-Palomo; Antonio Sanchez; Cristina Aguayo; David Gutierrez Abad; Maria Carmen Viñu...; Diego Benavent; Miren Taberna; Daniel Arumi; Miguel Ángel Hern...
• Keywords: 03 medical and health sciences; 0302 clinical medicine; 3. Good health

e18744 Background: Evidence regarding the clinical predictors of bleeding risk in patients with cancer and venous thromboembolism (VTE) is lacking. Our aim was to develop a predictive model assess major (MB) anticoagulant-treated active during first 6 months following VTE diagnosis. Methods: Observational, retrospective, multicenter study based on secondary analysis unstructured data electronic health records (EHRs). Using EHRead technology, Natural Language Processing (NLP) machine learning (ML), were collected from EHRs 9 Spanish hospitals between 2014 2018. The population comprised all adult diagnosis under anticoagulant treatment no history MB. This downsampled prevent bias class imbalance. A total 94 patient characteristics explored, Random Forest (RF) feature selection performed identify most relevant predictors. Multiple algorithms used train different prediction models, which subsequently validated hold-out dataset. best performance metrics (i.e., ROC-AUC) selected as final model. Results: Among source 2,893,208 patients, 21,227 identified EHRs. Of these, 53.9% men, median age (Q1, Q3) 70 (59,80) years. duration follow up across 0.7 (0.11, 2.03) During period, estimated in-hospital prevalence cancer-related 5.8 %. common type at baseline deep vein thrombosis (68.2 % patients), followed by pulmonary embolism (28.4%). frequent primary cancers colorectal (10.1%) lung (8.5 %). trained RF approach yielded performance, ROC-AUC = 0.7. MB identified: hemoglobin levels, presence metastasis, patient’s age, platelet count, leukocyte serum creatinine levels. Conclusions: use NLP extract information for anticoagulated VTE. These results may improve prevention management these patients.