Enhancing evidence-based medicine with natural language argumentative analysis of clinical trials

In the latest years, healthcare domain has seen an increasing interest in definition of intelligent systems to support clinicians their everyday tasks and activities. Among others, also field Evidence-Based Medicine is impacted by this twist, with aim combine reasoning frameworks proposed thus far mining algorithms extract structured information from clinical trials, guidelines, Electronic Health Records. paper, we go beyond state art proposing a new end-to-end pipeline address argumentative outcome analysis on trials. More precisely, our composed (i) Argument Mining module classify components (i.e., evidence claims trial) relations support, attack), (ii) identify effects improved, increased, decreased, no difference, occurrence) intervention trial, based PICO elements. We annotated dataset more than 500 abstracts Randomized Controlled Trials (RCT) MEDLINE database, leading labeled 4198 argument components, 2601 relations, 3351 outcomes five different diseases neoplasm, glaucoma, hepatitis, diabetes, hypertension). experiment deep bidirectional transformers combination neural architectures LSTM, GRU CRF) obtain macro F1-score of.87 for component detection and.68 relation prediction, outperforming current state-of-the-art systems, of.80 classification.