Deep multitask ensemble classification of emergency medical call incidents combining multimodal data improves emergency medical dispatch

• DOI: 10.1101/2020.06.26.20123216
• Publication Date: 2020-06-27T05:35:50Z
• Authors: Pablo Ferri; Carlos Sáez; Antonio Félix-De ...; Javier Juan-Albar...; Vicent Blanes-Selva; Purificación Sánc...; Juan M García-Gómez
• Keywords: Triage

ABSTRACT Objective To develop a predictive model to aid non-clinical dispatchers classify emergency medical call incidents by their life-threatening level (yes/no), admissible response delay (undelayable, minutes, hours, days) and system jurisdiction (emergency system/primary care) in real time. Materials A total of 1 244 624 independent retrospective from the Valencian dispatch service Spain 2009 2012, comprising clinical features, demographics, circumstantial factors free text dispatcher observations. Methods deep multitask ensemble integrating four subnetworks, composed turn multi-layer perceptron modules, bidirectional long short-term memory units encoding representations transformers module. Results The showed micro F1 score 0.771 classification, 0.592 0.801 jurisdiction, obtaining performance increase 13.2%, 16.4% 4.5%, respectively, with regard current in-house triage protocol service. Discussion captures information present calls not considered existing protocol, but relevant carry out incident classification. Besides, results suggest that most this is Conclusion our knowledge, study presents development first learning undertaking Its adoption centers would potentially improve processes, resulting positive impact patient wellbeing health services sustainability.