23 Machine learning model surpassing medical dispatchers recognition of out-of-hospital cardiac arrest. (16th April 2018)

• Record Type: Journal Article
• Authors: Blomberg, SN; Folke, F; Ersbøll, AK; Lippert, FK
• Is Part Of: BMJ open. Volume 8:Supplement 1(2018)
• Page Start: A9
• Page End: A9
• Publication Date: 2018-04-16
• Subjects: Medicine -- Research -- Periodicals; 610.72
• DOI: 10.1136/bmjopen-2018-EMS.23
• Languages: English
• ISSNs: 2044-6055

Abstract

Abstract : Aim: The chance of surviving Out-of-Hospital Cardiac Arrest (OHCA) is highly correlated with medical dispatchers' recognition of the condition during emergency calls. 1, 2 We investigated if machine learning (ML) could surpass medical dispatchers by applying ML directly on the realtime dialogue between the caller and the dispatcher. Method: We retrieved all 1 61 650 emergency calls recorded for 2014 to the Emergency Medical Dispatch Centre Copenhagen (EMDC). From the Danish Cardiac Arrest Register (DCAR) and audit of callrecordings where resuscitative efforts were begun, we extracted information on all OHCA in 2014. Emergency medical services (EMS) witnessed cardiac arrests and damaged audiofiles/recordings were excluded. The data-set for analysis contained henceforth 1 58 330 non-OHCA calls and 2, 157 OHCA calls. Each OHCA could span several calls from multiple callers. Time-to-recognition by the ML-Model was calculated for each call and for the dispatchers found by auditing all OHCA calls. Results: The ML-Model reached a sensitivity of 93.1% (95% CI: 91.9 to 94.1) and a specificity of 98.0%. (95% CI: 97.9 to 98.1) on OHCA-calls. Sensitivity of the dispatchers was 72.9% (95% CI: 70.0 to 75.6). Specificity is unknown for dispatchers, as false positives are not registered. Time-to-recognition was significantly shorter for the ML-model (mean time-to-recognition 00:48 mm:ss, 95% CI: 00:46 to 00:50) compared to dispatchers (mean time-to-recognition 01:19 mm:ss, 95% CI: 01:13 to 01:25) (p<0.0001). Conclusion: In recordings of 1 61 650 calls to EMDC a ML-model could recognise a higher proportion of OHCA, compared to medical dispatchers. Furthermore, we found that the ML-model was significantly faster in recognising OHCA. References: . Viereck S, Møller TP, Ersbøll AK, Bækgaard JS, Claesson A, Hollenberg J, et al. Recognising out-of-hospital cardiac arrest during emergency calls increases bystander cardiopulmonary resuscitation and survival. Resuscitatio 2017;115:141–7. . Wissenberg M, Lippert FK, Folke F, Weeke P, Hansen CM, Christensen EF, et al. Association of national initiatives to improve cardiac arrest management with rates of bystander intervention and patient survival after out-of-hospital cardiac arrest. JAMA2013;310(13):1377–84. Conflict of interest: None Funding: Unrestricted research grant from TrygFoundation.