Ulster University Logo

Ensemble decision trees to predict if a profiled bystander can use an AED to deliver a shock

Bond, Raymond, O’Hare, Peter, Torney, Hannah, Davis, Laura, Delafont, Bruno, McReynolds, Hannah, McLister, Anna, McCartney, Ben, Di Maio, Rebecca, Finlay, Dewar, Guldenring, Daniel, McLaughlin, James and McEneaney, David (2015) Ensemble decision trees to predict if a profiled bystander can use an AED to deliver a shock. In: 7th Annual Translational Medicine Conference, Derry/Londonderry. CTRIC. 1 pp. [Conference contribution]

Full text not available from this repository.

URL: http://www.c-tric.com/tmed7/programmespeakers


BackgroundEach year cardiac arrest kills 60,000 people in the UK [1-2]. The use of an automatic external defibrillator (AED) in the first few minutes can increase the probability of survival from less than 5% to over 75%. However, there is a challenge to build AEDs that are ‘usable’ to all members of the public. This study investigated if we could predict whether a profiled bystander is likely to succeed in delivering a shock. MethodsA total of 362 subjects were recruited at a shopping mall and were asked to use an AED in a simulated emergency situation as facilitated by a ‘sensorised’ mannequin. During this we extracted a range of features (i.e. Age, Gender, Education, CPR/AED Training, Time-to-Place-Electrodes, Time-to-First-Shock). Using logistic regression, Odds ratios (ORs) were analysed to identify variables that decrease/increase the likelihood of delivering a shock. The dataset was also split into training (70%) and testing datasets (30%) to build and evaluate an ensemble C5.0 decision tree to predict if a person is likely to deliver a successful shock. ResultsThe variables (1) Time-to-First-Shock [OR= 8.97], (2) Prior Defibrillation Training [OR= 8.37], (3) Prior CPR Training [OR= 6.76], High school education [OR= 5.71]) have the largest ORs. This may indicate that users with these characteristics are likely to deliver a shock. However, no OR had statistical significance (p<0.05) indicating that successful delivery may not be easily predicted. Nevertheless, the decision tree (trials=10) had a 96.33% (CI =90.87, 98.99%) accuracy rate in predicting shock delivery success. Sensitivity was high (100%) but specificity was low (50%) and the accuracy did not have statistical significance in comparison to the No-Information Rate (p=0.09).ConclusionIt maybe difficult to predict if a profiled bystander is ‘unlikely’ to deliver a successful shock. This can indicate that the AED was somewhat intuitive independent of the user’s social/economic demography.

Item Type:Conference contribution (Poster)
Keywords:machine learning, health informatics, human computer interaction
Faculties and Schools:Faculty of Computing & Engineering
Faculty of Computing & Engineering > School of Computing and Mathematics
Faculty of Computing & Engineering > School of Engineering
Research Institutes and Groups:Engineering Research Institute
Engineering Research Institute > Nanotechnology & Integrated BioEngineering Centre (NIBEC)
Computer Science Research Institute > Smart Environments
Computer Science Research Institute
ID Code:32537
Deposited By: Dr Raymond Bond
Deposited On:03 Nov 2015 12:30
Last Modified:03 Nov 2015 12:30

Repository Staff Only: item control page