2012 PhysioNet Challenge: An artificial neural network to predict mortality in ICU patients and application of solar physics analysis methods

Advances in technology are improving the quality and quantity of data available in ICU, creating opportunities for the development of patient-specific predictive models to support clinical decision-making. The 2012 PhysioNet Computing in Cardiology Challenge is to develop a patient-specific model for predicting in-hospital mortality using data collected during the first 48 hours of an ICU stay. Our approach was to develop an algorithm incorporating an artificial neural network trained on features extracted from the patient data. We explored the stability of vital signs such as heart rate and blood pressure with a method previously used to detect frequency and intensity of solar `nanoflares´. The ability of the resulting model to predict outcomes of patients was evaluated. The model was most successful in Event 2 of the Challenge, receiving a score of 22.83 for Set B and 38.23 for Set C. For the model to be clinically useful and to improve on existing scoring systems such as SAPS, further work is needed.