Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.

© 2019 Creative Commons. Predicting the onset of sepsis from clinical data is challenging, as physiological and laboratory measurements are sampled at different frequencies and missing data are not randomly distributed. Our team ("CRASHers") propose a two-model approach, where the first predicts a probability of sepsis and the second estimates the uncertainty of these predictions. We then optimize a "decision rule" using both the probability and uncertainty to make the final prediction. A set of derived features was used to train a Gradient Boosting Machine (GBM) classification model to predict sepsis (within 6 hours). A second GBM regression model was trained to estimate the uncertainty of those predictions using a different set of derived features. Optimal hyperparameters for both models were determined using Bayesian optimisation with 5-fold cross validation (using 70% records from each training set). The outputs from both models were then combined using logistic regression (using 15% of records available) to re-calibrate the probability of sepsis. Due to an error in setting up the test environment for our entries, we did not obtain a valid score in the hidden test set. The combined model was evaluated on the remaining 15% of records available for training (i.e., our validation set). Our uncertainty-aware approach achieved a Utility score of 0.412 on our validation set.

Original publication




Conference paper

Publication Date