Mortality prediction in ICU: a methodological advance

In The Lancet Respiratory Medicine, Pirracchio and colleagues present a new approach to predicting mortality in intensive care units (ICUs). The investigators propose that instead of picking one of the many mortality prediction models available, an ensemble machine learning approach can be used (the non-parametric Super Learner) to leverage the individual candidate models from a pre-specified library, to produce an optimum prediction algorithm. This is an elegant idea that frees the user from making an arbitrary choice of model, and that also guarantees at least as good performance as any individual model within that library.