Figure 2
From: Predicting outcomes of acute kidney injury in critically ill patients using machine learning
Study workflow for mortality prediction task (survival analysis). (a) In the first scenario, we utilized a population of patients from the observational follow-up data (Ldata) to train ML and statistical models to predict mortality in patients who developed AKI stage 3 in the ICU. In this scenario, the censoring rate is 57.42%. Prediction performance was assessed using C-index and has been tested on an external test set for each model separately. (b) In the second scenario, we utilized a population of patients from the observational follow-up data plus the unlabeled data (Udata) to have a bigger training set and train ML and statistical models to predict mortality in patients who developed AKI stage 3 in the ICU. In this scenario, the censoring rate is 80.8%. Prediction performance was assessed using C-index and has been tested on an external test set for each model separately.
