Fig. 3From: Real-time machine learning model to predict short-term mortality in critically ill patients: development and international validationThe average Shapley values in the internal testing dataset over time. The Shapley values are first averaged by hours for each admission and then averaged across all admissions. The black line represents the average prediction score. The average Shapley values are added to the prediction score in the figure. A For event admissions, we collected the values 24 h before death. B In the case of normal admissions, we gathered the values 24 h before random points to mitigate selective biasBack to article page