Skip to main content
Intensive Care Medicine Experimental
Fig. 7 | Intensive Care Medicine Experimental

Fig. 7

From: Reinforcement learning for intensive care medicine: actionable clinical insights from novel approaches to reward shaping and off-policy model evaluation

Fig. 7

Comparison of action distributions between physician decisions and the optimal policy, segmented by patient outcome. The upper pair of heatmaps delineates the frequency of actions taken for survivors, contrasting actual physician choices with those suggested by the optimal policy. The lower pair of heatmaps mirrors this analysis for non-survivors. Across both sets, the x-axis categorises the level of PEEP and FiO2, while the y-axis sorts by FiO2 percentage. The NV label stands for non-invasively ventilated. The colour gradient represents the count of actions, with darker shades indicating higher frequencies

Back to article page