Fig. 9: Visual example of dynamic imputation and feature impact on mortality risk prediction. | npj Digital Medicine

Fig. 9: Visual example of dynamic imputation and feature impact on mortality risk prediction.

From: Unlocking the potential of real-time ICU mortality prediction: redefining risk assessment with continuous data recovery

Fig. 9

a Visualization of the 10 most important features for predicting patient mortality risk at each time point where at least one feature has a new measurement. Solid symbols represent new measurements, while hollow symbols indicate missing values imputed by the RealMIP. The red triangle marks the model’s predicted probability of mortality. The left axis shows the Z-score of each feature, and the right axis displays the predicted mortality probability. b Contribution of each feature to the risk prediction within a specific time point. Features in red increase the predicted risk of mortality (non-survival), while features in blue decrease it (survival).

Back to article page