Fig. 1: Flow diagrams of the validation stages for a wearable-based alert system and processing and modeling steps. | Nature Communications

Fig. 1: Flow diagrams of the validation stages for a wearable-based alert system and processing and modeling steps.

From: Development and validation of a clinical wearable deep learning based continuous inhospital deterioration prediction model

Fig. 1: Flow diagrams of the validation stages for a wearable-based alert system and processing and modeling steps.

(Diagram A) The three stages of clinical validation for a wearable-based alert system (Yellow Panel) Wearable data accuracy validation ensures the data is of sufficient accuracy to use for clinical decision making and this to end we align (1st step) and visualize differences (2nd step) using Bland-Altman plots, that visualize the difference between the two modalities (y-axis) vs their mean (x-axis). (Green Panel) Comparison of clinical alerts across measurement modalities demonstrates whether a wearable device provides an advantage over episodic monitoring, where we compare the numbers and timing of alerts for each modality and calculated the sources of these differences. (Blue Panel) Wearable clinical alert modeling and validation showing the feature and label construction and performance of this algorithm. (Diagram B) Feature and label selection, artifact rejection, signal processing and model training and testing steps followed to build a wearable device based clinical alert system. Source data are provided as a Source Data file.

Back to article page