Utilizing dynamic treatment information for MACE prediction of acute coronary syndrome

Figures

• 

  Fig. 1. Utilizing EHR data to support MACE prediction during ACS patients’ hospitalization

• 

  Fig. 2. Representing dynamic features found on each hospitalization day as a specific vector

• 

  Fig. 3. The framework of bi-directional LSTM used in this proposed MACE prediction model

• 

  Fig. 4. Number and frequency of top 30 treatment interventions in the collected EHR dataset

• 

  Fig. 5. Impact of the different length of stay on the performance of MACE prediction regarding (a) AUC and (b) Accuracy

• 

  Fig. 6. Impact of the different ratio of training data for MACE prediction, in terms of (a) AUC, and (b) Accuracy