Table 2 Performance of different classification models.represent averages over 1,000 trials.

From: Machine Learning Improves Risk Stratification After Acute Coronary Syndrome

Model

AUC

95% CI

LRHx

0.695

0.581–0.809

LRST

0.701*

0.587–0.814

LRHx+ST

0.734**

0.623–0.845

LRHx+MV

0.727

0.615–0.839

LRHx+HRV

0.720

0.607–0.832

LRHx+DC

0.705

0.591–0.818

TRS

0.670

0.555–0.786

RNN

0.689

0.575–0.803

ANN

0.743***

0.633–0.853

Metric

Mean

95% CI

NRI of ANN w.r.t. TRS

0.065

0.059–0.071

Patients Correctly Reclassified

87

86.4–87.6

  1. AUCs of different models (and the TIMI NSTE-ACS risk score) using the Cohort-1 dataset and 2 Category NRI of the best performing model relative to the TIMI Risk Score (TRS). AUC is area under the curve; CI is confidence interval; MV is morphologic variability, HRV is heart rate variability LF/HF (see Materials and Methods); DC is deceleration capacity; TRS is the TIMI NSTE-ACS Risk Score; RNN is recurrent neural network. AUCs represent averages over 1,000 trials. NRI is two-category net reclassification index. Values *Improvement over LRHx significant with p < 0.05. **Improvement over LRHx, LRST, LRHx+MV, LRHx+HRV, and LRHx+DC significant with p < 0.001. ***Improvement over all other models significant with p < 0.001.
Back to article page