Table 3.
Predictor variables in ASCVD models
| Rank | ANN-based ASCVD prediction model | One minus AUC after permutations |
|---|---|---|
| 1 | Age | 0.6458121 |
| 2 | SBP | 0.7290980 |
| 3 | V2.R Area | 0.7464859 |
| 4 | V2.Max R Amplitude | 0.7522701 |
| 5 | I.T Area (Full) | 0.7565553 |
| 6 | V2.S Area | 0.7570991 |
| 7 | V4.Max S Amplitude | 0.7575463 |
| 8 | V3.QRS Area | 0.7575922 |
| 9 | CR | 0.7578144 |
| 10 | I.T Duration | 0.7582132 |
| 11 | V6.T Area (Full) | 0.7585335 |
| 12 | V6.T Area | 0.7590182 |
| 13 | eGFR | 0.7591497 |
| 14 | I.T Peak Amplitude | 0.7594988 |
| 15 | GLU | 0.7600205 |
| 16 | V2.Max S Amplitude | 0.7602828 |
| 17 | V3.Max S Amplitude | 0.7606117 |
| 18 | V2.QRS Area | 0.7607344 |
| 19 | V6.Max R Amplitude | 0.7615041 |
| 20 | Peak E Wave Velocity | 0.7618305 |
| 21 | WBC | 0.7618512 |
| 22 | UA | 0.7618903 |
| 23 | I.P Area (Full) | 0.7619013 |
| 24 | aVR.T Area | 0.7619031 |
| 25 | DBP | 0.7619409 |
| 26 | V1.QRS Area | 0.7619643 |
| 27 | V3.S Area | 0.7620318 |
| 28 | I.T Area | 0.7622241 |
| 29 | V2.T Duration | 0.7623373 |
| 30 | V6.QRS Area | 0.7628234 |
The importance of each feature was quantified using the permutation feature importance method which measures the importance of a feature by calculating the decrease in the model’s performance (area under the ROC curve) after permuting its values. The higher their values, the more important the feature is. Features are sorted according to permutation importance
Abbreviations: ANN Artificial Neural Network, SBP Systolic Blood Pressure, CR creatinine, eGFR Estimated Glomerular Filtration Rate, GLU glucose, WBC White Blood Cell, UA Uric Acid, DBP Diastolic Blood Pressure