Table 3.
Accuracy, sensitivity, and specificity for bidirectional GRU RNN models in the 4-visit scenario for all 10 complications of DM2
| Complication | Accuracy | Sensitivity | Specificity |
|---|---|---|---|
| Angina pectoris | 0.796 ± 0.024 | 0.862±0.019 | 0.698±0.014 |
| Atherosclerosis | 0.756 ± 0.003 | 0.791±0.012 | 0.718±0.014 |
| ICHD | 0.835 ± 0.005 | 0.886±0.014 | 0.787±0.012 |
| Depressive disorder | 0.820 ± 0.005 | 0.848±0.009 | 0.792±0.010 |
| Hearing impairment | 0.734 ± 0.017 | 0.743±0.016 | 0.722±0.012 |
| MI | 0.733 ± 0.011 | 0.806±0.021 | 0.652±0.012 |
| Nephropathy | 0.768 ± 0.012 | 0.826±0.017 | 0.654±0.021 |
| Neuropathy | 0.746 ± 0.053 | 0.795±0.041 | 0.701±0.049 |
| PVD | 0.767 ± 0.002 | 0.774±0.005 | 0.753±0.011 |
| Retinopathy | 0.796 ± 0.014 | 0.799±0.007 | 0.792±0.018 |
Values are mean ± SD.
DM2: type 2 diabetes mellitus; GRU: gated recurrent unit; ICHD: ischemic chronic heart disease; MI: myocardial infarction; PVD: peripheral vascular disease; RNN: recurrent neural network.