Table 4.
Results of hyperglycemia prediction (Mean ± SD) evaluated on 12 clinical T1D subjects.
| Prediction horizons | 15 min | 30 min | 45 min | 60 min | 60 min (Baseline45) |
|---|---|---|---|---|---|
| Accuracy (%) | 96.75 ± 0.99 | 93.22 ± 1.24 | 90.06 ± 1.05 | 87.20 ± 1.95 | 85.54 ± 3.13 |
| Sensitivity (%) | 95.32 ± 2.16 | 91.25 ± 4.75 | 88.48 ± 7.87 | 86.62 ± 7.81 | 91.58 ± 3.52 |
| Specificity (%) | 96.95 ± 1.14 | 92.62 ± 2.89 | 87.61 ± 5.10 | 82.59 ± 7.96 | 77.21 ± 4.74* |
| Precision (%) | 94.62 ± 2.37 | 90.51 ± 3.20 | 87.43 ± 4.76 | 85.11 ± 5.83 | 79.68 ± 12.49* |
| MCC score | 0.92 ± 0.02 | 0.84 ± 0.02 | 0.77 ± 0.03 | 0.70 ± 0.05 | 0.68 ± 0.06 |
| MD (mg/dL) | 13.00 ± 2.30 | 28.69 ± 5.16 | 40.05 ± 8.04 | 47.62 ± 10.33 | 61.42 ± 16.26** |
MCC Matthews correlation coefficient. MD mean deviation from true glucose levels for missed predicted hyperglycemic events. The best performance of the baseline methods (PKM45) is presented.
The significance is indicated as *p < 0.05, **p < 0.01.