Table 2.
Results of subgroup analysis of C-index by fracture site and machine learning type
| Subgroup | Training dataset | Validation dataset | ||
| N | C-statistic (95% CI) | N | C-statistic (95% CI) | |
| Fracture site | ||||
| Vertebral fracture | 15 | 0.80 (0.74, 0.87) | 6 | 0.87 (0.71, 1.00) |
| Hip fracture | 20 | 0.76 (0.72, 0.81) | 9 | 0.73 (0.65, 0.81) |
| Multi-site fracture | 31 | 0.70 (0.67, 0.72) | 17 | 0.71 (0.65, 0.76) |
| Model type | ||||
| LR | 26 | 0.75 (0.72, 0.78) | 7 | 0.80 (0.73, 0.87) |
| ANN | 4 | 0.73 (0.64, 0.82) | 3 | 0.66 (0.62, 0.70) |
| CNN | 2 | 0.95 (0.94, 0.96) | 1 | 0.98 (0.94, 1.00) |
| RF | 3 | 0.70 (0.68, 0.72) | 3 | 0.66 (0.59, 0.73) |
| SVM | 5 | 0.72 (0.60, 0.85) | 3 | 0.78 (0.59, 0.96) |
| DT | 2 | 0.78 (0.56, 0.99) | 1 | 0.69 (0.67, 0.70) |
| NB | 2 | 0.74 (0.39, 1.00) | – | |
| kNN | 1 | 0.51 (0.46, 0.55) | – | |
| Survival model | 13 | 0.70 (0.69, 0.74) | 9 | 0.68 (0.67, 0.69) |
| Boosted tree | 5 | 0.71 (0.69, 0.74) | 3 | 0.70 (0.69, 0.71) |
| Ensemble learning | 1 | 0.72 (0.71, 0.73) | ||
| Other DL | 2 | 0.97 (0.96, 0.97) | 1 | 0.82 (0.77, 0.87) |
| Overall | 66 | 0.75 (0.72, 0.78） | 32 | 0.75 (0.71, 0.78) |
ANN, artificial neural network; CNN, convolutional neural network; DL, deep learnimg model; DT, decision tree; kNN, k-nearest neighbour; LR, logistic regression; NB, Naive Bayes; RF, random forests; SVM, support vector machine.