Table 2.
The comparison of performance of machine learning models in gestational diabetes mellitus (GDM) prediction applied to different subgroups.
| Subgroup | Models (N=30), n (%) | AUROCa | Sensitivity (95% CI) | Specificity (95% CI) | PLRb (95% CI) | NLRc (95% CI) | DORd (95% CI) |
| Overall | 30 (100) | 0.8492 | 0.69 (0.68-0.69) | 0.75 (0.75-0.75) | 4.02 (3.13-5.17) | 0.31 (0.26-0.38) | 13.78 (9.53-19.94) |
| 0-13 weeks before diagnosis | 16 (53) | 0.8667 | 0.74 (0.73-0.75) | 0.64 (0.64-0.64) | 3.89 (2.92-5.19) | 0.28 (0.22-0.36) | 16.55 (9.52-28.77) |
| 14-28 weeks before diagnosis | 14 (47) | 0.8365 | 0.64 (0.63-0.65) | 0.85 (0.84-0.85) | 3.90 (2.76-5.53) | 0.35 (0.25-0.48) | 11.67 (7.59-18.02) |
| With GDM history | 11 (37) | 0.8759 | 0.67 (0.66-0.68) | 0.85 (0.85-0.86) | 5.29 (3.39-8.25) | 0.28 (0.18-0.44) | 19.82 (11.49-34.13) |
| Without GDM history | 19 (63) | 0.8330 | 0.70 (0.66-0.68) | 0.65 (0.64-0.65) | 3.12 (2.52-3.86) | 0.35 (0.30-0.41) | 8.27 (5.14-13.29) |
| Logistic regression | 19 (63) | 0.8151 | 0.71 (0.70-0.72) | 0.67 (0.67-0.67) | 3.04 (2.37-3.89) | 0.37 (0.32-0.43) | 8.73 (5.99-12.73) |
| Non–logistic regression | 11 (37) | 0.8891 | 0.66 (0.65-0.67) | 0.85 (0.85-0.86) | 6.80 (4.45-10.37) | 0.24 (0.15-0.38) | 31.85 (15.93-63.69) |
aAUROC: area under receiver operating characteristic curve.
bPLR: positive likelihood ratio.
cNLR: negative likelihood ratio.
dDOR: diagnostic odds ratio.