Comparison of Machine Learning Methods and Conventional Logistic Regressions for Predicting Gestational Diabetes Using Routine Clinical Data: A Retrospective Cohort Study

Table 3

Performance of the cutoff points of 0.3 and 0.7 for the GBDT model in predicting adverse pregnancy outcomes.

Cutoff points	Development cohort; Percent (95% CI)	Validation cohort; Percent (95% CI)
0.3
Negative predictive value	100% (92.7%-100%)	52.4% (32.4%-71.7%)
Positive predictive value	50.1% (48.7%-51.5%)	49.9% (47.8%-52.1%)
Sensitivity	100% (99.8%-100%)	99.0% (98.3%-99.5%)
Specificity	2% (1.5%-2.6%)	1% (0.6%-1.9%)
Positive likelihood ratio	1.02% (1.01%-1.03%)	1% (0.99%-1.01%)
Negative likelihood ratio	0 (0-nan)	1% (0.15%-1.85%)
0.7
Negative predictive value	51.4% (50.0%-52.8%)	50.9% (48.7%-53.0%)
Positive predictive value	83.0% (76.1%-88.2%)	79.2% (66.5%-88.0%)
Sensitivity	5% (4.2%-6.0%)	4% (3.0%-5.4%)
Specificity	99% (98.5%-99.3%)	99% (98.1%-99.4%)
Positive likelihood ratio	5 (4.57-5.43)	4 (3.33-4.67)
Negative likelihood ratio	0.96 (0.95-0.97)	0.97 (0.96-0.98)