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

Table 2

Performance of the cutoff points of 0.3 and 0.7 for the GBDT model in predicting GDM.

Cutoff points	Development cohort; Percent (95% CI)	Validation cohort; Percent (95% CI)
0.3
Negative predictive value	82.40% (79.90%-84.70%)	74.10% (69.50%-78.20%)
Positive predictive value	51.3% (49.80%-52.90%)	52.60% (50.20%-54.90%)
Sensitivity	92% (90.80%-93.10%)	90% (88.00%-91.70%)
Specificity	30% (28.30%-31.80%)	26% (23.50%-28.70%)
Positive likelihood ratio	1.31 (1.29-1.34)	1.22 (1.18-1.26)
Negative likelihood ratio	0.27 (0.11-0.42)	0.39 (0.17-0.60)
0.7
Negative predictive value	59.30% (57.80%-60.70%)	56.1% (53.90%-58.30%)
Positive predictive value	86.60% (82.90%-89.60%)	93.2% (88.20%-96.10%)
Sensitivity	16% (14.50%-17.60%)	15% (12.90%-17.30%)
Specificity	98% (97.40%-98.50%)	99% (98.20%-99.40%)
Positive likelihood ratio	8 (7.72-8.28)	15 (14.38-15.61)
Negative likelihood ratio	0.86 (0.84-0.88)	0.86 (0.83-0.89)