Comparison of Machine Learning Methods and Conventional Logistic Regressions for Predicting Gestational Diabetes Using Routine Clinical Data: A Retrospective Cohort Study
Table 1
Baseline characteristics.
Characteristic
Development set ()
Validation set ()
GDM ()
Control ()
value
GDM ()
Control ()
value
Maternal characteristics
Maternal age (years)
<20
1 (0.44%)
7 (0.26%)
<0.001
1 (0.11%)
0
<0.001
20-34.9
2051 (90.95%)
2522 (95.35%)
856 (92.34)
1105 (94.20%)
35-40
159 (7.05%)
88 (3.33%)
51 (5.50%)
50 (4.26%)
≥40
9 (0.40%)
3 (0.11%)
5 (0.54%)
1 (0.09%)
BMI (kg/m2)
<25
1290 (57.2%)
1920 (72.6%)
<0.001
559 (60.30%)
853 (72.7%)
<0.001
25-29.9
450 (20.0%)
269 (10.2%)
161 (17.40%)
110 (9.4%)
>30
80 (3.5%)
26 (1.0%)
29 (3.10%)
7 (0.6%)
Education status
<College
740 (33.9%)
957 (35.2%)
0.41
372 (37.2%)
383 (34.8%)
0.28
College
1142 (52.4%)
1363 (50.1%)
0.08
491 (49.1%)
539 (49.0%)
0.97
>College
209 (9.6%)
299 (11.0%)
0.12
99 (9.9%)
132 (12.0%)
0.12
Smoking
30 (1.4%)
31 (1.1%)
0.44
14 (1.4%)
15 (1.4%)
0.85
Nulliparous
1840 (81.60%)
2194 (82.95%)
0.44
763 (82.31%)
973 (82.95)
0.67
Prior macrosomia
22 (1.0%)
15 (0.57%)
0.10
10 (1.08%)
3 (0.26%)
0.02
Prior preterm delivery
22 (1.0%)
17 (0.64%)
0.18
7 (0.76%)
10 (0.85%)
0.80
Prior GDM
20 (0.89%)
0
<0.001
12 (1.30%)
0
<0.001
Family history of diabetes
21 (0.93%)
9 (0.34%)
0.008
14 (1.51%)
3 (0.36%)
0.001
Biochemical data
3-Triglyceride
<0.001
<0.001
Uric acid
<0.001
<0.001
Glycosylated hemoglobin
<0.001
<0.001
Alkaline phosphatase
0.008
0.07
Total cholesterol
<0.001
0.18
Lactic dehydrogenase
0.18
0.18
Fasting blood glucose
<0.001
<0.001
AFP concentration
<0.001
0.001
Fibrinogen
<0.001
<0.001
High-density lipoprotein
<0.001
0.04
Data are the (%) or . values indicate differences between groups calculated using the two-sample Wilcoxon rank-sum (Mann-Whitney) test for continuous variables and the Pearson test or ANOVA for categorical variables, with trend tests if appropriate. The “missing” category was not included in statistical tests. For characteristics that had no “missing” category, the data were 100% complete. Maternal age was defined as age at recruitment into the study. Maternal BMI was recorded at middle pregnancy when Down’s syndrome screening was performed.
