Introduction
State of the art
Materials and methods
Database
Variable/ Feature | Non-GDM women (n = 1,382) Mean (IQR) | GDM women (n = 229) Mean (IQR) | Acquisition (GW) |
---|---|---|---|
Age | 27.64 (23–32) | 31.11 (27–36) | 4–20 |
Pregnancy Type | 1.01 (1–1) | 1.02 (1–1) | 4–20 |
Maternal Weight (first control) [kg] | 71.62 (60–81) | 81.77 (69–92) | 4–20 |
Height [m] | 1.59 (1.55–1.63) | 1.59 (1.55–1.63) | 4–20 |
BMI (Body Mass Index) (first control) | 28.18 (24.03–31.64) | 32.17 (28.16–35.83) | 4–20 |
Gravidity | 1.24 (0–2) | 1.69 (0–2) | 4–20 |
Parity | 1.02 (0–2) | 1.38 (0–2) | 4–20 |
Abortions | 0.22 (0–0) | 0.32 (0–0) | 4–20 |
Vaginal deliveries | 0.79 (0–1) | 1.03 (0–2) | 4–20 |
Caesarean deliveries | 0.22 (0–0) | 0.34 (0–1) | 4–20 |
Stillbirths | 0.01 | 0.03 | 4–20 |
First trimester fasting glycemia [mg/dL] (1TFG) | 77.22 (72–83) | 87.12 (80–93) | 4–12 |
OGTT (fasting) [mg/dL] | 74.28 (69–81) | 95.48 (86–101) | 24–28 |
OGTT (2 h) [mg/dL] | 99.39 (84–114) | 142.87 (120–171) | 24–28 |
(%) | (%) | ||
Tobacco | 7.74 | 11.79 | 4–20 |
Alcohol | 3.62 | 4.80 | 4–20 |
Illicit Drugs | 2.89 | 0.87 | 4–20 |
Cardiac Disease | 0.65 | 0.44 | 4–20 |
Biliary Disease | 1.01 | 2.18 | 4–20 |
Urinary Tract Disease | 2.32 | 4.80 | 4–20 |
Chronic kidney Disease | 0.36 | 0.00 | 4–20 |
Inflammatory bowel Disease | 0.07 | 0.44 | 4–20 |
Chronic lung diseases | 2.31 | 3.05 | 4–20 |
Systemic lupus erythematosus /Antiphospholipid antibody syndrome | 0.14 | 0.44 | 4–20 |
Psychiatric Disorders | 1.88 | 3.49 | 4–20 |
Endocrine Disorders | 0.36 | 0.87 | 4–20 |
Gynecological Disorders | 3.40 | 7.42 | 4–20 |
Epilepsy | 1.09 | 0.44 | 4–20 |
Insulin Resistance | 2.46 | 6.99 | 4–20 |
Hypothyroidism | 4.05 | 9.17 | 4–20 |
Chronic Hypertension | 4.70 | 12.66 | 4–20 |
Antihypertensive Drugs | 3.55 | 10.04 | 4–20 |
Data augmentation
DA\Columns | Age (Years) | 1TFG (mg/dL) | Height (cm) | Weight (kg) | BMI |
---|---|---|---|---|---|
Expert original range | ± 2 | ± 5 | ± 3 | ± 5 | * |
Limited Expert range | ± 1 | ± 1 | ± 1 | ± 2 | * |
Prediction models
Model implementation and hyperparameters
Hyperparameter | Used by | Ranges [lower bound, upped bound] |
---|---|---|
“var_smoothing” | Gaussian Naïve Bayes | [1e-10, 1e-7] |
“alpha” | Bernoulli Naïve Bayes | [1e-10, 1] |
“criterion” | Decision Tree, Random Forest, Extra Trees, Balanced Random Forest | “gini”, “entropy” |
“max_depth” | Decision Tree, Extreme Gradient Boosting | [1, 20] |
“max_leaf_nodes” | Decision Tree | [6, 384] |
“splitter” | Decision Tree | “best”, “random” |
“kernel” | SVM | “linear”, “poly”, “rbf”, “sigmoid” |
“degree” | SVM | [1, 3] |
“decision_function_shape” | SVM | “ovo”, “ovr” |
“C” | SVM, Logistic Regression | [0.0001, 10] |
“solver” | Multi-Layer Perceptron | “sgd”, “adam” |
“hidden_layer_sizes” | Multi-Layer Perceptron | [8, 256], hidden layers: [1, 10] |
“activation” | Multi-Layer Perceptron | “logistic”, “tanh”, “relu” |
“learning_rate_init” | Multi-Layer Perceptron | [0.001, 0.1] |
“max_iter | Multi-Layer Perceptron | 20000 |
“early_stopping” | Multi-Layer Perceptron | True, False |
“learning_rate” | Multi-Layer Perceptron | “constant”, “invscaling”, “adaptive” |
“algorithm” | K-Nearest Neighbors | “auto”, “ball_tree”, “kd_tree”, “brute” |
“leaf_size” | K-Nearest Neighbors | [1, 30] |
“p” | K-Nearest Neighbors | [1, 4] |
“n_neighbors” | K-Nearest Neighbors | [1, 25] |
“solver” | Logistic Regression | “newton-cg”, “lbfgs”, “liblinear”, “sag”, “saga” |
“n_estimators” | Random Forest, Extra Trees, Balanced Random Forest, Extreme Gradient Boosting, Light Gradient Boosting Machine | [10, 2000] |
“eta” | Extreme Gradient Boosting | [0.001, 0.3] |
“booster” | Extreme Gradient Boosting | “gbtree”, “gblinear”, “dart” |
“gamma” | Extreme Gradient Boosting | [0, 1] |
“boosting” | Light Gradient Boosting Machine | “gbdt”, “rf”, “dart”, “goss” |
“learning_rate” | Light Gradient Boosting Machine | [0.001, 0.1] |
Model evaluation
Results
Population characteristics
Variable selection
Ranking | F-Test ANOVA | Chi-Square | Mutual Information | BRF |
---|---|---|---|---|
1 | 1TFG | 1TFG | 1TFG | 1TFG |
2 | BMI | Maternal Weight | BMI | BMI |
3 | Maternal Weight | BMI | Age | Maternal Weight |
4 | Age | Age | Antihypertensive Drugs | Age |
5 | Chronic Hypertension | Gravidity | Maternal Weight | Height |
6 | Gravidity | Chronic Hypertension | Inflammatory Bowel Disease | Gravidity |
7 | Antihypertensive Drugs | Parity | Illicit Drugs | Parity |
8 | Parity | Antihypertensive Drugs | Chronic Kidney Disease | Vaginal Deliveries |
9 | Insulin Resistance | Abortions | Urinary Tract Disease | Abortions |
10 | Hypothyroidism | Vaginal Deliveries | Insulin Resistance | Cesarean Deliveries |
11 | Vaginal Deliveries | Insulin Resistance | Psychiatric Disorders | Hypothyroidism |
12 | Abortions | Hypothyroidism | Cardiac Disease | Chronic Hypertension |
Model performance
Model | Model Type | Number of input variables | Data Augmentation | Accuracy | Sensitivity | Specificity | Recall Macro | AUC ROC | FP | FN | FP + FN |
1 | MLP | 12 | w/o DA | 0.3994 | 1 | 0.3169 | 0.6585 | 0.8189 | 194 | 0 | 194 |
2 | MLP | 10 | DA EO | 0.3715 | 1 | 0.2852 | 0.6426 | 0.7741 | 203 | 0 | 203 |
3 | MLP | 11 | DA LE | 0.3715 | 1 | 0.2852 | 0.6426 | 0.7890 | 203 | 0 | 203 |
4 | MLP | 11 | DA LE | 0.3653 | 1 | 0.2782 | 0.6391 | 0.7874 | 205 | 0 | 205 |
5 | MLP | 8 | DA LE | 0.5511 | 0.9744 | 0.4930 | 0.7337 | 0.8002 | 144 | 1 | 145 |
6 | SVM | 5 | DA LE | 0.5480 | 0.9744 | 0.4894 | 0.7319 | 0.8161 | 145 | 1 | 146 |
7 | SVM | 5 | DA LE | 0.5480 | 0.9744 | 0.4894 | 0.7319 | 0.8161 | 145 | 1 | 146 |
8 | MLP | 4 | DA EO | 0.5387 | 0.9744 | 0.4789 | 0.7266 | 0.8052 | 148 | 1 | 149 |
9 | SVM | 5 | DA EO | 0.6068 | 0.9487 | 0.5599 | 0.7543 | 0.8234 | 125 | 2 | 127 |
10 | MLP | 4 | DA EO | 0.5759 | 0.9487 | 0.5246 | 0.7367 | 0.8159 | 135 | 2 | 137 |
11 | MLP | 3 | w/o DA | 0.5728 | 0.9487 | 0.5211 | 0.7349 | 0.8165 | 136 | 2 | 138 |
12 | MLP | 4 | DA LE | 0.5728 | 0.9487 | 0.5211 | 0.7349 | 0.8082 | 136 | 2 | 138 |
13 | SVM | 5 | DA EO | 0.6130 | 0.9231 | 0.5704 | 0.7468 | 0.8234 | 122 | 3 | 125 |
14 | MLP | 6 | w/o DA | 0.6006 | 0.9231 | 0.5563 | 0.7397 | 0.8221 | 126 | 3 | 129 |
15 | MLP | 8 | DA EO | 0.6006 | 0.9231 | 0.5563 | 0.7397 | 0.8183 | 126 | 3 | 129 |
16 | LR | 3 | DA EO | 0.6006 | 0.9231 | 0.5563 | 0.7397 | 0.8159 | 126 | 3 | 129 |
17 | MLP | 5 | DA LE | 0.6594 | 0.8974 | 0.6268 | 0.7621 | 0.8199 | 106 | 4 | 110 |
18 | MLP | 5 | w/o DA | 0.6594 | 0.8974 | 0.6268 | 0.7621 | 0.8146 | 106 | 4 | 110 |
19 | MLP | 5 | DA LE | 0.6563 | 0.8974 | 0.6232 | 0.7603 | 0.8178 | 107 | 4 | 111 |
20 | MLP | 7 | DA LE | 0.6563 | 0.8974 | 0.6232 | 0.7603 | 0.8118 | 107 | 4 | 111 |
21 | MLP | 7 | DA LE | 0.6873 | 0.8718 | 0.6620 | 0.7669 | 0.8160 | 96 | 5 | 101 |
22 | MLP | 10 | DA LE | 0.6811 | 0.8718 | 0.6549 | 0.7634 | 0.8078 | 98 | 5 | 103 |
23 | MLP | 9 | DA LE | 0.6780 | 0.8718 | 0.6514 | 0.7616 | 0.8137 | 99 | 5 | 104 |
24 | MLP | 9 | DA EO | 0.6749 | 0.8718 | 0.6479 | 0.7598 | 0.8137 | 100 | 5 | 105 |
25 | MLP | 6 | DA LE | 0.7090 | 0.8462 | 0.6901 | 0.7681 | 0.8142 | 88 | 6 | 94 |
26 | MLP | 9 | DA EO | 0.7090 | 0.8462 | 0.6901 | 0.7681 | 0.8022 | 88 | 6 | 94 |
27 | MLP | 10 | w/o DA | 0.7028 | 0.8462 | 0.6831 | 0.7646 | 0.8063 | 90 | 6 | 96 |
28 | MLP | 9 | DA EO | 0.7028 | 0.8462 | 0.6831 | 0.7646 | 0.8022 | 90 | 6 | 96 |
29 | SVM | 12 | w/o DA | 0.7554 | 0.8205 | 0.7465 | 0.7835 | 0.8135 | 72 | 7 | 79 |
30 | SVM | 12 | w/o DA | 0.7461 | 0.8205 | 0.7359 | 0.7782 | 0.8135 | 75 | 7 | 82 |
31 | SVM | 7 | DA LE | 0.7399 | 0.8205 | 0.7289 | 0.7747 | 0.8143 | 77 | 7 | 84 |
32 | SVM | 7 | DA LE | 0.7368 | 0.8205 | 0.7254 | 0.7729 | 0.8143 | 78 | 7 | 85 |
33 | SVM | 7 | DA LE | 0.7399 | 0.7949 | 0.7324 | 0.7636 | 0.8143 | 76 | 8 | 84 |
34 | SVM | 10 | DA LE | 0.7337 | 0.7949 | 0.7254 | 0.7601 | 0.8173 | 78 | 8 | 86 |
35 | MLP | 5 | DA EO | 0.7276 | 0.7949 | 0.7183 | 0.7566 | 0.8120 | 80 | 8 | 88 |
36 | MLP | 9 | DA EO | 0.7245 | 0.7949 | 0.7148 | 0.7548 | 0.8068 | 81 | 8 | 89 |
Model | Model Type | Number of input variables | Data Augmentation | Accuracy | Sensitivity | Specificity | Recall Macro | AUC ROC | FP | FN | FP + FN |
---|---|---|---|---|---|---|---|---|---|---|---|
37 | MLP | 15 | DA LE | 0.3003 | 1 | 0.2042 | 0.6021 | 0.8210 | 226 | 0 | 226 |
38 | SVM | 15 | w/o DA | 0.5697 | 0.9744 | 0.5141 | 0.7442 | 0.7872 | 138 | 1 | 139 |
39 | MLP | 13 | DA LE | 0.5820 | 0.9487 | 0.5317 | 0.7402 | 0.8093 | 133 | 2 | 135 |
40 | SVM | 15 | w/o DA | 0.6099 | 0.9231 | 0.5669 | 0.7450 | 0.7872 | 123 | 3 | 126 |
41 | MLP | 13 | w/o DA | 0.6409 | 0.8974 | 0.6056 | 0.7515 | 0.8152 | 112 | 4 | 116 |
42 | MLP | 14 | DA LE | 0.7059 | 0.8718 | 0.6831 | 0.7774 | 0.7968 | 90 | 5 | 95 |
43 | MLP | 15 | DA LE | 0.7214 | 0.8462 | 0.7042 | 0.7752 | 0.7988 | 84 | 6 | 90 |
44 | SVM | 15 | DA EO | 0.7337 | 0.8205 | 0.7218 | 0.7712 | 0.8125 | 79 | 7 | 86 |
45 | SVM | 15 | DA EO | 0.7461 | 0.7949 | 0.7394 | 0.7672 | 0.8125 | 74 | 8 | 82 |
Model Number | Model Type | Number of input variables | Data Augmentation | Accuracy | Sensitivity | Specificity | Recall Macro | ROC | FP | FN | FP + FN |
1 w/DA | MLP | 12 | DA EO | 0.3313 | 1 | 0.2394 | 0.6197 | 0.7505 | 216 | 0 | 216 |
1 w/o DA | MLP | 12 | w/o DA | 0.3994 | 1 | 0.3169 | 0.6585 | 0.8189 | 194 | 0 | 194 |
5 w/DA | MLP | 8 | DA LE | 0.5511 | 0.9744 | 0.4930 | 0.7337 | 0.8002 | 144 | 1 | 145 |
5 w/o DA | MLP | 8 | w/o DA | 0.4303 | 0.9744 | 0.3556 | 0.6650 | 0.8172 | 183 | 1 | 184 |
9 w/DA | SVM | 5 | DA EO | 0.6068 | 0.9487 | 0.5599 | 0.7543 | 0.8234 | 125 | 2 | 127 |
9 w/o DA | SVM | 5 | w/o DA | 0.4396 | 0.9487 | 0.3697 | 0.6592 | 0.8221 | 179 | 2 | 181 |
13 w/DA | SVM | 5 | DA EO | 0.6130 | 0.9231 | 0.5704 | 0.7468 | 0.8234 | 122 | 3 | 125 |
13 w/o DA | SVM | 5 | w/o DA | 0.5913 | 0.9231 | 0.5458 | 0.7344 | 0.8221 | 129 | 3 | 132 |
17 w/DA | MLP | 5 | DA LE | 0.6594 | 0.8974 | 0.6268 | 0.7621 | 0.8199 | 106 | 4 | 110 |
17 w/o DA | MLP | 5 | w/o DA | 0.5944 | 0.8974 | 0.5528 | 0.7251 | 0.8202 | 127 | 4 | 131 |
25 w/DA | MLP | 6 | DA LE | 0.7090 | 0.8462 | 0.6901 | 0.7681 | 0.8142 | 88 | 6 | 94 |
25 w/o DA | MLP | 6 | w/o DA | 0.6099 | 0.8462 | 0.5775 | 0.7118 | 0.8156 | 120 | 6 | 126 |
29 w/DA | SVM | 12 | DA LE | 0.7368 | 0.8205 | 0.7254 | 0.7729 | 0.8129 | 78 | 7 | 85 |
29 w/o DA | SVM | 12 | w/o DA | 0.7554 | 0.8205 | 0.7465 | 0.7835 | 0.8135 | 72 | 7 | 79 |
33 w/DA | SVM | 7 | DA LE | 0.7399 | 0.7949 | 0.7324 | 0.7636 | 0.8143 | 76 | 8 | 84 |
33 w/o DA | SVMa | 7 | w/o DA | 0.5635 | 0.8205 | 0.5282 | 0.6743 | 0.7852 | 134 | 7 | 141 |
33 w/o DA | SVMa | 7 | w/o DA | 0.6161 | 0.7692 | 0.5951 | 0.6822 | 0.7852 | 115 | 9 | 124 |
Discussion
Comparison with state of the art
Models | Accuracy | Sensitivity | Specificity | Recall Macro | AUC ROC |
---|---|---|---|---|---|
DNN, 7 Variables [20] | - | 0.7 | 0.69 | 0.695* | 0.77 |
LR, 5 Continuous Variables [21] | - | 0.61 | 0.80 | 0.705* | 0.766 |
LGBM, 9 questions (Variables) [22] | - | - | - | - | 0.799 |
RF, Dimension Reduction, 6 Variables [25] | 0.789 | 0.651 | 0.813 | 0.732* | 0.777 |
LR, 4 Variables [26] | - | - | - | - | 0.70 |
1 Variable ** [27] | - | 0.490 | 0.676 | 0.583* | 0.608 |
RECPAM, 3 Variables [28] | - | 0.89 | 0.40 | 0.645* | - |
2 Variables ** [30] | - | 0.51 | 0.81 | 0.660* | 0.71 |
NN, 4 Variables, IADPSG Criteria [31] | - | - | - | - | 0.73 |
Ours 1 MLP 12 Variables No DA | 0.3994 | 1 | 0.3169 | 0.6585 | 0.8189 |
Ours 5 MLP 8 Variables DA LE | 0.5511 | 0.9744 | 0.4930 | 0.7337 | 0.8002 |
Ours 9 SVM 5 Variables DA EO | 0.6068 | 0.9487 | 0.5599 | 0.7543 | 0.8234 |
Ours 13 SVM 5 Variables DA EO | 0.6130 | 0.9231 | 0.5704 | 0.7468 | 0.8234 |
Ours 17 MLP 5 Variables DA EO | 0.6594 | 0.8974 | 0.6268 | 0.7621 | 0.8199 |
Ours 21 MLP 7 Variables DA LE | 0.6873 | 0.8718 | 0.6620 | 0.7669 | 0.8160 |
Ours 25 MLP 6 Variables DA LE | 0.7090 | 0.8462 | 0.6901 | 0.7681 | 0.8142 |
Ours 29 SVM 12 Variables No DA | 0.7554 | 0.8205 | 0.7465 | 0.7835 | 0.8135 |
Ours 33 SVM 7 Variables DA LE | 0.7399 | 0.7949 | 0.7324 | 0.7636 | 0.8143 |
Models | Input Variables |
---|---|
DNN, 7 Variables [20] | Age, Previous GDM, Family history of diabetes in a first-degree relative, Multiple pregnancy, FPG, HBA1C, Triglyceride |
LR, 5 Continuous Variables [21] | Age, pre-pregnancy BMI, FPG and Triglyceride |
LGBM, 9 questions (Variables) [22] | Age, Weight and Height, Familiar history of diabetes in first-degree relatives, High cholesterol, Miscarriage, PCOS, Pre-diabetes, Heart Diseases, GDM or High BP before current pregnancy, HBA1C, Previous birth (Yes or No), if yes, number of times and GCT or OGTT in that pregnancy if they are available |
RF, Dimension Reduction, 6 Variables [25] | Age, pre-pregnancy BMI, abdomen circumference in the first trimester, gravidity, PCOS, irregular menstruation and family history of diabetes |
LR, 4 Variables [26] | Age, BMI, FPG, Familiar history of diabetes in first-degree relatives |
1 Variable * [27] | FPG |
RECPAM, 3 Variables [28] | BMI, FPG, Familiar history of diabetes in first-degree relatives |
2 Variables * [30] | BMI, fasting glucose |
NN, 4 Variables, IADPSG Criteria [31] | Mean arterial blood pressure, Age, Previous history of GDM, Ethnicity |
Ours 1 MLP 12 Variables No DA | Age, Weight, BMI, Illicit Drugs, Cardiac Diseases, Urinal Tract Diseases, Psychiatric Disorders, Chronic Kidney Diseases, Inflammatory bowel disease, Insulin Resistance, Use of Antihypertensive drugs, FPG |
Ours 5 MLP 8 Variables DA LE | Age, Weight, BMI, Illicit Drugs, Chronic Kidney Diseases, Inflammatory bowel disease, Use of Antihypertensive drugs, FPG |
Ours 9 SVM 5 Variables DA EO | Age, Weight, BMI, Gravidity, FPG |
Ours 13 SVM 5 Variables DA EO | Age, Weight, BMI, Gravidity, FPG |
Ours 17 MLP 5 Variables DA EO | Age, Weight, BMI, Gravidity, FPG |
Ours 21 MLP 7 Variables DA LE | Age, Weight, BMI, Gravidity, Parity, Chronic Hypertension, FPG |
Ours 25 MLP 6 Variables DA LE | Age, Weight, BMI, Inflammatory bowel disease, Use of Antihypertensive drugs, FPG |
Ours 29 SVM 12 Variables No DA | Age, Weight, BMI, Illicit Drugs, Cardiac Diseases, Urinal Tract Diseases, Psychiatric Disorders, Chronic Kidney Diseases, Inflammatory bowel disease, Insulin Resistance, Use of Antihypertensive drugs, FPG |
Ours 33 SVM 7 Variables DA LE | Age, Weight, BMI, Gravidity, Chronic Hypertension, Use of Antihypertensive drugs, FPG |