Abstract
Introduction
Gestational diabetes mellitus (GDM) significantly affects pregnancy outcomes. Therefore, it is crucial to develop prediction models since they can guide timely interventions to reduce the incidence of GDM and its associated adverse effects.
Methods
A total of 554 pregnant women were selected and their sociodemographic characteristics, clinical data and dietary data were collected. Dietary data were investigated by a validated semi-quantitative food frequency questionnaire (FFQ). We applied random forest mean decrease impurity for feature selection and the models are built using logistic regression, XGBoost, and LightGBM algorithms. The prediction performance of different models was compared by accuracy, sensitivity, specificity, area under curve (AUC) and Hosmer–Lemeshow test.
Results
Blood glucose, age, pre-pregnancy body mass index (BMI), triglycerides and high-density lipoprotein cholesterol (HDL) were the top five features according to the feature selection. Among the three algorithms, XGBoost performed best with an AUC of 0.788, LightGBM came second (AUC = 0.749), and logistic regression performed the worst (AUC = 0.712). In addition, XGBoost and LightGBM both achieved a fairly good performance when dietary information was included, surpassing their performance on the non-dietary dataset (0.788 vs 0.718 in XGBoost; 0.749 vs 0.726 in LightGBM).
Conclusion
XGBoost and LightGBM algorithms outperform logistic regression in predicting GDM among Chinese pregnant women. In addition, dietary data may have a positive effect on improving model performance, which deserves more in-depth investigation with larger sample size.
Introduction
Gestational diabetes mellitus (GDM) refers to carbohydrate intolerance resulting in hyperglycemia of variable severity with onset or first recognition during pregnancy (1, 2). It has become one of the main reasons that seriously affect the pregnancy outcomes (3, 4, 5). The incidence of GDM in China is about 20% and is increasing year by year (6, 7). Though GDM significantly affects the health of pregnant women and their fetuses, the diagnosis of GDM in China now relies on the 75 g oral glucose tolerance test (OGTT) (8, 9). While OGTT is usually performed between the 24th and 28th weeks of pregnancy, it is obviously too late to prevent the occurrence of GDM. A meta-analysis (10) shows that an early intervention (e.g. 15 weeks) can reduce the incidence of GDM, and the earlier the intervention, the better the effect. Therefore, it is important to screen high-risk individuals for GDM at an earlier stage to protect the health of pregnant women and their fetuses from adverse pregnancy outcomes.
GDM is commonly associated with an array of factors, including obesity, a familial history of diabetes and the presence of polycystic ovary syndrome, among others (11, 12, 13). In recent years, these factors have been considered as the predictive factors in the research of the GDM prediction model. For example, in a cohort study of pregnant women in Tianjin, China, age, waist circumference, hip circumference, alanine aminotransferase (ALT), pre-pregnancy BMI, systolic blood pressure, family history of diabetes mellitus and fasting blood glucose were selected as predictive indexes (14). Researchers in Israel and South Korea have carried out similar studies (15, 16).
However, we believe that there is still room for optimization of current machine learning models related to GDM. As a matter of fact, while diets directly affect many important physical indexes such as BMI, blood pressure, serum cholesterol, etc., dietary data are missing in current studies of GDM prediction. In terms of pathogenesis, most GDM are caused by obesity and chronic insulin resistance (17), and these problems are mostly caused by bad dietary habits. Existing studies have shown that dietary factors such as dietary patterns, unreasonable energy ratio of the three major nutrients and excessive intake of carbohydrates are all risk factors for GDM (18, 19, 20, 21, 22, 23). Given the pathogenesis foundation, this study aims to fill the gap by incorporating dietary data as predictive factors for an improved and integrated predictive model. By analyzing complex dietary patterns in patient data and integrating these patterns with other clinical features, the gestational diabetes prediction models can assist clinicians in providing personalized dietary guidance to pregnant women at high risk of GDM.
Methods
Data collection
We recruited participants from the Seventh People’s Hospital Affiliated with Shanghai University of Chinese Medicine in Shanghai, China, between January 2018 and June 2021. Our inclusion criteria for pregnant women were (i) natural conception, (ii) singleton pregnancy and (iii) the ability to read, understand, and sign informed consent. We excluded pregnant women who had been diagnosed with diabetes prior to pregnancy, as well as those with severe heart, brain, liver, or kidney diseases, gestational thyroid disease, mental disorders, placenta previa, any infectious diseases, autoimmune diseases or malignant tumors. Ultimately, a total of 554 pregnant women were enrolled in the study.
We collected clinical data from these pregnant women, focusing particularly on routine blood tests and blood lipid examinations conducted between the 10th and 12th weeks of pregnancy. We also carried out dietary surveys and universal screening for GDM at around the 24th week of pregnancy. According to the diagnostic criteria of GDM in China, which refer to the 75g oral glucose tolerance test (OGTT), 152 pregnant women were diagnosed with GDM (24, 25, 26). In addition, a validated semi-quantitative food frequency questionnaire (FFQ) consisting of 222 food-related items was used to assess participants’ dietary intakes through face-to-face interviews. For each item in the FFQ, three aspects were recorded: whether the item was consumed, the usual frequency of consumption (times per day, week, month or year) and the estimated amount consumed per occasion, expressed in either the local unit ‘liang’ for weight (1 liang = 50 g) or a standard cup for volume (1 cup = 250 mL). All researchers conducting the food survey were dietitians with over 3 years of experience in hospitals and received training before the study. To enhance the accuracy of participants’ food recall, food models and pictures were also used. We estimated participants’ average daily intake of oil, salt and sugar based on their household’s total consumption of these condiments. Finally, we expressed all food data as daily intake per person and calculated individual nutrient intakes using a nutritional dietary management information system developed by Shanghai Zhending Health Technology Co., Ltd (27, 28). We collected 78 features in total, but only 77 were used in the actual modeling, as one feature was excluded due to excessive missing values.
Written informed consent was obtained from all study participants. To protect individuals’ privacy, we replaced sensitive information, particularly names, with codes. During data collection, codes were assigned based on hospital admission time and bed number. For statistical analysis, only these codes were retained, and the names of the pregnant women were excluded. This study was registered at www.chictr.org.cn. (ChiCTR1900027764) and was approved by the Medical Ethics Committee of the Seventh People’s Hospital affiliated with Shanghai University of Traditional Chinese Medicine (2020-7th-HIRB-016).
Model building
Missing values handling
We handled missing values in the dataset by removing features with more than 10% missing data, and excluding them from modeling. For the remaining data, missing values in continuous variables were imputed using mean values, while categorical variables were assigned default values: parity was set to 1, and both education level and family history of diabetes were set to zero. The count of missing values for all features is provided in the Supplementary Table 1 (see section on supplementary materials given at the end of this article).
Comparison of the sociodemographic characteristics between the control group and the GDM group.
| Characteristics | Control (402) | GDM (152) | P |
|---|---|---|---|
| Age/year | 28.43 ± 4.42 | 30.79 ± 4.68 | <0.001 |
| Height/cm | 161.00 ± 4.84 | 159.85 ± 4.55 | 0.005 |
| Pre-pregnancy Weight/kg | 57.02 ± 10.05 | 60.83 ± 12.30 | 0.001 |
| Pre-pregnancy BMI/kg∙m0-1-2 | 21.93 ± 3.44 | 23.98 ± 6.17 | <0.001 |
| Parity/n (%) | 0.072 | ||
| Primiparity | 209 (52.0) | 66 (43.4) | |
| Multipara | 193 (48.0) | 86 (56.6) | |
| Number of pregnancies/n (%) | 0.017 | ||
| Primigravida | 169 (42.1) | 47 (30.9) | |
| Multigravida | 233 (57.9) | 105 (69.1) | |
| Education level/n (%) | 0.981 | ||
| Below college | 317 (78.9) | 120 (78.9) | |
| College and above | 85 (21.1) | 32 (21.1) | |
| Exercise/n (%) | 0.836 | ||
| Yes | 282 (70.2) | 108 (71.1) | |
| No | 120 (29.8) | 44 (28.9) | |
| History of miscarriages/n (%) | 0.295 | ||
| Yes | 142 (35.4) | 61 (40.2) | |
| No | 260 (64.6) | 91 (59.8) | |
| Family history of diabetes/n (%) | 0.012 | ||
| Yes | 34 (8.5) | 24 (15.8) | |
| No | 368 (91.5) | 128 (84.2) | |
| History of diabetes/n (%) | <0.001 | ||
| Yes | 3 (0.7) | 16 (10.5) | |
| No | 399 (99.3) | 136 (89.5) |
GDM, gestational diabetes mellitus.
Feature selection
In this study, random forest mean decrease impurity was applied for feature selection.
Two datasets
Dietary data are important but not studied in predictive models for GDM. Therefore, in this study, we created two datasets containing different features. The features in one dataset included sociodemographic and clinical data; the other dataset included sociodemographic data, clinical data and dietary data. We compared the predictive performance of the models constructed with these two datasets separately. In this way, we can find out whether dietary data can improve the performance of the models.
Model training
Logistic regression, XGBoost and LightGBM algorithms were utilized. To maximize the use of the limited data, a five-fold cross-validation was performed to split the training and test sets, and Grid Search was employed to find the optimal parameters.
Model evaluation
Accuracy (Ac), sensitivity (Sn), specificity (Sp), areas under the curve (AUC) of receiver operating characteristic (ROC) and Hosmer–Lemeshow test (HL test) were used to evaluate the performance of the models. All the construction and evaluation of the models were conducted in Python (version 3.11.9) using Visual Studio Code (version 1.91.1).
Statistical analysis
Descriptive analysis was performed to determine the numbers and percentages for sociodemographic data. Group differences in sociodemographic, clinical, and dietary data were analyzed using an independent samples t-test, chi-square test, or Mann–Whitney U test. All analyses were performed using IBM SPSS Statistics, version 29 and statistical significance was defined as P < 0.05.
Results
Baseline characteristics
A total of 554 pregnant women were included in the final data analysis. The quantitative data were expressed as mean ± s.d., and the qualitative data were expressed as n (%). Among the 554 participants, there were 402 pregnant women in the control group and 152 pregnant women in the GDM group. The results showed that there were significant differences in age, height, pre-pregnancy weight, pre-pregnancy BMI, number of pregnancies, family history, and diabetes history between the control group and the GDM group (P < 0.001, P = 0.005, P = 0.001, P < 0.001, P = 0.017, P = 0.012, P < 0.001). There were no significant differences in parity, education level, exercise and history of miscarriages between the two groups. The results are detailed in Table 1.
Feature selection
After cleaning the data and eliminating redundant variables, we extracted a total of 18 out of 77 features by random forest and arranged these features as follows (Fig. 1). Blood glucose, age, pre-pregnancy BMI, triglycerides and HDL were the top five features. In dietary indexes, vitamin E (δ-E), livestock meat, vitamin E (α-E), aquatic products, selenium and grain contributed remarkably to the outcome.
The features selected by random forest mean decrease impurity. Cr, serum creatinine; NLR, neutrophil lymphocyte ratio; TC, total cholesterol; TG, triglyceride.
Citation: Endocrine Connections 13, 12; 10.1530/EC-24-0169
Model evaluation
Discrimination of different models
The ROC curves and confusion matrixes of the three models and the two datasets are shown in Fig. 2 and Supplementary Figure 1. A table of Ac, Sp, Sn and AUC is also provided (Table 2). The logistic regression (LR) model has the lowest performance (AUC = 0.712) and the conclusions of two datasets are almost the same. Of the other two models, XGBoost model performs better with an AUC of 0.788, while the AUC of LightGBM model is 0.749. Moreover, in both the XGBoost and LightGBM models, dietary dataset outperforms non-dietary dataset (0.788 vs 0.718 in XGBoost; 0.749 vs 0.726 in LightGBM), which indicates that dietary data may improve the prediction accuracy of the models.
The ROC curves of the three GDM models. LGB, LightGBM; LR, logistic regression; XGB, XGBoost.
Citation: Endocrine Connections 13, 12; 10.1530/EC-24-0169
Predictive performance of the three GDM models.
| Algorithm | Dataset | Accuracy | Sensitivity | Specificity | AUC (95% CI) |
|---|---|---|---|---|---|
| LR | Dietary | 0.755 | 0.270 | 0.938 | 0.712 (0.673, 0.752) |
| Non-dietary | 0.747 | 0.230 | 0.943 | 0.711 (0.665, 0.759) | |
| XGB | Dietary | 0.769 | 0.322 | 0.938 | 0.788 (0.751, 0.824) |
| Non-dietary | 0.744 | 0.283 | 0.918 | 0.718 (0.685, 0.750) | |
| LGB | Dietary | 0.747 | 0.309 | 0.913 | 0.749 (0.700, 0.798) |
| Non-dietary | 0.764 | 0.362 | 0.918 | 0.726 (0.690, 0.767) |
LGB, LightGBM; LR, logistic regression; XGB, XGBoost.
Calibration of different models
The HL test was used to test the calibration of the LR, XGB and LGB models (Table 3 and Supplementary Figure 2). Most models are well-calibrated since their P values are over 0.05. However, the P-value of the LGB models with dietary data is below 0.05, which indicates that the model’s calibration is poor.
The calibration of the three GDM models.
| Algorithm | Dataset | HL-chi2 | P |
|---|---|---|---|
| LR | Dietary | 9.47 | 0.304 |
| Non-dietary | 6.44 | 0.598 | |
| XGB | Dietary | 12.06 | 0.149 |
| Non-dietary | 11.64 | 0.168 | |
| LGB | Dietary | 21.66 | 0.006 |
| Non-dietary | 8.67 | 0.371 |
HL, Hosmer–Lemeshow test; LGB, LightGBM; LR, logistic regression; XGB, XGBoost.
Discussion
Our study used three different machine learning algorithms (logistic regression, XGBoost, LightGBM), two different datasets (dietary dataset and non-dietary dataset) and eighteen different features (sociodemographic data, clinical data, and dietary data) to establish predictive models for GDM. As a result, overall, models with the dietary dataset performed better than models with the non-dietary dataset, and the XGBoost algorithm was the best of the three.
Our results show that incorporating dietary data into the models improves their predictive performance to some extent. As to XGBoost and LightGBM, the AUC values of the models with dietary datasets are better than those of the models with non-dietary datasets (0.788 vs 0.718 in XGBoost; 0.749 vs 0.726 in LightGBM). The reason for this could be that many dietary factors are potentially associated with the development of GDM. A prospective cohort study in Western China (29) found that a high protein-low starch diet was associated with a lower risk for GDM among women who were overweight at pre-pregnancy, which indicated that dietary patterns can affect the incidence of GDM. We also found a significant difference in the energy contribution ratio of the three macronutrients between the GDM group and the control group (Supplementary Table 2). Many other studies (18, 19, 20, 21, 22, 23) also explored the relationship between diet and GDM from the perspective of dietary patterns and nutrients and have obtained conclusions that possess statistical significance. However, in the logistic regression models, the performance of the dietary dataset was nearly the same as that of the non-dietary dataset. This may be due to the fact that logistic regression is better at dealing with linear problems, whereas for datasets with weak linear relationships such as GDM, the performance of the model is poor and unstable.
Among the algorithms, XGBoost (AUC = 0.788 with the dietary dataset) performed the best, followed by LightGBM (AUC = 0.749 with the dietary dataset), while LR had relatively low predictive performance (AUC = 0.712 with the dietary dataset). This could be because, as mentioned in the previous paragraph, our dataset is complex, and logistic regression algorithms are not well-suited for handling non-linear relationships. Consequently, XGBoost and LightGBM perform better in prediction. When we compare the two again, XGBoost performs better and avoids overfitting. In contrast, LightGBM uses a leaf-wise growth strategy, which can result in deeper decision trees and a higher likelihood of overfitting, especially in models with smaller datasets. In our study, LightGBM very likely caused overfitting, which may explain why it underperforms compared to XGBoost. In addition, LightGBM’s advantages over XGBoost include lower memory usage and faster runtime.
When it comes to the calibration of the three models, LR is the best calibrated, this may be because logistic regression models the probability of a binary outcome directly using the logistic function. This means that the output is a probability that can be interpreted as the confidence of the prediction, which aligns naturally with the concept of calibration. But that doesn’t make much sense since it has a poor AUC value. Meanwhile, as mentioned above, the P value of the LightGBM model with dietary data is under 0.05, which indicates that although the model was able to distinguish the high-risk status of GDM, the specific risk probabilities provided by the model are not good (30). The reason for the poor calibration of LightGBM is most likely due to the occurrence of overfitting. LightGBM can build very deep decision trees, which can capture complex patterns in the data. However, deeper trees and a larger number of features are likely to make the model overly sensitive to specific details in the training data. This can lead the model to seek out complex relationships between features and the target variable, causing it to learn the noise in the dataset, ultimately resulting in overfitting. In contrast, when the number of features decreases (the non-dietary model), the calibration of LightGBM becomes better. Besides reducing the number of features, another way to address this issue is to increase the sample size. When it comes to XGBoost, the P-values for the HL test were all above 0.05, indicating good calibration. In summary, the XGBoost model with the dietary data has the highest AUC value while guaranteeing excellent calibration and is the best model in our study.
The models we built have similar or even higher AUC values compared to other studies that predict gestational diabetes (14, 31, 32, 33, 34). Though these studies also added features that are the main factors of GDM, such as pre-pregnancy BMI, age, history of GDM, etc. (35, 36, 37, 38), none of them takes dietary data into account and has a lower dimension than ours. This is probably the reason why our models have better performance. While at the same time, there are other models that have higher AUC values than ours, mostly higher than 0.80 (15, 39). This may be because they have a larger sample size. In the study conducted in Israel (15), researchers collected data from electronic health records (EHRs) and used the built-in algorithm to predict GDM. As a result, they had an AUC value of 0.85. Due to the convenience of EHRs, they have a sample size of 588,662 pregnant women and their diagnosis of GDM is different from ours, which may also be the reason that their model has a better performance. Meanwhile, Wang et al.’s study (39) also obtained an AUC value of more than 0.80. In their study, they used a score-scaled model, logistic regression model, decision tree model, and random forest model to predict GDM. However, in theory, the predictive performance of our XGBoost model and the LightGBM model are no worse, or even better. The reason for this may be that they included some variables in the model that we didn’t have, such as the positivity of thyroid-related antibodies and gestational weight gain (35, 36, 38, 40).
Our research has several strengths and limitations. A key strength is the integration of dietary data, which can enhance the performance of machine-learning models used to predict the onset of GDM. This means that incorporating dietary information can help build more accurate predictive models. However, our study has two primary limitations. First, the small sample size may have contributed to the suboptimal calibration of the LightGBM model. Increasing the sample size could improve the model’s calibration and overall predictive performance. Second, the dietary survey was conducted around the 24th week of pregnancy, later than the clinical data collection. Although the FFQ reflects dietary intake over the past 24 weeks and offers some insights into early pregnancy diet, it weakened the application value of predictive models. The significance of our research lies in the discovery that introducing dietary data is beneficial for predicting GDM. In future studies, an earlier dietary survey time point (such as about 10 weeks of early pregnancy) and a more accurate artificial intelligence dietary survey method can be explored to improve prediction accuracy.
In conclusion, this study established three predictive models of GDM to help clinicians provide intervention and dietary guidance to pregnant women at high risk of GDM. Among them, the model used XGBoost algorithm and dietary dataset performed best. It is a significant finding that dietary data do improve the predictive performance of the models. The feature selection methods and machine learning algorithms used in this paper can also help the following researchers build more accurate predictive models, especially with a larger sample size.
Supplementary materials
This is linked to the online version of the paper at https://doi.org/10.1530/EC-24-0169.
Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the study reported.
Funding
National Natural Science Foundation of China (81573140); 2022 Multidisciplinary Internal Research Grant at Texas State University; THRC/CHERR Faculty Fellowship Funding.
Ethics approval and consent to participate
This study was registered at www.chictr.org.cn (ChiCTR1900027764) and was approved by the Medical Ethics Committee of the Seventh People’s Hospital affiliated to Shanghai University of Traditional Chinese Medicine (2020-7th-HIRB-016). Written informed consent was obtained from all study participants.
Availability of data and materials
Although the datasets generated during the study are not publicly available due to limitations of ethical approval, researchers can get them from the corresponding author on reasonable request.
Authors’ contribution statement
KY, JZ: study design and concept; TD, LP: programming and manuscript drafting; TD: data analysis and result interpretation; JJ, HW, KY: data collection and critical revision of the manuscript; and all authors: read and approved the final manuscript. None of the authors reported a conflict of interest related to the study.
Acknowledgements
We sincerely thank all the women who participated in this study and their families for their cooperation. We also thank all the students and researchers who participated in the dietary survey. This study could not have been completed without your help.
References
- 1↑
Sweeting A, Wong J, Murphy HR, & Ross GP. A clinical update on gestational diabetes mellitus. Endocrine Reviews 2022 43 763–793. (https://doi.org/10.1210/endrev/bnac003)
- 2↑
Freinkel N, Metzger BE, Phelps RL, Dooley SL, Ogata ES, Radvany RM, & Belton A. Gestational diabetes mellitus. Heterogeneity of maternal age, weight, insulin secretion, HLA antigens, and islet cell antibodies and the impact of maternal metabolism on pancreatic B-cell and somatic development in the offspring. Diabetes 1985 34(Supplement 2) 1–7. (https://doi.org/10.2337/diab.34.2.s1)
- 3↑
Chiefari E, Arcidiacono B, Foti D, & Brunetti A. Gestational diabetes mellitus: an updated overview. Journal of Endocrinological Investigation 2017 40 899–909. (https://doi.org/10.1007/s40618-016-0607-5)
- 4↑
Lowe WL, Scholtens DM, Kuang A, Linder B, Lawrence JM, Lebenthal Y, McCance D, Hamilton J, Nodzenski M, Talbot O, et al.Hyperglycemia and adverse pregnancy outcome follow-up study (HAPO FUS): maternal gestational diabetes mellitus and childhood glucose metabolism. Diabetes Care 2019 42 372–380. (https://doi.org/10.2337/dc18-1646)
- 5↑
Lowe WL, Scholtens DM, Lowe LP, Kuang A, Nodzenski M, Talbot O, Catalano PM, Linder B, Brickman WJ, Clayton P, et al.Association of gestational diabetes with maternal disorders of glucose metabolism and childhood adiposity. JAMA 2018 320 1005–1016. (https://doi.org/10.1001/jama.2018.11628)
- 6↑
He Z, Xie H, Liang S, Tang Y, Ding W, Wu Y, & Ming W-K. Influence of different diagnostic criteria on gestational diabetes mellitus incidence and medical expenditures in China. Journal of Diabetes Investigation 2019 10 1347–1357. (https://doi.org/10.1111/jdi.13008)
- 7↑
Li G, Wei T, Ni W, Zhang A, Zhang J, Xing Y, & Xing Q. Incidence and risk factors of gestational diabetes mellitus: a prospective cohort study in Qingdao, China. Frontiers in Endocrinology 2020 11 636. (https://doi.org/10.3389/fendo.2020.00636)
- 8↑
Chinese Medical Association. Guideline of diagnosis and treatment of hyperglycemia in pregnancy (2022) [part one] (in Chinese). Chinese Journal of Obstetrics and Gynecology 2022 57 3–12. (https://doi.org/10.3760/cma.j.cn112141-20210917-00528)
- 9↑
Huang J, Li Y, Liu M & & Wu Comparison of the 2022 Chinese guidelines for the diagnosis and management of hyperglycemia in pregnancy with the American Diabetes Association guidelines for the diagnosis and management of diabetes mellitus associated with pregnancy (in Chinese). Journal of International Obstetrics and Gynecology 2022 49 691–699. (https://doi.org/10.12280/gjfckx.20220262)
- 10↑
Song C, Li J, Leng J, Ma RC, & Yang X. Lifestyle intervention can reduce the risk of gestational diabetes: a meta-analysis of randomized controlled trials. Obesity Reviews 2016 17 960–969. (https://doi.org/10.1111/obr.12442)
- 11↑
McIntyre HD, Catalano P, Zhang C, Desoye G, Mathiesen ER & & Damm Gestational diabetes mellitus. Nature Reviews Disease Primers 2019 5 47. (https://doi.org/10.1038/s41572-019-0098-8)
- 12↑
Toulis KA, Goulis DG, Kolibianakis EM, Venetis CA, Tarlatzis BC, & Papadimas I. Risk of gestational diabetes mellitus in women with polycystic ovary syndrome: a systematic review and a meta-analysis. Fertility and Sterility 2009 92 667–677. (https://doi.org/10.1016/j.fertnstert.2008.06.045)
- 13↑
Chu SY, Callaghan WM, Kim SY, Schmid CH, Lau J, England LJ, & Dietz PM. Maternal obesity and risk of gestational diabetes mellitus. Diabetes Care 2007 30 2070–2076. (https://doi.org/10.2337/dc06-2559a)
- 14↑
Liu H, Li J, Leng J, Wang H, Liu J, Li W, Liu H, Wang S, Ma J, Chan JC, et al.Machine learning risk score for prediction of gestational diabetes in early pregnancy in Tianjin, China. Diabetes/Metabolism Research and Reviews 2021 37 e3397. (https://doi.org/10.1002/dmrr.3397)
- 15↑
Artzi NS, Shilo S, Hadar E, Rossman H, Barbash-Hazan S, Ben-Haroush A, Balicer RD, Feldman B, Wiznitzer A, & Segal E. Prediction of gestational diabetes based on nationwide electronic health records. Nature Medicine 2020 26 71–76. (https://doi.org/10.1038/s41591-019-0724-8)
- 16↑
Lee SM, Hwangbo S, Norwitz ER, Koo JN, Oh IH, Choi ES, Jung YM, Kim SM, Kim BJ, Kim SY, et al.Nonalcoholic fatty liver disease and early prediction of gestational diabetes mellitus using machine learning methods. Clinical and Molecular Hepatology 2022 28 105–116. (https://doi.org/10.3350/cmh.2021.0174)
- 17↑
Buchanan TA, Xiang AH, & Page KA. Gestational diabetes mellitus: risks and management during and after pregnancy. Nature Reviews. Endocrinology 2012 8 639–649. (https://doi.org/10.1038/nrendo.2012.96)
- 18↑
Du HY, Jiang H, O K, Chen B, Xu LJ, Liu SP, Yi JP, He GS, & Qian X. Association of dietary pattern during pregnancy and gestational diabetes mellitus: a prospective cohort study in Northern China. Biomedical and Environmental Sciences 2017 30 887–897. (https://doi.org/10.3967/bes2017.119)
- 19↑
Tryggvadottir EA, Medek H, Birgisdottir BE, Geirsson RT, & Gunnarsdottir I. Association between healthy maternal dietary pattern and risk for gestational diabetes mellitus. European Journal of Clinical Nutrition 2016 70 237–242. (https://doi.org/10.1038/ejcn.2015.145)
- 20↑
Asadi M, Shahzeidi M, Nadjarzadeh A, Hashemi Yusefabad H, & Mansoori A. The relationship between pre-pregnancy dietary patterns adherence and risk of gestational diabetes mellitus in Iran: a case-control study. Nutrition and Dietetics 2019 76 597–603. (https://doi.org/10.1111/1747-0080.12514)
- 21↑
Ojo O, Ojo OO, Wang X-H, & Adegboye ARA. The effects of a low GI diet on cardiometabolic and inflammatory parameters in patients with type 2 and gestational diabetes: a systematic review and meta-analysis of randomised controlled trials. Nutrients 2019 11. (https://doi.org/10.3390/nu11071584)
- 22↑
Mustad VA, Huynh DTT, López-Pedrosa JM, Campoy C, & Rueda R. The role of dietary carbohydrates in gestational diabetes. Nutrients 2020 12. (https://doi.org/10.3390/nu12020385)
- 23↑
Schoenaker DAJM, Mishra GD, Callaway LK, & Soedamah-Muthu SS. The role of energy, nutrients, foods, and dietary patterns in the development of gestational diabetes mellitus: a systematic review of observational studies. Diabetes Care 2016 39 16–23. (https://doi.org/10.2337/dc15-0540)
- 24↑
Wei Y, & Yang H. Perspectives on diagnostic strategies for hyperglycemia in pregnancy - dealing with the barriers and challenges in China. Diabetes Research and Clinical Practice 2018 145 84–87. (https://doi.org/10.1016/j.diabres.2018.04.005)
- 25↑
International Association of Diabetes and Pregnancy Study Groups Consensus Panel, Metzger BE, Gabbe SG, Persson B, Buchanan TA, Damm PA, Damm P, Dyer AR, Kitzmiler JL, Hod M, et al.International association of diabetes and pregnancy study groups recommendations on the diagnosis and classification of hyperglycemia in pregnancy. Diabetes Care 2010 33 676–682. (https://doi.org/10.2337/dc09-1848)
- 26↑
Wei Y, Yang H, Zhu W, Yang H, Li H, Yan J, & Zhang C. International Association of Diabetes and Pregnancy Study Group criteria is suitable for gestational diabetes mellitus diagnosis: further evidence from China. Chinese Medical Journal 2014 127 3553–3556. (https://doi.org/10.3760/cma.j.issn.0366-6999.20140898)
- 27↑
Qin C, Guo Y, Pei P, Du H, Yang L, Chen Y, Shen X, Shi Z, Qi L, Chen J, et al.The relative validity and reproducibility of food frequency questionnaires in the China Kadoorie biobank study. Nutrients 2022 14. (https://doi.org/10.3390/nu14040794)
- 28↑
Liu Y-H, Lu L-P, Yi M-H, Shen C-Y, Lu G-Q, Jia J, & Wu H. Study on the correlation between homocysteine-related dietary patterns and gestational diabetes mellitus:a reduced-rank regression analysis study. BMC Pregnancy and Childbirth 2022 22 306. (https://doi.org/10.1186/s12884-022-04656-5)
- 29↑
Mak JKL, Pham NM, Lee AH, Tang L, Pan X-F, Binns CW, & Sun X. Dietary patterns during pregnancy and risk of gestational diabetes: a prospective cohort study in Western China. Nutrition Journal 2018 17 107. (https://doi.org/10.1186/s12937-018-0413-3)
- 30↑
Alba AC, Agoritsas T, Walsh M, Hanna S, Iorio A, Devereaux PJ, McGinn T, & Guyatt G. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. JAMA 2017 318 1377–1384. (https://doi.org/10.1001/jama.2017.12126)
- 31↑
Wei Y, He A, Tang C, Liu H, Li L, Yang X, Wang X, Shen F, Liu J, Li J, et al.Risk prediction models of gestational diabetes mellitus before 16 gestational weeks. BMC Pregnancy and Childbirth 2022 22 889. (https://doi.org/10.1186/s12884-022-05219-4)
- 32↑
Zheng T, Ye W, Wang X, Li X, Zhang J, Little J, Zhou L, & Zhang L. A simple model to predict risk of gestational diabetes mellitus from 8 to 20 weeks of gestation in Chinese women. BMC Pregnancy and Childbirth 2019 19 252. (https://doi.org/10.1186/s12884-019-2374-8)
- 33↑
Wu Y, Ma S, Wang Y, Chen F, Zhu F, Sun W, Shen W, Zhang J, & Chen H. A risk prediction model of gestational diabetes mellitus before 16 gestational weeks in chinese pregnant women. Diabetes Research and Clinical Practice 2021 179 109001. (https://doi.org/10.1016/j.diabres.2021.109001)
- 34↑
Wu Y-T, Zhang C-J, Mol BW, Kawai A, Li C, Chen L, Wang Y, Sheng J-Z, Fan J-X, Shi Y, et al.Early prediction of gestational diabetes mellitus in the chinese population via advanced machine learning. Journal of Clinical Endocrinology and Metabolism 2021 106 e1191–e1205. (https://doi.org/10.1210/clinem/dgaa899)
- 35↑
Santos S, Voerman E, Amiano P, Barros H, Beilin LJ, Bergström A, Charles MA, Chatzi L, Chevrier C, Chrousos GP, et al.Impact of maternal body mass index and gestational weight gain on pregnancy complications: an individual participant data meta-analysis of European, North American and Australian cohorts. BJOG 2019 126 984–995. (https://doi.org/10.1111/1471-0528.15661)
- 36↑
Kim SY, England L, Wilson HG, Bish C, Satten GA, & Dietz P. Percentage of gestational diabetes mellitus attributable to overweight and obesity. American Journal of Public Health 2010 100 1047–1052. (https://doi.org/10.2105/AJPH.2009.172890)
- 37↑
Li Y, Ren X, He L, Li J, Zhang S, & Chen W. Maternal age and the risk of gestational diabetes mellitus: a systematic review and meta-analysis of over 120 million participants. Diabetes Research and Clinical Practice 2020 162 108044. (https://doi.org/10.1016/j.diabres.2020.108044)
- 38↑
Yen I-W, Lee C-N, Lin M-W, Fan K-C, Wei J-N, Chen K-Y, Chen S-C, Tai Y-Y, Kuo C-H, Lin C-H, et al.Overweight and obesity are associated with clustering of metabolic risk factors in early pregnancy and the risk of GDM. PLoS One 2019 14 e0225978. (https://doi.org/10.1371/journal.pone.0225978)
- 39↑
Wang N, Guo H, Jing Y, Song L, Chen H, Wang M, Gao L, Huang L, Song Y, Sun B, et al.Development and validation of risk prediction models for gestational diabetes mellitus using four different methods. Metabolites 2022 12. (https://doi.org/10.3390/metabo12111040)
- 40↑
Luo J, Wang X, Yuan L, & Guo L. Association of thyroid disorders with gestational diabetes mellitus: a meta-analysis. Endocrine 2021 73 550–560. (https://doi.org/10.1007/s12020-021-02712-2)
