This analysis is based on prospective data collected between 2017 and 2021 in the Mutaba’ah Study, the most extensive multicenter mother and child cohort study in the UAE. It recruits and follows up mother–baby pairs from the Emirati population in Al Ain city, which has the largest proportion of Emiratis in the country. The eligibility criteria for recruitment into the Mutaba’ah study include being an 18-year-old and above pregnant woman from the Emirati population. The study was approved by the United Arab Emirates University Human Research Ethics Committee (ERH-2017-5512) and the Abu Dhabi Health Research and Technology Ethics Committee (DOH/CVDC/2022/72). All participants provided written informed consent.

For this study, we included the Mutaba’ah study participants with singleton pregnancies screened with a 75-g 2-h oral glucose tolerance test (OGTT) at 24 to 32 weeks between May 2017 and March 2021. Those with at least two OGTT readings were included, and those with preexisting or newly diagnosed diabetes (fasting plasma glucose (FPG) ≥7 mmol/L and/or 2-h OGTT ≥ 11·1 mmol/L) were excluded.

Details of the Mutaba’ah study, including the recruitment process, have been published elsewhere (29). Trained research assistants approached eligible pregnant women during their antenatal care visits to any of the participating hospitals. The assistants administer the questionnaires during the first trimester. After delivery, clinical data, including anthropometric measurements, OGTT results, other laboratory results, and data on the index pregnancy, delivery, and outcomes, were extracted from the medical records. The extraction was automated by the IT department of the hospital.

GDM screening is standardized across the three recruiting hospitals as they all follow the Abu Dhabi Department of Health (DOH) guidelines. Standard quality is achieved through internal and external quality control. All pregnant women during routine antenatal care are offered universal GDM screening with 75-g 2-h OGTT at 24 to 28 weeks. Before 24 weeks of gestation, they undergo screening for preexisting diabetes using fasting plasma glucose (FPG) or HbA1C tests, and those positive are being co-managed with endocrinologists. The OGTT procedures are similar across the hospitals. The pregnant women are requested to fast for 8 to 10 h from the night before testing. A venous blood sample is drawn by an expert phlebotomist using standard practice. The women are then given 75 g oral glucose (Trutol, 10 fluid ounces (296 ml) dextrose beverage, Nerl Diagnostic, Rhode Island, USA). After 1 h, a venous sample is taken, and again after 2 h. The samples taken are immediately processed to avoid preanalytic glycolysis. Glucose analysis is done using the enzymatic reference method with hexokinase (Hexokinase G6PDH/UV Roche Cobas^® c500/c303, CA, USA).

This study assessed four GDM diagnostic criteria, namely, the IADPSG, NICE 2015, WHO 1999, and ADIPS 1998. These criteria are among the commonly used by doctors in the UAE as highlighted by different studies (24, 25). Standard definitions of the criteria are given in Supplementary Table S1 .

A sample size of 1,438 pregnancies will allow the detection of a true relative risk of approximately 1·73 (30) of developing large for gestational age (LGA) babies in GDM patients (using IADPSG) with 80% power and at 95% confidence level, accounting for a 20% attrition rate (Fleiss with CC).

The outcomes assessed were large for gestational age (LGA) and a composite outcome (three maternal and three newborn adverse outcomes). The LGA was defined as a newborn’s birth weight above the 90th percentile for gestational age and sex. The LGA was determined using the reference from the US Centers for Disease Control and Prevention growth chart and the method described by Vidmar S. I. and colleagues in 2013 (31). The composite outcome was defined as having one or more from LGA, neonatal intensive care unit (NICU) admission, abnormal APGAR score, caesarean delivery, premature delivery, and preeclampsia. These were selected based on the evidence from a meta-analysis of adverse perinatal outcomes among more than 7.5 million pregnancies and the most commonly found in the Arab population (32, 33). See Supplementary Material (p 2) for the operational definition of each outcome variable.

We summarized continuous variables using means with standard deviations (SD) and categorical variables using frequencies with percentages (%). Logistic regressions were conducted to assess the associations between the four GDM diagnostic criteria and the adverse perinatal outcomes (LGA and composite outcome). Generalized linear models (GLM) were used to estimate the adjusted relative risks (RR) and risk differences (RD) with 95% confidence intervals for these associations. Logistic regressions were also conducted to assess the associations between OGTT (FPG, 1 h, and 2 h) readings (used as continuous variables) and the outcomes. Regression results were reported using odds ratios with 95% confidence intervals. For the OGTT readings, we calculated odds ratios per unit change (1 mmol/L) in fasting, 1-h, and 2-h plasma glucose levels.

For the multiple regressions, two models (model 1 and model 2) were used. Model 1 was constant for all outcomes and adjusted for baseline maternal characteristics, including age, gravidity, body mass index (BMI) at booking, education, employment, family history of type 2 DM, previous GDM, and study center. Model 2 adjusted for model 1 plus lifestyle factors and other factors. For LGA, model 2 included model 1 plus maternal smoking, passive smoking, physical activity, antepartum hemorrhage, and previous macrosomia. For the composite outcome, a family history of hypertension was added to this list. See Supplementary Material (p 2) for the operational definition of each exposure variable. Missing data in this study were handled using multiple imputations by the chained equation (MICE) method with the number of imputed datasets specified at m = 100.

Following multiple regression with the OGTT results, post-estimation analysis was conducted and expressed using the marginsplot to show the predictive margins for the fitted, adjusted model, which was then used to make predictions. The method adopted by the UCLA Statistical Consulting Group (34), which utilizes multiple imputed data, was used. New cutoff values were identified using the adjusted predictions. New GDM diagnostic criteria was proposed for this population based on these cutoffs.

The new criteria was compared with two others (IADPSG and NICE 2015) in terms of GDM cumulative incidence, adjusted odds ratio (aOR) for outcomes with 95% confidence intervals, and agreements using kappa statistics. These two criteria were selected because they are commonly used in the UAE and were also found to be the most inclusive in GDM diagnosis for this population (3).

Further evaluation of the new criteria and the IADPSG was done using:

The AUC assessed the performance and discrimination of the IADPSG and the new criteria in two separate models containing established risk factors (models 1 and 2) of the outcomes. Results were presented as a graph using the ROC curves and showing the AUC with 95% confidence intervals for both models. AUC values of 0·9–1·0 show that tests have excellent quality; 0·8–0·9 very good quality; 0·7–0·8 good (acceptable) quality; 0·6–0·7 satisfactory; 0·5–0·6 unsatisfactory. The test of equality for the two models was conducted using the chi-square test. P-value was significant at <0·05.

The continuous NRI was used to assess the clinical utility of the new criteria by assessing the incremental value in its risk predictions of the outcomes. Results were presented as the proportions of reclassified cases and non-cases based on risk predictions with their NRI values. The IDI showed the mean difference in discrimination slopes of the two models (extended and traditional) for both criteria. Results were reported as Absolute IDI (standard errors—SE) and Relative IDI (%) (35).

All analyses were conducted using STATA statistical software version 16·1 (StataCorp LLC, College Station, TX, USA).

The mean maternal age of participants at booking was 30.4 ± 6.0 years, and the mean booking body mass index (BMI) was 27.7 ± 5.6 kg/m². 53.4% of the participants had above high school education, and 30.7% were employed. Their mean fasting plasma glucose (FPG) was 4.6 ± 0.4 mmol/L, 1-h OGTT was 8.0 ± 1.9 mmol/L, and 2-h OGTT was 6.5 ± 1.6 mmol/L. Table 1 shows the descriptives for maternal characteristics.

Table 2 summarizes the adverse perinatal outcomes among the participants. 17.4% of the participants had large for gestational age (LGA) babies, and 42.1% had the composite outcome (at least one of the six specified outcomes). The descriptives of the six specified outcomes (LGA, NICU admission, abnormal APGAR score, caesarean delivery, premature delivery, and preeclampsia) are highlighted in Supplementary Table S2 .

Table 3 shows the associations between the GDM diagnostic criteria (IADPSG, NICE 2015, WHO 1999, and ADIPS 1998) and the outcomes. The IADPSG criteria had the highest odds ratio for the LGA (aOR 1.77, 95% CI 1.36–2.29) compared with the remaining three criteria after adjusting for models 1 and 2. The IADPSG criteria also had the highest adjusted odds ratio for the composite outcome (aOR 1·49, 95% CI 1.19–1.86).

Risk analysis results are reported in Supplementary Table S3 , and the outcomes’ adjusted relative risks (RR) and risk differences (RD) were compared for the four criteria. It confirmed the IADPSG criteria as the strongest predictor for LGA (aRR 1·55, 95% CI 1.27–1.88) and composite outcome (aRR 1.22, 95% CI 1.09–1.35). The analysis showed that the excess risk of LGA identified by the IADPSG criteria was 9%, whereas it was 6% for the other three criteria. Similarly, for the composite outcome, the RD was highest with the IADPSG criteria.

Table 4 shows the adjusted odds ratio denoting the strength of association between the FPG, 1-h, and 2-h OGTT results and the outcomes (LGA and composite outcome). FPG had the strongest positive association with both outcomes compared with 1-h and 2-h OGTT. Following adjustments for models 1 and 2, only the association between FPG and the composite outcome remained significant (aOR 1.67, 95% CI 1.21–2.28).

Only FPG showed a significant association with the composite outcome after adjusting for models 1 and 2 ( Table 4 ); hence, post-estimation was limited to its analysis. Figure 1 shows the post-estimation prediction graphed by marginsplot showing the fitted, adjusted model for FPG and composite outcome. The graph shows that the association with the composite outcome starts to become positive at a maternal fasting glucose level of 5.00 mmol/L (aOR—1·01 approximately). At the recommended aOR of 1.75 (as specified by the IADPSG consensus panel), this population’s mean maternal FPG level was approximately 6.00 mmol/L.

Based on the above prediction, we developed new GDM diagnostic criteria for the Emirati population using the FPG level of 6.0 mmol/L. For the 1-h and 2-h OGTT, thresholds of the IADPSG criteria were maintained, as it was found to be the best predictor of adverse perinatal outcomes in this population compared with other existing criteria. The new criteria was defined as maternal fasting plasma glucose (FPG) of ≥6.0 mmol/L and/or 1-hr OGTT of ≥10.0 mmol/L and/or 2-hr OGTT of ≥8·5 mmol/L following the 75-g 2-h oral glucose tolerance test.

Our study’s strengths include being multicentered and conducted in a large Emirati population of the UAE, thereby increasing its generalizability and power. Rigorous methodological approaches were employed to ensure good internal validity. Regression models addressed potential predictors while keeping the bias/variance issue in mind. Finally, risk and prediction analyses provided a more relevant result for translation into clinical practice.

Limitations in this study include using non-probability (consecutive) sampling, which might affect the representativeness of our sample. Nevertheless, this issue was mitigated by the multicentered nature of our study. We did not assess all adverse perinatal outcomes due to the unavailability of the data. However, we included the primary GDM adverse perinatal outcomes expected in this population (49). The use of medical records for some variables provided incomplete data; however, we conducted a missing data analysis as described in the methods section.

Firstly, our study highlighted that the IADPSG was the best predictor of adverse perinatal outcomes out of the four commonly used GDM diagnostic criterion (IADPSG, NICE 2015, WHO 1999, and ADIPS 1998) in the UAE. Secondly, from the study data, new evidence-based GDM diagnostic criteria was developed based on the risk of adverse perinatal outcomes, which we found to be a more optimal diagnostic tool in the Emirati population than the other criteria. The new criteria could improve GDM care and reduce the burden of its perinatal complications better than the current clinical practice in this population. Following clinical trials and cost-effectiveness studies in multiethnic settings, the new criteria could be adopted widely. A multi-sectoral approach is needed to ensure the translation of this research into practice.