Records were obtained for 25,789 births; 25,543 records were included in the analysis after exclusion of pregnancies resulting in miscarriage (n = 59) or termination (n = 65) and records with no birthweight information (n = 3), duplicate data (n = 20) and data consistent with overt diabetes (RPG ≥11.1 mmol/l at booking; n = 99). Of these, only 17,736 pregnancies had a documented RPG at booking. Those without RPG measurements recorded have been described more thoroughly elsewhere [17].
Baseline characteristics of women are described according to the presence or absence of GDM according to the IADPSG criteria (Table 1). As expected, women with GDM present had higher rates of obesity (BMI ≥30 kg/m2), higher age at delivery and were more likely to give birth to a macrosomic or large-for-gestational-age (LGA) infant compared with women who did not have GDM.
Table 1 Characteristics of all pregnancies and those identified as GDM-positive and GDM-negative according to the IADPSG criteria
The ability of the RPG to predict GDM was tested using ROC curves (Fig. 1). RPG was able to predict GDM according to the IADPSG (n = 1,181 positive diagnoses, n = 884 with RPG; AUC 0.81; 95% CI 0.80, 0.83), NICE 2015 (n = 1,055 positive diagnoses, n = 806 with RPG; AUC 0.81; 95% CI 0.79, 0.83), WHO 1999 (n = 1,016 positive diagnoses; n = 775 with RPG; AUC 0.81; 95% CI 0.79, 0.83) and Modified WHO 1999 (n = 1,025 positive diagnoses; n = 782 with RPG; AUC 0.80; 95% CI 0.78, 0.83) criteria.
Using a cut-off value of RPG ≥7.5 mmol/l (135 mg/dl), which produced best overall performance of sensitivity and specificity, RPG was able to predict GDM diagnosis using IADPSG (sensitivity 0.70, specificity 0.90), NICE 2015 (sensitivity 0.69, specificity 0.89), WHO 1999 (sensitivity 0.69, specificity 0.89) and Modified WHO 1999 (sensitivity 0.69, specificity 0.89) criteria. In this dataset of 17,736 pregnancies with RPG data, 15,396 women (86.81%) fell below this threshold and 2,340 fell above the threshold (13.19%).
As the clinical value of RPG would be in excluding women who do not need further investigation for GDM, a higher cut-off value of ≥8.5 mmol/l (153 mg/dl) was also assessed to maximise specificity while providing acceptable sensitivity. At this level, RPG was able to predict GDM according to IADPSG (sensitivity 0.43, specificity 0.97), NICE 2015 (sensitivity 0.42, specificity 0.96), WHO 1999 (sensitivity 0.42, specificity 0.96) and Modified WHO 1999 (sensitivity 0.42, specificity 0.96) criteria. In this dataset of 17,736 pregnancies with recorded RPG, 16,789 women fell below this threshold and 947 fell above the threshold.
As the range of RPG values was considerable in women who later developed GDM (see Fig. 1e), a cut-off point of around ≥4.7 mmol/l (85 mg/dl) was required to give a sensitivity of 90% using any diagnostic criteria. Of the 3,863 women who had values <4.7 mmol/l, 68 (1.76%) were eventually diagnosed with GDM according to IADPSG criteria.
Theoretically, adopting an RPG screening policy in this population using IADPSG criteria with a cut-off point of ≥7.5 mmol/l (135 mg/dl) would have identified 2,340 women as being at high risk of GDM (Fig. 2; 615 [26.3%] were later found to be positive for GDM). This screening policy would also have identified 15,396 women as being at low risk of GDM (of whom 15,127 [98.3%] were negative for GDM). However, this low-risk group contained 269 women who were confirmed positive for GDM later in pregnancy (30.4% of cases of GDM). Interestingly, our data suggests that these 269 women might not have been readily identified using risk-factor-based screening methods as approximately 38.7% were of normal pre-pregnancy BMI (<25 kg/m2). They had a 33.7% risk of having an LGA infant (Fig. 2).
The use of RPG at booking compared favourably with other screening strategies in current clinical use (Fig. 3a, b). For example, maternal pre-pregnancy BMI and maternal age were both inferior at predicting GDM using the IADPSG criteria (n = 1,181 positive diagnoses, n = 884 with RPG; BMI AUC 0.65, 95% CI 0.63, 0.67; age AUC 0.60, 95% CI 0.59, 0.62). Maternal age ≥30 years predicted IADPSG GDM with a sensitivity of 74.8% and a specificity of 38.9%. Maternal pre-pregnancy BMI ≥30 k/m2 predicted IADPSG GDM with a sensitivity of 0.30 and a specificity of 0.87. Combining the risk factors age and BMI with RPG did not improve the overall predictive ability compared with using RPG alone (Fig. 3c–f) when using thresholds of RPG ≥7.5 mmol/l, age ≥30 years and BMI ≥30 kg/m2. However, combining age and BMI (Fig. 3c), RPG and age (Fig. 3d) or RPG and BMI (Fig. 3e) gave an improvement in test sensitivity to 0.83–0.95, but at the cost of reducing the overall ROC AUC.