Sensitivity, specificity, and diagnostic accuracy of WHO 2013 criteria for diagnosis of gestational diabetes mellitus in low risk early pregnancies: international, prospective, multicentre cohort study

Participants

In this multicentre prospective study on diagnostic accuracy, we included participants aged 18-45 years presenting in early pregnancy to six university or cantonal departments in Austria, Germany, and Switzerland. All participants with singleton pregnancies who wished to undergo first trimester screening received an information leaflet about the study and those interested were recruited. The first trimester scan confirmed or corrected gestational age. Participants were excluded if they had pre-existing diabetes, chronic infectious diseases, such as hepatitis or HIV infection, chronic liver, kidney, or heart disease, if they had previous bariatric surgery, or were receiving metformin or acetylsalicylic acid because of a history of hypertensive disease in a previous pregnancy. Further exclusion criteria were fetal genetic, chromosomal, or morphological abnormalities which required further clarification.

Sample collection

A sampling protocol was distributed to all participating centres. The main focus of the protocol was on the type of tubes for the OGTT, and the aliquots, times, and details of centrifugation for aliquoting the serum and plasma samples.

Test methods: oral glucose tolerance test

Participants were universally screened with the 75 g OGTT at 12-15 weeks of gestation. Participants were advised to eat and drink normally for the two days before the test but not to eat, drink (only a few sips of water), or smoke on the morning of the test. No physical activity was allowed during the test. The first measurement of serum glucose levels was performed after an overnight fasting period of at least 10 hours, between 8 am and 12 pm. Then, after intake of the 75 g glucose load in 250-300 mL of water, blood samples were taken one hour and two hours postload for determination of glucose levels. To minimise the effects of glycolysis in vivo, all centres were asked to send the samples to the laboratory directly. Only natriumfluorid tubes with citrate buffer were used for the study.

Plasma levels of glucose were measured by an automated colorimetric enzymatic method with the hexokinase-glucose-6-phosphate-dehydrogenase method (GLUC3 test by Roche or the Dimension Vista Hexokinase test) and analysed by the Hitachi-Roche cobas modular analyser (Roche Diagnostics, Rotkreuz, Switzerland) or Siemens Dimension Vista analyser (Siemens Healthcare, Aarau, Switzerland). Both tests have a imprecision value of <1.25% and bias value of <1.23%. The tests are ISO17025 accredited and include the need for external quality control (inter-laboratory comparisons). All participting laboratories were ISO 17025:2017 or ISO15189 accredited.

Participants and healthcare providers were blinded to the results of the early 75 g OGTT. Values were unblinded by hospital laboratories if the fasting plasma glucose concentration was ≥7.0 mmol/L (≥126 mg/dL) or the random or two hour value was ≥11.1 mmol/L (≥200 mg/dL), defined as pre-existing diabetes by the American Diabetes Association.16 The diagnosis of pre-existing diabetes mellitus had to be confirmed by raised levels of glycated haemoglobin A_1c of ≥6.5%. Participants with plasma concentrations of glucose <2.5 mmol/L (≤45 mg/dL) were unblinded for further clarification and underwent a second reference standard 75 g OGTT at 24-28 weeks of gestation.

The WHO 2013 criteria (fasting plasma glucose concentration ≥5.1 mmol/L (≥92 mg/dL), and ≥10.0 mmol/L (≥180 mg/dL) for one hour and >8.5 mmol/L (≥153 mg/dL) for two hour postload plasma glucose concentrations) were used to diagnose gestational diabetes mellitus if at least one value was abnormal.4 Participants with a diagnosis of gestational diabetes mellitus were treated according to the recommendations of the American Diabetes Association17 and, if targets could not be reached in 1-2 weeks after changes in lifestyle, insulin was started to control hyperglycaemia.

Prenatal data recording

Personal and family history, height, weight, gravidity, parity, blood pressure, urine dipstick, and medical complications were recorded for each study participant. Also, prenatal care visits, inpatient stays, delivery details, and postpartum visits were recorded in a clinical data management application (secuTrial) maintained by the Clinical Trial Unit, Basel.

Study outcome

We determined the predictability of gestational diabetes mellitus with a 75 g OGTT in early pregnancy based on the WHO 2013 criteria. We tested several proposed cut-off values according to previously published data for fasting (≥5.1 mmol/L (≥92 mg/dL), ≥5.3 mmol/L (≥95 mg/dL), ≥5.7 mmol/L (≥103 mg/dL), and ≥6.1 mmol/L (≥110 mg/dL)) and for postprandial glucose values at one hour (≥8.9 mmol/L (≥160 mg/dL) and ≥10.0 mmol/L (≥180 mg/dL)) and at two hours (≥7.1 mmol/L (≥128 mg/dL), ≥7.5 mmol/L (≥135 mg/dL), ≥7.8 mmol/L (≥140 mg/dL), and ≥8.5 mmol/L (≥153 mg/dL)).14 18 19 We chose a sample size of 748 participants (assuming a prevalence of 10.9% (n=65) for gestational diabetes mellitus) with a dropout rate of 15% to predict the development of gestational diabetes mellitus with an early 75 g OGTT with or without additional biomarkers. The published study protocol provides full details.15

Sample size considerations

We calculated sample size based on the area under the curve of a newly proposed screening method combining the 75 g OGTT with new biomarkers, such as glycosylated fibronectin. The power calculation was performed with a proposed true area under the curve of 0.9 with a lower boundary of 0.8 (95% confidence interval >0.8) which gave a power of 90% and an α level of 5%. Offsetting a dropout rate of 15%, the sample size was 748.15 The study was not powered to determine the diagnostic accuracy of specific cut-off values.

Statistical analysis

Summary statistics of patient characteristics are reported as mean (standard deviation) or median (interquartile range) for continuous variables, and as frequencies and percentages for categorical variables. The analysis set included only participants with complete early and late OGTTs (n=636) and hence there were no missing values in the six variables relevant for this analysis. For each diagnostic test we reported sensitivity, specificity, overall diagnostic accuracy (percentage of correct diagnoses), and positive and negative predictive values (with 95% confidence intervals). Positive and negative predictive values were derived from the observed prevalence in the analysis set. Positive and negative likelihood ratios, and diagnostic odds ratios, were estimated as prevalence independent measures. We also reported summary statistics for early and late OGTTs dependent on the development of gestational diabetes mellitus. Correlations between early and late OGTTs were assessed by Spearman’s rank correlation coefficient.

We made no adjustment for multiplicity, unless otherwise indicated in the manuscript. Heterogeneity between centres was evaluated with a meta-analysis for the WHO 2013 criteria, and forest plots for the diagnostic measures were derived. Models were fitted with centre as a random effect (random intercept); logistic regression was used for proportions (sensitivity, specificity, accuracy, and positive and negative predictive values), and the Mantel-Haenszel method was used for positive and negative likelihood ratios. For the generalised linear mixed models, no weights for centres are provided with this approach. Statistical analyses were performed with R (version 4.0.3, 2020) and related packages.20

Patient and public involvement

Patients or the public were not involved in the design of the study, interpretation of the results, or writing of the manuscript. A lay summary of the results will be published on the funders’ websites. The main findings of the cohort study will be distributed to our study participants in a research newsletter.

Characteristics of participants

Among 829 eligible pregnant individuals recruited for the study, 636 were included in the analysis with a complete early and late OGTT data set (figure 1) between 1 May 2016 and 31 January 2019. Seventy four (12%) of the 636 participants developed gestational diabetes mellitus diagnosed by a late OGTT. The prevalence of gestational diabetes mellitus in the six university or cantonal centres varied depending on sample size (online supplemental table S4). Table 1 summarises the characteristics of the study cohort.

Request permission

Individuals with gestational diabetes mellitus tended to have a higher median pre-gravid body mass index (25.41, interquartile range 22.04-30.5) than those with no gestational diabetes mellitus (22.68, 20.52-25.66), and a higher body mass index at the study visit (26.17, 23.1-30.97 v 23.61, 21.23-26.36). The two groups were comparable for maternal age, gravidity, parity, and gestational age at study visit.

Diagnostic performance of early oral glucose tolerance test

Table 2 summarises glucose concentrations from the early and late OGTTs. Forty six of 636 (7.3%) OGTT results were abnormal based on the WHO 2013 criteria in early pregnancy. Participants who had a diagnosis of gestational diabetes mellitus later, at 24-28 weeks of gestation, tended to have higher glucose levels in the early OGTT and markedly higher values in the late OGTT than those with no gestational diabetes mellitus. Fasting plasma glucose concentrations did not differ in the early OGTT (median 4.4, interquartile range 4.2-4.6 mmol/L) versus the late OGTT (4.4, 4.1-4.6 mmol/L) in the non-gestational diabetes mellitus group. Only 2.5% of participants (14/562) showed glucose concentrations ≥5.1 mmol/L (≥92 mg/dL) in early pregnancy in the non-gestational diabetes mellitus group compared with 20.3% (15/74) in the gestational diabetes mellitus group.

Diagnostic values in the late OGTT in the gestational diabetes mellitus group were low (median fasting plasma glucose 5.1, interquartile range 4.7-5.3 mmol/L; one hour postload 10.0, 8.1-10.7 mmol/L; two hour postload 7.4, 6.0-8.3 mmol/L), in the range of milder degrees of gestational diabetes mellitus. Glucose levels from the early and late OGTTs largely overlapped between the non-gestational diabetes mellitus and gestational diabetes mellitus groups, and the overlap was more pronounced for fasting plasma glucose concentrations than for the one and two hour postload glucose values (figure 2).

Request permission

Request permission

Correlation of early versus late OGTT was moderate (figure 3; Spearman’s rank correlation: r=0.51 for fasting plasma glucose concentions, r=0.55 for one hour postload, and r=0.55 for two hour postload glucose concentrations). Twenty six of 74 participants had a diagnosis of gestational diabetes mellitus (35.1%) by an early OGTT with the WHO 2013 criteria (n=15 by fasting value of 5.1 mmol/L and n=11 by postprandial levels of 10.0 (n=9) and 8.5 mmol/L (n=9)). Twenty participants without the later diagnosis of gestational diabetes mellitus had abnormal values in early pregnancy (n=14 with fasting values >5.1 mmolL/L). Online supplemental table S3 shows the cross tables and diagnostic statistics.

The WHO 2013 criteria (the standard for screening for gestational diabetes mellitus at 24-28 weeks of gestation) showed low sensitivity (0.35, 95% confidence interval 0.24 to 0.47) and high specificity (0.96, 0.95 to 0.98) in early pregnancy. These results gave a high overall accuracy of 0.89 (0.87 to 0.92), with a positive predictive value of 0.57 (0.41 to 0.71), negative predictive value of 0.92 (0.89 to 0.94), positive likelihood ratio of 9.97 (5.81 to 16.78), negative likelihood ratio of 0.67 (0.57 to 0.80), and diagnostic odds ratio of 14.68 (7.64 to 28.21).

Online supplemental figure S4 shows the receiver operator characteristic curves with corresponding area under the curves for glucose concentrations. Based on only fasting glucose values and increasing the cut-off to 5.3 mmol/L and 5.7 mmol/L resulted in higher positive predictive values. Because only one of the three participants with a fasting glucose concentration ≥6.1 mmol/L developed gestational diabetes mellitus later in pregnancy, however, increasing the cut-off for the fasting glucose value to 6.1 mmol/L gave a low positive predictive value of 0.33. Based on all three values and lowering only the postload levels (cut-off values of 5.1, 8.9, and 7.8 mmol/L) increased sensitivity (0.53, 95% confidence interval 0.41 to 0.64) and negative predictive value (0.94, 0.91 to 0.95) but decreased specificity (0.91, 0.88 to 0.93) and positive predictive value (0.42, 0.32 to 0.53), with a false positive rate of 9%. Online supplemental figure S5A-C shows the box plots of the OGTTs for the six university or cantonal centres. To deal with potential heterogenity between centres, we performed meta-analyses and produced forest plots for the diagnostic measures derived from the WHO 2013 criteria (online supplemental figure S6).

Missing values and incomplete oral glucose tolerance tests

Thirty eight participants had an incomplete late OGTT and therefore a final diagnosis could not be made. Online supplemental table S5 provides summary statistics of the available glucose measurements for these 38 participants. A sensitivity analysis (adding all participants with missing values or incomplete OGTTs to the full analysis set with complete late OGTTs) was performed to derive a final diagnosis. This approach resulted in an additional 27 participants in the sensitivity analysis (663 compared with 636 in the full analysis set) and two more participants with gestational diabetes mellitus (76 v 74 in the full analysis set). Diagnostic performance measures were estimated based on the respective available data for each measurement time (totals differ because of varying patterns of missingness; online supplemental tables S5-S7). The results of the sensitivity analysis were consistent with the full analysis set.

Principal findings

In our observational multicentre study, we tested the WHO 2013 criteria prospectively during screening in the first trimester, at 12-15 weeks of gestation, in a low risk population. In the cohort with a prevalence of gestational diabetes mellitus of 12%, the WHO criteria had low sensitivity (0.35, 95% confidence interval 0.24 to 0.47) and high specificity (0.96, 0.95 to 0.98), giving high overall accuracy (0.89, 0.87 to 0.92). In a previous study, the WHO 2013 criteria were tested at a mean of 15.2±3 weeks of gestation retrospectively in a high risk cohort with obesity (body mass index ≥29) for the DALI (Vitamin D and Lifestyle Intervention for Gestational Diabetes Mellitus Prevention) study, and interventions of healthy eating or physical activity, or both, were implemented afterwards.21 In the DALI cohort of 1035 participants, gestational diabetes mellitus was diagnosed by a 75 g OGTT in early pregnancy in 22.9% of participants. Of these early abnormal OGTTs, 78.5% had abnormal fasting plasma glucose concentrations of ≥5.1 mmol/L. The researchers did not compare the results of both OGTTs, however, and the results of the first OGTT were not blinded to participants and healthcare providers. Therefore, the late OGTTs might have been influenced by the interventions implemented after the early OGTT. Early and late OGTTs were compared prospectively in 146 individuals with a diagnosis of early onset gestational diabetes mellitus in Japan; 47% (69/146) had normal OGTTs at 24-28 weeks of gestation, indicating a high false positive rate.22 An early 75 g OGTT based on the WHO 2013 criteria was performed in 378/1401 early pregnancies (29.6%) and 170 (12.1%) had pathological results.23 In those with a diagnosis of gestational diabetes mellitus, early treatment was started immediately and hence the OGTT results in early and late pregnancy could not be compared.

Setting the cut-off value for fasting plasma glucose concentrations at ≥5.1 mmol/L might raise concerns of potential overdiagnosis, because fasting plasma glucose levels decrease until 20 weeks of gestation.12 We showed that the results of early and late OGTTs, especially in the non-gestational diabetes mellitus group, were comparable and only 2.5% of participants had a fasting plasma glucose level above this threshold in the non-gestational diabetes mellitus group versus 20.3% in the gestational diabetes mellitus group. Increasing the cut-off value for fasting glucose slightly improved test performance (positive predictive value was increased although the negative predictive value was only slightly reduced) and effectiveness (diagnostic odds ratio). The best approach in our cohort was to lower the one hour postload cut-off values to ≥8.9 mmol/L and the two hour postload cut-off values to ≥7.8 mmol/L, which gave a detection rate of 53% at a false positive rate of 10%. We are not aware of any study comparing the results of the 75 g OGTT in early and late pregnancy in a setting without increased risks. Studies reporting screening in the first trimester were usually in individuals with a high risk of developing gestational diabetes and were based on fasting plasma glucose and postprandial glucose concentrations as a screening method,24 thereby potentially missing a substantial proportion of gestational diabetes mellitus in the general population.

The 2013 WHO cut-off values were derived from the HAPO study published in 2008.3 The HAPO study reported a linear relation between maternal fasting and postload glucose values and perinatal adverse outcomes, such as birth weight >90th centile, levels of C peptide in cord blood >90th centile, and neonatal body fat >90th centile. No one glucose value was strongly correlated with the outccomes and no one value was superior to others in predicting a diagnosis of gestational diabetes mellitus. The defined cut-off values represent the average glucose values at which the odds of developing these perinatal adverse outcomes were increased by 1.75-fold, based on fully adjusted logistic regression models. Only one cut-off value was needed for a positive result on screening and to diagnose gestational diabetes mellitus. The no threshold effect of the 75 g OGTT lowers reproducibility, however, which could lead to missclassification by the reference test. Also, many factors can influence the reproducibility and accuracy of tests before, during, and after analysis (eg, time of fasting maternal diet, smoking, excercise or stress, type of sample, collection tube, storage, and transportation).25 26 A Chinese study showed overall reproducibility of only 65.6% in men and non-pregnant women between two OGTTs performed in a six week interval.27 Reproducibility could not be improved even in the high risk group with increased levels of glycated haemoglobin A_1c, high body mass index, or high waist-to-hip ratio. Only one small study reported low overall reproducibility (74.2%) with a 75 g OGTT in pregnancy in a sub-Saharan African population.28 These results are important for the interpretation of our results, and the reported low reproducibility of the 75 g OGTT could result in misclassification of the diagnosis of gestational diabetes mellitus in the predefined period of screening (24-28 weeks of gestation) and might also have an effect on the results of the 75 g OGTT in early pregnancy. Because the 75 g OGTT is the reference standard in pregnancy, these difficulties cannot be overcome at the moment, but biomarkers of glucose metabolism or continuous glycaemic monitoring might improve diagnostic reproducibility and accuracy, and are under evaluation.15 29 30

Strengths and weaknesses of the study

The strengths of our study were the prospective design, multicentre approach, and low risk setting, showing the real effect of universal early screening in a general population. Also, participants, study midwives, doctors, and investigators were blinded to the early test results and therefore the significance and efficacy of the therapeutic interventions were unclear, but we consider this blinding to be a strength. Ongoing randomised controlled studies need to clarify these questions, which were not the focus of this study. For conciseness, we did not further investigate maternal characteristics and comorbidities between participants with early or late gestational diabetes mellitus and therefore we cannot describe an early onset phenotype for gestational diabetes mellitus. We also did not perform prediction models and did not analyse the WHO 2013 criteria in terms of maternal and neonatal outcomes. These topics will be analysed separately. Finally, the study population of mainly white participants might limit the generalisability of the findings to more diverse populations.

Comparisons with other studies

Individuals with abnormal glucose values in early to mid-pregnancy seem to have a higher risk of adverse pregnancy outcomes. A study showed that participants with an abnormal OGTT at 18-20 weeks of gestation and confirmed gestational diabetes mellitus at 24-28 weeks of gestation were more likely to have metabolic disorders, deliver infants that were large for gestational age, and have a higher risk of fetal hyperinsulinaemia.31 In a retrospective high risk cohort, higher maternal and neonatal adverse outcomes, including a larger proportion of preterm births, caesarean sections, and neonates large for gestational age, were identified.32 Children born to participants having an early diagnosis of gestational diabetes mellitus were more likely to develop metabolic disorders and obesity.33 Another study also showed that glucose targets and weight gain could be more easily met if the diagnosis was made by a standard late OGTT rather than during early pregnancy, which hindered the reduction of complications in early diagnosed gestational diabetes mellitus.32 Milder degrees of hyperglycaemia, lower than the threshold for pre-exisitng diabetes, but diagnosed before 24-28 weeks of gestation, might be a phenotype of gestational diabetes mellitus with higher risks for adverse outcomes, and individuals with a diagnosis of gestational diabetes mellitus after 24 weeks of gestation seem to have a milder phenotype. Therefore, a treatment approach for the early diagnosed, gestational diabetes mellitus phenotype might be justified but more challenging, and the benefits of an early intervention are lacking.11

EGGO (Early Gestational Diabetes Screening in the Gravid Obese Woman),34 a randomised controlled trial published in 2020, investigated early screening (14-20 weeks of gestation) with a two step screening approach (first 50 g glucose challenge test followed by 100 g OGTT based on the criteria of Carpenter and Coustan) in a population with obesity. The study reported no improvement in the primary outcome (macrosomia defined as birth weight >4000 g) in 29 individuals with early diagnosed, gestational diabetes mellitus (<20 weeks of gestation). Group size was low and underpowered for the early intervention group. Obesity in particular is an independent and the most prevalent risk factor for neonatal macrosomia, which might not be improved easily by glycaemic control alone. A Danish study about lifestyle interventions in women with obesity also reported no improvement in primary obstetric and metabolic outcomes.35

In the recently published TOBOGM trial (The Treatment of Booking Gestational Diabetes Mellitus),36 participants with at least one risk factor for hyperglycaemia were tested at 4-20 weeks of gestation (mean 15.6) with a two hour postload 75 g OGTT, with participants randomised to immediate (n=400) or deferred treatment (n=393; dietary advice or drug treatments). The immediate intervention caused a modest improvement in adverse neonatal outcomes (birth at <37 weeks of gestation, birth trauma, birth weight >4500 g, respiratory distress, phototherapy, stillbirth, neonatal death, or shoulder dystocia; 24.9% v 30.5%, adjusted relative risk 0.82, 95% confidence interval 0.68 to 0.98) and no substantial differences in pregnancy related hypertension or neonatal lean body fat. A flaw of the study could be that the adverse outcomes of preterm birth, respiratory distress syndrome, or phototherapy were not strongly related to hyperglycaemia. Secondary outcomes, such as infants large for gestational age (16.8% v 19.6%) and neonatal hypoglycaemia ≤40 mg/dL within 72 hours (18.9% v 22.7%), were improved by a slight increase in small for gestational age infants (12% v 9.2%), especially in pregnant individuals with lower glycaemic ranges. The higher rate of infants small for gestational age might be a possitble harm, but neonatal pH or Apgar status, or the possible increased admissions to the neonatal intensive care unit in the pilot study in this group of neonates was not further evaluated in the main study.37

Implications for clinicians and policy makers

Early diagnosis of gestational diabetes mellitus is important and treatment is challenging. Few studies have shown a sufficient effect of early diagnosis and intervention in those most at risk of later complications or have explored the cost effectiveness of early diagnosis and treatment. Our study indicated that the 75 g OGTT as a screening tool in early pregnancy is not sensitive enough when applying the WHO 2013 criteria, and the phenotype of early gestational diabetes mellitus cannot easily be described. Ongoing randomised controlled trials on the efficacy of early interventions to prevent maternal, and fetal and neonatal, adverse outcomes might clarify the importance of early screening for gestational diabetes mellitus.

Conclusions

The results of our prospective low risk cohort study indicated that the WHO 2013 criteria need to be modified to detect at least half of pregnant individuals with a later diagnosis of gestational diabetes mellitus. The early onset gestational diabetes mellitus phenotype seems to be associated with poorer pregnancy and neonatal outcomes and treatment is more challenging. Further research should include validation of our findings in different populations and investigation of the effect of early lifestyle and drug interventions in individuals with an early diagnosis versus a late diagnosis of gestational diabetes mellitus.