Identifying women with prediabetes or early type 2 diabetes in a high-risk group of relatively young women post-GDM is important for the planning of a potential next pregnancy and for implementing appropriate measures to prevent the development of type 2 diabetes and its complications. Our study showed that prediabetes was highly prevalent and more so among South Asian than Nordic women. Both the HbA_1c and the compound diagnostic criteria detected a pattern of excess risk in the South Asian compared to the Nordic women. However, HbA_1c seemed to capture this ethnic difference just as well as other tests and speculatively better. Adding the measurement of FPG to HbA_1c, identified more women at risk and might be an option to consider in women planning further pregnancies. Important predictors for actionable HbA_1c were South Asian ethnicity, GDM before index pregnancy, use of glucose-lowering drugs in pregnancy, higher age, and higher in-pregnancy FPG. Whilst pre-pregnancy BMI did not feature as a significant predictor, when it was replaced by actual WHR, obesity appeared as a key modifiable risk factor.

Prevalence of prediabetes and diabetes

In our population, applying the ADA-HbA_1c cut-offs of 39 mmol/mol (5.7%) identified 58.3% and 22.2% of South Asian and Nordic women with prediabetes or diabetes, of whom most had pre-diabetes. Hence, only two South Asian or five Nordic women would need to be tested to identify one woman with an actionable ADA-HbA_1c result. The comparable percentages by the WHO-HbA_1c cut-offs of 42 mmol/mol (6.0%) were 25.8 and 6.5%.

Similar to our findings in the Nordic population, a few studies [23,24,25] have reported a prevalence of 16-19% (applying the ADA-HbA_1c cut-points) for the diagnosis of prediabetes or diabetes in mostly white women 1–2.5 years post-GDM. Comparable data in the South Asian population is, so far, scarce.

As current guidelines differ regarding which test to use as post-GDM screening [7, 9, 26], we also assessed the prevalence of prediabetes and diabetes including the OGTT data. We then found that 87.8% and 65.7% of the South Asian and Nordic qualified for this diagnosis. These high prevalence rates, however, seems rather unrealistic in terms of future workload for prevention. One reason for this increase may be that some women have isolated peaks of postprandial glucose that do not translate into higher HbA_1c or into long-term diabetes complications [10, 11]. Although some studies defend the use of OGTT as a predictor of diabetes outcomes, they do not apply to our younger prediabetic population [27, 28]. Another possibility is that, despite our best efforts, perhaps some did not fast for as long as requested before their OGTT, once again reiterating the importance of using tests which have better analytical performance and are less prone to pre-analytical factors, which may vary by ethnicity. Even so, both the use of stand-alone HbA_1c and different compound diagnostic criteria confirmed a higher prevalence among the South Asian than the Nordic women (Fig. 1).

Two previous studies performed in India and Ireland with similar design to ours [13] also reported high (albeit somewhat lower than the present study) prevalence rates of prediabetes or diabetes post-GDM (by the compound diagnostic ADA-criteria), 57.7% and 18.4%, respectively. In these studies, the IADPSG criteria were used for diagnosing GDM, in contrast to our WHO 99 (4.8%) or modified IADPSG (93%) criteria with higher cut-off values for the diagnosis. This might explain our higher prevalence rates, as a higher cut-off values are associated with higher conversion rates to diabetes [14, 29]. Nevertheless, the comparable proportion of Nordic women that developed prediabetes or diabetes in our study was unexpectedly high (65.7% vs. 18.4% in the Irish study)), since the Irish women had similar age and higher mean BMI (32.4 vs 29.3 kg/m²) and waist circumference (100.1 vs 96.1 cm). A Finnish study [30] reported the prevalence of prediabetes or diabetes to be > 50% 6 years post-GDM, which is in broad accordance with our results, acknowledging the differences in the GDM criteria applied.

A high prevalence of diabetes short time after pregnancy in South Asian women is consistent with current litterature, reporting a cumulative conversion rate to diabetes of 10.5% and 22.0% within ~4 and 5 years post-GDM on the basis of OGTT data [14]. In Western women, however, studies have indicated a lower cumulative conversion rate of 2.3% and 5.8% within 5 and ~10 years [13, 30]. Notwithstanding this, the early post-GDM development of diabetes in both populations, yet meaningfully lower by the HbA_1c, underscores the importance of annually screening for prediabetes and diabetes post-GDM.

Comparing different modes for diagnosing prediabetes and diabetes with diagnosis based on OGTT

Our study resonates strongly with previous work [13, 14, 31], acknowledging that adding a measurement of FPG to HbA_1c increases the diagnostic sensitivity for prediabetes and diabetes significantly. We also found that more women were identified by applying the ADA rather than the WHO-IEC criteria. Interestingly, the benefits of adding HbA_1c to FPG was only significant in the South Asian group. This finding emphasise that HbA_1c may efficiently capture the pattern of excess risk in South Asians, and is consistent with studies showing higher HbA_1c levels in South Asian than white women despite lower FPG [32]. Accordingly, this could partially explain the difference in HbA_1c between the ethnic groups for the best performance of the combined FPG and HbA_1c in our study (South Asian group 5.6 mmol/l and 38 mmol/mol (5.6%) vs 5.6 mmol/l and 37 mmol/mol (5.5%) in the Nordic group) (Additional file 2: Table S2), again with the assumption that OGTT is gold standard, which can now be strongly debated.

Whether adding FPG to HbA_1c identifies more women at risk for high glucose levels in subsequent pregnancies, and thereby adverse pregnancy outcomes [33,34,35], are not answered by our study and deserve further studies. In clinical practice, we support the use of HbA_1c testing for women post-GDM.

Predictors for prediabetes or diabetes by ethnicity

In our multivariate analysis, South Asian ethnicity was the strongest risk factor for an actionable HbA_1c post-GDM. Although this finding was not unexpected and is supported by recent litterature, these studies all held the assumption that OGTT is gold standard for the diagnosis of prediabetes or diabetes [14, 36, 37]. To the best of our knowledge, this is the first study that sought to identify determinants for actionable ADA-HbA_1c. And by doing so, it was reassuring to see that the predictors for glucose deterioration were consistent regardless of criteria used [30, 36, 38] (Table 2).

Of note, intake of glucose-lowering drugs did not impact the risk of actionable HbA_1c in women with GDM before index pregnancy, contrary to findings in women without GDM before index pregnancy, emphasizing the strong impact of previous GDM itself as a risk factor. This finding indicates that a GDM-pregnancy may result in a severe deterioration of the glucose metabolism, as stated by others [39], fitting with the high GDM recurrence rate in following pregnancies [40].

Overweight and obesity are well known risk factors for prediabetes and diabetes, and in our study 86% (82/95) of the South Asians, and 81% (17/21) of the Nordic women identified with actionable HbA_1c were either overweight or obese before the pregnancy. Type 2 diabetes is strongly associated with excess total and ectopic fat in all ethnicities, and the greater risk in South Asians may be linked to larger visceral fat and lower skeletal muscle mass for a given BMI [41]. Accordingly, we found larger WHR in South Asian than in Nordic women at similar BMI, and an increased risk for actionable HbA_1c by WHR, but not by BMI. There is also an emerging suggestion that South Asians have more rapid genetically-determined beta cell deterioration [42]; consistent with our findings that the increased risk of actionable HbA_1c among South Asians was not greatly attenuated after adjustment for WHR. Whatever the mechanism, we strongly encourage weight redcution in overweight women post-GDM, as modest weight loss can prevent diabetes [43] and greater weight loss can reverse diabetes [44]. Such weight loss reduces liver fat, and may improve beta cell function in whites, at least [45], though weight loss led to similar or even greater diabetes remission in a Qatari population [46].

Strengths and limitations

The major strengths of this multicentre study is the inclusion of a relatively large sample size from two different ethnic groups living in the same area and cared for in the same healthcare setting. Futhermore, the application of different definitions for prediabetes that might make comparison between studies easier is also a major strength. Finally, reporting on ethnic difference in HbA_1c-based prevalence of prediabetes and diabetes post-GDM is novel and important, especially as rates of diabetes and obesity are rasing worldwide, especially in “metabolically higher risk” South Asians communities.

Our study has limitations. First, we only recruited women referred to hospital for treatment of GDM. Our findings are, therefore, not applicable to women with diet-treated GDM cared for in primary healthcare.

Second, we cannot exclude a selection bias due to the low participation rate (45% and 14% in the South Asian and the Nordic groups). We, therefore, compared significant baseline characteristics in the regression analysis of women who did vs. a randomly selected subgroup of women who did not participate in the study (100 South Asian and 100 Nordic women) (Additional file 2: Table S4). Among the South Asian women no difference in age, pre-pregnancy BMI, in-pregnancy glucose values, the use of glucose-lowering drugs, GDM before index pregnancy, or first-degree relatives with diabetes were found. The participating Nordic women were older than non-participants, but the other characteristics did not differ. The older age among participating Nordic women might have led to an overestimation of the proportion of women with actionable HbA_1c in this group. Third, differences in the recruitment procedures may have introduced a selection bias between the ethnic groups. After sending invitation letters to eligible women, only South Asian women received a telephone invitation to enhance recruitment. It is therefore possible that Nordic women with higher in-pregnancy glucose levels may have been more likely to respond to the invitation. Speaking against this is the fact that a higher percentage of South Asian than Nordic women were using glucose-lowering drugs in pregnancy and by minimal differences in the in-pregnancy glucose levels between the ethnic groups (Table 1). Fourth, our analyses did not include assessments of dietary habits and physical activity that may have important impact on the prevalence rates in the two groups. Our study had limited power to detect ethnic difference in the prevalence of diabetes. Finally, causality cannot be inferred from a cross-sectional study, and a single measurement of HbA_1c or OGTT is not sufficient for a diagnosis of diabetes in asymptomatic women.