This study identified that the HOMA2-B/HOMA2-S precision categorisation of GDM was robust to changes in specimen processing, but that these indices provided little additional diagnostic or prognostic information beyond that which is already available using existing antenatal monitoring. However, we identified a novel subgroup (accounting for 25% of patients with GDM) with an isolated postprandial defect in glucose handling and normal pregnancy outcomes; these women had comparable values of HOMA2-S and HOMA2-B to those for normoglycaemic pregnant women.

Although HOMA2-B/HOMA2-S precision categorisation of GDM makes sense from a pathophysiological perspective, in practice neither HOMA2-S nor HOMA2-B are strong predictors of GDM or pregnancy outcomes (Tables 2 and 3). We found that, despite being an imperfect predictor, maternal BMI shows equivalent or better predictive capability over pregnancy outcomes compared with HOMA scores, and is routinely measured in clinical practice without the need for insulin determination. Our data suggest that BMI may be a more successful and equitable precision marker than mathematical measures of insulin secretion or resistance. This is unsurprising as BMI in normoglycaemic pregnant women is associated with increased rates of operative delivery [20], babies who are large for gestational age [21, 22], hypertensive disorders [22, 23] and preterm birth [24]. In women with diabetes, BMI also correlates with the onset and severity of hyperglycaemia [25, 26] and the severity of insulin resistance [27], which are additive risk factors for suboptimal pregnancy outcomes [28]. BMI offers practical advantages as a prognostic marker, being widely used, easily measured and cost-effective, with minimal resources and training needed.

Strengths and weaknesses

We recruited an antenatal population of women with one or more risk factors for GDM, who had specimens taken using standard and enhanced protocols at 0 and 120 min during a 75 g antenatal OGTT. We calculated multiple indices for insulin resistance and insulin secretory function but were limited to indices that use these timepoints. Additional sampling at 60 min may have improved the performance of indices of insulin secretion and allowed comparison with GDM diagnosed using other diagnostic criteria, as there is mounting evidence for the value of the 60 min test [29, 30]. Our standard specimen processing procedure was designed to replicate clinical care and local sample pre-analytical and analytical procedures at each of our nine sites. While this introduced variation, all laboratories used in this study were accredited under the UK Accreditation Service, and can demonstrate acceptable analytical performance during the study period. Citrate tubes were not used for glucose assessment as these were not widely available at the time of study initiation [31].

Only 281 of the 1308 women had GDM, although all women in our population had at least one risk factor for GDM, which may have influenced glucose, insulin, HOMA2-S and HOMA2-B values for the NGT population. This is important as it reflects the thresholds used (< 25th centile of normal population) for HOMA2-S and HOMA2-B, and thus may have changed the number of women allocated to the insulin-resistant and insulin-insufficient subgroups. We considered that it was more robust to use the enhanced glucose measurements for calculation of HOMA scores and categorisation into groups. However, clinical diagnosis was based on standard processing of venous glucose levels, causing heterogeneity within our group, as not all women had access to treatment. Treatment itself may reduce the predictive capability of any measure obtained at 28 weeks, but this cannot be avoided while conducting an ethical study.

Relevance to other studies

Previous work by Powe et al used HOMA2-S and HOMA2-B thresholds of the 25th centile to categorise women with GDM into insulin-resistant and insulin-insufficient subgroups, or both [4]. In their population in the USA, the proportions of women with insulin-resistant GDM (51%), insulin-insufficient GDM (30%) were similar to those for our cohort; however, the proportion of women categorised as having GDM with both insulin resistance and insulin insufficiency differed (18% GDM-mixed [the category referred to as ‘GDM-both’ in our study]), and only one woman was unclassified, compared with 71 (25.3%) in our study [4]. In contrast to the study by Powe et al and our work, Benhalima et al found a greater proportion of women with insulin-resistant GDM (83%), defined as having a Matsuda index below the 50th percentile of that for women with NGT [5]. A study by Liu et al in China found that 17% of women with GDM could be categorised as insulin-resistant, 21% were categorised as insulin-insufficient and 38% were categorised as mixed. Liu et al used thresholds of the 25th percentile of the Matsuda index and oral disposition index to distinguish insulin sensitivity and insulin secretion [32]. Similarly to our study, Liu et al also identified a group of women with GDM and normal indices (representing 23.8% of women with GDM), but these participants were not further characterised. While use of HOMA2-S and HOMA2-B can identify women with GDM and near-normal indices, suggesting valid pathophysiological classification, they probably miss isolated postprandial insulin sensitivity defects, indicating imperfect detection of more subtle abnormalities. Differences in approaches to screening for hyperglycaemia in pregnancy, diagnostic criteria for GDM, the definition of NGT and the population prevalence of obesity may influence the thresholds for insulin resistance and insulin insufficiency, creating differences between the performance of this precision approach in different populations.

Use of the HOMA scores is undoubtedly convenient, but using improved measures of insulin resistance or insulin secretion may improve the performance of this precision approach. Cohen et al attempted to validate the HOMA-IR score longitudinally in six women who were assessed using euglycaemic–hyperinsulinaemic clamps in pregnancy and postnatally [11]. They identified large inter-individual differences in the relationship between HOMA-IR and glucose disappearance on the clamp, and concluded that the HOMA-IR score lacked sensitivity for individual use (R2 = 0.45) [11]. Kirwan et al found the Matsuda index to be most closely associated with clamp-derived insulin resistance estimates in pregnancy, with HOMA2-S showing the weakest associations (R2 = 0.52–0.61) [10]. However, we found similar results when using a Matsuda–Stumvoll classification instead of HOMA2-S and HOMA2-B. Despite these limitations, HOMA scores are widely used in the pregnancy literature, for example as primary outcomes in a randomised controlled trial [33] and to assess change in response to lifestyle interventions [34], and are associated with relevant pregnancy outcomes, including the development of GDM [35, 36].

Although the optimal criteria for precision categorisation of women with GDM are still unclear, there is a growing body of evidence suggesting that insulin-resistant GDM is more strongly associated with suboptimal pregnancy outcomes, including pre-eclampsia, preterm delivery, LGA babies and neonatal hypoglycaemia, independently of maternal blood glucose levels or obesity [5, 32]. In the study by Benhalima et al, women with high levels of insulin resistance had higher rates of preterm delivery and neonatal hypoglycaemia, despite adjustment for confounding factors such as BMI and gestational weight gain [5]. Liu et al found that women with GDM with both insulin resistance and insulin insufficiency had a higher risk of LGA babies, but there was no other difference between the groups [32]. In this study, we also found that insulin-resistant GDM was associated with LGA babies. However, BMI offered similar prognostic information to the HOMA2-S score, and is more accessible in clinical practice internationally. Women with insulin-insufficient GDM appear to have pregnancy outcomes that are comparable to those of the normoglycaemic population [32], but women with altered insulin secretion still appear to have substantial risks of type 2 diabetes postpartum [29, 30].

Main implications for future work

With rising population levels of obesity in many countries internationally, there is an urgent need to improve risk prediction in women with GDM, allowing time and resources to be targeted to the highest-risk pregnancies and to type 2 diabetes prevention [2, 3]. While novel precision approaches may help, widely available measurements such as BMI and glucose concentrations during an OGTT already provide rich prognostic information and could be used more effectively, as highlighted by a recent publication from the joint ADA/EASD Precision Medicine in Diabetes Initiative [2]. A binary diagnosis for GDM with a single treatment pathway overlooks the severity of hyperglycaemia and obesity, which are key determinants of outcomes in GDM.

A key question remains: what might we reasonably aim to achieve with a precision approach? The recent consensus report from Tobias et al as part of the ADA/EASD Precision Medicine in Diabetes Initiative highlights GDM prevention, diagnosis, treatment and prognosis as key areas where precision medicine can improve care [1]. Our study found that women with insulin-resistant GDM had higher rates of Caesarean delivery and gave birth to babies of higher birthweight. However, LGA babies can already be predicted using ultrasound measurements, and delivery modality is more likely to be affected by maternal BMI than maternal insulin resistance.

Conclusion

Use of a precision medicine approach comprising HOMA2-S and HOMA2-B did not add useful diagnostic or prognostic information in a population of pregnant women at risk of GDM. However, it did identify a novel subgroup of GDM women with an isolated postprandial defect in glucose handling, which was not associated with substantial impairments in HOMA2-S or HOMA2-B. Further work is needed to identify a successful precision medicine approach to GDM.