Multi-omics gestational diabetes models show promise but need clinical validation

First-trimester blood markers can predict gestational diabetes mellitus with encouraging accuracy, but the evidence remains largely confined to small derivation cohorts and has not yet established that biomarker-guided care improves outcomes. A 2026 narrative review assessed conventional lipid measures, metabolomic and lipidomic signatures, gut-microbiome features, thyroid indicators and environmental exposures, finding that omics-enhanced models reported area under the receiver operating characteristic curve values of approximately 0.83–0.95 at the development stage. Clinical factors alone generally produced lower values of 0.63–0.69, while routine lipid markers and composite indices reached about 0.75–0.82. These comparisons suggest incremental information, but most models lack external validation, and the most spectacular performance estimates came from studies with especially small samples.

The motivation for earlier prediction is straightforward. Gestational diabetes mellitus, defined as hyperglycaemia first detected during pregnancy that does not meet thresholds for overt diabetes, is diagnosed in an estimated one in seven pregnancies worldwide, with substantial variation among populations and criteria. A meta-analysis using International Association of Diabetes and Pregnancy Study Groups criteria estimated prevalence in mainland China at 14.8% (95% CI 12.8%–16.7%). Standard diagnosis relies on a 75-g oral glucose tolerance test at 24–28 weeks, after placental hormones have driven increasing insulin resistance and beta cells have had to compensate for underlying metabolic dysfunction. The Treatment of Booking GDM trial found a modest reduction in a composite neonatal outcome when early-onset gestational diabetes, diagnosed before 20 weeks, was treated immediately rather than deferred or untreated: 24.9% versus 30.5%, an adjusted risk difference of −5.6 percentage points. This result supports earlier action but does not yet prove that a first-trimester biomarker test can guide treatment safely and effectively.

Clinical risk factors remain important, but inexpensive measurements may be the most immediately deployable route to early stratification. In a prospective cohort of 32,411 pregnant women in Changsha, the highest versus lowest quartiles of first-trimester triglycerides, total cholesterol and remnant cholesterol were associated with gestational diabetes, with odds ratios of 1.65 (95% CI 1.36–1.99), 1.65 (1.24–2.21) and 1.40 (1.19–1.64), respectively. Those lipids also predicted pre-eclampsia, indicating that they may reflect shared metabolic vulnerability rather than a GDM-specific process. A separate Chinese development and external-validation study combined body mass index, triglycerides, total cholesterol, LDL cholesterol and HbA1c, reporting AUCs of 0.772 in training, 0.754 after internal validation and 0.759 in an independent cohort of 390 women. A Korean registry study found that an index combining fasting plasma glucose, triglycerides, liver enzymes and body mass index achieved an AUC of 0.802 (95% CI 0.72–0.88), while also being associated with non-alcoholic fatty liver disease.

Deeper molecular profiling has identified recurring signals that map onto the mechanisms already implicated in gestational diabetes: insulin resistance, lipotoxicity, inflammation and hepatic steatosis. In the Finnish EDDIE population cohort, long-chain fatty acids and medium- or long-chain acylcarnitines correlated with glycaemia, while HOMA-IR tracked with tyrosine and phospholipids. The Indian STRiDE prospective cohort identified 49 associated metabolites, chiefly phosphatidylcholines, sphingomyelins and triacylglycerols; a random-forest panel of eight metabolites produced an AUC of 0.880 (95% CI 0.809–0.951), with a more conservative bootstrap mean of 0.83 (0.77–0.89). Profiling of 180 women in a TOBOGM substudy detected a distinct lipid profile in 89 women with early-onset gestational diabetes, including higher diacylglycerols, fatty acids and ethanolamine-containing phospholipids, and lower lysophosphatidylcholines, ether-linked phosphatidylcholines and hexosylceramides. Because those samples were collected at the diagnostic oral glucose tolerance test, however, the result represents a concurrent signature rather than a prospective prediction model.

The strongest direct evidence that molecular data can improve established risk assessment comes from within-cohort comparisons. In the Korean Fatty Liver in Pregnancy registry, clinical factors alone yielded an AUC of 0.631 (95% CI 0.566–0.693), whereas clinical factors plus 16 selected metabolites raised it to 0.849 (0.797–0.892; p<0.001). An exploratory mediation analysis suggested that associations involving glycocholic acid, butanoyl carnitine and uric acid were partly accounted for by concurrently measured hepatic steatosis, with estimated mediated proportions of 9.7%–31.9%. Because the mediator and exposures were measured at the same time, the analysis cannot establish temporality or exclude unmeasured confounding. In STRiDE, conventional risk factors produced AUCs of 0.655 and 0.688, while their combination with the eight-metabolite panel reached 0.881. Another exploratory analysis of just 20 GDM cases and 185 controls reported AUCs of 0.95 for a model including body mass index, insulin, soluble CD163 and tumour necrosis factor-alpha, and 0.94 for a model excluding body mass index but adding pregnancy-associated plasma protein A. Those estimates were never independently validated and are best regarded as upper bounds rather than expected performance.

Across studies, the clearest molecular convergence involves phospholipid remodelling and glycerolipid accumulation: diacylglycerols and triacylglycerols tend to rise while lysophosphatidylcholines are depleted. This axis has appeared in Finnish, Indian and Australian-led studies using different analytical platforms, making it more reproducible than many amino-acid or bile-acid findings. Free fatty acids also occupy a mechanistic junction between lipid metabolism and insulin resistance, but their interpretation is complicated by recent diet and substantial individual variability. Other emerging signals broaden the biological frame. A nine-cohort analysis of 2,717 mothers and infants identified a core gut microbial guild associated with gestational diabetes risk and disease progression, although it remains an exploratory marker rather than a validated predictor. In a nested case-control study with serial sampling, first-trimester L-carnitine and trimethylamine were positively associated with GDM, but the strongest joint effects involving trimethylamine N-oxide-related metabolites appeared later in pregnancy.

Thyroid function and environmental exposure provide additional avenues of inquiry, although neither is ready for implementation. In a prospective two-centre cohort of 1,486 women, the first-trimester free T4/TSH ratio was higher among women who developed GDM, with the association most pronounced in pregnancies carrying a male fetus. A Belgian cohort separately linked lower first-trimester free T4 with subsequent insulin use among women with GDM. In a Wuhan birth-cohort study comprising 286 case-control pairs, gestational exposure to several non-phthalate plasticisers was associated with greater GDM risk, with partial mediation through the plasma lipidome. These findings show that gestational diabetes signals extend beyond glucose and routine lipids. They do not, however, establish that measuring thyroid function or plasticiser exposure improves prediction, and the review identified 145 candidate first-trimester biomarkers across 133 studies with marked differences in sampling windows, platforms, ethnicity, adiposity adjustment and reporting.

Translation requires more than a high AUC. Most studies emphasize discrimination and rarely report calibration, which measures whether predicted risks match observed event rates. Clinical utility is also largely untested: the review said no study had yet shown through decision-curve analysis that biomarker-guided management provides greater net benefit than universal oral glucose tolerance testing or clinical factors alone. At a gestational diabetes prevalence of 15%, a model with 80% sensitivity and 80% specificity would have a positive predictive value of only about 41%, illustrating how false positives could multiply surveillance burdens. Small derivation cohorts, variable quantification methods, inconsistent use of false-discovery-rate control, limited adjustment for maternal adiposity and the risk of overfitting make the apparent performance gradient from 0.63 to 0.95 hypothesis-generating rather than definitive.

The proposed path forward combines larger studies with simpler clinical outputs. Priorities include multi-ethnic prospective cohorts with standardised first-trimester sampling, harmonised platforms and absolute rather than relative quantification; head-to-head external validation of clinical, routine-lipid, omics and combined models; transparent reporting under TRIPOD, with formal risk-of-bias assessment using tools such as PROBAST; and evaluation of low-cost panels containing only three to eight analytes. Asian populations warrant particular attention because they carry a disproportionate burden of gestational diabetes and may develop metabolic disease at lower body mass index, but resource-diverse health systems will also require explicit cost-effectiveness and implementation testing. The final missing step is a trial that recruits women using a validated first-trimester panel and randomly assigns those classified as high risk to early lifestyle or pharmacological intervention or standard care, with perinatal and longer-term maternal and child outcomes. Until that closed loop is completed, multi-omics offers a promising route to earlier risk identification, not yet a proven clinical service.

Subject of Research: Clinical translation of first-trimester multi-omics biomarkers and prediction models for gestational diabetes mellitus

Article Title: From biomarker discovery to prediction models: clinical translation of multi-omics for early prediction of gestational diabetes mellitus

Article References: Yan, H. (2026). From biomarker discovery to prediction models: clinical translation of multi-omics for early prediction of gestational diabetes mellitus. Frontiers in Endocrinology, 17, Article 2007266. https://doi.org/10.3389/fendo.2026.2007266

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.2007266

Keywords: gestational diabetes, first-trimester prediction, multi-omics, metabolomics, lipidomics, gut microbiome, biomarkers, prediction models, insulin resistance, pregnancy, biomarker, discovery