Blood metabolite profiling outperforms BMI in predicting pregnancy complications

By Hugo Francisco de SouzaReviewed by Susha Cheriyedath, M.Sc.Dec 21 2025

A large international study shows that metabolomic signatures in maternal blood, especially later in pregnancy, reveal hidden metabolic risk and predict gestational diabetes and preeclampsia more accurately than BMI alone.

Study: A metabolomic signature of maternal BMI is associated with pregnancy complications across two independent pregnancy cohorts. Image Credit: ibragimova / Shutterstock

In a recent study published in the journal Communications Medicine, researchers analyzed blood samples from two large, independent cohorts to identify specific metabolomic signatures associated with maternal BMI. The study leveraged machine learning to identify a profile of 46 metabolites that correlated with BMI and showed stronger associations with certain pregnancy complications than BMI alone.

The study further identified a subset of 16 metabolites that, in model-based analyses, statistically mediated the relationship between obesity and diabetes, suggesting that targeted blood profiling may help refine prenatal risk stratification.

Rising Obesity and Pregnancy Risk

The global rise in obesity, particularly in Western countries, has been accompanied by an increase in high-risk pregnancies. Maternal obesity has long been associated with complications such as gestational diabetes mellitus (GDM) and preeclampsia.

Clinicians typically rely on pre-pregnancy BMI to estimate these risks. However, BMI reflects only height and weight and does not capture the underlying metabolic state. As a result, individuals with a normal BMI may still carry metabolic risk, while some individuals with a higher BMI may be metabolically healthy.

Metabolomics as a Biological Lens

To address these limitations, researchers are increasingly turning to metabolomics - the study of small molecules circulating in the blood that reflect metabolic activity. Metabolomic profiling offers a more precise biological snapshot of metabolic health and may better capture pregnancy-related metabolic stress than anthropometric measures alone.

Cohorts, Sampling, and Machine Learning Approach

The study analyzed data from two independent pregnancy cohorts: the Copenhagen Prospective Studies on Asthma in Childhood (COPSAC) in Denmark and the Vitamin D Antenatal Asthma Reduction Trial (VDAART) in the United States.

Blood plasma samples were processed using untargeted liquid chromatography–tandem mass spectrometry (LC-MS/MS), enabling detection of hundreds of metabolites. A machine learning model based on sparse partial least squares regression was applied to identify metabolite patterns associated with BMI and pregnancy outcomes.

The Danish COPSAC2010 cohort, which included blood samples from 684 women at mid-pregnancy (24 weeks), served as the discovery cohort. The VDAART cohort, consisting of 775 women with samples collected in early (10–18 weeks) and late (32–38 weeks) pregnancy, was used for validation.

Metabolic Profiles Predict Pregnancy Complications

Across both cohorts, LC-MS/MS identified 640 candidate metabolites associated with maternal BMI and pregnancy complications. Machine learning analyses distilled these into a robust 46-metabolite signature linked to adverse outcomes, particularly gestational diabetes and preeclampsia. Key contributors included sphingolipids involved in cell signaling and metabolites related to vitamin A metabolism.

In the discovery cohort, higher BMI was associated with gestational diabetes (odds ratio [OR] 1.90), but the metabolite score was a stronger predictor (OR 2.47). Importantly, while BMI alone did not significantly predict preeclampsia, the metabolite score did.

Timing, Validation, and Mediation Findings

Validation analyses in the VDAART cohort confirmed the robustness of the metabolic signature across populations. The timing of sample collection proved critical. Metabolite scores measured in late pregnancy were strongly predictive of both preeclampsia and gestational diabetes, whereas early pregnancy scores were substantially less informative.

Mediation analyses identified 16 metabolites that partially explained the association between obesity and gestational diabetes. Plant-derived metabolites, such as carotene diol, were associated with a lower risk of diabetes, whereas lipid-related metabolites, including ceramides and sphingomyelins, were associated with an increased risk.

A separate machine learning model using only these 16 metabolites outperformed a BMI-only model in predicting gestational diabetes, as assessed by likelihood ratio testing.

Implications for Prenatal Risk Assessment

The findings highlight the limitations of BMI as a standalone predictor of pregnancy complications and suggest that metabolomic profiling may offer a more nuanced and biologically meaningful approach. Combining BMI with metabolite-based risk scores may improve the prediction of gestational diabetes and preeclampsia.

Although observational and conducted in high-resource settings, the study supports further investigation into integrating blood-based metabolomic screening into prenatal care. With additional validation and comparison to existing screening tools, such approaches could help identify high-risk pregnancies earlier and enable more personalized monitoring and intervention.