Introduction Gestational diabetes mellitus (GDM) is generally identified by measuring elevated maternal glycemic responses to an oral glucose load in late pregnancy (> 0.6 term). However, our preliminary study suggests that GDM could be identified with a high predictive accuracy (96%) in the first trimester (< 0.35 term) by characteristic changes in the metabolite profile of maternal urine. (Koos and Gornbein, American Journal of Obstetrics and Gynecology 224:215.e1–215.e7, 2021). The gestational decrease in insulin sensitivity and the accompanying perturbations of the maternal metabolome suggest that a distinguishing urinary metabolite algorithm could differ in later gestation. Objectives This study was carried out (1) to identify the metabolites of late-pregnancy urine that are independently associated with GDM, (2) to select a metabolite subgroup for a predictive model for the disorder, (3) to compare the predictive accuracy of this late pregnancy algorithm with the model previously established for early pregnancy, and (4) to determine whether the late urinary markers of GDM likely contribute to the late pregnancy decline in insulin sensitivity. Methods This observational nested case–control study comprised a cohort of 46 GDM patients matched with 46 control subjects (CON). Random urine samples were collected at ≥ 24 weeks’ gestation and were analyzed by a global metabolomics platform. A consensus of three multivariate criteria was used to distinguish GDM from CON subjects, and a classification tree of selected metabolites was utilized to compute a model that separated GDM vs CON. Results The GDM and CON groups were similar with respect to maternal age, pre-pregnancy BMI and gestational age at urine collection [GDM 30.8 <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11306_2025_2314_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">\(\pm \)</EquationSource> </InlineEquation> 3.6(SD); CON [30.5 ± 3.6] weeks as they were matched by these variables. Three multivariate criteria identified eight metabolites simultaneously separating GDM from CON subjects, comprising five markers of mitochondrial dysfunction and three of inflammation/oxidative stress. A five-level classification tree incorporating four of the eight metabolites predicted GDM with an unweighted accuracy of 89%. The model derived from early pregnancy urine also had a high predictive accuracy (85.9%). Conclusion The late pregnancy urine metabolites independently linked to GDM were markers for diminished insulin sensitivity and glucose-stimulated insulin release. The high predictive accuracy of the models in both early and late pregnancy in this cohort supports the notion that a urinary metabolite phenotype may separate GDM vs CON across both early and late gestation. A large validation study should be conducted to affirm the accuracy of this noninvasive and time-efficient technology in identifying GDM.

Gestational age and models for predicting gestational diabetes mellitus

期刊论文
出版模式：

开放获取
发表日期：

2025年9月26日

查看全文

Aisling Murphy,
Jeffrey Gornbein,
Ophelia Yin,
Brian Koos

摘要

Introduction

Gestational diabetes mellitus (GDM) is generally identified by measuring elevated maternal glycemic responses to an oral glucose load in late pregnancy (> 0.6 term). However, our preliminary study suggests that GDM could be identified with a high predictive accuracy (96%) in the first trimester (< 0.35 term) by characteristic changes in the metabolite profile of maternal urine. (Koos and Gornbein, American Journal of Obstetrics and Gynecology 224:215.e1–215.e7, 2021). The gestational decrease in insulin sensitivity and the accompanying perturbations of the maternal metabolome suggest that a distinguishing urinary metabolite algorithm could differ in later gestation.

Objectives

This study was carried out (1) to identify the metabolites of late-pregnancy urine that are independently associated with GDM, (2) to select a metabolite subgroup for a predictive model for the disorder, (3) to compare the predictive accuracy of this late pregnancy algorithm with the model previously established for early pregnancy, and (4) to determine whether the late urinary markers of GDM likely contribute to the late pregnancy decline in insulin sensitivity.

Methods

This observational nested case–control study comprised a cohort of 46 GDM patients matched with 46 control subjects (CON). Random urine samples were collected at ≥ 24 weeks’ gestation and were analyzed by a global metabolomics platform. A consensus of three multivariate criteria was used to distinguish GDM from CON subjects, and a classification tree of selected metabolites was utilized to compute a model that separated GDM vs CON.

Results

The GDM and CON groups were similar with respect to maternal age, pre-pregnancy BMI and gestational age at urine collection [GDM 30.8 \(\pm \) 3.6(SD); CON [30.5 ± 3.6] weeks as they were matched by these variables. Three multivariate criteria identified eight metabolites simultaneously separating GDM from CON subjects, comprising five markers of mitochondrial dysfunction and three of inflammation/oxidative stress. A five-level classification tree incorporating four of the eight metabolites predicted GDM with an unweighted accuracy of 89%. The model derived from early pregnancy urine also had a high predictive accuracy (85.9%).

Conclusion

The late pregnancy urine metabolites independently linked to GDM were markers for diminished insulin sensitivity and glucose-stimulated insulin release. The high predictive accuracy of the models in both early and late pregnancy in this cohort supports the notion that a urinary metabolite phenotype may separate GDM vs CON across both early and late gestation. A large validation study should be conducted to affirm the accuracy of this noninvasive and time-efficient technology in identifying GDM.