Do Mass Spectrometry-Derived Metabolomics Improve the Prediction of Pregnancy-Related Disorders? Findings from a UK Birth Cohort with Independent Validation.
Smith, Gordon CS
Lawlor, Deborah A
MetadataShow full item record
McBride, N., Yousefi, P., Sovio, U., Taylor, K., Vafai, Y., Yang, T., Hou, B., et al. (2021). Do Mass Spectrometry-Derived Metabolomics Improve the Prediction of Pregnancy-Related Disorders? Findings from a UK Birth Cohort with Independent Validation.. Metabolites, 11 (8) https://doi.org/10.3390/metabo11080530
Many women who experience gestational diabetes (GDM), gestational hypertension (GHT), pre-eclampsia (PE), have a spontaneous preterm birth (sPTB) or have an offspring born small/large for gestational age (SGA/LGA) do not meet the criteria for high-risk pregnancies based upon certain maternal risk factors. Tools that better predict these outcomes are needed to tailor antenatal care to risk. Recent studies have suggested that metabolomics may improve the prediction of these pregnancy-related disorders. These have largely been based on targeted platforms or focused on a single pregnancy outcome. The aim of this study was to assess the predictive ability of an untargeted platform of over 700 metabolites to predict the above pregnancy-related disorders in two cohorts. We used data collected from women in the Born in Bradford study (BiB; two sub-samples, n = 2000 and n = 1000) and the Pregnancy Outcome Prediction study (POPs; n = 827) to train, test and validate prediction models for GDM, PE, GHT, SGA, LGA and sPTB. We compared the predictive performance of three models: (1) risk factors (maternal age, pregnancy smoking, BMI, ethnicity and parity) (2) mass spectrometry (MS)-derived metabolites (n = 718 quantified metabolites, collected at 26-28 weeks' gestation) and (3) combined risk factors and metabolites. We used BiB for the training and testing of the models and POPs for independent validation. In both cohorts, discrimination for GDM, PE, LGA and SGA improved with the addition of metabolites to the risk factor model. The models' area under the curve (AUC) were similar for both cohorts, with good discrimination for GDM (AUC (95% CI) BiB 0.76 (0.71, 0.81) and POPs 0.76 (0.72, 0.81)) and LGA (BiB 0.86 (0.80, 0.91) and POPs 0.76 (0.60, 0.92)). Discrimination was improved for the combined models (compared to the risk factors models) for PE and SGA, with modest discrimination in both studies (PE-BiB 0.68 (0.58, 0.78) and POPs 0.66 (0.60, 0.71); SGA-BiB 0.68 (0.63, 0.74) and POPs 0.64 (0.59, 0.69)). Prediction for sPTB was poor in BiB and POPs for all models. In BiB, calibration for the combined models was good for GDM, LGA and SGA. Retained predictors include 4-hydroxyglutamate for GDM, LGA and PE and glycerol for GDM and PE. MS-derived metabolomics combined with maternal risk factors improves the prediction of GDM, PE, LGA and SGA, with good discrimination for GDM and LGA. Validation across two very different cohorts supports further investigation on whether the metabolites reflect novel causal paths to GDM and LGA.
External DOI: https://doi.org/10.3390/metabo11080530
This record's URL: https://www.repository.cam.ac.uk/handle/1810/329912
Attribution 4.0 International
Licence URL: https://creativecommons.org/licenses/by/4.0/