Dataset Information

Diagnostic utility of clinicodemographic, biochemical and metabolite variables to identify viable pregnancies in a symptomatic cohort during early gestation.

ABSTRACT:

Study question: Can a novel machine learning pipeline create predictive models that identify a composite biomarker to differentiate live normally sited pregnancies (LNSP) from non-viable pregnancies in women presenting with pain and/or bleeding in early gestation?

Summary answer: Serum metabolite levels and biochemical markers from a single blood sample possess modest predictive utility in differentiating LNSP from non-viable pregnancies upon first presentation, which is improved by variable selection and combination using machine learning.

What is known already: Around 15% of intrauterine pregnancies are lost in the first trimester. A further 1–2% of pregnancies are located outside of the endometrial cavity and these ectopic pregnancies (EPs) are the leading cause of maternal mortality in the first trimester. Early pregnancy loss can cause significant morbidity when bleeding or infection occurs. Symptoms of pregnancy loss and EP are very similar (including pain and bleeding); however, these symptoms are also common in LNSP. To date, no biomarkers have been identified to differentiate LNSP from non-viable pregnancies.

Study design, size, duration: This is a prospective cohort study that included 370 participants who attended the early pregnancy assessment unit (EPAU) at Liverpool Women’s Hospital between 5/11/2018 and 08/12/2021 with pain and/or bleeding ≤10 weeks of gestation.

Participants/materials, setting, methods: A single blood sample was prospectively collected from participants prior to final clinical diagnosis of pregnancy outcome: LNSP, miscarriage, pregnancy of unknown location (PUL) or tubal EP (tEP). Human chorionic gonadotrophin β (β-hCG) and progesterone concentrations were measured in plasma. Serum samples were subjected to untargeted metabolomics profiling using nuclear magnetic resonance (NMR) spectroscopy. Metabolite abundances underwent statistical analyses together with patient demographic data and biochemical markers. The cohort was randomly split into train and validation data sets. The train data set was subjected to variable selection using linear mixed models and 10-fold cross-validation with Least Absolute Shrinkage and Selection Operator (LASSO) selection. Random forest models were constructed using selected metabolites, biochemical markers, and metadata variables.

Main results and the role of chance: β-hCG and progesterone concentrations were significantly higher in the LNSP group compared with miscarriage, PUL and tEP. Thirty-two unique metabolites were identified in serum, along with 70 unlabelled signals of unknown origin. Of these, 21 metabolite signals exhibited significantly different abundances in the LNSP group when compared with miscarriage, PUL and tEP. When miscarriage, PUL and tEP were grouped together into combined adverse outcomes (CAO), 15 metabolites were significantly different compared with LNSP. LASSO selection identified nine metabolite signals as key discriminators of LNSP versus CAO. Addition of covariates β-hCG, progesterone, participant age, BMI and gestational age improved group separation. Random forest models were constructed using stable metabolite signals alone, or in combination with plasma hormone concentrations and demographic data. Using the independent validation data set. Model performance was poor when all four pregnancy outcomes were included, regardless of variable inclusion (accuracy = 0.48–0.58). When comparing LNSP with CAO, a model with stable metabolite signals only demonstrated a modest predictive accuracy (0.68), which was comparable to a model of β-hCG and progesterone (0.71). The best model for LNSP prediction comprised stable metabolite signals and hormone concentrations (accuracy = 0.79), while the addition of clinicodemographic variables did not improve the model further.

Large scale data: NMR data is available via the EMBL-EBI MetaboLights repository (ID: MTBLS6219).

Limitations, reasons for caution: Women with LNSPs presented at the EPAU with pain and/or bleeding. The presence of symptoms may influence the metabolome of this group versus LNSPs without symptoms, thus limiting the translation of these findings. Furthermore, environmental factors were not controlled (e.g., fasting status), making it likely that cohort heterogeneity was enhanced. Finally, a larger study should be conducted to establish the robustness of composite biomarkers for LNSP diagnosis in an independent cohort.

Wider implications of the findings: A diagnostic test to confirm LNSP and thus exclude potentially life-threatening pregnancy outcomes would be invaluable in emergency situations.

Study funding/competing interest(s): This work was funded by a Wellbeing of Women Project Grant (RG2137). C.J.H. is supported by the Wellbeing of Women (RG2137), SRI/Bayer and Wellcome Trust IFFS3. N.T. is supported by an Academic Clinical Lectureship from the National Institute of Health Research (NIHR). P.B. was supported by an NIHR Academic Clinical Fellowship. P.J.D. was supported by a Clinical Research Fellowship from the Liverpool Women’s Hospital NHS Foundation Trust. D.K.H. is supported by the Wellbeing of Women (RG2137) and A.M. and D.K.H. are supported by an MRC Clinical Research Training Fellowship (MR/V007238/1). A.W.H. has received honoraria for consultancy for Ferring, Roche, Nordic Pharma, Gesynta, Joii and Abbvie. A.W.H. has received payment for presentations from Theramex and Gideon Richter. The remaining authors have no competing interests to report.

INSTRUMENT(S): Nuclear Magnetic Resonance (NMR) -

PROVIDER: MTBLS6219 | MetaboLights | 2025-08-14

REPOSITORIES: MetaboLights

ACCESS DATA

Dataset's files

Source:

Items per page:

1 - 5 of 320

Publications

Diagnostic utility of clinicodemographic, biochemical and metabolite variables to identify viable pregnancies in a symptomatic cohort during early gestation.

Hill Christopher J CJ Phelan Marie M MM Dutton Philip J PJ Busuulwa Paula P Maclean Alison A Davison Andrew S AS Drury Josephine A JA Tempest Nicola N Horne Andrew W AW Gutiérrez Eva Caamaño EC Hapangama Dharani K DK

Scientific reports 20240515 1

A significant number of pregnancies are lost in the first trimester and 1-2% are ectopic pregnancies (EPs). Early pregnancy loss in general can cause significant morbidity with bleeding or infection, while EPs are the leading cause of maternal mortality in the first trimester. Symptoms of pregnancy loss and EP are very similar (including pain and bleeding); however, these symptoms are also common in live normally sited pregnancies (LNSP). To date, no biomarkers have been identified to differenti ...[more]

PMID: 38750192

Similar Datasets

Project description:Objective To characterise the vaginal microbiota, mucosal metabolome and host immune response in early pregnancy and investigate its relationship with adverse pregnancy outcomes, including ectopic pregnancy, within the context of pregnancies of unknown location (PUL). DesignProspective cohort study Setting Queen Charlotte’s and Chelsea Hospital, Imperial College Healthcare NHS Trust, London, UK. Population Ninety-one pregnancies of which 22 patients had a favourable outcome of a viable intrauterine pregnancy (VIUP). The remainder had an adverse outcome including 15 with a non-viable intrauterine pregnancy (NVIUP, i.e. miscarriage), 26 a failed PUL (FPUL), 20 an ectopic pregnancy (EP) and 8 a persistent PUL (PPUL). Methods Two matching pairs of vaginal swab samples were collected from women as early as four weeks gestation in pregnancies that resulted in an ectopic pregnancy, miscarriage or viable intrauterine pregnancy pregnancies matched for age, gestation and body mass index. Sequencing of the V1-V2 region of the 16S rRNA gene amplicon was used to characterize and compare vaginal bacterial compositions. The second of the pair of vaginal swabs was used sequentially first for DESI-MS direct-on swab untargeted metabolite profiling followed by extraction of the protein content for quantitative analysis of chemokine and cytokine levels using a 15-plex Luminex immune-profiling assay. Results Adverse final pregnancy outcomes were associated with reduced Lactobacillus spp. abundance (100% vs. 75.4%, p-value=9.83 x 10-3) and higher Shannon α-diversity (p-value=1.10 x 10-3) when compared to viable pregnancies. This association was independent of vaginal bleeding and observed prior to the diagnosis of the final pregnancy outcome (i.e. before the pregnancy was visible on ultrasound). Ectopic pregnancy had an even stronger association with Lactobacillus spp. depletion when compared to viable pregnancies (30% vs 0%, p-value =5.52 x 10-3). Although a strong immune mediator signature, which included MMP-1, IL-6, CCL-2/MCP-1, TNF-α, amongst others was observed in adverse pregnancy outcomes, vaginal bleeding was identified as a major confounder and adjustment of models for bleeding removed the association with outcome. Vaginal bleeding was also found to impact metabolic profiles, increasing the abundances of multiple lipid species. The prediction of outcome based on metabolic profiles was not possible but metabolic profiles were predictive with high accuracy of vaginal microbial composition at the genera level (L. dominant vs. L. deplete), and metabolic correlates of host immune activation were identified, mainly composed of lipids. Conclusions Early pregnancy vaginal microbiome communities dominated by L. crispatus or L. gasseri were observed in women with a PUL who go onto have a viable intrauterine pregnancy. Conversely, a vaginal microbiota deplete in Lactobacillus spp or dominated by L. iners is associated with a diagnosis of ectopic pregnancy in a PUL population. These findings suggest that vaginal microbiota composition is a risk factor for ectopic pregnancy. Vaginal bleeding is an inevitable cofounding factor that must be taken into consideration when performing multi-omic analysis of vaginal mucosal samples in similar clinical populations. Immune and metabolic profiles were particularly impacted by bleeding and bleeding could greatly impact the diagnostic usefulness of immune marker profiling. Further studies are required to clarify the role of microbes and infection in implantation and ectopic pregnancy, as well as determine the mechanistic pathways by which sub-optimal vaginal microbial composition increases risk. Through the integration of metataxonomics, metabolomic and immune profiling data obtained from corresponding samples, our findings demonstrate the robust predictive capacity of specific metabolome signatures. These signatures enable the simultaneous prediction of both the composition of the vaginal microbiome and the inflammatory status of the host, even in the presence of bleeding. The data derived from direct on-swab metabolic profiling using DESI-MS holds promise for swiftly stratifying the risk of early pregnancy loss by rapidly assessing the dynamics between the vaginal microbiota and host. Further validation, however, is essential for future studies to confirm this potential. Linked cross omic data sets:Meta-taxonomics data associated with this study are available in the European Nucleotide Archive (ENA): accession number PRJEB72306.

			Action	DRS
	1.zip	Other
	10.zip	Other
	100.zip	Other
	101.zip	Other
	102.zip	Other

Dataset Information

Diagnostic utility of clinicodemographic, biochemical and metabolite variables to identify viable pregnancies in a symptomatic cohort during early gestation.

Dataset's files

Publications

Diagnostic utility of clinicodemographic, biochemical and metabolite variables to identify viable pregnancies in a symptomatic cohort during early gestation.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets