Predicting spontaneous preterm birth: Integrating maternal vaginal and gut microbiome profiles with individual risk factors

Predicting spontaneous preterm birth: Integrating maternal vaginal and gut microbiome profiles with individual risk factors | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Predicting spontaneous preterm birth: Integrating maternal vaginal and gut microbiome profiles with individual risk factors Lars Engstrand, Nicole Wagner, Unnur Gudnadottir, Fredrik Boulund, and 13 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7621941/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Preterm birth remains the leading cause of neonatal morbidity and mortality globally, with rates staying unchanged despite advances in obstetric and neonatal care. Inflammation and infection are key mechanisms implicated in preterm birth, with the maternal microbiome emerging as a potential predictor and therapeutic target. This study aimed to predict spontaneous preterm birth by analysing vaginal and faecal microbiomes (collected at <20 weeks and 28–30 weeks of gestation) together with extensive questionnaire data. Leveraging the largest microbiome cohort for preterm birth to date (collected in Sweden, 2017–2021, with 132 cases of spontaneous preterm birth), we developed a machine learning-based prediction model, achieving an AUROC of 0.89. This work underscores the significant potential of early prediction for preterm birth, highlighting that accurate prediction relies on the integration of lifestyle, health status, and microbiome composition. Our results provide a pathway for developing targeted prevention strategies. Health sciences/Biomarkers/Predictive markers Health sciences/Biomarkers/Diagnostic markers Preterm birth microbiome machine learning pregnancy prediction Figures Figure 1 Figure 2 Figure 3 Introduction Preterm birth (PTB) is the leading cause of neonatal morbidity and mortality, ranging from 4-16% worldwide [1, 2] and approximately 6% in Sweden [3]. Predicting PTB is difficult [4]; however, early detection of women at risk could advance preventive interventions, decrease complications, and improve outcomes. PTB is defined as delivery before 37 weeks of gestation, further classified as extremely preterm (<28 completed weeks), very preterm (28-31 completed weeks) and moderate to late PTB (32-36 completed weeks) [1, 5]. PTB can commence through induction, caesarean (C-) section, or spontaneous birth with intact membranes or after preterm premature rupture of membranes (PPROM) [6]. Known risk factors for spontaneous preterm birth (sPTB) include previous PTB, multiple pregnancies (being pregnant with more than one foetus), a history of cervical insufficiency, inflammation, and intrauterine infection [6], although most (two-thirds) sPTB occurs without evident cause [7]. Other sPTB risk factors include low maternal gestational weight gain and short interpregnancy interval following miscarriage [8]. Currently, there is no consensus about the association between the composition of the vaginal microbiome (the collection of bacteria, viruses, and fungi in the vagina) and sPTB, as previous studies are based on smaller cohorts and often single sampling. However, the microbiomes of the vagina and gut during pregnancy are highly dynamic. In the vagina, both alpha-diversity (within each sample) and beta-diversity (between samples) are gradually reduced during pregnancy, with Lactobacillus species becoming dominant, likely related to the significant increase in circulating oestrogens [9, 10]. Low diversity is generally considered ideal in the vaginal microbiome, in contrast to the gut, where high diversity generally is associated with health. In contrast, the alpha diversity in the gut decreases during pregnancy while beta diversity increases, suggesting that the gut microbiome becomes unique for each pregnant woman [11]. A few recent studies on sPTB found no association with microbiome characteristics [12-14], while others reported a link between sPTB and bacterial species commonly associated with vaginal dysbiosis (such as BVAB1, Sneathia spp., Gardnerella vaginalis and Prevotella spp. ) [13, 15-17], or with microbially derived metabolites [18]. Our recent network meta-analysis of vaginal microbial composition and sPTB concluded that women with non-Lactobacilli dominance were at higher risk for sPTB compared to women with Lactobacillus crispatus -dominant vaginal microbiome [19], in agreement with two other meta-analyses [20, 21]. Vaginal dysbiosis, defined as relative lack of Lactobacillus species with excesses of other anaerobic bacteria in the vagina, also has been associated with higher risk of sexually transmitted infections [22], poor fertility treatment outcome [23], human papillomavirus infection and gynaecological cancers [24, 25]. Such dysbiosis may lead to inflammation, locally on the cervix or systemically through, e.g. a leaky gut phenotype, which could, in turn, trigger sPTB onset. Recent studies investigating the risk of sPTB have explored either the maternal oral [26] or vaginal [27] microbiome, with prediction accuracies ranging from 72 to 87%. Factors including gynaecological comorbidities, dietary habits and smoking also are known to be associated with both the maternal microbiome and the risk of PTB [6]. As such, we propose a novel machine-learning framework for the predictive modelling of sPTB, incorporating a longitudinal approach and accounting for the combined effects of microbiome data and a wide range of health and lifestyle factors. The present study leverages data from a large pregnancy microbiome cohort, including faecal and vaginal samples, detailed health data and unique linkage to high-quality registries. Results A total of 5,439 participants completed the first questionnaire and gave consent for registry data collection. Of these, 3,973 also provided the first microbiome samples prior to gestation week 20 (Supplementary Figure S3). Of 3,973 participants who provided a total of 12,902 oral, vaginal, and faecal samples, 132 (3.3%) participants had sPTB (cases), of which 9% had extreme sPTB, 11% had very sPTB, 11% had moderate sPTB, and 68% had late sPTB (Table 1). These 132 cases were matched 1:2 with 264 controls who gave birth at term (> 38 completed weeks). The associations between health (questionnaire) data and sPTB were assessed through univariable and multivariable analyses for the whole cohort and the case-control sub cohort (Tables S1 and S2). For the entire cohort, multivariable logistic regression showed that having had previous PTB (aOR 9.15, 95%CI 2.98-24.92) or using gastrointestinal medication (such as proton pump inhibitors and laxatives) during early pregnancy (aOR 3.40, 95%CI 1.10-8.91) was associated with increased risk of sPTB. No variables were significant for the case-control subset in univariable or multivariable logistic regression. Vaginal microbiome The vaginal microbiome taxonomic composition of women who developed sPTB and controls had the same dominant species. At both time points, the vaginal microbiome was predominantly characterized by high abundance of Lactobacillus species, particularly L. crispatus, L. iners and L. jensenii , as well as Gardnerella vaginalis (Supplementary Figure 2). The largest difference was the presence of Megasphaera lorna and Alloscardovia microlens in some sPTB cases, while absent from control samples (Supplementary Figure S4). As measured by the Shannon index, alpha-diversity of cases and controls did not differ significantly (Figure 1E). However, considering sPTB classification, participants in the extreme (p = 0.031, Figure 1A), very (p=0.048, Figure 1B) and moderate (p = 0.030, Figure 1C) sPTB subgroups, significantly differed from controls in alpha diversity. At both timepoints, no taxa significantly differed between cases and controls. However, at the second time point, subjects with moderate sPTB had significantly higher alpha diversity (p= 0.041) than controls (Figure 1D). Permutational Analysis of variance (PERMANOVA) was employed to examine the relationship between the microbiome (combining cases and controls) and variables derived from health data, to select variables to include in the machine learning models (Supplementary Table S3). For vaginal samples at timepoint one, there were significant associations between microbiome composition and parity, conception type, snus (snuff – smokeless tobacco) use, BMI prior to and during pregnancy, endometriosis and fiber in the diet (Figure 1F). At point two, significant associations were found between microbiome and iron supplementation and the number of previous pregnancies, parity, BMI prior to and during pregnancy, and snus (snuff) usage were found (Figure 1G). Faecal microbiome At both time points, the faecal microbiome exhibited high diversity in both cases and controls, with the most abundant species being Eubacterium rectale, Ruminococcus bromine , and multiple species from genera Bifidobacterium , Faecalibacterium , Blautia , and Clostridium . No significant change in alpha diversity was found between the two time points for either cases or controls (Figure S2). However, Shannon’s index for moderate sPTB at timepoint two significantly differed from controls (p=0.041; Figure 2A, 2B). Richness and evenness significantly differed between very sPTB and controls (Richness p=0.046, evenness p=0.041) at timepoint one. When combining cases and controls, faecal microbiome composition was associated with regularity of menstruation, socioeconomic score, Pregnancy-Unique Quantification of Emesis (PUQE) rating, Bristol stool form scale, alcohol usage three months before pregnancy, BMI and several dietary factors at timepoint one (Figure 2C; univariable PERMANOVA). For timepoint two, menstruation factors (regularity and pain), BMI, several dietary factors, Bristol rating and SES score all were significantly associated to faecal microbiome composition (Figure 2D). All factors assessed in PERMANOVA are shown in Supplementary Table S3. For analysis of differential abundance using ANCOM-BC2, samples were split into cases and controls with BMI and antibiotic use during pregnancy as covariates to control for confounding. Eight species were found to show a significant (Benjamini-Hochberg (BH)-adjusted) log-fold change >0.5 at timepoint one and six at timepoint two. At timepoint one, Alistipes dispar , Bifidobacterium animalis and Intestinibacter bartlettii were more abundant in cases, while Prevotella timonensis , Christensenella sp. Marseille P3954, Eubacteriales SGB15145 and Rothia mucilaginosa were more abundant in the controls (Figure 2E). For timepoint two, only Schaalia turicensis was more abundant in cases, while GGB9063 SGB13982 (family Clostridiaceae ), Eubacteriales SGB15145, Rheinheimera SGB14999, GGB2980 SGB3962 (order Eubacteriales) and Lachnospiraceae bacterium BX10 were more abundant in controls (Figure 2F). Predictive modelling Random Forest (RF), Neural Net and Support Vector Machines (SVM) were all used to predict sPTB with data from timepoint one. RF demonstrated the highest accuracy, with the area under the receiver-operator characteristic curve (AUROC) of 0.51 for vaginal microbiome only and up to 0.77, combining both microbiomes, health data, and register data (Figure 3, Table 2). In contrast, for Neural Nets, the best AUROC values ranged from 0.3 (microbiome only) to 0.47 (including patient-level data). For SVM, the model for the microbiome alone yielded an AUROC of 0.39, and including contextual data increased it to 0.41. The RF model was refined through a second round of training, excluding variables that did not contribute to the first model. Additionally, since previous work indicated that adding all alpha-diversity measures as predictors improved the models, we used a similar approach [27] (Figure 3A, 3B). Table 2 contains all AUROC, sensitivity, and specificity of the RF models. Using the best model for timepoint one (denoised fecal and vaginal species along with health data and alpha diversity combined), the health data variables, vaginal and fecal species as most important for the model were identified. This model achieved an AUROC of 0.89 and a sensitivity of 0.76 (Figure 3B, Table 2), with a negative predictive value (NPV) of 0.82 and a positive predictive value (PPV) of 0.87. The ten most important patient-level variables were previous repeated pregnancy loss, previous intra-uterine foetal death, Bristol rating, previous miscarriage, blood pressure, maternal age, previous PTB, smoking, endometriosis and PCOS (Figure 3D). The 10 most critical vaginal species in the combined model were Coriobacteriales bacterium DNF00809 , unclassified Lachnospiraceae, Cutibacterium acnes, Campylobacter ureolyticus, Lawsonella SGB3665 , Cutibacterium granulosum, Corynebacterium college, Companilactobacillus custom, Lactobacillus kefiranofaciens and Blautia glucanase (Figure 3E). The 10 most important faecal species were found to be Blautia hydrogenotrophica, Porphyromonas SGB1980 , Anaerobutyricum hallii, Lachnospiraceae, GGB3537 SGB4727, Eisenbergiella massiliensis, Akkermansia muciniphila, GGB3160 SGB4174 (phylum Firmicutes), Clostridiaceae GGB9782 SGB15403, Schaalia turicensis, and Oscillospiraceae GGB9707 SGB15229 (Figure 3F). A similar approach was used to refine the initial RF models for time point two. A model also was built using the difference between the CLR values of the species present at both time points as well as the health data. The models from time point two did not perform as well as those from time point one (Table 2). The AUROC curves and most important variables from time point two are in Supplementary Figure S5. Discussion This study sought to test the value of human microbiome characteristics for predicting sPTB, while considering the multifaceted and complex aetiology underlying microbial effects. Effective prediction, which could guide preventive measures, would greatly benefit maternal and infant health outcomes. Our study uniquely explores both vaginal and faecal microbiomes, together with an extensive questionnaire covering health, pregnancy history, and lifestyle factors and leveraged these background data with samples collected at two time points in pregnancy to construct robust prediction models using a variety of approaches. Recently, machine learning methods have been used to create prediction models for PTB using vaginal microbiome metabolites [18], vaginal [27, 28] and oral microbiome [26] or other health-related variables [29-33] with AUROCs ranging from 0.69-0.81. Currently, the best prediction model published for PTB (AUROC 0.89) used clinical indicators derived from hospital tests and background variables from late pregnancy but did not include microbiome data [34]. Our models using vaginal microbiome, faecal microbiome, or health data separately yielded suboptimal prediction rates, ranging from 52% to 74%. However, integrating all datasets and subsequent feature selection from early pregnancy and alpha diversity indices achieved an AUROC of 89%, higher than previously proposed models for sPTB [18, 26-31]. Data from later in pregnancy yielded lower accuracy (AUROC 77%). While seemingly counter-intuitive, a recent meta-analysis found that the vaginal microbiome is a better predictor of early preterm birth than later [21]. Since the later timepoint comes after the earliest preterm deliveries, we also had less power using only data collected after week 28. Nevertheless, from a clinical perspective, earlier detection of women at risk is most beneficial, as there is a longer window of opportunity to modulate the risk. Among the species important for the prediction model, many with the highest importance were from the gut; half of the top 10 were metagenomically assigned genomes (MAGs), underscoring the role of the gut microbiome in pregnancy outcomes. S. turicensis has previously been linked to adverse pregnancy and postpartum outcomes [35]. A. municipal also has been associated with increased preeclampsia risk in mouse models, which may contribute to preterm birth (PTB) [36]. Previous studies have shown that a Lactobacillus-dominated is associated with a lower PTB risk [19]. In contrast, a highly diverse vaginal microbiome dominated by Gardnerella , Atopobium and Sneathia has been associated with increased risk of PTB [19, 37, 38]. Interestingly, in our best model, none of these species were among the 10 most important for classifying cases and controls. Of note is that the addition of alpha diversity measures improved prediction, which could mean that the overall state of the vaginal and faecal microbiome is as important as the species present. The maternal microbiome adapts [9, 10] as a healthy pregnancy progresses; lack of such adaptability might increase PTB risk. The most important strength of this study is the comprehensive and longitudinal sample collection from a large representative national cohort, with both biological samples and extensive data on individual risk factors. These could be used in advanced statistical predictive modelling with a balanced data set to compensate for rare outcomes in our population. The machine learning approaches permit exploring this complex dataset with minimal information loss. The models were optimized using leave-one-out cross-validation, bootstrapping, and repeated cross-validation techniques. Another strength is the comparison of multiple machine learning algorithms, where the AUROC values and accuracy varied greatly, highlighting the importance of testing several methods when creating prediction models. Nevertheless, the accuracy of the models may vary depending on socio-geographics. We studied a relatively homogenous cohort, composed primarily of Swedish-born women aged 25-35 and of normal BMI. In future studies, stratification by parity will be needed. Direct validation in an external cohort is currently limited by methodological heterogeneity across studies, particularly in sample collection protocols and timing. These differences reduce comparability and make replication efforts less informative at this stage. Including alpha-diversity metrics and changes between time points limit the potential translatability of this model to clinical settings. However, our findings may help identify key players in sPTB, which could be used as prognostic markers. Better understanding of microbiome involvement in inducing sPTB can improve obstetric care, e.g., by modifying a high-risk microbiome using dietary interventions or live biotherapeutics. In conclusion, this study shows that sPTB can be predicted early in pregnancy using a machine-learning model that integrates vaginal and gut microbiome profiles with individual risk factors. Combining these data types, the model achieved high predictive accuracy, emphasizing the power of a multidimensional approach to understanding PTB risk. These findings highlight the potential of microbiome-informed prediction tools to identify pregnancies at risk and pave the way for developing targeted preventive strategies to reduce PTB rates. Online methods Study design and cohort selection Training and testing data in this manuscript stems from the Swedish Maternal Microbiome (SweMaMi) study [39]. Inclusion criteria were ongoing pregnancy, minimum age of 18, living in Sweden with a personal identification number and speaking Swedish or English [39]. Although the data collection did not include questions on gender, and although we acknowledge that not all pregnant individuals identify as women, we used the term women to describe participants in this study. All study participants provided informed consent, which could be freely withdrawn. The studies complied with the Declaration of Helsinki and the General Data Protection Regulation (GDPR). The SweMaMi study protocol was approved by the Regional Board of Ethics, Stockholm, Sweden (2017/1118-31, amendments 2020-01629 and 2023-00875-02). Data collection Data were collected at three time points (gestational weeks 10-20, 28-30 and 5-8 weeks after calculated delivery date) between November 2017 and November 2021 using home sampling and online questionnaires. The personal identification numbers were also linked to Swedish quality registries for additional information. Questionnaires and registries Web-based questionnaires were filled out twice during pregnancy and once postpartum [39]. These questionnaires included information about pregnancy and health characteristics; parity, height and pre-pregnancy weight, perceived Stress (the Perceived Stress Scale (PSS-4) [40]), depressive symptoms (the Edinburgh Postnatal Depression Scale (EPDS) [41]), method of conception (assisted or not), pregnancy history (previous PTB, recurrent pregnancy loss or intrauterine foetal death (IUFD)), dietary habits (such as frequency of dietary fibre intake and healthy diet (based on intake of fruit, vegetables, whole grain bread, sugar- and sugar free drinks)), intake of iron supplements, smoking and use of snuff (smokeless tobacco), parity, any pregnancy problems in current pregnancy (gestational diabetes, thyroid disease, high blood pressure, hyperemesis gravidarum, vaginal bleeding, heartburn, symphysis pubis dysfunction or other), pregnancy-unique quantification of emesis (PUQE) score [42], drug use during pregnancy, gynaecological health, regular menstruation, endometriosis or polycystic ovary syndrome (PCOS). To supplement the health data, information from the Swedish Pregnancy Register [43] (a quality registry) was collected regarding gestational length (completed weeks and days of gestation) and delivery onset (spontaneous, induction, or planned C-section). Registry data on pregnancy length and delivery mode were prioritized over questionnaire answers to classify cases and controls. Microbiome samples Samples for vaginal and faecal microbiome analysis were collected between weeks 10-20 of gestation and again at weeks 28-30. Vaginal swabs and faecal samples were self-collected by participants and stored in DNA/RNA shield (Zymo Research, California, USA), which ensures sample stability at room temperature during transport, after which the samples were stored at -80°C [39]. DNA was extracted by combining bead-beating and chemical lysis as previously described [39]. A total of 5870 vaginal and 3940 faecal samples were shotgun sequenced using an MGI T7 sequencer with PE150 reads. The raw paired-end reads were trimmed, filtered and annotated using our in-house Snakemake (v4.8.1) workflow StaG-mwc (v0.5.1) [44] [45]. The workflow used fastp (v0.23.2) with default settings for adapter removal and quality trimming [46]. Kraken2 (v2.1.2) [47] annotation against the GRCh38 human genome was used for host DNA removal [48]. Taxonomic annotation was performed with MetaPhlAn 4.0 [49]. Taxonomic tables were cleaned using the R decontam package [50] in prevalence mode and with a threshold of 0.5. Additionally, any species with less than 100 raw counts across all samples were removed from the data. Data analysis All statistical analyses were performed in R (2022.02.4+500) [51]. All plots were created using the R package ggplot2 v3.4.2 [52]. A threshold of p-value <0.05 was set for all tests unless otherwise stated. Correction for multiple testing was done with the Benjamini-Hochberg (BH) procedure where stated. Case-control selection Cases were selected as the 132 women who gave birth spontaneously before 37 completed weeks (preterm) and had both filled in the first questionnaire and sent in the first faecal samples. Each case was matched with two controls, defined as women who gave birth after 38completed weeks of pregnancy. Matching was performed using the R package MatchIt [53] with the nearest neighbour method. The variables included in the matching were age (continuous), pre-pregnancy BMI (continuous), pregnancy week at sampling (continuous), nulliparity (yes/no) and diagnosed PCOS (yes/no). Diversity metrics Alpha diversity was assessed using Shannon’s diversity, Pielou’s evenness and Observed Species Richness [54]. Wilcoxon was used to test for differences in alpha diversity between cases and controls. sPTB cases were also investigated as subcategories of sPTB (extreme, very, moderate and late sPTB). Beta-diversity For beta-diversity analyses, data was transformed through the addition of a pseudo count of 10E-7 and submitted to a centred-log ratio (CLR) transformation using the R package compositions v2.0-8 [55] before being submitted to the verdict function (Vegan package v2.6-4 [54]) with Euclidean distances. To determine whether the health data collected from the questionnaires (Table S3) were associated with the microbiome profile, a Permutational analysis of variance (PERMANOVA) analysis was performed using the Adonis2 function [56] from the R package Vegan v2.6-4 [54]. Given the unbalanced nature of many categorical variables, we performed PERMANOVA using random subsampling with replacement, using the R package Scutr v0.1.2 [57]. Ten iterations of subsampling were done, and if six or more of the iterations returned a significant difference (the rule of majorities) post-correction with the BH method, the variable was considered important. For multivariable models, since the order of factors in the PERMANOVA model may affect results, random shuffling of variables was performed, using either 999 permutations (when including all available health data) or 99 permutations (when including only the ones significant in univariable analysis). In either case, a rule of majorities was adopted. Differential abundance Differential abundance analyses, including adjustment for maternal BMI, parity and antibiotic usage, were performed at each time point for both vaginal and faecal microbiome data using ANCOM-BC2 v2.0.1 [55, 58]. Prediction modelling Classification methods included the Random Forest (RF) algorithm [59] using the R Caret package v6.0-94 [60], along with the randomForest package v4.7-1,1 [61], Support Vector Machines (SVM) using the R Caret package v6.0.94 [60], and Neural Networks using the R package net v7.3.18 [62]. Initially, the classification analysis was carried out on CLR-transformed microbiome data and health data separately, as well as combined. Variables were included in the health data dataset if they were significant in at least one univariable and multivariable logistic regression (see Supplementary information), univariable PERMANOVA or if they had previously been described to associate with the gut or vaginal microbiota. This latter category includes all chronic diseases, medications and diet, as well as variables describing fertility and HPV infection or prevention, including a binary variable for vaginal dysbiosis, defined as <60% abundance of genus Lactobacillus . After the first round of machine learning, which showed RF to be the most promising approach, it was repeated, including only variables whose importance in the first round was >0 (function VarImp in R package Caret). These included both microbiome and health or registry data. A third run was performed using these same VarImp variables and alpha-diversity metrics (Shannon, Pielou, Richness and Inverse Simpson). A final machine learning run was conducted for the second timepoint only, incorporating all the previous information and the difference in CLR transformed relative abundance of bacterial species. We performed leave-one-out validation for all models, cross-validation with 10 repeats and bootstrapping with 10,000 repeats. Given the unbalanced nature of the key variable (case and control), we used the R scutr package v0.1.2 [63] to randomly subsample the controls to equal the number of cases, and we ran the classifiers 10 times with a randomly subsampled control set for each loop. We used an 80:20 division for the test and training sets (respectively). We tested each classification on all 10 random test sets. ROC curves and plots were generated using pROC v1.18.4 [64]. Results are reported as AUROC values, sensitivity, specificity, negative prediction values (proportion of truly negative tests) and positive prediction values (proportion of truly positive tests). The models' confidence interval, sensitivity and specificity were obtained using the confusion matrix function of the R Caret package v6.0-94 [60]. Declarations Code availability The main code used in the work is available on the CTMR GitHub page, which is accessible at https://github.com/ctmrbio/ML_swemami. Data availability The metagenomic sequences have been submitted to ENA (project number PRJEB81814). Acknowledgements We are grateful to the participants of the SweMaMi study. Furthermore, we would like to thank all Centre for Translational Microbiome Research members who contributed to the SweMaMi study during sample collection, extraction, and sequencing, especially Alexandra Pennhag, Marica Hamsten, and Maike Seifert. We also want to acknowledge that Ferring Pharmaceuticals has funded this work with an unrestricted research grant to LE. The study was partly financed by grants from the Swedish Research Council (Dnr 2023-02868 to NB), the Swedish state under the agreement between the Swedish government and the county councils (the ALF-agreement, Region Stockholm, DNR 20200471 to EF), Åke Wiberg foundation (Dnr M21-0153 to EF), the Gillbergska Foundation (to EF), and SciLifeLab & Wallenberg Data Driven Life Science Program (KAW 2020.0239, to LWH). References World Health Organization (WHO). Preterm birth . 2023 10 May 2023; Available from: isir.is. Ohuma, E.O., et al., National, regional, and global estimates of preterm birth in 2020, with trends from 2010: a systematic analysis. Lancet, 2023. 402 (10409): p. 1261-1271. National Board of Health and Welfare, Statistics on Pregnancies, Deliveries and Newborn Infants [Graviditeter, förlossningar och nyfödda barn] . 2023, Socialstyrelsen: https://www.socialstyrelsen.se/. Jacobsson, B., et al., Prediktion, prevention och behandlingsmetoder [Prediction, prevention and treatment options] Läkartidningen, 2019. Spong, C.Y., Defining "term" pregnancy: recommendations from the Defining "Term" Pregnancy Workgroup. JAMA, 2013. 309 (23): p. 2445-6. Goldenberg, R.L., et al., Epidemiology and causes of preterm birth. Lancet, 2008. 371 (9606): p. 75-84. Vogel, J.P., et al., The global epidemiology of preterm birth. Best Pract Res Clin Obstet Gynaecol, 2018. 52 : p. 3-12. Mitrogiannis, I., et al., Risk factors for preterm birth: an umbrella review of meta-analyses of observational studies. BMC Med, 2023. 21 (1): p. 494. Romero, R., et al., The composition and stability of the vaginal microbiota of normal pregnant women is different from that of non-pregnant women. Microbiome, 2014. 2 (1): p. 4. Oliver, A., et al., Cervicovaginal Microbiome Composition Is Associated with Metabolic Profiles in Healthy Pregnancy. mBio, 2020. 11 (4). Koren, O., et al., Host remodeling of the gut microbiome and metabolic changes during pregnancy. Cell, 2012. 150 (3): p. 470-80. Pace, R.M., et al., Complex species and strain ecology of the vaginal microbiome from pregnancy to postpartum and association with preterm birth. Med (N Y), 2021. 2 (9): p. 1027-1049. Chan, D., et al., Microbial-driven preterm labour involves crosstalk between the innate and adaptive immune response. Nat Commun, 2022. 13 (1): p. 975. Liu, X., et al., Vaginal flora during pregnancy and subsequent risk of preterm birth or prelabor rupture of membranes: a nested case-control study from China. BMC Pregnancy Childbirth, 2023. 23 (1): p. 244. Fettweis, J.M., et al., The vaginal microbiome and preterm birth. Nat Med, 2019. 25 (6): p. 1012-1021. Baud, A., et al., Microbial diversity in the vaginal microbiota and its link to pregnancy outcomes. Sci Rep, 2023. 13 (1): p. 9061. Downes, K.L., et al., 15: Gardnerella vaginalis and spontaneous preterm birth: New insights. American Journal of Obstetrics and Gynecology, 2018. 218 (1, Supplement): p. S12-S13. Kindschuh, W.F., et al., Preterm birth is associated with xenobiotics and predicted by the vaginal metabolome. Nat Microbiol, 2023. 8 (2): p. 246-259. Gudnadottir, U., et al., The vaginal microbiome and the risk of preterm birth: a systematic review and network meta-analysis. Sci Rep, 2022. 12 (1): p. 7926. Kosti, I., et al., Meta-Analysis of Vaginal Microbiome Data Provides New Insights Into Preterm Birth. Front Microbiol, 2020. 11 : p. 476. Huang, C., et al., Meta-analysis reveals the vaginal microbiome is a better predictor of earlier than later preterm birth. BMC Biol, 2023. 21 (1): p. 199. Edwards, V.L., et al., The Cervicovaginal Microbiota-Host Interaction Modulates Chlamydia trachomatis Infection. mBio, 2019. 10 (4). Koedooder, R., et al., The vaginal microbiome as a predictor for outcome of in vitro fertilization with or without intracytoplasmic sperm injection: a prospective study. Hum Reprod, 2019. 34 (6): p. 1042-1054. Brusselaers, N., et al., Vaginal dysbiosis and the risk of human papillomavirus and cervical cancer: systematic review and meta-analysis. Am J Obstet Gynecol, 2019. 221 (1): p. 9-18 e8. Cheng, L., et al., Vaginal microbiota and human papillomavirus infection among young Swedish women. NPJ Biofilms Microbiomes, 2020. 6 (1): p. 39. Hong, Y.M., et al., Predicting preterm birth using machine learning techniques in oral microbiome. Sci Rep, 2023. 13 (1): p. 21105. Golob, J.L., et al., Microbiome preterm birth DREAM challenge: Crowdsourcing machine learning approaches to advance preterm birth research. Cell Rep Med, 2024. 5 (1): p. 101350. Park, S., et al., Predicting preterm birth through vaginal microbiota, cervical length, and WBC using a machine learning model. Front Microbiol, 2022. 13 : p. 912853. Arabi Belaghi, R., J. Beyene, and S.D. McDonald, Prediction of preterm birth in nulliparous women using logistic regression and machine learning. PLoS One, 2021. 16 (6): p. e0252025. Belaghi, R.A., Prediction of preterm birth in multiparous women using logistic regression and machine learning approaches. Sci Rep, 2024. 14 (1): p. 21967. Kassahun, E.A., et al., Development and validation of a simplified risk prediction model for preterm birth: a prospective cohort study in rural Ethiopia. Sci Rep, 2024. 14 (1): p. 4845. Ding, L., et al., Prediction of preterm birth using machine learning: a comprehensive analysis based on large-scale preschool children survey data in Shenzhen of China. BMC Pregnancy Childbirth, 2024. 24 (1): p. 810. Huang, C., et al., Predicting preterm birth using electronic medical records from multiple prenatal visits. BMC Pregnancy Childbirth, 2024. 24 (1): p. 843. Chen, Y., et al., Development and validation of a spontaneous preterm birth risk prediction algorithm based on maternal bioinformatics: A single-center retrospective study. BMC Pregnancy Childbirth, 2024. 24 (1): p. 763. Rajasekera, T.A., et al., Stress and depression-associated shifts in gut microbiota: A pilot study of human pregnancy. Brain Behav Immun Health, 2024. 36 : p. 100730. Liu, W., et al., Protective Effect of Akkermansia muciniphila on the Preeclampsia-Like Mouse Model. Reprod Sci, 2023. 30 (9): p. 2623-2633. Talukdar, D., et al., Previse preterm birth in early pregnancy through vaginal microbiome signatures using metagenomics and dipstick assays. iScience, 2024. 27 (11): p. 111238. Tabatabaei, N., et al., Vaginal microbiome in early pregnancy and subsequent risk of spontaneous preterm birth: a case-control study. Bjog-an International Journal of Obstetrics and Gynaecology, 2019. 126 (3): p. 349-358. Fransson, E., et al., Cohort profile: the Swedish Maternal Microbiome project (SweMaMi) - assessing the dynamic associations between the microbiome and maternal and neonatal adverse events. BMJ Open, 2022. 12 (10): p. e065825. Cohen, S., T. Kamarck, and R. Mermelstein, A global measure of perceived stress. J Health Soc Behav, 1983. 24 (4): p. 385-96. Cox, J.L., J.M. Holden, and R. Sagovsky, Detection of postnatal depression. Development of the 10-item Edinburgh Postnatal Depression Scale. Br J Psychiatry, 1987. 150 : p. 782-6. Ebrahimi, N., et al., Nausea and vomiting of pregnancy: using the 24-hour Pregnancy-Unique Quantification of Emesis (PUQE-24) scale. J Obstet Gynaecol Can, 2009. 31 (9): p. 803-807. Stephansson, O., et al., The Swedish Pregnancy Register - for quality of care improvement and research. Acta Obstet Gynecol Scand, 2018. 97 (4): p. 466-476. F. Boulund, A.A., ctmrbio/stag-mwc: StaG v0.5.1 . 2022, Zenodo. Koster, J. and S. Rahmann, Snakemake--a scalable bioinformatics workflow engine. Bioinformatics, 2012. 28 (19): p. 2520-2. Chen, S., et al., fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics, 2018. 34 (17): p. i884-i890. Wood, D.E., J. Lu, and B. Langmead, Improved metagenomic analysis with Kraken 2. Genome Biol, 2019. 20 (1): p. 257. Fuchs, S., Drechsel, Oliver, Kraken2 Database (Human, SARS-CoV2) . 2021, Zenodo. Truong, D.T., et al., Microbial strain-level population structure and genetic diversity from metagenomes. Genome Res, 2017. 27 (4): p. 626-638. Davis, N.M., et al., Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data. Microbiome, 2018. 6 (1): p. 226. RStudio Team, RStudio: Integrated Development for R . 2020, RStudio, PBC, Boston, MA. Wickham, H., ggplot2: Elegant Graphics for Data Analysis . 2016, Springer-Verlag New York. Ho D, I.K., King G, Stuart E, MatchIt: Nonparametric Preprocessing for Parametric Causal Inference. Journal of Statistical Software, 2011. Jari Oksanen, G.L.S., F. Guillaume Blanchet, Roeland Kindt, Pierre Legendre, Peter R., et al., vegan: Community Ecology Package . 2022, R package version 2.6-4. Lin, H., M. Eggesbo, and S.D. Peddada, Linear and nonlinear correlation estimators unveil undescribed taxa interactions in microbiome data. Nat Commun, 2022. 13 (1): p. 4946. Martin Henry H. Stevens, J.O. Permutational Multivariate Analysis of Variance Using Distance Matrices . 2023; Available from: https://search.r-project.org/CRAN/refmans/vegan/html/adonis.html. Ganz, K., scutr: Balancing Multiclass Datasets for Classification Tasks . 2021, R package version 0.1.2. Lin, H. and S.D. Peddada, Analysis of compositions of microbiomes with bias correction. Nat Commun, 2020. 11 (1): p. 3514. Ho, T.K. Random decision forests . in 3rd international conference on document analysis and recognition 1995. Kuhn, M., Building Predictive Models in R Using the caret Package. Journal of Statistical Software, 2008. A. Liaw, M.W., Classification and Regression by randomForest. R News, 2002. 2(3) : p. 18-22. Ripley, W.N.V.a.B.D., Modern Applied Statistics with S . Fourth ed. 2002: Springer. Ganz, K., scutr: Balancing Multiclass Datasets for Classification Tasks . 2021. Xavier Robin, N.T., Alexandre Hainard, Natalia Tiberti, Frédérique Lisacek, Jean-Charles Sanchez, Markus Müller, pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics, 2011. 12 : p. 77. Tables Table 1. Descriptive characteristics of participants in the preterm group and the matched control group in early pregnancy (questionnaire answered before 20 weeks of gestation). Controls All spontaneous preterm birth cases Total (N=264) (N=132) (N=396) Age 35 63 (24 %) 31 (23 %) 94 (24 %) BMI* prior to pregnancy 25 Overweight 83 (31 %) 42 (32 %) 125 (32 %) Swedish born 238 (90 %) 122 (92 %) 360 (91 %) University education 53 (20 %) 53 (20 %) 79 (20 %) Socioeconomic score High 95 (36 %) 50 (38 %) 145 (37 %) Low 169 (64 %) 82 (62 %) 251 (63 %) Parity Multiparous 126 (48 %) 64 (48 %) 190 (48 %) Primiparous 138 (52 %) 68 (52 %) 206 (52 %) Conception Assisted 35 (13 %) 19 (14 %) 54 (14 %) Natural 229 (87 %) 113 (86 %) 342 (86 %) Previous IUFD** 1 (0 %) 3 (2 %) 4 (1 %) Previous preterm birth 4 (2 %) 5 (4 %) 9 (2 %) Previous RPL*** 15 (6 %) 4 (3 %) 19 (5 %) Regular menstruation 215 (81 %) 112 (85 %) 327 (83 %) Diagnosed PCOS**** 18 (7 %) 9 (7 %) 27 (7 %) Diagnosed endometriosis 13 (5 %) 8 (6 %) 21 (5 %) Daily fibre 219 (83 %) 115 (87 %) 334 (84 %) Healthy diet 195 (74 %) 99 (75 %) 294 (74 %) Intake of iron during early pregnancy 1 (0 %) 1 (1 %) 2 (1 %) Snuff/smokeless tobacco use during pregnancy 0 (0 %) 1 (1 %) 1 (0 %) Smoking during pregnancy 1 (0 %) 2 (2 %) 3 (1 %) Any drugs used during pregnancy 113 (43 %) 56 (42 %) 169 (43 %) Multiple drugs used during pregnancy 96 (36 %) 47 (36 %) 143 (36 %) Thyroid medication 31 (12 %) 9 (7 %) 40 (10 %) Allergy and antihistamines 29 (11 %) 15 (11 %) 44 (11 %) Neurological medication 27 (10 %) 13 (10 %) 40 (10 %) Gastrointestinal medication 0 (0 %) 5 (4 %) 5 (1 %) Diabetes medication 1 (0 %) 1 (1 %) 2 (1 %) Antibiotics 0 (0 %) 0 (0 %) 0 (0 %) EPDS § score mean (SD) 7.9 (± 6.6) 10 (± 5.9) 7.6 (± 6.4) PSS ¤ score mean (SD) 7.3 (± 3.0) 6.1 (± 3.7) 7.1 (± 3.0) Bristol stool scale group Fast transit 64 (24%) 38 (29%) 102 (26%) Normal transit 52 (20%) 25 (19%) 77 (19%) Slow transit 81 (31%) 30 (23%) 111 (28%) Various transit 67 (25%) 39 (30%) 106 (27%) *Body mass index, **Intrauterine foetal demise, ***Recurrent pregnancy loss, ****Polycystic ovary syndrome, § Edinburgh Postnatal Depression Scale, ¤ Perceived stress scale Table 2. AUROC values (with confidence intervals) for Random Forest machine learning models of the ten random test sets. The best model is shown in bold. Timepoint one Timepoint two AUC (95% CI) Sensitivity Specificity AUC (95% CI) Sensitivity Specificity Vaginal microbiome 0.51 (0.36-0.68) 0.47 0.57 0.54 (0.27-0.62) 0.19 0.67 Faecal microbiome 0.59 (0.36-0.68) 0.32 0.70 0.59 (0.38-0.73) 0.38 0.72 Health data 0.74 (0.61-0.88) 0.63 0.87 0.68 (0.56-0.87) 0.88 0.61 All combined 0.77 (0.61-0.88) 0.58 0.91 0.70 (0.49-0.83) 0.63 0.70 Denoised microbiome and health data 0.86 (0.81-0.99) 0.87 0.90 0.75 (0.43-0.77) 0.38 0.80 Denoised microbiome, alpha diversity indices and health data 0.89 (0.68-0.94) 0.76 0.90 0.77 (0.52-0.84) 0.56 0.80 Additional Declarations There is NO Competing Interest. Supplementary Files PredicitingSPTBSUPPLEMENTFinal1.docx Predicting Spontaneous Preterm Birth: Integrating Maternal Vaginal and Gut Microbiome Profiles with Individual Risk Factors RSEngstrand.pdf Reporting Summary Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7621941","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":519558683,"identity":"887eed77-5832-4adf-8c9d-4e1bce172f13","order_by":0,"name":"Lars Engstrand","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6klEQVRIiWNgGAWjYBACPlQum40BGyEtbKgstjTStRw2IOgwNvazjz8wVNyxmz+/9+HngrLzxnwM7A8f4NXCk24mwXDmWfKGY+zG0jPO3TZjY+AxxmsVG0MaGwNj2+FkAzY2Bmnetts2QC1sEni18D9j/sD473CyfBsb82/etnNALezPf+DVIpHGIMHYcNiO4RgbG9CWA0CHMZjh0wHU8oxNIuHY4QSDY2ls1jznko3ZmHmM8TqMnz+N+cOHmsP28s3HmG/zlNkZzm9vf/gBrzUgkMDAkNgA5zETVA8B9kSqGwWjYBSMgpEIAKQrOsPe+5zPAAAAAElFTkSuQmCC","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":true,"prefix":"","firstName":"Lars","middleName":"","lastName":"Engstrand","suffix":""},{"id":519558684,"identity":"ba871c0a-9928-4eb5-8a7a-806704ff37fc","order_by":1,"name":"Nicole Wagner","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Nicole","middleName":"","lastName":"Wagner","suffix":""},{"id":519558685,"identity":"14320296-c877-4657-a257-c2919a0d3040","order_by":2,"name":"Unnur Gudnadottir","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Unnur","middleName":"","lastName":"Gudnadottir","suffix":""},{"id":519558686,"identity":"72bf0e6f-1ee0-490d-9aad-685eeaab3de9","order_by":3,"name":"Fredrik Boulund","email":"","orcid":"https://orcid.org/0000-0002-3806-323X","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Fredrik","middleName":"","lastName":"Boulund","suffix":""},{"id":519558687,"identity":"2788a045-59a2-4d5b-85a5-5bb9417a4fe1","order_by":4,"name":"Luisa Hugerth","email":"","orcid":"","institution":"Uppsala University","correspondingAuthor":false,"prefix":"","firstName":"Luisa","middleName":"","lastName":"Hugerth","suffix":""},{"id":519558688,"identity":"29d97e98-1ac6-495d-97cb-96e8a9de5d33","order_by":5,"name":"Tatjana Pavlenko","email":"","orcid":"","institution":"Uppsala University","correspondingAuthor":false,"prefix":"","firstName":"Tatjana","middleName":"","lastName":"Pavlenko","suffix":""},{"id":519558689,"identity":"761b201e-c001-4059-8b20-cd1dd4f384e9","order_by":6,"name":"Anusha Antony","email":"","orcid":"","institution":"Uppsala University","correspondingAuthor":false,"prefix":"","firstName":"Anusha","middleName":"","lastName":"Antony","suffix":""},{"id":519558690,"identity":"08269ddf-18da-46d2-b214-036576a0a68d","order_by":7,"name":"Stefanie Prast-Nielsen","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Stefanie","middleName":"","lastName":"Prast-Nielsen","suffix":""},{"id":519558691,"identity":"1ed383f9-6a9f-4d34-b5b6-716a7fa5a180","order_by":8,"name":"Gustav Ahlström","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Gustav","middleName":"","lastName":"Ahlström","suffix":""},{"id":519558692,"identity":"3dc6b74e-a021-47a2-8024-ba209d7168d9","order_by":9,"name":"Bianka Csitari","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Bianka","middleName":"","lastName":"Csitari","suffix":""},{"id":519558693,"identity":"10c5a853-3988-4fd5-bae1-b2db28323bf3","order_by":10,"name":"Alkistis Skalkidou","email":"","orcid":"","institution":"Uppsala University","correspondingAuthor":false,"prefix":"","firstName":"Alkistis","middleName":"","lastName":"Skalkidou","suffix":""},{"id":519558694,"identity":"9b9df4f0-02c1-447d-80ce-b8b1f88805a5","order_by":11,"name":"Eva Wiberg Itzel","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Eva","middleName":"Wiberg","lastName":"Itzel","suffix":""},{"id":519558695,"identity":"9d79d6ff-07e3-4b29-afc8-fa83e7ba6521","order_by":12,"name":"Maria Gloria Dominguez","email":"","orcid":"","institution":"Rutgers University","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"Gloria","lastName":"Dominguez","suffix":""},{"id":519558696,"identity":"808496be-6d0a-44c1-bbc6-123a842420c4","order_by":13,"name":"Martin Blaser","email":"","orcid":"https://orcid.org/0000-0003-2447-2443","institution":"Rutgers University","correspondingAuthor":false,"prefix":"","firstName":"Martin","middleName":"","lastName":"Blaser","suffix":""},{"id":519558697,"identity":"44e3fb7d-f4d9-4438-9843-023cf66ebccc","order_by":14,"name":"Nele Brusselaers","email":"","orcid":"https://orcid.org/0000-0003-0137-447X","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Nele","middleName":"","lastName":"Brusselaers","suffix":""},{"id":519558698,"identity":"15f16930-c074-4a3b-8e67-4e16120d76e5","order_by":15,"name":"Emma Fransson","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Emma","middleName":"","lastName":"Fransson","suffix":""},{"id":519558699,"identity":"dcb29fe6-05bc-404d-a9de-e4b5efae6ea5","order_by":16,"name":"Ina Schuppe-Koistinen","email":"","orcid":"","institution":"Karolinska Institutet","correspondingAuthor":false,"prefix":"","firstName":"Ina","middleName":"","lastName":"Schuppe-Koistinen","suffix":""}],"badges":[],"createdAt":"2025-09-15 14:53:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7621941/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7621941/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92144226,"identity":"613963ed-666b-4e11-9108-ee40add05c55","added_by":"auto","created_at":"2025-09-25 06:40:20","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":155093,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAlpha- and beta-diversity analysis of vaginal samples. Shannon diversity between cases and (a) extreme PTB, (b) very PTB, (c) moderate PTB at timepoint one (pregnancy week 10-20), and (d) cases and moderate PTB at timepoint two (pregnancy week 28-30. (e) All p-values for comparison of alpha-diversity between PTB cases and controls, with significant values highlighted in pink. Significant (p\u0026lt;0.05) PERMANOVA results when comparing health data variables and the (f) vaginal microbiome at timepoint one and (g) timepoint two with Goodness-of-fit (R2) on the x-axis and p-values in colour gradient.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7621941/v1/5f83d777e31e30492dc18b8b.png"},{"id":92144228,"identity":"a35859d6-6708-42f8-9a38-aae81c618a2b","added_by":"auto","created_at":"2025-09-25 06:40:20","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":254750,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAlpha- and beta-diversity and differential abundance analysis for faecal samples. Comparison of Shannon diversity for (a) moderate PTB and controls, and (b) all p-values for comparison of PTB to controls with significant results highlighted in pink. Significant (p\u0026lt;0.05) PERMANOVA results when comparing health data variables and the faecal microbiome at (c) timepoint one and (d) timepoint two. Differential abundance analyses at (e) timepoint one and (f) timepoint two. Significantly different species (FDR-adjusted p \u0026lt; 0.05) are coloured in blue and named in the figure.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7621941/v1/f4a9fa0743c8f0e03a0a5c0c.png"},{"id":92144229,"identity":"dddf9b73-bdac-4a46-b4db-c3978333cc2a","added_by":"auto","created_at":"2025-09-25 06:40:20","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":166434,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eRandom Forest models for timepoint one with the (a) full dataset and (b) denoised data. Most important variables ranked in the best model for (c) health data variables, (d) vaginal taxa and (e) faecal taxa.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7621941/v1/4809c14102d29fd9380b512d.png"},{"id":94600302,"identity":"b5b51e34-8dd7-46d1-a0e9-63e6fbd3c130","added_by":"auto","created_at":"2025-10-28 19:15:58","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1295394,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7621941/v1/60059a60-9bda-47e6-9c72-5148d0398cec.pdf"},{"id":92144231,"identity":"7aba8e00-5746-49b6-9644-9611846827da","added_by":"auto","created_at":"2025-09-25 06:40:20","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2476951,"visible":true,"origin":"","legend":"Predicting Spontaneous Preterm Birth: Integrating Maternal Vaginal and Gut Microbiome Profiles with Individual Risk Factors","description":"","filename":"PredicitingSPTBSUPPLEMENTFinal1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7621941/v1/29a62258e7895ecd0802df28.docx"},{"id":92144230,"identity":"c4c34e32-9f3e-47b8-846b-e624f299fee4","added_by":"auto","created_at":"2025-09-25 06:40:20","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":2158110,"visible":true,"origin":"","legend":"Reporting Summary","description":"","filename":"RSEngstrand.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7621941/v1/dc157081418d72af53511075.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Predicting spontaneous preterm birth: Integrating maternal vaginal and gut microbiome profiles with individual risk factors","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePreterm birth (PTB) is the leading cause of neonatal morbidity and mortality, ranging from 4-16% worldwide [1, 2] and approximately 6% in Sweden [3]. Predicting PTB is difficult [4]; however, early detection of women at risk could advance preventive interventions, decrease complications, and improve outcomes.\u003c/p\u003e\n\u003cp\u003ePTB is defined as delivery before 37 weeks of gestation, further classified as extremely preterm (\u0026lt;28 completed weeks), very preterm (28-31 completed weeks) and moderate to late PTB (32-36 completed weeks) [1, 5]. PTB can commence through induction, caesarean (C-) section, or spontaneous birth with intact membranes or after preterm premature rupture of membranes (PPROM) [6]. Known risk factors for spontaneous preterm birth (sPTB) include previous PTB, multiple pregnancies (being pregnant with more than one foetus), a history of cervical insufficiency, inflammation, and intrauterine infection [6], although most (two-thirds) sPTB occurs without evident cause [7]. Other sPTB risk factors include low maternal gestational weight gain and short interpregnancy interval following miscarriage [8].\u003c/p\u003e\n\u003cp\u003eCurrently, there is no consensus about the association between the composition of the vaginal microbiome (the collection of bacteria, viruses, and fungi in the vagina) and sPTB, as previous studies are based on smaller cohorts and often single sampling. However, the microbiomes of the vagina and gut during pregnancy are highly dynamic. In the vagina, both alpha-diversity (within each sample) and beta-diversity (between samples) are gradually reduced during pregnancy, with \u003cem\u003eLactobacillus\u0026nbsp;\u003c/em\u003especies becoming dominant, likely related to the significant increase in circulating oestrogens [9, 10]. Low diversity is generally considered ideal in the vaginal microbiome, in contrast to the gut, where high diversity generally is associated with health. In contrast, the alpha diversity in the gut decreases during pregnancy while beta diversity increases, suggesting that the gut microbiome becomes unique for each pregnant woman [11].\u003c/p\u003e\n\u003cp\u003eA few recent studies on sPTB found no association with microbiome characteristics [12-14], while others reported a link between sPTB and bacterial species commonly associated with vaginal dysbiosis (such as BVAB1, \u003cem\u003eSneathia spp., Gardnerella vaginalis\u0026nbsp;\u003c/em\u003eand \u003cem\u003ePrevotella spp.\u003c/em\u003e) [13, 15-17], or with microbially derived metabolites [18]. Our recent network meta-analysis of vaginal microbial composition and sPTB concluded that women with non-Lactobacilli dominance were at higher risk for sPTB compared to women with \u003cem\u003eLactobacillus crispatus\u003c/em\u003e-dominant vaginal microbiome [19], in agreement with two other meta-analyses [20, 21]. Vaginal dysbiosis, defined as relative lack of \u003cem\u003eLactobacillus\u0026nbsp;\u003c/em\u003especies with excesses of other anaerobic bacteria in the vagina, also has been associated with higher risk of sexually transmitted infections [22], poor fertility treatment outcome [23], human papillomavirus infection and gynaecological cancers [24, 25]. Such dysbiosis may lead to inflammation, locally on the cervix or systemically through, e.g. a leaky gut phenotype, which could, in turn, trigger sPTB onset.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRecent studies investigating the risk of sPTB have explored either the maternal oral [26] or vaginal [27] microbiome, with prediction accuracies ranging from 72 to 87%. Factors including gynaecological comorbidities, dietary habits and smoking also are known to be associated with both the maternal microbiome and the risk of PTB [6]. As such, we propose a novel machine-learning framework for the predictive modelling of sPTB, incorporating a longitudinal approach and accounting for the combined effects of microbiome data and a wide range of health and lifestyle factors. The present study leverages data from a large pregnancy microbiome cohort, including faecal and vaginal samples, detailed health data and unique linkage to high-quality registries.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eA total of 5,439 participants completed the first questionnaire and gave consent for registry data collection. Of these, 3,973 also provided the first microbiome samples prior to gestation week 20 (Supplementary Figure S3). Of 3,973 participants who provided a total of 12,902 oral, vaginal, and faecal samples, 132 (3.3%) participants had sPTB (cases), of which 9% had extreme sPTB, 11% had very sPTB, 11% had moderate sPTB, and 68% had late sPTB (Table 1). These 132 cases were matched 1:2 with 264 controls who gave birth at term (\u0026gt; 38 completed weeks).\u003c/p\u003e\n\u003cp\u003eThe associations between health (questionnaire) data and sPTB were assessed through univariable and multivariable analyses for the whole cohort and the case-control sub cohort (Tables S1 and S2). For the entire cohort, multivariable logistic regression showed that having had previous PTB (aOR 9.15, 95%CI 2.98-24.92) or using gastrointestinal medication (such as proton pump inhibitors and laxatives) during early pregnancy (aOR 3.40, 95%CI 1.10-8.91) was associated with increased risk of sPTB. No variables were significant for the case-control subset in univariable or multivariable logistic regression.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eVaginal microbiome\u003c/p\u003e\n\u003cp\u003eThe vaginal microbiome taxonomic composition of women who developed sPTB and controls had the same dominant species. At both time points, the vaginal microbiome was predominantly characterized by high abundance of \u003cem\u003eLactobacillus\u0026nbsp;\u003c/em\u003especies, particularly \u003cem\u003eL. crispatus, L. iners\u0026nbsp;\u003c/em\u003eand\u003cem\u003e\u0026nbsp;L. jensenii\u003c/em\u003e, as well as \u003cem\u003eGardnerella vaginalis\u0026nbsp;\u003c/em\u003e(Supplementary Figure 2). The largest difference was the presence of \u003cem\u003eMegasphaera lorna\u003c/em\u003e and \u003cem\u003eAlloscardovia microlens\u003c/em\u003e in some sPTB cases, while absent from control samples (Supplementary Figure S4).\u003c/p\u003e\n\u003cp\u003eAs measured by the Shannon index, alpha-diversity of cases and controls did not differ significantly (Figure 1E). However, considering sPTB classification, participants in the extreme (p = 0.031, Figure 1A), very (p=0.048, Figure 1B) and moderate (p = 0.030, Figure 1C) sPTB subgroups, significantly differed from controls in alpha diversity. At both timepoints, no taxa significantly differed between cases and controls. However, at the second time point, subjects with moderate sPTB had significantly higher alpha diversity (p= 0.041) than controls (Figure 1D).\u003c/p\u003e\n\u003cp\u003ePermutational Analysis of variance (PERMANOVA) was employed to examine the relationship between the microbiome (combining cases and controls) and variables derived from health data, to select variables to include in the machine learning models (Supplementary Table S3). For vaginal samples at timepoint one, there were significant associations between microbiome composition and parity, conception type, snus (snuff \u0026ndash; smokeless tobacco) use, BMI prior to and during pregnancy, endometriosis and fiber in the diet (Figure 1F). At point two, significant associations were found between microbiome and iron supplementation and the number of previous pregnancies, parity, BMI prior to and during pregnancy, and snus (snuff) usage were found (Figure 1G).\u003c/p\u003e\n\u003cp\u003eFaecal microbiome\u003c/p\u003e\n\u003cp\u003eAt both time points, the faecal microbiome exhibited high diversity in both cases and controls, with the most abundant species being \u003cem\u003eEubacterium rectale, Ruminococcus bromine\u003c/em\u003e, and multiple species from genera \u003cem\u003eBifidobacterium\u003c/em\u003e, \u003cem\u003eFaecalibacterium\u003c/em\u003e, \u003cem\u003eBlautia\u003c/em\u003e, and \u003cem\u003eClostridium\u003c/em\u003e.\u003c/p\u003e\n\u003cp\u003eNo significant change in alpha diversity was found between the two time points for either cases or controls (Figure S2). However, Shannon\u0026rsquo;s index for moderate sPTB at timepoint two significantly differed from controls (p=0.041; Figure 2A, 2B). Richness and evenness significantly differed between very sPTB and controls (Richness p=0.046, evenness p=0.041) at timepoint one.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWhen combining cases and controls, faecal microbiome composition was associated with regularity of menstruation, socioeconomic score, Pregnancy-Unique Quantification of Emesis (PUQE) rating, Bristol stool form scale, alcohol usage three months before pregnancy, BMI and several dietary factors at timepoint one (Figure 2C; univariable PERMANOVA). For timepoint two, menstruation factors (regularity and pain), BMI, several dietary factors, Bristol rating and SES score all were significantly associated to faecal microbiome composition (Figure 2D). All factors assessed in PERMANOVA are shown in Supplementary Table S3.\u003c/p\u003e\n\u003cp\u003eFor analysis of differential abundance using ANCOM-BC2, samples were split into cases and controls with BMI and antibiotic use during pregnancy as covariates to control for confounding. Eight species were found to show a significant (Benjamini-Hochberg (BH)-adjusted) log-fold change \u0026gt;0.5 at timepoint one and six at timepoint two. At timepoint one, \u003cem\u003eAlistipes dispar\u003c/em\u003e, \u003cem\u003eBifidobacterium animalis\u003c/em\u003e and \u003cem\u003eIntestinibacter bartlettii\u003c/em\u003e were more abundant in cases, while \u003cem\u003ePrevotella timonensis\u003c/em\u003e, \u003cem\u003eChristensenella\u003c/em\u003e \u003cem\u003esp.\u0026nbsp;\u003c/em\u003eMarseille P3954, \u003cem\u003eEubacteriales\u003c/em\u003e SGB15145 and \u003cem\u003eRothia mucilaginosa\u003c/em\u003e were more abundant in the controls (Figure 2E).\u003c/p\u003e\n\u003cp\u003eFor timepoint two, only \u003cem\u003eSchaalia turicensis\u003c/em\u003e was more abundant in cases, while GGB9063 SGB13982 (family \u003cem\u003eClostridiaceae\u003c/em\u003e), Eubacteriales SGB15145, \u003cem\u003eRheinheimera\u0026nbsp;\u003c/em\u003eSGB14999, GGB2980 SGB3962 (order Eubacteriales) and \u003cem\u003eLachnospiraceae\u0026nbsp;\u003c/em\u003ebacterium BX10 were more abundant in controls (Figure 2F).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePredictive modelling\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRandom Forest (RF), Neural Net and Support Vector Machines (SVM) were all used to predict sPTB with data from timepoint one. RF demonstrated the highest accuracy, with the area under the receiver-operator characteristic curve (AUROC) of 0.51 for vaginal microbiome only and up to 0.77, combining both microbiomes, health data, and register data (Figure 3, Table 2). In contrast, for Neural Nets, the best AUROC values ranged from 0.3 (microbiome only) to 0.47 (including patient-level data). For SVM, the model for the microbiome alone yielded an AUROC of 0.39, and including contextual data increased it to 0.41.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe RF model was refined through a second round of training, excluding variables that did not contribute to the first model. Additionally, since previous work indicated that adding all alpha-diversity measures as predictors improved the models, we used a similar approach [27] (Figure 3A, 3B). Table 2 contains all AUROC, sensitivity, and specificity of the RF models.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUsing the best model for timepoint one (denoised fecal and vaginal species along with health data and alpha diversity combined), the health data variables, vaginal and fecal species as most important for the model were identified. This model achieved an AUROC of 0.89 and a sensitivity of 0.76 (Figure 3B, Table 2), with a negative predictive value (NPV) of 0.82 and a positive predictive value (PPV) of 0.87. The ten most important patient-level variables were previous repeated pregnancy loss, previous intra-uterine foetal death, Bristol rating, previous miscarriage, blood pressure, maternal age, previous PTB, smoking, endometriosis and PCOS (Figure 3D). The 10 most critical vaginal species in the combined model were\u003cem\u003e\u0026nbsp;\u003c/em\u003eCoriobacteriales bacterium DNF00809\u003cem\u003e,\u0026nbsp;\u003c/em\u003eunclassified \u003cem\u003eLachnospiraceae, Cutibacterium acnes, Campylobacter ureolyticus, Lawsonella\u0026nbsp;\u003c/em\u003eSGB3665\u003cem\u003e, Cutibacterium granulosum, Corynebacterium college, Companilactobacillus custom, Lactobacillus kefiranofaciens\u0026nbsp;\u003c/em\u003eand\u003cem\u003e\u0026nbsp;Blautia glucanase\u0026nbsp;\u003c/em\u003e(Figure 3E). The 10 most important faecal species were found to be \u003cem\u003eBlautia hydrogenotrophica, Porphyromonas\u0026nbsp;\u003c/em\u003eSGB1980\u003cem\u003e, Anaerobutyricum hallii, Lachnospiraceae,\u0026nbsp;\u003c/em\u003eGGB3537 SGB4727, \u003cem\u003eEisenbergiella massiliensis, Akkermansia muciniphila,\u0026nbsp;\u003c/em\u003eGGB3160 SGB4174 (phylum Firmicutes), \u003cem\u003eClostridiaceae\u0026nbsp;\u003c/em\u003eGGB9782 SGB15403, \u003cem\u003eSchaalia turicensis,\u0026nbsp;\u003c/em\u003eand \u003cem\u003eOscillospiraceae\u0026nbsp;\u003c/em\u003eGGB9707 SGB15229 (Figure 3F).\u003c/p\u003e\n\u003cp\u003eA similar approach was used to refine the initial RF models for time point two. A model also was built using the difference between the CLR values of the species present at both time points as well as the health data. The models from time point two did not perform as well as those from time point one (Table 2). The AUROC curves and most important variables from time point two are in Supplementary Figure S5.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study sought to test the value of human microbiome characteristics for predicting sPTB, while considering the multifaceted and complex aetiology underlying microbial effects. Effective prediction, which could guide preventive measures, would greatly benefit maternal and infant health outcomes. Our study uniquely explores both vaginal and faecal microbiomes, together with an extensive questionnaire covering health, pregnancy history, and lifestyle factors and leveraged these background data with samples collected at two time points in pregnancy to construct robust prediction models using a variety of approaches. Recently, machine learning methods have been used \u0026nbsp;to create prediction models for PTB using vaginal microbiome metabolites [18], vaginal [27, 28] and oral microbiome [26] or other health-related variables [29-33] with AUROCs ranging from 0.69-0.81. Currently, the best prediction model published for PTB (AUROC 0.89) used clinical indicators derived from hospital tests and background variables from late pregnancy but did not include microbiome data [34].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur models using vaginal microbiome, faecal microbiome, or health data separately yielded suboptimal prediction rates, ranging from 52% to 74%. However, integrating all datasets and subsequent feature selection from early pregnancy and alpha diversity indices achieved an AUROC of 89%, higher than previously proposed models for sPTB [18, 26-31]. Data from later in pregnancy yielded lower accuracy (AUROC 77%). While seemingly counter-intuitive, a recent meta-analysis found that the vaginal microbiome is a better predictor of early preterm birth than later [21]. Since the later timepoint comes after the earliest preterm deliveries, we also had less power using only data collected after week 28. Nevertheless, from a clinical perspective, earlier detection of women at risk is most beneficial, as there is a longer window of opportunity to modulate the risk.\u003c/p\u003e\n\u003cp\u003eAmong the species important for the prediction model, many with the highest importance were from the gut; half of the top 10 were metagenomically assigned genomes (MAGs), underscoring the role of the gut microbiome in pregnancy outcomes. \u003cem\u003eS. turicensis\u003c/em\u003e has previously been linked to adverse pregnancy and postpartum outcomes [35]. \u003cem\u003eA. municipal\u003c/em\u003e also has been associated with increased preeclampsia risk in mouse models, which may contribute to preterm birth (PTB) [36]. Previous studies have shown that a Lactobacillus-dominated is associated with a lower PTB risk [19].\u003c/p\u003e\n\u003cp\u003eIn contrast, a highly diverse vaginal microbiome dominated by \u003cem\u003eGardnerella\u003c/em\u003e, \u003cem\u003eAtopobium\u003c/em\u003e and \u003cem\u003eSneathia\u003c/em\u003e has been associated with increased risk of PTB [19, 37, 38]. Interestingly, in our best model, none of these species were among the 10 most important for classifying cases and controls. Of note is that the addition of\u0026nbsp;alpha diversity measures improved prediction, which could mean that the overall state of the vaginal and faecal microbiome is as important as the species present. The maternal microbiome adapts\u0026nbsp;[9, 10]\u0026nbsp;as a healthy pregnancy progresses; lack of such adaptability might increase PTB risk.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe most important strength of this study is the comprehensive and longitudinal sample collection from a large representative national cohort, with both biological samples and extensive data on individual risk factors. These could be used in advanced statistical predictive modelling with a balanced data set to compensate for rare outcomes in our population. The machine learning approaches permit exploring this complex dataset with minimal information loss. The models were optimized using leave-one-out cross-validation, bootstrapping, and repeated cross-validation techniques. Another strength is the comparison of multiple machine learning algorithms, where the AUROC values and accuracy varied greatly, highlighting the importance of testing several methods when creating prediction models. Nevertheless, the accuracy of the models may vary depending on socio-geographics. We studied a relatively homogenous cohort, composed primarily of Swedish-born women aged 25-35 and of normal BMI. In future studies, stratification by parity will be needed. Direct validation in an external cohort is currently limited by methodological heterogeneity across studies, particularly in sample collection protocols and timing. These differences reduce comparability and make replication efforts less informative at this stage.\u003c/p\u003e\n\u003cp\u003eIncluding alpha-diversity metrics and changes between time points limit the potential translatability of this model to clinical settings. However, our findings may help identify key players in sPTB, which could be used as prognostic markers. Better understanding of microbiome involvement in inducing sPTB can improve obstetric care, e.g., by modifying a high-risk microbiome using dietary interventions or live biotherapeutics.\u003c/p\u003e\n\u003cp\u003eIn conclusion, this study shows that sPTB can be predicted early in pregnancy using a machine-learning model that integrates vaginal and gut microbiome profiles with individual risk factors. Combining these data types, the model achieved high predictive accuracy, emphasizing the power of a multidimensional approach to understanding PTB risk. These findings highlight the potential of microbiome-informed prediction tools to identify pregnancies at risk and pave the way for developing targeted preventive strategies to reduce PTB rates.\u003c/p\u003e"},{"header":"Online methods","content":"\u003cp\u003eStudy design and cohort selection\u003c/p\u003e\n\u003cp\u003eTraining and testing data in this manuscript stems from the Swedish Maternal Microbiome (SweMaMi) study [39]. Inclusion criteria were ongoing pregnancy, minimum age of 18, living in Sweden with a personal identification number and speaking Swedish or English\u0026nbsp;[39]. Although the data collection did not include questions on gender, and although we acknowledge that not all pregnant individuals identify as women, we used the term women to describe participants in this study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll study participants provided informed consent, which could be freely withdrawn. The studies complied with the Declaration of Helsinki and the General Data Protection Regulation (GDPR). The SweMaMi study protocol was approved by the Regional Board of Ethics, Stockholm, Sweden (2017/1118-31, amendments 2020-01629 and 2023-00875-02).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData were collected at three time points (gestational weeks 10-20, 28-30 and 5-8 weeks after calculated delivery date) between November 2017 and November 2021 using home sampling and online questionnaires. The personal identification numbers were also linked to Swedish quality registries for additional information.\u003c/p\u003e\n\u003cp\u003eQuestionnaires and registries\u003c/p\u003e\n\u003cp\u003eWeb-based questionnaires were filled out twice during pregnancy and once postpartum [39]. These questionnaires included information about pregnancy and health characteristics; parity, height and pre-pregnancy weight, perceived Stress (the Perceived Stress Scale (PSS-4) [40]), depressive symptoms (the Edinburgh Postnatal Depression Scale (EPDS) [41]), method of conception (assisted or not), pregnancy history (previous PTB, recurrent pregnancy loss or intrauterine foetal death (IUFD)), dietary habits (such as frequency of dietary fibre intake and healthy diet (based on intake of fruit, vegetables, whole grain bread, sugar- and sugar free drinks)), intake of iron supplements, smoking and use of snuff (smokeless tobacco), parity, any pregnancy problems in current pregnancy (gestational diabetes, thyroid disease, high blood pressure, hyperemesis gravidarum, vaginal bleeding, heartburn, symphysis pubis dysfunction or other), pregnancy-unique quantification of emesis (PUQE) score [42], drug use during pregnancy, gynaecological health, regular menstruation, endometriosis or polycystic ovary syndrome (PCOS).\u003c/p\u003e\n\u003cp\u003eTo supplement the health data, information from the Swedish Pregnancy Register [43] (a quality registry) was collected regarding gestational length (completed weeks and days of gestation) and delivery onset (spontaneous, induction, or planned C-section). Registry data on pregnancy length and delivery mode were prioritized over questionnaire answers to classify cases and controls.\u003c/p\u003e\n\u003cp\u003eMicrobiome samples\u003c/p\u003e\n\u003cp\u003eSamples for vaginal and faecal microbiome analysis were collected between weeks 10-20 of gestation and again at weeks 28-30. Vaginal swabs and faecal samples were self-collected by participants and stored in DNA/RNA shield (Zymo Research, California, USA), which ensures sample stability at room temperature during transport, after which the samples were stored at -80°C [39]. DNA was extracted by combining bead-beating and chemical lysis as previously described [39]. A total of 5870 vaginal and 3940 faecal samples were shotgun sequenced using an MGI T7 sequencer with PE150 reads.\u003c/p\u003e\n\u003cp\u003eThe raw paired-end reads were trimmed, filtered and annotated using our in-house Snakemake (v4.8.1) workflow StaG-mwc (v0.5.1) [44] [45]. The workflow used fastp (v0.23.2) with default settings for adapter removal and quality trimming [46]. Kraken2 (v2.1.2) [47] annotation against the GRCh38 human genome was used for host DNA removal [48]. Taxonomic annotation was performed with MetaPhlAn 4.0 [49]. Taxonomic tables were cleaned using the R decontam package [50] in prevalence mode and with a threshold of 0.5. Additionally, any species with less than 100 raw counts across all samples were removed from the data.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eData analysis\u003c/p\u003e\n\u003cp\u003eAll statistical analyses were performed in R (2022.02.4+500) [51]. All plots were created using the R package ggplot2 v3.4.2 [52]. A threshold of p-value \u0026lt;0.05 was set for all tests unless otherwise stated. Correction for multiple testing was done with the Benjamini-Hochberg (BH) procedure where stated.\u003c/p\u003e\n\u003cp\u003eCase-control selection\u003c/p\u003e\n\u003cp\u003eCases were selected as the 132 women who gave birth spontaneously before 37 completed weeks (preterm) and had both filled in the first questionnaire and sent in the first faecal samples. Each case was matched with two controls, defined as women who gave birth after 38completed weeks of pregnancy. Matching was performed using the R package MatchIt [53] with the nearest neighbour method. The variables included in the matching were age (continuous), pre-pregnancy BMI (continuous), pregnancy week at sampling (continuous), nulliparity (yes/no) and diagnosed PCOS (yes/no).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eDiversity metrics\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAlpha diversity was assessed using Shannon’s diversity, Pielou’s evenness and Observed Species Richness [54]. Wilcoxon was used to test for differences in alpha diversity between cases and controls. sPTB cases were also investigated as subcategories of sPTB (extreme, very, moderate and late sPTB). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBeta-diversity\u003c/p\u003e\n\u003cp\u003eFor beta-diversity analyses, data was transformed through the addition of a pseudo count of 10E-7 and submitted to a centred-log ratio (CLR) transformation using the R package compositions v2.0-8 [55]\u0026nbsp;before being submitted to the verdict function (Vegan package v2.6-4\u0026nbsp;[54]) with Euclidean distances.\u003c/p\u003e\n\u003cp\u003eTo determine whether the health data collected from the questionnaires (Table S3) were associated with the microbiome profile, a Permutational analysis of variance (PERMANOVA) analysis was performed using the Adonis2 function [56] from the R package Vegan v2.6-4 [54]. Given the unbalanced nature of many categorical variables, we performed PERMANOVA using random subsampling with replacement, using the R package Scutr v0.1.2 [57]. Ten iterations of subsampling were done, and if six or more of the iterations returned a significant difference (the rule of majorities) post-correction with the BH method, the variable was considered important. For multivariable models, since the order of factors in the PERMANOVA model may affect results, random shuffling of variables was performed, using either 999 permutations (when including all available health data) or 99 permutations (when including only the ones significant in univariable analysis). In either case, a rule of majorities was adopted.\u003c/p\u003e\n\u003cp\u003eDifferential abundance\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDifferential abundance analyses, including adjustment for maternal BMI, parity and antibiotic usage, were performed at each time point for both vaginal and faecal microbiome data using ANCOM-BC2 v2.0.1 [55, 58].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ePrediction modelling\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eClassification methods included the Random Forest (RF) algorithm [59] using the R Caret package v6.0-94 [60], along with the randomForest package v4.7-1,1 [61], Support Vector Machines (SVM) using the R Caret package v6.0.94 [60], and Neural Networks using the R package net v7.3.18 [62].\u003c/p\u003e\n\u003cp\u003eInitially, the classification analysis was carried out on CLR-transformed microbiome data and health data separately, as well as combined. Variables were included in the health data dataset if they were significant in at least one univariable and multivariable logistic regression (see Supplementary information), univariable PERMANOVA or if they had previously been described to associate with the gut or vaginal microbiota. This latter category includes all chronic diseases, medications and diet, as well as variables describing fertility and HPV infection or prevention, including a binary variable for vaginal dysbiosis, defined as \u0026lt;60% abundance of genus \u003cem\u003eLactobacillus\u003c/em\u003e.\u003c/p\u003e\n\u003cp\u003eAfter the first round of machine learning, which showed RF to be the most promising approach, it was repeated, including only variables whose importance in the first round was \u0026gt;0 (function VarImp in R package Caret). These included both microbiome and health or registry data. A third run was performed using these same VarImp variables and alpha-diversity metrics (Shannon, Pielou, Richness and Inverse Simpson). A final machine learning run was conducted for the second timepoint only, incorporating all the previous information and the difference in CLR transformed relative abundance of bacterial species. We performed leave-one-out validation for all models, cross-validation with 10 repeats and bootstrapping with 10,000 repeats.\u003c/p\u003e\n\u003cp\u003eGiven the unbalanced nature of the key variable (case and control), we used the R scutr package v0.1.2 [63] to randomly subsample the controls to equal the number of cases, and we ran the classifiers 10 times with a randomly subsampled control set for each loop. We used an 80:20 division for the test and training sets (respectively). We tested each classification on all 10 random test sets. ROC curves and plots were generated using pROC v1.18.4 [64]. Results are reported as AUROC values, sensitivity, specificity, negative prediction values (proportion of truly negative tests) and positive prediction values (proportion of truly positive tests). The models' confidence interval, sensitivity and specificity were obtained using the confusion matrix function of the R Caret package v6.0-94 [60].\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe main code used in the work is available on the CTMR GitHub page, which is accessible at https://github.com/ctmrbio/ML_swemami.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe metagenomic sequences have been submitted to ENA (project number PRJEB81814).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe are grateful to the participants of the SweMaMi study.\u003c/p\u003e\n\u003cp\u003eFurthermore, we would like to thank all Centre for Translational Microbiome Research members who contributed to the SweMaMi study during sample collection, extraction, and sequencing, especially Alexandra Pennhag, Marica Hamsten, and Maike Seifert.\u003c/p\u003e\n\u003cp\u003eWe also want to acknowledge that Ferring Pharmaceuticals has funded this work with an unrestricted research grant to LE. The study was partly financed by grants from the Swedish Research Council (Dnr 2023-02868 to NB), the Swedish state under the agreement between the Swedish government and the county councils (the ALF-agreement, Region Stockholm, DNR 20200471 to EF), Åke Wiberg foundation (Dnr M21-0153 to EF), the Gillbergska Foundation (to EF), and SciLifeLab \u0026amp; Wallenberg Data Driven Life Science Program (KAW 2020.0239, to LWH).\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eWorld Health Organization (WHO). \u003cem\u003ePreterm birth\u003c/em\u003e. 2023 10 May 2023; Available from: isir.is.\u003c/li\u003e\n \u003cli\u003eOhuma, E.O., et al., \u003cem\u003eNational, regional, and global estimates of preterm birth in 2020, with trends from 2010: a systematic analysis.\u003c/em\u003e Lancet, 2023. \u003cstrong\u003e402\u003c/strong\u003e(10409): p. 1261-1271.\u003c/li\u003e\n \u003cli\u003eNational Board of Health and Welfare, \u003cem\u003eStatistics on Pregnancies, Deliveries and Newborn Infants [Graviditeter, f\u0026ouml;rlossningar och nyf\u0026ouml;dda barn]\u003c/em\u003e. 2023, Socialstyrelsen: https://www.socialstyrelsen.se/.\u003c/li\u003e\n \u003cli\u003eJacobsson, B., et al., \u003cem\u003ePrediktion, prevention och behandlingsmetoder [Prediction, prevention and treatment options]\u0026nbsp;\u003c/em\u003eL\u0026auml;kartidningen, 2019.\u003c/li\u003e\n \u003cli\u003eSpong, C.Y., \u003cem\u003eDefining \u0026quot;term\u0026quot; pregnancy: recommendations from the Defining \u0026quot;Term\u0026quot; Pregnancy Workgroup.\u003c/em\u003e JAMA, 2013. \u003cstrong\u003e309\u003c/strong\u003e(23): p. 2445-6.\u003c/li\u003e\n \u003cli\u003eGoldenberg, R.L., et al., \u003cem\u003eEpidemiology and causes of preterm birth.\u003c/em\u003e Lancet, 2008. \u003cstrong\u003e371\u003c/strong\u003e(9606): p. 75-84.\u003c/li\u003e\n \u003cli\u003eVogel, J.P., et al., \u003cem\u003eThe global epidemiology of preterm birth.\u003c/em\u003e Best Pract Res Clin Obstet Gynaecol, 2018. \u003cstrong\u003e52\u003c/strong\u003e: p. 3-12.\u003c/li\u003e\n \u003cli\u003eMitrogiannis, I., et al., \u003cem\u003eRisk factors for preterm birth: an umbrella review of meta-analyses of observational studies.\u003c/em\u003e BMC Med, 2023. \u003cstrong\u003e21\u003c/strong\u003e(1): p. 494.\u003c/li\u003e\n \u003cli\u003eRomero, R., et al., \u003cem\u003eThe composition and stability of the vaginal microbiota of normal pregnant women is different from that of non-pregnant women.\u003c/em\u003e Microbiome, 2014. \u003cstrong\u003e2\u003c/strong\u003e(1): p. 4.\u003c/li\u003e\n \u003cli\u003eOliver, A., et al., \u003cem\u003eCervicovaginal Microbiome Composition Is Associated with Metabolic Profiles in Healthy Pregnancy.\u003c/em\u003e mBio, 2020. \u003cstrong\u003e11\u003c/strong\u003e(4).\u003c/li\u003e\n \u003cli\u003eKoren, O., et al., \u003cem\u003eHost remodeling of the gut microbiome and metabolic changes during pregnancy.\u003c/em\u003e Cell, 2012. \u003cstrong\u003e150\u003c/strong\u003e(3): p. 470-80.\u003c/li\u003e\n \u003cli\u003ePace, R.M., et al., \u003cem\u003eComplex species and strain ecology of the vaginal microbiome from pregnancy to postpartum and association with preterm birth.\u003c/em\u003e Med (N Y), 2021. \u003cstrong\u003e2\u003c/strong\u003e(9): p. 1027-1049.\u003c/li\u003e\n \u003cli\u003eChan, D., et al., \u003cem\u003eMicrobial-driven preterm labour involves crosstalk between the innate and adaptive immune response.\u003c/em\u003e Nat Commun, 2022. \u003cstrong\u003e13\u003c/strong\u003e(1): p. 975.\u003c/li\u003e\n \u003cli\u003eLiu, X., et al., \u003cem\u003eVaginal flora during pregnancy and subsequent risk of preterm birth or prelabor rupture of membranes: a nested case-control study from China.\u003c/em\u003e BMC Pregnancy Childbirth, 2023. \u003cstrong\u003e23\u003c/strong\u003e(1): p. 244.\u003c/li\u003e\n \u003cli\u003eFettweis, J.M., et al., \u003cem\u003eThe vaginal microbiome and preterm birth.\u003c/em\u003e Nat Med, 2019. \u003cstrong\u003e25\u003c/strong\u003e(6): p. 1012-1021.\u003c/li\u003e\n \u003cli\u003eBaud, A., et al., \u003cem\u003eMicrobial diversity in the vaginal microbiota and its link to pregnancy outcomes.\u003c/em\u003e Sci Rep, 2023. \u003cstrong\u003e13\u003c/strong\u003e(1): p. 9061.\u003c/li\u003e\n \u003cli\u003eDownes, K.L., et al., \u003cem\u003e15: Gardnerella vaginalis and spontaneous preterm birth: New insights.\u003c/em\u003e American Journal of Obstetrics and Gynecology, 2018. \u003cstrong\u003e218\u003c/strong\u003e(1, Supplement): p. S12-S13.\u003c/li\u003e\n \u003cli\u003eKindschuh, W.F., et al., \u003cem\u003ePreterm birth is associated with xenobiotics and predicted by the vaginal metabolome.\u003c/em\u003e Nat Microbiol, 2023. \u003cstrong\u003e8\u003c/strong\u003e(2): p. 246-259.\u003c/li\u003e\n \u003cli\u003eGudnadottir, U., et al., \u003cem\u003eThe vaginal microbiome and the risk of preterm birth: a systematic review and network meta-analysis.\u003c/em\u003e Sci Rep, 2022. \u003cstrong\u003e12\u003c/strong\u003e(1): p. 7926.\u003c/li\u003e\n \u003cli\u003eKosti, I., et al., \u003cem\u003eMeta-Analysis of Vaginal Microbiome Data Provides New Insights Into Preterm Birth.\u003c/em\u003e Front Microbiol, 2020. \u003cstrong\u003e11\u003c/strong\u003e: p. 476.\u003c/li\u003e\n \u003cli\u003eHuang, C., et al., \u003cem\u003eMeta-analysis reveals the vaginal microbiome is a better predictor of earlier than later preterm birth.\u003c/em\u003e BMC Biol, 2023. \u003cstrong\u003e21\u003c/strong\u003e(1): p. 199.\u003c/li\u003e\n \u003cli\u003eEdwards, V.L., et al., \u003cem\u003eThe Cervicovaginal Microbiota-Host Interaction Modulates Chlamydia trachomatis Infection.\u003c/em\u003e mBio, 2019. \u003cstrong\u003e10\u003c/strong\u003e(4).\u003c/li\u003e\n \u003cli\u003eKoedooder, R., et al., \u003cem\u003eThe vaginal microbiome as a predictor for outcome of in vitro fertilization with or without intracytoplasmic sperm injection: a prospective study.\u003c/em\u003e Hum Reprod, 2019. \u003cstrong\u003e34\u003c/strong\u003e(6): p. 1042-1054.\u003c/li\u003e\n \u003cli\u003eBrusselaers, N., et al., \u003cem\u003eVaginal dysbiosis and the risk of human papillomavirus and cervical cancer: systematic review and meta-analysis.\u003c/em\u003e Am J Obstet Gynecol, 2019. \u003cstrong\u003e221\u003c/strong\u003e(1): p. 9-18 e8.\u003c/li\u003e\n \u003cli\u003eCheng, L., et al., \u003cem\u003eVaginal microbiota and human papillomavirus infection among young Swedish women.\u003c/em\u003e NPJ Biofilms Microbiomes, 2020. \u003cstrong\u003e6\u003c/strong\u003e(1): p. 39.\u003c/li\u003e\n \u003cli\u003eHong, Y.M., et al., \u003cem\u003ePredicting preterm birth using machine learning techniques in oral microbiome.\u003c/em\u003e Sci Rep, 2023. \u003cstrong\u003e13\u003c/strong\u003e(1): p. 21105.\u003c/li\u003e\n \u003cli\u003eGolob, J.L., et al., \u003cem\u003eMicrobiome preterm birth DREAM challenge: Crowdsourcing machine learning approaches to advance preterm birth research.\u003c/em\u003e Cell Rep Med, 2024. \u003cstrong\u003e5\u003c/strong\u003e(1): p. 101350.\u003c/li\u003e\n \u003cli\u003ePark, S., et al., \u003cem\u003ePredicting preterm birth through vaginal microbiota, cervical length, and WBC using a machine learning model.\u003c/em\u003e Front Microbiol, 2022. \u003cstrong\u003e13\u003c/strong\u003e: p. 912853.\u003c/li\u003e\n \u003cli\u003eArabi Belaghi, R., J. Beyene, and S.D. McDonald, \u003cem\u003ePrediction of preterm birth in nulliparous women using logistic regression and machine learning.\u003c/em\u003e PLoS One, 2021. \u003cstrong\u003e16\u003c/strong\u003e(6): p. e0252025.\u003c/li\u003e\n \u003cli\u003eBelaghi, R.A., \u003cem\u003ePrediction of preterm birth in multiparous women using logistic regression and machine learning approaches.\u003c/em\u003e Sci Rep, 2024. \u003cstrong\u003e14\u003c/strong\u003e(1): p. 21967.\u003c/li\u003e\n \u003cli\u003eKassahun, E.A., et al., \u003cem\u003eDevelopment and validation of a simplified risk prediction model for preterm birth: a prospective cohort study in rural Ethiopia.\u003c/em\u003e Sci Rep, 2024. \u003cstrong\u003e14\u003c/strong\u003e(1): p. 4845.\u003c/li\u003e\n \u003cli\u003eDing, L., et al., \u003cem\u003ePrediction of preterm birth using machine learning: a comprehensive analysis based on large-scale preschool children survey data in Shenzhen of China.\u003c/em\u003e BMC Pregnancy Childbirth, 2024. \u003cstrong\u003e24\u003c/strong\u003e(1): p. 810.\u003c/li\u003e\n \u003cli\u003eHuang, C., et al., \u003cem\u003ePredicting preterm birth using electronic medical records from multiple prenatal visits.\u003c/em\u003e BMC Pregnancy Childbirth, 2024. \u003cstrong\u003e24\u003c/strong\u003e(1): p. 843.\u003c/li\u003e\n \u003cli\u003eChen, Y., et al., \u003cem\u003eDevelopment and validation of a spontaneous preterm birth risk prediction algorithm based on maternal bioinformatics: A single-center retrospective study.\u003c/em\u003e BMC Pregnancy Childbirth, 2024. \u003cstrong\u003e24\u003c/strong\u003e(1): p. 763.\u003c/li\u003e\n \u003cli\u003eRajasekera, T.A., et al., \u003cem\u003eStress and depression-associated shifts in gut microbiota: A pilot study of human pregnancy.\u003c/em\u003e Brain Behav Immun Health, 2024. \u003cstrong\u003e36\u003c/strong\u003e: p. 100730.\u003c/li\u003e\n \u003cli\u003eLiu, W., et al., \u003cem\u003eProtective Effect of Akkermansia muciniphila on the Preeclampsia-Like Mouse Model.\u003c/em\u003e Reprod Sci, 2023. \u003cstrong\u003e30\u003c/strong\u003e(9): p. 2623-2633.\u003c/li\u003e\n \u003cli\u003eTalukdar, D., et al., \u003cem\u003ePrevise preterm birth in early pregnancy through vaginal microbiome signatures using metagenomics and dipstick assays.\u003c/em\u003e iScience, 2024. \u003cstrong\u003e27\u003c/strong\u003e(11): p. 111238.\u003c/li\u003e\n \u003cli\u003eTabatabaei, N., et al., \u003cem\u003eVaginal microbiome in early pregnancy and subsequent risk of spontaneous preterm birth: a case-control study.\u003c/em\u003e Bjog-an International Journal of Obstetrics and Gynaecology, 2019. \u003cstrong\u003e126\u003c/strong\u003e(3): p. 349-358.\u003c/li\u003e\n \u003cli\u003eFransson, E., et al., \u003cem\u003eCohort profile: the Swedish Maternal Microbiome project (SweMaMi) - assessing the dynamic associations between the microbiome and maternal and neonatal adverse events.\u003c/em\u003e BMJ Open, 2022. \u003cstrong\u003e12\u003c/strong\u003e(10): p. e065825.\u003c/li\u003e\n \u003cli\u003eCohen, S., T. Kamarck, and R. Mermelstein, \u003cem\u003eA global measure of perceived stress.\u003c/em\u003e J Health Soc Behav, 1983. \u003cstrong\u003e24\u003c/strong\u003e(4): p. 385-96.\u003c/li\u003e\n \u003cli\u003eCox, J.L., J.M. Holden, and R. Sagovsky, \u003cem\u003eDetection of postnatal depression. Development of the 10-item Edinburgh Postnatal Depression Scale.\u003c/em\u003e Br J Psychiatry, 1987. \u003cstrong\u003e150\u003c/strong\u003e: p. 782-6.\u003c/li\u003e\n \u003cli\u003eEbrahimi, N., et al., \u003cem\u003eNausea and vomiting of pregnancy: using the 24-hour Pregnancy-Unique Quantification of Emesis (PUQE-24) scale.\u003c/em\u003e J Obstet Gynaecol Can, 2009. \u003cstrong\u003e31\u003c/strong\u003e(9): p. 803-807.\u003c/li\u003e\n \u003cli\u003eStephansson, O., et al., \u003cem\u003eThe Swedish Pregnancy Register - for quality of care improvement and research.\u003c/em\u003e Acta Obstet Gynecol Scand, 2018. \u003cstrong\u003e97\u003c/strong\u003e(4): p. 466-476.\u003c/li\u003e\n \u003cli\u003eF. Boulund, A.A., \u003cem\u003ectmrbio/stag-mwc: StaG v0.5.1\u003c/em\u003e. 2022, Zenodo.\u003c/li\u003e\n \u003cli\u003eKoster, J. and S. Rahmann, \u003cem\u003eSnakemake--a scalable bioinformatics workflow engine.\u003c/em\u003e Bioinformatics, 2012. \u003cstrong\u003e28\u003c/strong\u003e(19): p. 2520-2.\u003c/li\u003e\n \u003cli\u003eChen, S., et al., \u003cem\u003efastp: an ultra-fast all-in-one FASTQ preprocessor.\u003c/em\u003e Bioinformatics, 2018. \u003cstrong\u003e34\u003c/strong\u003e(17): p. i884-i890.\u003c/li\u003e\n \u003cli\u003eWood, D.E., J. Lu, and B. Langmead, \u003cem\u003eImproved metagenomic analysis with Kraken 2.\u003c/em\u003e Genome Biol, 2019. \u003cstrong\u003e20\u003c/strong\u003e(1): p. 257.\u003c/li\u003e\n \u003cli\u003eFuchs, S., Drechsel, Oliver, \u003cem\u003eKraken2 Database (Human, SARS-CoV2)\u003c/em\u003e. 2021, Zenodo.\u003c/li\u003e\n \u003cli\u003eTruong, D.T., et al., \u003cem\u003eMicrobial strain-level population structure and genetic diversity from metagenomes.\u003c/em\u003e Genome Res, 2017. \u003cstrong\u003e27\u003c/strong\u003e(4): p. 626-638.\u003c/li\u003e\n \u003cli\u003eDavis, N.M., et al., \u003cem\u003eSimple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data.\u003c/em\u003e Microbiome, 2018. \u003cstrong\u003e6\u003c/strong\u003e(1): p. 226.\u003c/li\u003e\n \u003cli\u003eRStudio Team, \u003cem\u003eRStudio: Integrated Development for R\u003c/em\u003e. 2020, RStudio, PBC, Boston, MA.\u003c/li\u003e\n \u003cli\u003eWickham, H., \u003cem\u003eggplot2: Elegant Graphics for Data Analysis\u003c/em\u003e. 2016, Springer-Verlag New York.\u003c/li\u003e\n \u003cli\u003eHo D, I.K., King G, Stuart E, \u003cem\u003eMatchIt: Nonparametric Preprocessing for Parametric Causal Inference.\u003c/em\u003e Journal of Statistical Software, 2011.\u003c/li\u003e\n \u003cli\u003eJari Oksanen, G.L.S., F. Guillaume Blanchet, Roeland Kindt, Pierre Legendre, Peter R., et al., \u003cem\u003evegan: Community Ecology Package\u003c/em\u003e. 2022, R package version 2.6-4.\u003c/li\u003e\n \u003cli\u003eLin, H., M. Eggesbo, and S.D. Peddada, \u003cem\u003eLinear and nonlinear correlation estimators unveil undescribed taxa interactions in microbiome data.\u003c/em\u003e Nat Commun, 2022. \u003cstrong\u003e13\u003c/strong\u003e(1): p. 4946.\u003c/li\u003e\n \u003cli\u003eMartin Henry H. Stevens, J.O. \u003cem\u003ePermutational Multivariate Analysis of Variance Using Distance Matrices\u003c/em\u003e. 2023; Available from: https://search.r-project.org/CRAN/refmans/vegan/html/adonis.html.\u003c/li\u003e\n \u003cli\u003eGanz, K., \u003cem\u003escutr: Balancing Multiclass Datasets for Classification Tasks\u003c/em\u003e. 2021, R package version 0.1.2.\u003c/li\u003e\n \u003cli\u003eLin, H. and S.D. Peddada, \u003cem\u003eAnalysis of compositions of microbiomes with bias correction.\u003c/em\u003e Nat Commun, 2020. \u003cstrong\u003e11\u003c/strong\u003e(1): p. 3514.\u003c/li\u003e\n \u003cli\u003eHo, T.K. \u003cem\u003eRandom decision forests\u003c/em\u003e. in \u003cem\u003e3rd international conference on document analysis and recognition\u0026nbsp;\u003c/em\u003e1995.\u003c/li\u003e\n \u003cli\u003eKuhn, M., \u003cem\u003eBuilding Predictive Models in R Using the caret Package.\u003c/em\u003e Journal of Statistical Software, 2008.\u003c/li\u003e\n \u003cli\u003eA. Liaw, M.W., \u003cem\u003eClassification and Regression by randomForest.\u003c/em\u003e R News, 2002. \u003cstrong\u003e2(3)\u003c/strong\u003e: p. 18-22.\u003c/li\u003e\n \u003cli\u003eRipley, W.N.V.a.B.D., \u003cem\u003eModern Applied Statistics with S\u003c/em\u003e. Fourth ed. 2002: Springer.\u003c/li\u003e\n \u003cli\u003eGanz, K., \u003cem\u003escutr: Balancing Multiclass Datasets for Classification Tasks\u003c/em\u003e. 2021.\u003c/li\u003e\n \u003cli\u003eXavier Robin, N.T., Alexandre Hainard, Natalia Tiberti, Fr\u0026eacute;d\u0026eacute;rique Lisacek, Jean-Charles Sanchez, Markus M\u0026uuml;ller, \u003cem\u003epROC: an open-source package for R and S+ to analyze and compare ROC curves.\u003c/em\u003e BMC Bioinformatics, 2011. \u003cstrong\u003e12\u003c/strong\u003e: p. 77.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cem\u003eTable 1. Descriptive characteristics of participants in the preterm group and the matched control group in early pregnancy (questionnaire answered before 20 weeks of gestation).\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"596\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eControls\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAll spontaneous preterm birth cases\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e(N=264)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e(N=132)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e(N=396)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026lt; 25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e5 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e3 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e8 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;25-35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e196 (74 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e98 (74 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e294 (74 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026gt; 35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e63 (24 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e31 (23 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e94 (24 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eBMI* prior to pregnancy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026lt; 18.5 Underweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e5 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e2 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e7 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;18.5-25 Normal weight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e157 (59 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e75 (57 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e232 (59 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026gt; 25 Overweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e83 (31 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e42 (32 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e125 (32 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eSwedish born\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e238 (90 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e122 (92 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e360 (91 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eUniversity education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e53 (20 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e53 (20 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e79 (20 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eSocioeconomic score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; High\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e95 (36 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e50 (38 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e145 (37 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003e\u0026nbsp; Low\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e169 (64 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e82 (62 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e251 (63 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eParity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eMultiparous\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e126 (48 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e64 (48 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e190 (48 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003ePrimiparous\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e138 (52 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e68 (52 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e206 (52 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eConception\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eAssisted\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e35 (13 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e19 (14 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e54 (14 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eNatural\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e229 (87 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e113 (86 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e342 (86 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003ePrevious IUFD**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e1 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e3 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e4 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003ePrevious preterm birth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e4 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e5 (4 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e9 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003ePrevious RPL***\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e15 (6 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e4 (3 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e19 (5 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eRegular menstruation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e215 (81 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e112 (85 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e327 (83 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eDiagnosed PCOS****\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e18 (7 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e9 (7 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e27 (7 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eDiagnosed endometriosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e13 (5 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e8 (6 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e21 (5 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eDaily fibre\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e219 (83 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e115 (87 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e334 (84 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eHealthy diet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e195 (74 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e99 (75 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e294 (74 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eIntake of iron during early pregnancy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e1 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e1 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e2 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eSnuff/smokeless tobacco use during pregnancy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e0 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e1 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e1 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eSmoking during pregnancy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e1 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e2 (2 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e3 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eAny drugs used during pregnancy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e113 (43 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e56 (42 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e169 (43 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eMultiple drugs used during pregnancy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e96 (36 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e47 (36 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e143 (36 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eThyroid medication\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e31 (12 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e9 (7 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e40 (10 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eAllergy and antihistamines\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e29 (11 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e15 (11 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e44 (11 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eNeurological medication\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e27 (10 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e13 (10 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e40 (10 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eGastrointestinal medication\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e0 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e5 (4 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e5 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eDiabetes medication\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e1 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e1 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e2 (1 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eAntibiotics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e0 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e0 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e0 (0 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eEPDS\u003csup\u003e\u0026sect;\u003c/sup\u003e score mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e7.9 (\u0026plusmn; 6.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e10 (\u0026plusmn; 5.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e7.6 (\u0026plusmn; 6.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003ePSS\u003csup\u003e\u0026curren;\u003c/sup\u003e score mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e7.3 (\u0026plusmn; 3.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e6.1 (\u0026plusmn; 3.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e7.1 (\u0026plusmn; 3.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eBristol stool scale group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eFast transit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e64 (24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e38 (29%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e102 (26%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eNormal transit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e52 (20%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e25 (19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e77 (19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eSlow transit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e81 (31%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e30 (23%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e111 (28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 212px;\"\u003e\n \u003cp\u003eVarious transit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e67 (25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003e39 (30%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003e106 (27%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;*Body mass index, **Intrauterine foetal demise, ***Recurrent pregnancy loss, ****Polycystic ovary syndrome, \u0026sect; Edinburgh Postnatal Depression Scale, \u0026curren; Perceived stress scale\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTable 2. AUROC values (with confidence intervals) for Random Forest machine learning models of the ten random test sets. The best model is shown in bold.\u003c/em\u003e\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"600\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTimepoint one\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTimepoint two\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eAUC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eAUC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003eVaginal microbiome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.51 (0.36-0.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.54 (0.27-0.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003eFaecal microbiome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.59 (0.36-0.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.59 (0.38-0.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003eHealth data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.74 (0.61-0.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.68 (0.56-0.87)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.61\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003eAll combined\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.77 (0.61-0.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.70 (0.49-0.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003eDenoised microbiome and health data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.86 (0.81-0.99)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.75 (0.43-0.77)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDenoised microbiome, alpha diversity indices and health data\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.89 (0.68-0.94)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.76\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.90\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e0.77 (0.52-0.84)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Preterm birth, microbiome, machine learning, pregnancy, prediction","lastPublishedDoi":"10.21203/rs.3.rs-7621941/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7621941/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Preterm birth remains the leading cause of neonatal morbidity and mortality globally, with rates staying unchanged despite advances in obstetric and neonatal care. Inflammation and infection are key mechanisms implicated in preterm birth, with the maternal microbiome emerging as a potential predictor and therapeutic target. This study aimed to predict spontaneous preterm birth by analysing vaginal and faecal microbiomes (collected at \u003c20 weeks and 28–30 weeks of gestation) together with extensive questionnaire data. Leveraging the largest microbiome cohort for preterm birth to date (collected in Sweden, 2017–2021, with 132 cases of spontaneous preterm birth), we developed a machine learning-based prediction model, achieving an AUROC of 0.89. This work underscores the significant potential of early prediction for preterm birth, highlighting that accurate prediction relies on the integration of lifestyle, health status, and microbiome composition. Our results provide a pathway for developing targeted prevention strategies.","manuscriptTitle":"Predicting spontaneous preterm birth: Integrating maternal vaginal and gut microbiome profiles with individual risk factors","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-25 06:40:15","doi":"10.21203/rs.3.rs-7621941/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-microbiology","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"nmicrobiol","sideBox":"Learn more about [Nature Microbiology](http://www.nature.com/nmicrobiol/)","snPcode":"","submissionUrl":"","title":"Nature Microbiology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"a925f15b-7f34-41b4-a150-78a01262b1a8","owner":[],"postedDate":"September 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":55219661,"name":"Health sciences/Biomarkers/Predictive markers"},{"id":55219662,"name":"Health sciences/Biomarkers/Diagnostic markers"}],"tags":[],"updatedAt":"2026-02-26T12:46:47+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-25 06:40:15","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7621941","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7621941","identity":"rs-7621941","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}