A Multi-Site Study to Develop and Validate First Trimester, Circulating Microparticle Biomarkers for Tiered Risk Stratification

Background:

Despite much research, advances in early prediction of spontaneous preterm birth (sPTB) has been slow. The evolving field of circulating microparticle (CMP) biology may identify novel blood-based, and clinically useful, biomarkers.

Objective:

To test the ability of a previously identified, 7-marker set of CMP-derived proteins from the first trimester of pregnancy, in the form of an in vitro diagnostic multivariate index assay (IVDMIA), to stratify pregnant patients according to their risk for sPTB.

Study Design:

We employed a previously validated set of CMP protein biomarkers, utilizing mass spectrometry assays and a nested case-control design in a subset of participants from the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be (nuMoM2b). We evaluated these biomarkers in the form of an IVDMIA to predict risk for sPTB at different gestational ages. Plasma samples collected at 9- to 13-weeks’ gestation were analyzed. The IVDMIA assigned subjects to one of three sPTB risk categories: low risk (LR), moderate risk (MR), or high risk (HR). Independent validation on a set-aside set confirmed the IVDMIA’s performance in risk stratification.

Results:

Samples from 400 participants from the nuMoM2b cohort were used for the study; of these, 160 delivered <37 weeks and 240 delivered at term. Through Monte Carlo simulation in which the validation results were adjusted based on actual weekly sPTB incidence rates in the nuMoM2b cohort, the IVDMIA stratifications demonstrated statistically significant differences among the risk groups in time-to-event (birth) analysis (p < 0.0001). The incidence-rate adjusted cumulative risks of sPTB at ≤ 32 weeks’ gestation were 0.4%, 1.6%, and 7.5%, respectively for the LR, MR, and HR groups, respectively. Compared to the LR group, the corresponding risk ratios (RR) of the IVDMIA assigned MR and HR group were 4.25 (95% CI 2.2 to 7.9) and 19.92 (95% CI 10.4 to 37.4), respectively.

Conclusion:

A first trimester CMP protein biomarker panel can be used to stratify risk for sPTB at different gestational ages. Such a multi-tiered stratification tool could be used to assess risk early in pregnancy to enable timely clinical management and interventions, and, ultimately, to enable the development of tailored care pathways for sPTB prevention.

Participants and specimens

The subjects in the current study were from the nuMoM2b study cohort¹³ selected using a nested case-control design. The nuMoM2b study was approved by the local Institutional Review Boards at each of eight clinical study sites and the data coordinating center and all participants provided informed consent. The study was registered on clinicaltrials.gov (NCT01322529). In brief, participants in the nuMoM2b cohort were recruited in the first trimester, provided informed consent for the study, and had study visits in the 1^st (Visit 1: gestational age 6 weeks 0 days to 13 weeks 6 days), 2^nd (Visit 2: gestational age 16 weeks 0 days to 21 weeks 6 days), and late 2^nd-early 3^rd (Visit 3: gestational age 22 weeks 0 days to 29 weeks 6 days) trimesters, and at the time of delivery (Visit 4). During study visits, which were conducted in English or Spanish, multiple questionnaires and psychosocial instruments were completed, and biological specimens were obtained ¹³. Plasma samples were prepared within two hours of drawing using a standardized Manual of Operations (available upon request). All specimens were stored at −80 degrees Celsius until analyzed. After delivery, certified chart abstractors extracted delivery outcomes from the medical records. Preterm birth was defined as any delivery less than 37 weeks+0 days gestation. Spontaneous preterm birth was defined as any preterm birth, including both preterm labor and preterm premature ruptured membranes. Medically indicated or iatrogenic preterm delivery (such as for preeclampsia) were excluded. Gestational age was determined using first trimester ultrasound in conjunction with last menstrual period ¹³. All plasma samples obtained for this analysis were from Visit 1 in the first trimester.

CMP Enrichment

We enriched for CMPs from plasma by Size Exclusion Chromatography (SEC), using methods that have been reported in prior publications ^8,12,13,16. Anonymized EDTA plasma samples labeled by study numbers known only to nuMoM2b investigators, with the technical team blinded to case or control status, were randomly assorted and shipped on dry ice to NX Prenatal, Inc. (Bellaire, TX). Briefly, size exclusion chromatography (SEC) columns were manufactured by AmericanBio, Inc. (Exosome Isolation Columns; Canton, MA). The columns were packed by with 2% Sepharose 4B-CL (bead size range 45-165 μm, pore size 42-70 nm) from Cytiva (Marlborough, MA) to a total packed volume of 10 mL and delivered to NX Prenatal under ambient shipping conditions. Once received by NX Prenatal, the columns were stored at 2–8 °C until use. Prior to use, the columns were allowed to equilibrate to room temperature and subsequently washed with 30 mL of the Exosome Elution Reagent. EDTA plasma samples were thawed, and 0.5 mL of plasma was applied and allowed to incorporate into the Exosome Isolation Column. CMP-associated protein enrichment was carried out by SEC and eluted using the Exosome Elution Reagent. The plasma samples were not filtered, diluted, or pretreated prior to application to the columns. Following the incorporation of the sample into the column, sequential 0.5 mL volumes of Exosome Elution Reagent were added, and 0.5 mL fractions were collected. As previously published, the eluted fractions yielded two peaks ^10,14,19. The CMPs were captured in the column void volumes and resolved from the highly-abundant, soluble protein peak ²⁰. To minimize batch effects, samples were processed with block-randomization ²¹. The CMP-containing fractions (0.5 mL aliquots per fraction) were pooled for each individual sample and a total protein measurement was performed using the Pierce^™ BCA Protein Assay Kit (ThermoFisher Scientific, Waltham, MA). An aliquot containing a total protein of 50 μg from each individual CMP-isolate pool was then immediately processed for mass spectrometry (MS) analysis; the remainder was stored at −80 °C pending completion of all CMP-isolate processing.

Sample Lysis and Preparation for Liquid Chromatography-Tandem Mass Spectrometry

A total of 440 Plasma-enriched CMP samples (400 cases and controls + 40 QC samples) were processed and used for the quantification of biomarkers by Liquid Chromatography-Tandem Mass Spectrometry (LC-MS/MS). 50 μg of total protein from each sample was denatured with 8M urea, reduced using dithiothreitol (ThermoFisher Scientific, Waltham, MA), alkylated with iodoacetamide (ThermoFisher Scientific, Waltham, MA)²², and digested overnight with trypsin (ThermoFisher Scientific, Waltham, MA). 10 μL of acetic acid was added to each sample to quench the trypsin digestion. The resulting samples were cleaned using low protein binding, filter tube, spin columns by spinning them for 30 minutes at 15000 RPM in a centrifuge. 1.5 μL of Internal Standards (heavy isotope-labeled synthetic tryptic peptides) were then added to each eluent sample from a spin column; then 200 μL of that spin column sample was added to injection glass vials for LC-MS/MS analysis.

Multiple-Reaction Monitoring LC-MS Analysis

Quantitative assays for the target proteins were performed by multiple-reaction monitoring LC-MS (MRM-LC-MS/MS) analysis using the ThermoFisher Scientific Vanquish UHPLC system coupled to a tandem TSQ Altis^™ Triple Quadrupole mass spectrometer equipped with a heated electrospray ionization source (H-ESI). The MRM-LC-MS/MS chromatography was carried out using a linear gradient for Solvent A (LC-MS Grade Water with 0.1% Formic Acid) and Solvent B (LC-MS Grade Acetonitrile with 0.1% Formic Acid) as shown in Table S1. Additional details were provided in Supplementary Material.

Study Design and Sample Utilization

The study included 400 samples from the nuMoM2b cohort selected with a 2:3 nested case-control design. The selection of cases was block-randomized to satisfy a pre-defined number of cases in sPTB subcategories. Spontaneous preterm birth was defined as delivery that occurred at < 37 weeks gestational age subsequent to spontaneous onset of preterm labor or premature rupture of the membranes (PROM) or fetal membrane prolapse. This outcome as well as the gestational age at birth was recorded by certified, trained chart abstractors at each of the nuMoM2b clinical sites. Further details are contained in the study methods paper ¹³. We adopted a propensity score method to select and match cases and controls. This did not match based on individual variables, but allowed us to match controls to cases more globally based on all the matching variables. The entire set of matching variables included gestational age when specimen was drawn, maternal age, race/ethnicity group (collapsing Asian and Other together due to sample size), study site, and smoking status during the 3 months prior to pregnancy. A 2:3 case:control match was accomplished. Propensity score distributions are included in supplementary material. Only participants who had live births were included.

Based on available throughput, the 400 samples were divided into two batches for processing and data generation with block-randomization stratified on sPTB and full-term subcategories. Within each batch the samples were processed and analyzed again in block-randomized order stratified for cases and controls.

Plasma CMP protein biomarkers comprising hemopexin (HEMO), fibulin 1 (FBLN1), inter-alpha-trypsin inhibitor heavy chain 2 (ITIH2), serotransferrin (TRFE), inhibitor of C1 (IC1), inter-alpha-trypsin inhibitor heavy chain 4 (ITIH4), and lecithin-cholesterol acyltransferase (LCAT) were measured by targeted multiple reaction monitoring mass spectrometry (MRM-MS/MS). These proteins were identified in previous work as the most predictive for preterm birth.^10,14 Approximately 60% (n = 241) of the 400 study subjects were randomly selected with stratification on sPTB/FT subcategories for IVDMIA model development. The remaining 159 subjects were used as an independent validation sample set. Demographic and clinical characterization of the development set and the independent validation set are summarized in Tables 1A and 1B, respectively.

The choice of study sample size was constrained by available eligible sPTB cases due to the relatively low prevalence. For power estimation, if we considered the 159 validation samples in three clinical groups: gestational age < 32 weeks (n = 15), 32 to < 37 weeks (n = 38), and full-term (n = 106) and in three IVDMIA assigned risk categories, the validation samples would achieve a power > 71% at a two-sided significance level of 0.05 with an assumed effect size of 0.25.

Model Derivation and Statistical Analysis

Input variables to the IVDMIA model included HEMO, FBLN1, ITIH2, TRFE, IC1, ITIH4, LCAT, and body mass index (BMI). Missing BMI values for 11 out of the 400 subjects were imputed with the constant 25 kg/m². Protein biomarker measurements were rescaled by median absolute deviation normalization. BMI was normalized by division by 25.

Optimal model structures and training hyperparameters were determined through extensive Monte Carlo cross-validation within the training dataset. The final derived IVDMIA included two multivariate models that are applied sequentially to stratify the test populations into 3-tiered risk categories: a rule-out model identifies a subset of test population as low-risk (LR), a second rule-in model identifies a small portion of the remaining test population as high-risk (HR). The remaining subjects are labeled as moderate risk (MR). The rule-out model was by design trained to achieve a high-sensitivity and hence a high negative predictive value (NPV) for subjects classified as LR. On the other hand, the rule-in model was aimed to capture a clinically meaning proportion of the sPTBs while maintaining a high specificity and hence a high positive predictive value (PPV) for subjects classified as HR.

Multiple models were derived using machine learning approaches including neural nets and extreme gradient boosting (XGBoost)^23,24 either to predict risk of binary outcomes (sPTB vs controls for given gestational age cutoffs) or to directly model gestational age as time-to-birth event with survival analysis models. To achieve the desired preference for high sensitivity or specificity, these models were themselves further combined to form the final rule-out and rule-in models.

Clinical performances of the IVDMIA were evaluated by time-to-events (births) analysis ²⁵ with comparison among the 3-tiered sPTB risk categories. Similar to that of survival analysis used to compare multiple treatment arms with respect to overall survival time, Kaplan-Meier plots from time-to-events analysis and corresponding risk tables were compared among the validation samples in the three IVDMIA predicted sPTB risk categories and tested for statistical significance by Log Rank test²⁵. In the risk table, “N at Risk” is the number of subjects in a risk group who had not given birth at the beginning of a particular gestation week, and Cumulative Events is the group’s cumulative number of births at the end of the week.

The validation result confirmed the risk stratification ability of the IVDMIA. However, the validation sample set was a case-control set where the distribution of gestation at delivery does not represent a true test population intended for the IVDMIA. In order to predict the clinical performance of the IVDMIA for its intended test population, Monte Carlo (MC) simulation analyses were performed. In each MC simulation, the validation sample set was repeatedly resampled 9,559 times (the number of subjects in the nuMoM2b cohort) with replacement based on subjects’ gestation weeks at delivery using probability sampling. The distribution used for probability sampling was estimated from the actual weekly gestation at sPTB/birth data of the entire nuMoM2b cohort. Results from Kaplan-Meier plots and risk tables from this MC simulation sample set were then used to predict the clinical performance of the IVDMIA. Risk tables from 500 repeated MC simulation analyses were aggregated to compute point estimates and confidence intervals of the risk table entries and additional calculated performance metrics. Among them, percent cumulative events represent the proportion of subjects in an IVDMIA-predicted sPTB risk group who had a sPTB during or before a given gestation week. It is therefore also the post-test prevalence or positive predictive value of sPTB during or before a given gestation week. Another clinically meaningful performance metrics is the risk ratio between HR and LR, or MR and LR of sPTB at or before a given gestation week.

As the rule-out and rule-in models in the IVDMIA are applied sequentially, receiver operator characteristic (ROC) curve analysis was performed using the rule-out model first and then estimating a second “conditional” ROC analysis using the rule-in model only on samples that were not assigned to LR. The final ROC curve was obtained by “fusing” together the portion of the first ROC curve up to the cutoff point of the rule-out model and the second conditional ROC curve rescaled by the proportion of cases and controls in the non-LR samples over those in the entire sample set, respectively. Fig. S1 shows such fused ROC curves with sPTB defined at different final delivery gestation weeks cutoffs. An R script for estimating the fused ROC curves is provided in Supplementary Information.

Statistical and model development calculations were carried out in the R statistical computational environment (version 2021.9.0.351) ²⁶. Reporting followed the TRIPOD Checklist for Prediction Model Development and Validation.

Table 2A.

Table 2B.

Table 3A.

Table 3B.

Table 3C.

a. Principal Findings

In the current study, we demonstrated that our previously reported ¹⁴, CMP-derived set of biomarkers, collected from blood samples as early as 9 - 13 weeks, continue to show potential as a first-trimester, risk stratification tool to predict the risk of sPTB in nulliparous patients. This biomarker set was developed as an in vitro diagnostic, multivariate index assay (IVDMIA) based on 7 CMP protein biomarkers that comprise two multivariate models working in tandem to first “rule-out” low-risk sPTB subjects followed by a “rule-in” step to identify patients at the highest risk for sPTB, especially very early sPTB. This strategy results in a three-tiered clinical stratification of pregnant nulliparas for risk of sPTB into LR, MR, or HR categories. As shown in Fig. 3 and Tables 3A and 3C, the test allows the separation of pregnant nulliparas into the three risk groups with statistically significant differences even after adjustment for weekly sPTB incidence rates in Kaplan-Meier curves representing time-to-events (births) cumulative distribution patterns. The estimated means and 95% CIs of the post-test risks of pre-term births showed that while the risks of the MR group were only slightly higher than the natural prevalence of sPTB at a given gestational age range in the nuMoM2b cohort, the risk differences between the LR and HR were statistically significant in both absolute terms and in risk ratios, with even greater differences for earlier sPTBs. With a risk ratio for early PTB <32 weeks of 19.92 (95% CI 10.38, 37.38), this risk tool has potential clinical uses.

b. Results in the Context of What is Known

A recent, comprehensive review of the literature uncovered 77 primary research articles that assessed blood-borne biomarkers predictive for sPTB ²⁷. Two hundred seventy-eight independent and distinct markers were found to be associated with sPTB in at least one of the studies; these markers were limited to asymptomatic populations rather than investigated among those with the signs and symptoms of labor. To date, the prevailing clinical “marker” of sPTB has been prior history of a PTB. As a single predictor, it has remained the most potent and oft-used variable for managing patients because of its pre-pregnancy association with sPTB, (mostly) verifiable ascertainment, and greater than 20% positive predictive value for a subsequent preterm birth ²⁸. The problem with this historical marker is that this information, or rather lack of it, is of no use in the case of nulliparous individual. Biomarkers that are minimally invasive, easily obtainable in the first trimester, and quickly evaluable for predicting preterm births in people bearing children for the first time could reasonably be expected to provide early intervention opportunities.

The risk ratios in our study were in general higher than those reported in the literature for many of the general maternal health factors, obstetric history, and anatomy/biomarkers ^29-32. The high negative likelihood ratio of LR for rule-out and the large contrast in sPTB risks between LR and HR groups seen in this selected sub-cohort could provide additional information in the clinical decision process for pregnancy management and treatment options for nulliparous patients to improve of overall pregnancy outcomes.

c. Clinical Implications

From a clinical perspective, this test could aid in provider counseling to patients very early in their pregnancy. Someone whose test put them in a HR tier would have a PPV for ≤ 35 week delivery of 13.3% (95% CI 11.41, 15.76), while someone in the LR group would have a NPV to deliver ≤ 35 weeks of 98.6%. Given that newborns born at earlier gestational ages have the highest rates of morbidity and mortality, the ability to determine risk early in these patients is potentially important. This may allow for clinicians and healthcare systems to allocate PTB prevention programs to those at highest need, while refraining from overutilization of resources for groups extremely unlikely to deliver preterm.

Risk stratification has been put forth as an effective way to identify care pathways for subgroups of patients to direct care for individuals while managing the overall outcomes for populations of patients ^{33, 34}. As applied to pregnant patients, the strategy could be used to personalize care plans for the relatively common adverse obstetric outcomes, such as preterm labor and preeclampsia, by segmenting the patients into a high-risk level and a medium-risk level, while distinguishing a significant proportion of patients who are at lowest risk who do not require a high-intensity support. As adjuncts to clinical history and examination, objective, biochemical markers that can assess fetal and maternal states and capture early signs of the varying underlying causes of sPTB ⁷ without increasing risk to the pregnancy, could serve as additional stratification tools to conserve resources for individuals who require active intervention. Such tools could also help define entry criteria for clinical trials evaluating and developing optimal care pathways. Because those molecular mechanisms will likely not be operant in all patients, profiling those molecular pathways can provide a way to fine-tune the personalization of care to subgroups of high-risk patients. The effective use of that information for personalized treatment could rationally be based on clinical trials designed around molecular information similar to the adaptive clinical trials proposed for prostate and breast cancers ^{34, 35}.

d. Research Implications

From a technical standpoint, our CMP-based biomarker strategy fulfills the goal of a minimally invasive way to sample derangements in maternal-fetal physiology and maternal systemic responses and the profiling of multiple pathways including those proposed for antecedents to sPTB ^{7, 10, 14}. However, it is the development of our multi-tiered model in the current work that furthers the goal of improved stratification for a clinical condition defined by a continuum of time to events. The approach we used, analogous to other clinical tests with dual cutoff such as the OncoType Dx molecular diagnostic test for breast cancer to stratify risk of recurrence and response to chemotherapy ^36-38, represents a potentially first in kind effort in the field of obstetrics using first trimester biochemical markers to better stratify patients at risk for sPTB. This strategy could also be applied to other adverse pregnancy outcomes. Such a clinical tool could potentially define a new taxonomy while enabling better clinical management and the activation of tailored clinical care pathways.

e. Strengths and Limitations

The current study included one of the largest numbers of extremely and very preterm cases reported for sPTB risk prediction studies. However, due to their extremely low incidence rates in the nuMoM2b cohort, we chose to have the size of the MC sample sets equal to that of the nuMoM2b cohort to ensure that the early sPTB cases from the validation set would have a chance to be selected in the MC sample set. The p-values from analyses comparing risk groups using MC sample sets (Table 3B and Fig. 3) reflected this choice of MC sample set size.

The current study recruited from eight geographically diverse U.S. sites covering a fairly broad range of populations. However, the study is limited by the relatively small sample size, particularly for very early preterm deliveries. While the current multi-tier method holds clinical promise as an early predictive risk-stratifying biomarker, the isolation of the CMPs and testing of the protein panel is not currently commercially available. Further work is needed to validate the test, make this test widely available, and deliver rapid results before clinical implementation. This is important, as there may be some degradation in test performance prospectively arising from our exclusion of medically indicated PTB in this cohort. Additionally, the cohort included only nulliparas and clearly none have a history of sPTB. The association between the serum assay and risk factors or clinical outcomes such as a second trimester short cervix would need to be analyzed in future cohorts.

f. Conclusion

In conclusion, a multi-tier model generated from a CMP-based protein biomarker panel demonstrated effective risk stratification of nulliparous pregnant individuals and the ability to capture very early preterm birth. Further clinical refinement and validation, coupled with trials of effective preventive therapies for those identified at the highest risk, is necessary. Findings from such studies may provide a novel means to reduce sPTB during early gestation.

AJOG-at-a-Glance.

1. Why was this study conducted? This study was conducted to characterize the ability of a first trimester circulating microparticle protein (CMP) panel to risk stratify patients for spontaneous preterm birth.

2. What are the key findings? The 7-protein panel can stratify individuals into low-, medium-, and high risk groups. The Risk Ratio for sPTB for high vs low risk individuals was 19.92 (95% CI 10.4 to 37.4). Additionally, the risk groups demonstrated statistically significant differences in time-to-event (birth) analysis of gestational age at birth (p < 0.0001).

3. What does this study add to what is already known? This study demonstrates the ability of a first trimester CMP protein biomarker panel to stratify risk for sPTB. The findings suggest that such a multi-tiered stratification tool could be used to assess risk early in pregnancy to enable timely clinical management and interventions, and, ultimately, to enable the development of tailored care pathways for sPTB prevention.

• 1.Katz J, Lee AC, Kozuki N, et al. Mortality risk in preterm and small-for-gestational-age infants in low-income and middle-income countries: a pooled country analysis. Lancet 2013;382:417–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 2.Liu L, Oza S, Hogan D, et al. Global, regional, and national causes of child mortality in 2000-13, with projections to inform post-2015 priorities: an updated systematic analysis. Lancet 2015;385:430–40. [DOI] [PubMed] [Google Scholar]
• 3.Shapiro-Mendoza CK, Tomashek KM, Kotelchuck M, et al. Effect of late-preterm birth and maternal medical conditions on newborn morbidity risk. Pediatrics 2008;121:e223–32. [DOI] [PubMed] [Google Scholar]
• 4.Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes of preterm birth. Lancet 2008;371:75–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 5.McElrath TF, Hecht JL, Dammann O, et al. Pregnancy disorders that lead to delivery before the 28th week of gestation: an epidemiologic approach to classification. Am J Epidemiol 2008;168:980–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 6.Gyamfi-Bannerman C, Ananth CV. Trends in spontaneous and indicated preterm delivery among singleton gestations in the United States, 2005-2012. Obstet Gynecol 2014;124:1069–74. [DOI] [PubMed] [Google Scholar]
• 7.Romero R, Dey SK, Fisher SJ. Preterm labor: one syndrome, many causes. Science 2014;345:760–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 8.Adam S, Elfeky O, Kinhal V, et al. Review: Fetal-maternal communication via extracellular vesicles - Implications for complications of pregnancies. Placenta 2017;54:83–88. [DOI] [PubMed] [Google Scholar]
• 9.Al-Nedawi K, Meehan B, Micallef J, et al. Intercellular transfer of the oncogenic receptor EGFRvIII by microvesicles derived from tumour cells. Nat Cell Biol 2008;10:619–24. [DOI] [PubMed] [Google Scholar]
• 10.Cantonwine DE, Zhang Z, Rosenblatt K, et al. Evaluation of proteomic biomarkers associated with circulating microparticles as an effective means to stratify the risk of spontaneous preterm birth. Am J Obstet Gynecol 2016;214:631 e1–31 e11. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 11.Di Vizio D, Kim J, Hager MH, et al. Oncosome formation in prostate cancer: association with a region of frequent chromosomal deletion in metastatic disease. Cancer Res 2009;69:5601–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 12.Ginestra A, La Placa MD, Saladino F, Cassara D, Nagase H, Vittorelli ML. The amount and proteolytic content of vesicles shed by human cancer cell lines correlates with their in vitro invasiveness. Anticancer Res 1998;18:3433–7. [PubMed] [Google Scholar]
• 13.Haas DM, Parker CB, Wing DA, et al. A description of the methods of the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (nuMoM2b). Am J Obstet Gynecol 2015;212:539 e1–39 e24. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 14.McElrath TF, Cantonwine DE, Jeyabalan A, et al. Circulating microparticle proteins obtained in the late first trimester predict spontaneous preterm birth at less than 35 weeks' gestation: a panel validation with specific characterization by parity. Am J Obstet Gynecol 2019;220:488 e1–88 e11. [DOI] [PubMed] [Google Scholar]
• 15.World Health O. Born too soon: the global action report on preterm birth. Geneva: World Health Organization, 2012. [Google Scholar]
• 16.Wortzel I, Dror S, Kenific CM, Lyden D. Exosome-Mediated Metastasis: Communication from a Distance. Dev Cell 2019;49:347–60. [DOI] [PubMed] [Google Scholar]
• 17.In: Behrman RE, Butler AS, eds. Preterm Birth: Causes, Consequences, and Prevention. Washington (DC), 2007. [PubMed] [Google Scholar]
• 18.Administration USFD. In Vitro Diagnostic Multivariate Index Assays - Draft Guidance for Industry, Clinical Laboratories, and FDA Staff, 2007.
• 19.McElrath TF, Cantonwine DE, Gray KJ, et al. Late first trimester circulating microparticle proteins predict the risk of preeclampsia < 35 weeks and suggest phenotypic differences among affected cases. Sci Rep 2020;10:17353. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 20.Ezrin AM, Brohman B, Willmot J, et al. Circulating serum-derived microparticles provide novel proteomic biomarkers of spontaneous preterm birth. Am J Perinatol 2015;32:605–14. [DOI] [PubMed] [Google Scholar]
• 21.Burger B, Vaudel M, Barsnes H. Importance of Block Randomization When Designing Proteomics Experiments. J Proteome Res 2021;20:122–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 22.Turapov OA, Mukamolova GV, Bottrill AR, Pangburn MK. Digestion of native proteins for proteomics using a thermocycler. Anal Chem 2008;80:6093–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 23.Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System KDD '16: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining ACM, 2016. [Google Scholar]
• 24.Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer; Number of pages. [Google Scholar]
• 25.D'Arrigo G, Leonardis D, Abd ElHafeez S, Fusaro M, Tripepi G, Roumeliotis S. Methods to Analyse Time-to-Event Data: The Kaplan-Meier Survival Curve. Oxid Med Cell Longev 2021;2021:2290120. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 26.R Core Team. R: A language and environment for statistical computing.: R Foundation for Statistical Computing, Vienna, Austria, 2021. [Google Scholar]
• 27.Hornaday KK, Wood EM, Slater DM. Is there a maternal blood biomarker that can predict spontaneous preterm birth prior to labour onset? A systematic review. PLoS One 2022;17:e0265853. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 28.Phillips C, Velji Z, Hanly C, Metcalfe A. Risk of recurrent spontaneous preterm birth: a systematic review and meta-analysis. BMJ Open 2017;7:e015402. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 29.Bolt LA, Chandiramani M, De Greeff A, Seed PT, Kurtzman J, Shennan AH. The value of combined cervical length measurement and fetal fibronectin testing to predict spontaneous preterm birth in asymptomatic high-risk women. J Matern Fetal Neonatal Med 2011;24:928–32. [DOI] [PubMed] [Google Scholar]
• 30.Hee L. Likelihood ratios for the prediction of preterm delivery with biomarkers. Acta Obstet Gynecol Scand 2011;90:1189–99. [DOI] [PubMed] [Google Scholar]
• 31.Meis PJ, Goldenberg RL, Mercer BM, et al. The preterm prediction study: risk factors for indicated preterm births. Maternal-Fetal Medicine Units Network of the National Institute of Child Health and Human Development. Am J Obstet Gynecol 1998;178:562–7. [DOI] [PubMed] [Google Scholar]
• 32.Rundell K, Panchal B. Preterm Labor: Prevention and Management. Am Fam Physician 2017;95:366–72. [PubMed] [Google Scholar]
• 33.Haas LR, Takahashi PY, Shah ND, et al. Risk-stratification methods for identifying patients for care coordination. Am J Manag Care 2013;19:725–32. [PubMed] [Google Scholar]
• 34.Wang G, Zhao D, Spring DJ, DePinho RA. Genetics and biology of prostate cancer. Genes Dev 2018;32:1105–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 35.Sparano JA, Gray RJ, Makower DF, et al. Clinical Outcomes in Early Breast Cancer With a High 21-Gene Recurrence Score of 26 to 100 Assigned to Adjuvant Chemotherapy Plus Endocrine Therapy: A Secondary Analysis of the TAILORx Randomized Clinical Trial. JAMA Oncol 2020;6:367–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 36.Cronin M, Sangli C, Liu ML, et al. Analytical validation of the Oncotype DX genomic diagnostic test for recurrence prognosis and therapeutic response prediction in node-negative, estrogen receptor-positive breast cancer. Clin Chem 2007;53:1084–91. [DOI] [PubMed] [Google Scholar]
• 37.Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifentreated, node-negative breast cancer. N Engl J Med 2004;351:2817–26. [DOI] [PubMed] [Google Scholar]
• 38.Sparano JA, Gray RJ, Makower DF, et al. Adjuvant Chemotherapy Guided by a 21-Gene Expression Assay in Breast Cancer. N Engl J Med 2018;379:111–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 39.Rogeberg M, Malerod H, Roberg-Larsen H, Aass C, Wilson SR. On-line solid phase extraction-liquid chromatography, with emphasis on modern bioanalysis and miniaturized systems. J Pharm Biomed Anal 2014;87:120–9. [DOI] [PubMed] [Google Scholar]
• 40.Kuklenyik Z, Calafat AM, Barr JR, Pirkle JL. Design of online solid phase extraction-liquid chromatography-tandem mass spectrometry (SPE-LC-MS/MS) hyphenated systems for quantitative analysis of small organic compounds in biological matrices. J Sep Sci 2011;34:3606–18. [DOI] [PubMed] [Google Scholar]

Supplementary Materials