You are here

Development and validation of a spontaneous preterm delivery predictor in asymptomatic women

Background

Preterm delivery remains the leading cause of perinatal mortality. Risk factors and biomarkers have traditionally failed to identify the majority of preterm deliveries.

Objective

To develop and validate a mass spectrometry–based serum test to predict spontaneous preterm delivery in asymptomatic pregnant women.

Study Design

A total of 5501 pregnant women were enrolled between 170/7 and 286/7 weeks gestational age in the prospective Proteomic Assessment of Preterm Risk study at 11 sites in the United States between 2011 and 2013. Maternal blood was collected at enrollment and outcomes collected following delivery. Maternal serum was processed by a proteomic workflow, and proteins were quantified by multiple reaction monitoring mass spectrometry. The discovery and verification process identified 2 serum proteins, insulin-like growth factor–binding protein 4 (IBP4) and sex hormone–binding globulin (SHBG), as predictors of spontaneous preterm delivery. We evaluated a predictor using the log ratio of the measures of IBP4 and SHBG (IBP4/SHBG) in a clinical validation study to classify spontaneous preterm delivery cases (<370/7 weeks gestational age) in a nested case-control cohort different from subjects used in discovery and verification. Strict blinding and independent statistical analyses were employed.

Results

The predictor had an area under the receiver operating characteristic curve value of 0.75 and sensitivity and specificity of 0.75 and 0.74, respectively. The IBP4/SHBG predictor at this sensitivity and specificity had an odds ratio of 5.04 for spontaneous preterm delivery. Accuracy of the IBP4/SHBG predictor increased using earlier case-vs-control gestational age cutoffs (eg, <350/7 vs ≥350/7 weeks gestational age). Importantly, higher-risk subjects defined by the IBP4/SHBG predictor score generally gave birth earlier than lower-risk subjects.

Conclusion

A serum-based molecular predictor identifies asymptomatic pregnant women at risk of spontaneous preterm delivery, which may provide utility in identifying women at risk at an early stage of pregnancy to allow for clinical intervention. This early detection would guide enhanced levels of care and accelerate development of clinical strategies to prevent preterm delivery.

Key words: biomarker, pregnancy, preterm birth, proteomics, IGFBP4, IBP4, SHBG.

Preterm birth (PTB), defined as delivery before 37 weeks of gestation, affects 15 million infants born each year, varying from approximately 5% to 18% of all births across different geographies worldwide.1 In the United States, it is the leading cause of neonatal death and the second-leading cause of death in children before age 5 years. PTB is also a major source of long-term health consequences, including chronic lung disease, hearing and visual impairments, and neurodevelopmental disabilities, such as cerebral palsy. The health-economic impact of PTB in 2005 in the United States was estimated to be above $26 billion,2 and costs continue to rise in most countries.3 and 4

Prior history of spontaneous preterm delivery (sPTD) is currently the single strongest predictor of subsequent PTD. After 1 prior sPTD, the probability of a second PTD is 30–50%.5, 6, and 7 Other maternal risk factors include black race, low maternal body mass index (BMI), and short cervical length.8 and 9 Amniotic fluid, cervicovaginal fluid, and serum biomarker studies to predict sPTD suggest that multiple molecular pathways are aberrant in women who ultimately deliver preterm.10, 11, 12, and 13

Despite intense research to identify at-risk women, PTD prediction algorithms based solely on clinical and demographic factors or using measured serum or vaginal biomarkers have not resulted in clinically useful tests.14, 15, 16, and 17 More accurate methods to identify women at risk during their first pregnancy and sufficiently early in gestation are needed to allow for clinical intervention. The purpose of the project was to develop a proteomic signature profile for the prediction of sPTD and to validate this profile in a separate independent sample of new subjects.

Materials and Methods

Subjects

The Proteomic Assessment of Preterm Risk (PAPR) study was conducted under a standardized protocol at 11 Institutional Review Board (IRB)-approved sites across the United States (Clinicaltrials.gov identifier: NCT01371019). Subjects were enrolled between 170/7 and 286/7 weeks gestational age (GA). Dating was established using a predefined protocol of menstrual dating confirmed by early ultrasound biometry, or ultrasound alone, to provide the best clinically estimated GA. BMI was derived from height and prepregnancy self-reported weight. Pregnancies with multiple gestations or with known or suspected major fetal anomalies were excluded. Pertinent information regarding subject demographic characteristics, past medical and pregnancy history, current pregnancy history, and concurrent medications was collected and entered into an electronic case report form. Following delivery, data were collected for maternal and infant outcomes and complications. All deliveries were classified by the study sites as term (≥370/7 weeks GA), spontaneous preterm (including preterm premature rupture of membranes [PPROM]), or medically indicated preterm births. Classification of preterm deliveries was subsequently adjudicated by the chief medical officer (D.H.) at Sera Prognostics, Inc, who was blinded to results from laboratory analysis. As indicated, discrepancies were clarified with the principal investigator at the study site. The adjudication occurred prior to locking down the validation database and conducting laboratory and statistical analysis.

Sample collection

Maternal blood was collected and processed as follows: a 10-minute room temperature clotting period, followed by immediate refrigerated centrifugation or placement in an ice-water bath at 4–8oC until centrifugation. Blood was centrifuged within 2.5 hours of collection and 0.5 mL serum aliquots were stored at −80oC until analyzed. Details regarding sample accessioning can be found in Supplementary Materials and Methods.

Predictor development principles

Development of the insulin-like growth factor–binding protein 4 (IBP4)/sex hormone–binding globulin (SHBG) predictor included independent and sequential discovery, verification, and validation steps consistent with Institute of Medicine (IOM) guidelines for best practices in “omics” research.18 Analytical validation preceded clinical validation sample analysis and included assessment of inter- and intrabatch precision, carry-over, and limit of detection.

The validation nested case-control analysis was performed on specimens from 81 sPTD cases and controls independent of discovery and verification. Validation sPTD cases were the last to be enrolled in PAPR and included samples from 9 sites in total, with 2 sites being unique to validation. Validation cases and controls underwent 100% on-site source document verification with each subject’s medical record prior to mass spectrometry (MS) analysis. This process ensured that all subjects satisfied the inclusion and exclusion criteria, as well as confirmed medical/pregnancy complications and GA at birth assignments for all subjects at time of sample collection and delivery. Detailed analysis protocols, including the validation study design, analysis plan, and a blinding protocol, were preestablished. Personnel were blinded to subject case, control, and GA at birth data assignments, with the exception of the director of clinical operations (DCO) and clinical data manager. The data analysis plan included prespecified validation claims and a protocol for double independent external analyses. Predictor scores, calculated as described below, were determined for all subject samples by a blinded statistician and subsequently confirmed by 2 external blinded statisticians, 1 of whom was university based (E. Mazzola) and the other an industry consultant (P. Kearney). Case, control, and GA data, linked to the predictor scores by the DCO, were then provided to the 2 external statisticians for analysis. Area under the receiver operating characteristic curve (AUROC) and significance testing results were then transferred back to the DCO. Transfer of data incorporated the use of the SUMPRODUCT19 function to ensure data integrity. To provide an audit trail of data from each subject through to validation results, real-time digital time-stamping was applied to analytical data, plans, and reports.

Validation study design

In the primary analysis, cases were defined as subjects with deliveries due to PPROM or spontaneous onset of labor with delivery <370/7 weeks GA. Controls were subjects who delivered at ≥370/7 weeks GA. Prior discovery and verification analyses investigated 44 candidate biomarkers using serum samples collected across broad GA (170/7 through 256/7 weeks GA) (Supplementary Materials and Methods). Discovery and verification identified an optimal narrow GA at blood draw (GABD) interval (190/7 through 216/7 weeks) and 2 proteins, IBP4, up-regulated in sPTD cases, and SHBG, down-regulated in sPTD cases, used in a ratio (IBP4/SHBG) as the best predictor by AUROC for sPTD (Supplementary Materials and Methods). In discovery and verification, subjects without extreme BMI values had improved classification performance by IBP4/SHBG (Supplementary Results, Appendix). Following discovery and verification analyses, we proceeded to analytical and clinical validation.

Validation sPTD cases totaled 18 subjects collected between 190/7 and 216/7 weeks GABD from a total available of 81 subjects between 170/7 and 286/7 weeks GA. Sets of controls, comprising 2 controls per sPTD case matched by GABD, were randomly selected using the R statistical program (R 3.0.2)20 and 21 and compared to the term delivery distribution as outlined in the 2012 National Vital Statistics Report22 using a chi-square test. Randomly created control sets (in groups of 10) were examined for sets yielding a P value approaching 1.0.

Our primary objective was to validate the performance of the IBP4/SHBG ratio as a predictor for sPTD using AUROC.20 and 23 To control the overall multiple testing error rate (α = 0.05), the fixed sequence approach24 and 25 was applied to GABD increments within the optimal interval (190/7 through 216/7 weeks GA) identified in discovery and verification with and without the application of a BMI stratification (Supplementary Materials and Methods). Significance was assessed by the Wilcoxon-Mann-Whitney statistic that tests equivalence to AUROC = 0.5 (random chance).26 and 27 For determinations of classification performance at GA boundaries other than <370/7 vs ≥370/7 weeks GA (eg, <360/7 vs ≥360/7, <350/7 vs ≥350/7), cases and controls were redefined as all subjects below and equal to/above the specific boundary, respectively.

Laboratory methods

A systems biology approach was employed to generate a highly multiplexed multiple reaction monitoring (MRM) MS assay (Supplementary Materials and Methods and Supplementary Results). The validation assay quantified proteotypic peptides specific to predictor proteins IBP4 and SHBG and other controls. Samples were processed in batches of 32, which were composed of clinical subjects (n = 24), pooled serum standards from healthy nonpregnant donors (HGS) (n = 3), pooled serum standards from healthy pregnant donors (pHGS) (n = 3), and phosphate-buffered saline that served as process controls (n = 2). For all analyses, serum samples were first depleted of high-abundance and nondiagnostic proteins using MARS-14 immunodepletion columns (Agilent Technologies, Santa Clara, CA), reduced with dithiothreitol, alkylated with iodoacetamide, and digested with trypsin. Heavy-labeled stable isotope standard (SIS) peptides (New England Peptide, Gardner, MA) were then added to samples, which were subsequently desalted and analyzed by reversed-phase liquid chromatography (LC)/MRM MS. SIS peptides were used for normalization by generating response ratios (RR), where the peak area of a peptide fragment ion (ie, transition) measured in serum was divided by that of the corresponding SIS transition spiked into the same serum sample.

The IBP4/SHBG predictor

The predictor score was defined as the natural log of the ratio of the IBP4 and SHBG peptide transition response ratios:

FORMULA:

stripin: si1.gif

where RR are the measured response ratios of the respective peptides.

Results

Figure 1 summarizes the distribution of study subjects in PAPR. Between March 2011 and August 2013, 5501 subjects were enrolled. Clinical and demographic data of the enrolled subjects by site are included in Supplementary Materials and Methods. As predefined in the protocol, 410 subjects (6.7%) were excluded from analysis owing to receiving progestogen therapy after the first trimester of pregnancy. An additional 120 subjects (2.2%) were excluded owing to early discontinuation, and 146 (2.7%) were lost to follow-up. A total of 4825 subjects were available for analysis. There were 533 PTDs: 248 (4.7%) spontaneous and 285 (5.9%) medically indicated. Compared to those who delivered at term, subjects with an sPTD were more likely to have had 1 or more prior PTDs and to have experienced bleeding after 12 weeks of gestation in the study pregnancy (Table 1). Characteristics of sPTD cases and term controls selected for the overall validation cohort were not significantly different from each other, with the exception that there were significantly more Hispanic controls (47.5% vs 33.3%, P = .035). Similarly, subjects selected for the validated window were largely representative of the study cohort as a whole (Table 1).

gr1

Figure 1 Distribution of subjects in the PAPR databaseA total of 5501 subjects were enrolled in the Proteomic Assessment of Preterm Risk (PAPR) study between 170/7 and 286/7 weeks gestational age (GA). A number of subjects (120) were discontinued, and another 146 subjects were lost to follow-up. Of the 5235 delivered subjects, 410 were excluded from these analyses owing to progestogen use. Of the 4825 subjects remaining, 4292 delivered at term, 248 experienced a spontaneous preterm delivery (sPTD), and 285 delivered preterm owing to medical indications. Following preanalytic exclusion of 31 subjects, 217 sPTDs were available for analysis and distributed among discovery, verification, and validation studies as shown.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Table 1 Maternal characteristics and pregnancy outcomes stratified by timing of delivery (sPTD and term)

Variables PAPR study Entire validation cohort (170/7 to 286/7 weeks) Validated window (191/7 to 206/7 weeks)
Cases n (%) (n = 217) Controls n (%) (n = 4292) P value Cases n (%) (n = 81) Controls n (%) (n = 162) P value Cases n (%) (n = 18) Controls n (%) (n = 36) P value
Maternal characteristics
 Maternal age at enrollment, y .245 .239 .387
 18–22 y 58 (26.7) 990 (23.1) 22 (27.2) 47 (29.0) 6 (33.3) 13 (36.1)
 23–27 y 56 (25.8) 1222 (28.5) 17 (21.0) 41 (25.3) 6 (33.3) 9 (25.0)
 28–32 y 54 (24.9) 1154 (26.9) 25 (30.9) 34 (21.0) 5 (27.8) 5 (13.9)
 33–37 y 31 (14.3) 692 (16.1) 9 (11.1) 30 (18.5) 1 (5.6) 7 (19.4)
 38 y or more 18 (8.3) 234 (5.5) 8 (9.9) 10 (6.2) 0 2 (5.6)
 Mean 28 28 28 28 25 27
 Median 27 27 28 27 25 25
 Interquartile range 22–32 23–32 21–32 22–32 21–30 22–33
 Body mass index, kg/m2 .380 .802 .959
 Less than 18.5 10 (4.7) 129 (3.1) 1 (1.3) 2 (1.3) 0 0
 18.5–24.9 78 (36.8) 1789 (42.3) 25 (31.3) 55 (34.6) 8 (44.4) 16 (45.7)
 25.0–29.9 54 (25.5) 1091 (25.8) 26 (32.5) 46 (28.9) 4 (22.2) 9 (25.7)
 30.0–34.9 39 (18.4) 617 (15.6) 17 (21.3) 25 (15.7) 3 (16.7) 4 (11.4)
 35.0–39.9 17 (8.0) 320 (7.6) 6 (7.5) 17 (10.7) 2 (11.1) 5 (14.3)
 Greater than 40.0 14 (6.6) 286 (6.7) 5 (6.3) 14 (8.8) 1 (5.6) 1 (2.9)
 Mean 27.8 27.5 28.4 29.1 28.2 27.4
 Median 26.5 25.7 27.4 27.8 26.5 27
 Interquartile range 22.7–31.8 22.3–31.1 23.6–32.0 23.4–32.4 23.8–33.7 22.3–30.6
 Education level <.0002 .201 .263
 Graduate degree 13 (6.0) 461 (10.9) 6 (7.7) 14 (8.7) 0 2 (5.7)
 College diploma 34 (15.8) 701 (16.6) 10 (12.6) 22 (13.8) 2 (11.1) 5 (14.3)
 Some college 51 (23.7) 936 (22.2) 19 (24.0) 23 (14.4) 1 (5.6) 5 (14.3)
 High school diploma/equivalent 46 (21.4) 1032 (24.5) 16 (20.2) 50 (31.3) 5 (27.8) 14 (40.0)
 Some high school 53 (24.6) 774 (18.4) 25 (31.6) 36 (22.5) 9 (50.0) 6 (17.1)
 9th grade or less 12 (5.8) 292 (6.9) 3 (3.8) 14 (8.7) 1 (5.6) 3 (8.6)
 Other 6 (2.8) 23 (0.6) 0 1 (0.6) 0 0
 Ethnicity .157 .035 .844
 Hispanic or Latino 89 (41.0) 1557 (36.3) 27 (33.3) 77 (47.5) 7 (38.9) 15 (41.7)
 Non-Hispanic or Latino 128 (59.0) 2735 (63.7) 54 (66.7) 85 (52.5) 11 (61.1) 21 (58.3)
 Race .887 .811 .319
 American Indian/Alaskan Native 1 (0.5) 29 (0.7) 0 2 (1.2) 0 1 (2.8)
 Asian 4 (1.8) 131 (3.1) 1 (1.2) 1 (0.6) 0 1 (2.8)
 Black or African-American 45 (20.7) 838 (19.5) 19 (23.5) 37 (22.8) 2 (11.1) 11 (30.6)
 Native Hawaiian or other Pacific Islander 0 12 (0.30) 0 2 (1.2) 0 1 (2.8)
 White 156 (71.9) 3101 (72.3) 58 (71.6) 114 (70.4) 16 (88.9) 22 (61.1)
 Other 11 (5.1) 193 (4.5) 3 (3.7) 6 (3.7) 0 0
Obstetrical characteristics
 Primigravida 64 (29.5) 1212 (28.2) .689 27 (33.3) 39 (24.1) .126 5 (27.8) 8 (22.2) .652
 Multigravida 153 (70.5) 3080 (71.8) 54 (66.7) 123 (75.9) 13 (72.2) 28 (77.8)
 Number of prior full-term deliveries .007 .326 .790
 1 or more 113 (73.8) 2538 (82.4) 40 (74.5) 102 (82.9) 10 (76.9) 22 (78.6)
 None 40 (26.2) 542 (17.6) 13 (24.5) 21 (17.1) 3 (23.1) 6 (21.4)
 Number of prior sPTDs <.0001 .221 .524
 1 or more 35 (22.9) 339 (11.0) 9 (16.7) 11 (8.9) 1 (7.7) 6 (21.4)
 None 118 (77.1) 2741 (89.0) 45 (83.3) 112 (91.1) 12 (92.3) 22 (78.6)
Lifestyle characteristics
 Smoking .412 .719 1.000
 Yes 34 (15.7) 588 (13.7) 15 (18.5) 27 (16.7) 3 (16.7) 6 (16.7)
 No 183 (84.3) 3704 (86.3) 66 (81.5) 135 (83.3) 15 (83.3) 30 (83.3)
 Illicit drugs .283 .628 .739
 Yes 16 (7.4) 242 (5.6) 6 (7.4) 15 (9.3) 2 (11.1) 3 (8.3)
 No 201 (92.6) 4050 (94.4) 75 (92.6) 147 (90.7) 16 (88.9) 33 (91.7)
 Alcohol .096 .628 .278
 Yes 20 (9.2) 273 (6.4) 6 (7.4) 15 (9.3) 4 (22.2) 4 (11.1)
 No 197 (90.8) 4018 (93.6) 75 (92.6) 147 (90.7) 14 (77.8) 32 (88.9)
 Alcohol use .108 .592 .278
 Yes (amount unknown) 3 (1.4) 39 (0.9) 0 2 (1.2) 0 0
 Social (occasional) 16 (7.4) 230 (5.4) 6 (7.4) 13 (8.0) 4 (22.2) 4 (11.1)
 Heavy (daily) 1 (0.5) 4 (0.09) 0 0 0 0
 No 197 (90.8) 4018 (93.6) 75 (92.6) 147 (90.7) 14 (77.8) 32 (88.9)
Medical characteristics
Bleeding during pregnancy after 12 wk .006 .360 .308
 Yes 21 (9.7) 228 (5.3) 7 (8.6) 9 (5.6) 0 2 (5.6)
 No 196 (90.3) 4064 (94.7) 74 (91.4) 153 (94.4) 18 (100.0) 34 (94.4)

Comparisons of clinical data between cases and controls were performed using chi-square test, Fisher exact test, or Mann-Whitney test, as appropriate (SAS System 9.4 and R 3.1.0).

Missing values are excluded in the frequency tables.

N, number of subjects; sPTD, spontaneous preterm delivery.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Validation analysis

In discovery and verification analyses, the ratio of IBP4 (up-regulated in sPTD)/SHBG (down-regulated in sPTD) and the interval between 190/7 and 216/7 weeks GA was identified as the best-performing sPTD predictor by AUROC and GA interval, respectively (Supplementary Results). For validation, a predefined fixed sequence approach validated the IBP4/SHBG predictor with and without BMI stratification, with optimal performance identified for the GA interval of 191/7 through 206/7 weeks. Without taking BMI into consideration, validated performance was AUROC = 0.67 (95% confidence interval [CI], 0.52–0.81) for 18 sPTD cases and 36 term controls (Supplementary Results). However, as expected, performance was improved with a BMI stratification of >22 and ≤37 kg/m2, which corresponded to an AUROC of 0.75 for 12 sPTD cases and 23 term controls (95% CI, 0.56–0.91) (Figure 2; and Supplementary Results). More detailed characterization of BMI stratification can be found in the Supplementary Results. Performance measures of sensitivity, specificity, AUROC, and odds ratios (ORs) were determined at varied case-vs-control boundaries (Table 2). For sPTD vs term birth (<370/7 vs ≥370/7 weeks), the sensitivity and specificity was 0.75 and 0.74, respectively, with an OR of 5.04 (95% CI, 1.4–18). The results at other boundaries are summarized in Table 2. Accuracy of the test improved at lower GA boundaries.

gr2

Figure 2 ROC performance of the IBP4/SHBG predictor in validationReceiver operating characteristic performance in the validation sample set. The plot graphs sensitivity (true-positive rate) vs 1-specificity (false-positive rate), where sPTD cases are defined as delivery <370/7 weeks GA and term controls are defined as delivery ≥370/7 weeks GA. The AUROC corresponds to 0.75 for the BMI-stratified validation subjects, (>22 and ≤37 kg/m2) comprising 35 subjects: 12 sPTD cases and 23 term controls.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Table 2 Performance of IBP4/SHBG predictor

GA boundary AUC (95% CI) Sensitivity Specificity OR (95% CI)
<37 vs ≥37 0.75 (0.56–0.91) 0.75 0.74 5.04 (1.4–18)
<36 vs ≥36 0.79 (0.53–0.99) 0.83 0.83 17.33 (2.2–138)
<35 vs ≥35 0.93 (0.81–1.00) 1 0.83 34.47 (1.7–699)

AUC, area under the curve; CI, confidence interval; GA, gestational age; OR, odds ratio.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

The prevalence adjusted positive predictive value (PPV), a measure of clinical risk, is shown as a function of predictor score in Figure 3. Stratification of subjects with increasing predictor scores occurs as PPV increases from a background value (population sPTD rate of 7.3% for singleton births in the United States)28 to relative risks of 2× (14.6%) and 3× (21.9%) (dashed lines) and higher (Figure 3). The distribution of IBP4/SHBG predictor score values for subjects color-coded by GA at birth category are shown in box plots in Figure 3. The earliest sPTD cases (<350/7 weeks GA) have higher predictor scores than late-term controls (≥390/7 weeks GA), while the scores for late sPTD cases (≥350/7 through <370/7 weeks GA) overlap with early-term controls (≥370/7 through <390/7 weeks GA) (Figure 3). Using the BMI-stratified risk curve in Figure 3, validation subjects were identified as high or low risk according to a predictor score cutoff corresponding to 2× relative risk (PPV of 14.6%, predictor score = −1.36655). The rates of births for the high- and low-risk groups were then displayed as events in a Kaplan-Meier analysis (Figure 4). From this analysis, those classified as high risk generally delivered earlier than those classified as low risk (P = .0004).

gr3

Figure 3 Stratification of validation subjects by the IBP4/SHBG predictorPrevalence-corrected positive predictive value (PPV) was plotted as a function of predictor score for the validation samples within the validated blood draw window and BMI >22 and ≤37 kg/m2 (35 subjects: 12 sPTD cases and 23 term controls). Horizontal dashed lines identify the average population risk of 7.3%, calculated as 75% of the singleton rate of PTD of 9.71%,28 and relative risks of 2× (14.6%) and 3× (21.9%). Vertical dashed lines identify corresponding predictor scores. The confidence interval about the PPV curve (gray shaded area) was estimated using 150 subjects from postdiscovery datasets (verification, validation, and prevalence controls) as described in Supplementary Materials and Methods. Confidence intervals about the PPV were calculated with the normal approximation of the error for binomial proportions.54 Box plots at the foot of the figure correspond to the distributions of predictor scores for subjects in the different gestational age at birth (GAB) categories identified in the legend. The PPV curve and the box plots share the same predictor score axis.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

gr4

Figure 4 Kaplan-Meier estimator of high- and low-risk groupsShown is the rate of events (births) as a function of GAB for high- and low-risk groups defined by the IBP4/SHBG predictor. Subjects at or above 2× the background risk (14.6%) were considered high risk, while those below 2× were considered low risk. Curves depict the rate of events, which are identified as vertical lines, as a function of time, indicated by horizontal lines.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Postvalidation analyses

Predictor performance was measured using a combination of subjects from the blinded verification (Supplementary Materials and Methods) and validation analyses within the optimal BMI and GA interval. The ROC curve for the combined sample set (16 sPTD cases and 34 term controls) is shown and corresponds to an AUROC of 0.72 (95% CI, 0.51–0.8) (Figure 5).

gr5

Figure 5 Predictor performance in combined datasetsAUROC performance in the verification and validation subjects (BMI >22 and ≤37 kg/m2) within the validated blood draw interval. ROC plots of sensitivity vs 1 − specificity, where sPTD cases were defined as delivery ≥370/7 weeks GA and term controls were defined as delivery ≤370/7 weeks GA. The AUROC corresponds to 0.72.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Comment

Using an “omics” approach, we developed a maternal serum predictor comprising the ratio of IBP4/SHBG levels at 19–20 weeks with a BMI interval of >22 and ≤37 kg/m2 that identified 75% of women destined for sPTD. Prior history of sPTD8 and 29 and cervical length measurements9 and 30 are considered the best measures of clinical risk to date; however, either individually or in combination, they fail to predict the majority of sPTDs.

An ideal sPTD prediction tool would be minimally invasive; would be performed early in gestation, coinciding with timing of routine obstetrical visits; and would accurately identify those at highest risk. Current “omics” studies suggest that perturbations in the physiological state of pregnancy can be detected in maternal serum analytes measured in sPTD subjects. “Omics” discovery studies in PTD have included proteomic,10, 12, 13, 31, 32, 33, and 34 transcriptomic,35, 36, and 37 genomic,38, 39, 40, 41, and 42 and metabolomic43 approaches. However, to date, none of these approaches has produced validated testing methods to reliably predict the risk of sPTD in asymptomatic women.

The current investigation builds on the previous approaches in several ways. We completed a large prospective and contemporaneous clinical study that allowed independent discovery, verification, and validation analyses, while adhering to IOM guidelines regarding “omics” test development. We constructed a large and standardized multiplexed proteomic assay to probe biological pathways of relevance in pregnancy. Our study size and relatively broad blood collection window (170/7 through 286/7 weeks GA) also enabled the identification of a GA interval in which there were marked alterations in protein concentrations between sPTD cases and term controls. Use of a low-complexity predictor model (ie, the ratio of 2 proteins) limited the pitfalls of over-fitting.

Application of the proteomic assay and model building led to the identification of a pair of critical proteins (IBP4 and SHBG) with consistently good predictive performance for sPTD. Despite the challenges of building a classifier for a condition attributed to multiple etiologies, the predictor demonstrated good performance in 3 independent studies at a cutoff of <370/7 vs ≥370/7 weeks GA. Importantly, accuracy of the predictor improved for earlier sPTDs (eg, <350/7 weeks GA), enabling the detection of those sPTDs with the greatest potential for morbidity. Subjects determined to be at high risk for sPTD using the IBP4/SHBG predictor delivered significantly earlier than subjects identified as low risk. Our findings suggest that IBP4 and SHBG may perform important functions related to the etiologies of sPTD and/or act as convergence points in relevant biological pathways.

IBP4 is a member of a family of insulin-like growth factor binding proteins (IBPs) that negatively regulate the insulin-like growth factors IGF1 and IGF2.44 IBP4 is expressed by syncytiotrophoblasts45 and is the dominant IBP expressed by extravillous trophoblasts.46 Compared to normal pregnancies, maternal IBP4 levels in early pregnancy are higher in pregnancies complicated by fetal growth restriction and preeclampsia.46 We speculate that elevated maternal serum IBP4 levels may reflect abnormal placentation and thus an increased risk of sPTD.

SHBG regulates the availability of biologically active unbound steroid hormones.47 Plasma SHBG levels increase 5- to 10-fold during pregnancy,48 and evidence exists for extrahepatic expression, including placental trophoblastic cells.49 Physiologically, SHBG levels negatively correlate with triglycerides, insulin levels, and BMI.50 BMI’s effect on SHBG levels may explain, in part, the improved predictive performance with BMI stratification.

Intraamniotic infection and inflammation have been associated with PTD, as have increased levels of proinflammatory cytokines, including TNF-α and IL1-β.51 and 52 SHBG transcription in liver is suppressed by IL1-β and NF-kB-mediated TNF-α signaling,50 a pathway implicated in initiation of normal and abnormal labor.53 Lower levels of SHBG in women destined for sPTD may be a result of infection and/or inflammation. Hence, SHBG may be critical for control of androgen and estrogen action in the placental-fetal unit in response to upstream inflammatory signals.

Despite the strengths of our study results, these findings need to be evaluated in the context of the study limitations. Universal transvaginal ultrasound measurement of cervical length (CL) was not performed routinely at the majority of our study centers and was available for fewer than one-third of study subjects. It will be of interest to assess whether CL measurements improve upon the proteomic predictor in future studies or, alternatively, if risk stratification by the IBP4/SHBG classifier identifies women that benefit most from serial CL measurements. There was an insufficient number of women with prior preterm delivery who were not being treated with progesterone to allow inclusion of this variable in the analysis. Therefore, women with prior sPTD should be treated according to national guidelines, which include prophylactic treatment with 17-alpha hydroxyprogesterone caproate. Owing to sample size limitations, a more complete assessment of confounders will require future studies. Finally, it will be intriguing to investigate the performance of the molecular predictor together with a BMI variable or perhaps in combination with other medical/pregnancy history and sociodemographic characteristics.

In conclusion, a predefined predictive test for sPTD based on serum measurements of IBP4 and SHBG in asymptomatic parous and nulliparous women was validated in a completely independent set of subjects. Further functional studies on these proteins, their gene regulation, and related pathways may help to elucidate the molecular and physiological underpinnings of sPTD. Application of this predictor should enable early and sensitive detection of women at risk of sPTD. This early detection may improve pregnancy outcomes through increased clinical surveillance as well as accelerate the development of clinical interventions for PTD prevention.

Acknowledgments

The authors wish to acknowledge the research teams at each of the 11 study sites. Individuals who contributed substantively to the work include, but are not limited to, Holly Lynn Boggan, MHA, and Kenreka Tiwan Yeadon, Medical University of South Carolina; Erika A. Campos, Kathia Pena, and Karen Dorman, RN, MS, The University of North Carolina Chapel Hill (Wakemed); Dawn Cline, RN, BSN, CCRC, The Ohio State University Medical Center; Laura Gebhardt, Baystate Medical Center; Ashley Vanneman and Stephanie Lynch, BSN, RN, CCRC, Christiana Care Health System; Bianca Flor Jimenez and Crystal Ramos, Maricopa Integrated Health System; Lorrie A. Mason, MSN, Regional Obstetrical Consultants; Leah McCoy, The University of Texas Medical Branch at Galveston; Ami Patel, BA, San Diego Perinatal Center; Monica Rincon, MD, CCRP, Oregon Health & Science University–OHSU; and biostatistics consultant Giovanni Parmigiani, PhD, Dana Farber Cancer Institute.

Appendix. Supplementary Materials and Methods

Discovery and verification subjects

Discovery and verification subjects were derived from the Proteomic Assessment of Preterm Risk (PAPR) study described in the Materials and Methods section.

Discovery and verification principles

Spontaneous preterm delivery (sPTD) cases were defined as described in Materials and Methods.

Discovery and verification of the predictor was conducted according to guidelines for best practices in “omics” research.1 Nested case-control analyses used sample sets completely independent of each other. Cases and controls selected for discovery and verification underwent central review for within-subject data discrepancies; no source document verification (SDV) with the medical record was performed. All sPTD cases and controls for discovery and verification were individually adjudicated by the chief medical officer, and discrepancies were clarified with the principal investigator at the clinical site. Detailed analysis protocols, including study designs, analysis plans, and a verification blinding protocol, were preestablished. Laboratory and data analysis personnel were blinded to verification subject’s case, control, and gestational age (GA) data assignments. Predictor scores, calculated as described below, were assigned to all subjects by an internal blinded statistician. Case, control, and GA data, linked to the predictor scores by the director of clinical operations (DCO), were provided to an independent university-based external statistician for analysis. Area under the receiver operating characteristic curve (AUROC) results were then transferred back to the DCO. Transfer of data utilized a SUMPRODUCT2 function in Excel to ensure maintenance of data integrity. To provide an audit trail of data from subjects through to verification results, digital time-stamping was applied to analytical data, plans, and reports.

Discovery and verification study design

One hundred and thirty-six sPTD cases were randomly distributed between discovery (n = 86) and verification (n = 50), collected from 170/7 through 286/7 weeks GA at blood draw (GABD) (Supplementary Table 1). Subjects used in discovery and verification were completely independent of each other and independent from those used in validation. Matched controls were identified for sPTD cases in discovery and verification, as described in Materials and Methods.

Prevalence analyses

Following discovery, verification, and validation analyses, additional term controls, not used in prior studies, were selected from the PAPR database and processed in the laboratory using the multiple reaction monitoring (MRM) mass spectrometry (MS) assay applied in validation and described in Materials and Methods. Using the Sampling package in R statistical software (version 3.0.3),3 and 4 sets of 187 subjects, without source data monitoring, were randomly selected from the validated GABD interval and compared via univariate statistical analyses (chi-square test) against the gestational age at birth (GAB) data from the 2012 National Vital Statistics Report (NVSR).5 Sets of controls most closely approximating the distribution of deliveries in the 2012 NVSR based on the best P value (approaching 1.0 with minimum acceptable value of .950) were then selected for comparison against the body mass index (BMI) distribution in the PAPR study as a whole. Using univariate statistical analyses (chi-square test) against the BMI data from the PAPR study database, the sets of controls most closely approximating the distribution of BMI (approaching 1.0 with minimum acceptable value of .950) and the distribution of delivery timing in the NVSR were selected and compared to the GABD of the validated blood draw samples. The set that most closely approximated all 3 distributions was selected as the subject set for the prevalence study. Predictor score values for verification, validation, and prevalence within the validation GABD interval and BMI restriction totaled 150 subjects. This composite dataset was used to obtain the best estimates of confidence intervals about the positive predictive value (PPV) curve in Figure 3. Confidence intervals about the PPV curve were calculated with the normal approximation of the error for binomial proportions.6

Sample accessioning

Biospecimen accessioning procedures included: (1) immediate 2-dimensional barcode labeling by study personnel on site following specimen processing, (2) visual inspection that specimens were received frozen on dry ice and entry into a computerized database upon accession by Sera Prognostics, (3) verification of specimen temperature throughout the shipping process from temperature tracking monitors, (4) comparison of specimen IDs from barcodes and site shipment inventories, and (5) immediate transfer of specimens from shipping containers (with dry ice) into −80°C freezers.

Laboratory methods

A systems biology approach was employed to generate a highly multiplexed MRM MS assay by iterative application of literature curation, targeted and untargeted proteomic discovery, and small-scale MRM MS analyses of subject samples. Initial curation was done manually and independently by 3 individuals using search terms including, but not limited to: preterm birth, pre-term birth, preeclampsia, placenta, placental gene expression, labor, preterm labor, premature rupture of membranes, PPROM, myometrial gene expression, and intra-amniotic infection. In subsequent rounds, larger-scale literature searches were performed using publicly available data obtained from PubMed. A Perl program executed keyword searches through National Center for Biotechnology Information’s public application programming interface, then downloaded the result sets in XML format. The resultant XML files were parsed by another Perl script yielding a list of PubMed identifiers (IDs). These PubMed IDs were then cross-referenced to Entrez Gene IDs using a gene2pubmed file. The Entrez Gene IDs were then filtered against a list of extracellular proteins annotated in Uniprot. The mature MRM MS assay, measuring 147 proteins, was applied in discovery and verification studies. For all analyses, serum samples were processed in the laboratory as described in Materials and Methods. Aliquots of pooled serum controls (pHGS) were used to calculate the interbatch analytical coefficient of variation for insulin-like growth factor–binding protein 4 (IBP4) and sex hormone–binding globulin (SHBG).

Normal ranges

In an analytical validation study, the details of which will be published separately, acceptable performance of each analyte was demonstrated for a range of protein responses. All clinical validation samples had protein responses within this range. Analyses of 1163 patient samples were used to develop historical means and standard deviations (SD) for protein responses. Sample acceptability criteria were set at ≤2.5 SD from the historical mean.

General predictor development strategy

A strategy was developed to avoid over-fitting and to overcome the dilution of biomarker performance expected across broad GA ranges owing to the dynamic nature of protein expression during pregnancy. Ratios of up-regulated over down-regulated analyte intensities were employed in predictor development. Such “reversals” are similar to the top-scoring pair and 2-gene classifier strategies.7 and 8 This approach allowed amplification of the diagnostic signal and self-normalization, as both proteins in a “reversal” underwent the same preanalytical and analytical processing steps. As a strategy to normalize peptide intensity measures in complex proteomics workflows, reversals are also similar to a recently introduced approach termed endogenous protein normalization (EPN).9 and 10 The number of candidate analytes used for model building was reduced by analytic criteria. Analytic filters included cutoffs for analytical precision, intensity, evidence of interference, sample processing order dependence, and preanalytical stability. The total number of analytes in any one predictor was limited to a single reversal, thus avoiding complex mathematical models. Predictor scores were defined as the natural log of a single reversal value, in which the reversal itself was a response ratio (defined in Materials and Methods). Lastly, predictive performance was investigated in narrow overlapping 3-week intervals of gestation.

Receiver operating characteristic curves

AUROC values and associated P values were calculated for reversals as described in Materials and Methods. The distribution and mean value for predictor AUROC in the combined discovery and verification set was calculated using a bootstrap sampling performed iteratively by selecting random sets of samples with replacement.11 The total number of selected samples at each iteration corresponded to the total available in the starting pool.

Supplementary Results

Discovery, verification, and validation subject characteristics are summarized in Supplementary Table 1. Distribution of subjects and select clinical variables by site are summarized in Supplementary Table 2. The percentage of subjects with 1 or more prior sPTDs in discovery sPTD cases were higher than in verification or validation, and other characteristics were largely consistent across the studies.

Discovery and verification analyses

Forty-four proteins were either up- or down-regulated in overlapping 3-week GA intervals and passed analytic filters (Supplementary Figure 1, Supplementary Table 3). All possible reversals were formed from the ratio of up- over down-regulated proteins and predictive performance by AUROC was tested in samples in each of the overlapping 3-week GA intervals using an R script. Performance for a subset of reversals displaying representative patterns is shown in Supplementary Figure 2. Waves of performance were evident: IBP4/SHBG and APOH/SHBG reversals possessed better AUROC values in early windows, while ITIH4/BGH3 and PSG2/BGH3 peaked later in gestation (Supplementary Figure 2). Some reversals had a consistent but moderate performance across the entire GA range (PSG2/PRG2) (Supplementary Figure 2). The top-performing reversal overall, formed from the up-regulated protein IBP4 and the down-regulated protein SHBG (IBP4/SHBG), had an AUROC = 0.74 in the interval from 190/7 through 216/7 (Supplementary Figure 2). AUROC performance of the IBP4/SHBG predictor increased to 0.79 when subjects were stratified by prepregnancy BMI <35 kg/m2 (Supplementary Table 4). Because of its consistently strong performance early in gestation (ie, 170/7 through 226/7 weeks GA) (Supplementary Figure 2) and potentially desirable clinical utility, the IBP4/SHBG predictor was selected for verification analysis.

The blinded IBP4/SHBG AUROC performance on verification samples was 0.77 and 0.79 for all subjects and BMI-stratified subjects, respectively, in good agreement with performance obtained in discovery (Supplementary Table 4). Following blinded verification, discovery and verification samples were combined for a bootstrap performance determination. A mean AUROC of 0.76 was obtained from 2000 bootstrap iterations (Supplementary Figure 3).

BMI validation analyses

The performance of the IBP4/SHBG predictor was evaluated at several cutoffs of BMI in the validation samples (Supplementary Table 5). AUROC-measured performance modestly improved by elimination of either very high (eg, >37 kg/m2) or low BMI (eg, ≤22 kg/m2). Stratification by a combination of those 2 cutoffs gave an AUROC of 0.75 (Supplementary Table 5).

Supplementary Table 1 Maternal characteristics and pregnancy outcomes stratified by timing of delivery (sPTD and term)

Variables Discovery Verification Validation Discovery vs verification Discovery vs validation Verification vs validation
Case n (%) (n = 86) Control n (%) (n = 172) P value Case n (%) (n = 50) Control n (%) (n = 150) P value Case n (%) (n = 81) Control n (%) (n = 162) P value P value P value P value
Maternal characteristics
 Maternal age at enrollment, y .245 .977 .239 .644 .594 .427
 18–22 y 26 (30.2) 39 (22.7) 10 (20.0) 21 (21.0) 22 (27.2) 47 (29.0)
 23–27 y 25 (29.1) 58 (33.7) 14 (28.0) 26 (26.0) 17 (21.0) 41 (25.3)
 28–32 y 14 (16.3) 44 (25.6) 15 (30.0) 27 (27.0) 25 (30.9) 34 (21.0)
 33–37 y 14 (16.3) 23 (13.4) 8 (16.0) 20 (20.0) 9 (11.1) 30 (18.5)
 38 y or more 7 (8.1) 8 (4.6) 3 (6.0) 6 (6.0) 8 (9.9) 10 (6.2)
 Mean 28 28 29 29 28 28
 Median 26 27 28 29 28 27
 Interquartile range 22–32 23–31 24–32 23–34 21–32 22–32
 Body mass index, kg/m2 .528 .722 .802 .869 .501 .729
 Less than 18.5 4 (4.8) 8 (4.7) 5 (10.4) 6 (6.0) 1 (1.3) 2 (1.3)
 18.5–24.9 33 (39.3) 82 (48.5) 20 (41.7) 39 (39.0) 25 (31.3) 55 (34.6)
 25.0–29.9 20 (23.8) 33 (19.5) 8 (16.7) 25 (25.0) 26 (32.5) 46 (28.9)
 30.0–34.9 14 (16.7) 22 (13.0) 8 (16.7) 14 (14.0) 17 (21.3) 25 (15.7)
 35.0–39.9 8 (9.5) 9 (5.3) 3 (6.3) 10 (10.0) 6 (7.5) 17 (10.7)
 Greater than 40.0 5 (5.9) 15 (9.0) 4 (8.3) 6 (6.0) 5 (6.3) 14 (8.8)
 Mean 27.5 26.9 27.2 27.4 28.4 29.1
 Median 26.1 24.6 24.8 25.8 27.4 27.8
 Interquartile Range 22.4–32.2 21.8–30.4 22–31.6 22–32.5 23.6–32.0 23.4–32.4
 Education level .220 .204 .201 .153 .161 .115
 Graduate degree 5 (5.8) 15 (8.7) 2 (4.0) 16 (16.2) 6 (7.7) 14 (8.7)
 College diploma 10 (11.6) 37 (21.5) 14 (28.0) 20 (20.2) 10 (12.6) 22 (13.8)
 Some college 19 (22.1) 41 (23.8) 13 (26.0) 18 (18.2) 19 (24.0) 23 (14.4)
 High school diploma/equivalent 23 (26.7) 35 (20.4) 7 (14.0) 19 (19.2) 16 (20.2) 50 (31.3)
 Some high school 18 (20.9) 31 (18.0) 10 (20.0) 19 (19.2) 25 (31.6) 36 (22.5)
 9th grade or less 6 (7.0) 10 (5.8) 3 (6.0) 7 (7.1) 3 (3.8) 14 (8.7)
 Other 5 (5.8) 3 (1.7) 1 (2.0) 0 0 1 (0.6)
 Ethnicity .210 .343 .035 .116 .564 .277
 Hispanic or Latino 40 (46.5) 66 (38.4) 22 (44.0) 36 (36.0) 27 (33.3) 77 (47.5)
 Non-Hispanic or Latino 46 (53.5) 106 (61.6) 28 (56.0) 64 (64.0) 54 (66.7) 85 (52.5)
 Race .173 .373 .811 .390 .602 .615
 American Indian/Alaskan Native 1 (1.1) 0 0 0 0 2 (1.2)
 Asian 1 (1.1) 9 (5.2) 2 (4.0) 3 (3.0) 1 (1.2) 1 (0.6)
 Black or African-American 20 (23.3) 41 (23.8) 6 (12.0) 21 (21.0) 19 (23.5) 37 (22.8)
 Native Hawaiian or other Pacific Islander 0 0 0 0 0 2 (1.2)
 White 62 (72.1) 112 (65.1) 36 (72.0) 70 (70.0) 58 (71.6) 114 (70.4)
 Other 2 (2.3) 10 (5.8) 6 (12.0) 6 (6.0) 3 (3.7) 6 (3.7)
Obstetrical characteristics
 Primigravida 21 (24.4) 52 (30.2) .328 16 (32.0) 33 (33.0) .902 27 (33.3) 39 (24.1) .126 .400 .724 .272
 Multigravida 65 (75.6) 120 (69.8) 34 (68.0) 67 (67.0) 54 (66.7) 123 (75.9)
 Number of prior full-term deliveries .141 .673 .326 .208 .134 .221
 1 or more 45 (71.4) 98 (81.7) 25 (73.5) 55 (82.1) 40 (74.5) 102 (82.9)
 None 18 (28.6) 22 (18.3) 9 (26.5) 12 (17.9) 13 (24.5) 21 (17.1)
 Number of prior sPTDs .060 .188 .221 .014 .018 .056
 1 or more 17 (26.2) 14 (11.7) 9 (26.5) 9 (13.4) 9 (16.7) 11 (8.9)
 None 48 (73.8) 106 (88.3) 25 (73.5) 58 (86.6) 45 (83.3) 112 (91.1)
Lifestyle characteristics
 Smoking .329 .728 .719 .328 .365 .622
 Yes 12 (14.0) 17 (9.9) 7 (14.0) 12 (12.0) 15 (18.5) 27 (16.7)
 No 74 (86.0) 155 (90.1) 43 (86.0) 88 (88.8) 66 (81.5) 135 (83.3)
 Illicit drugs .030 .824 .628 .125 .491 .794
 Yes 6 (7.0) 2 (1.1) 4 (8.0) 7 (7.0) 6 (7.4) 15 (9.3)
 No 80 (93.0) 170 (98.8) 46 (92.0) 93 (93.0) 75 (92.6) 147 (90.7)
 Alcohol .147 .171 .628 .052 .494 .781
 Yes 10 (11.6) 11 (6.4) 4 (8.0) 3 (3.0) 6 (7.4) 15 (9.3)
 No 76 (86.1) 161 (93.6) 46 (92.0) 97 (97.0) 75 (92.6) 147 (90.7)
 Alcohol use .410 .379 .592 .206 .728 .853
 Yes (amount unknown) 2 (2.3) 4 (2.3) 1 (2.0) 1 (1.0) 0 2 (1.2)
 Social (occasional) 7 (8.1) 6 (3.5) 3 (6.0) 2 (2.0) 6 (7.4) 13 (8.0)
 Heavy (daily) 1 (1.2) 1 (0.6) 0 0 0 0
 No 76 (86.1) 161 (93.6) 46 (92.0) 97 (97.0) 75 (92.6) 147 (90.7)
Medical characteristics
 Bleeding during pregnancy after 12 wk .101 .784 .360 .193 .065 .543
 Yes 12 (14.0) 13 (7.6) 2 (4.0) 5 (5.0) 7 (8.6) 9 (5.6)
 No 74 (86.0) 159 (92.4) 48 (96.0) 95 (95.0) 74 (91.4) 153 (94.4)

Comparisons of clinical data between cases and controls were performed using chi-square test or Fisher exact test or Mann-Whitney test, as appropriate (SAS System 9.4).

Missing values are excluded in the frequency tables.

N, number of subjects; sPTD, spontaneous preterm delivery.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Supplementary Table 2 Clinical and demographic data by PAPR trial site (enrolled PAPR n = 5501)

Variables Site #01 n (%) (n = 385) Site #02 n (%) (n = 684) Site #03 n (%) (n= 851) Site #04 n (%) (n = 395) Site #05 n (%) (n = 634) Site #06 n (%) (n = 468) Site #07 n (%) (n = 389) Site #08 n (%) (n = 960) Site #09 n (%) (n = 268) Site #10 n (%) (n = 256) Site #11 n (%) (n = 198)
Maternal age at enrollment, y
 18–22 y 44 (11.4) 120 (17.5) 212 (24.9) 107 (27.1) 214 (33.7) 124 (26.5) 23 (5.9) 271 (28.2) 22 (8.2) 53 (20.7) 51 (25.8)
 23–27 y 147 (38.2) 152 (22.2) 261 (30.7) 107 (27.1) 189 (29.8) 150 (32.1) 69 (17.7) 272 (28.3) 66 (24.6) 65 (25.4) 64 (32.3)
 28–32 y 132 (34.2) 228 (33.3) 219 (25.7) 97 (24.6) 117 (18.5) 99 (21.2) 140 (36.0) 235 (24.5) 90 (33.6) 60 (23.4) 56 (28.3)
 33–37 y 50 (13.0) 145 (21.2) 111 (10.0) 57 (14.4) 81 (12.8) 71 (15.2) 105 (27.0) 127 (13.2) 64 (23.9) 56 (21.9) 24 (12.1)
 38 y or more 12 (3.1) 39 (5.7) 48 (5.6) 27 (6.8) 33 (5.2) 24 (5.1) 52 (13.4) 55 (5.7) 26 (9.7) 22 (8.6) 3 (1.5)
Ethnicity
 Hispanic or Latino 53 (13.8) 17 (2.8) 348 (40.9) 329 (83.3) 325 (51.3) 11 (2.4) 58 (14.9) 629 (65.5) 92 (34.3) 9 (3.5) 31 (15.6)
 Non-Hispanic or Latino 332 (86.2) 665 (97.2) 503 (59.1) 66 (16.7) 309 (48.7) 457 (97.6) 331 (85.1) 331 (34.5) 176 (65.7) 247 (96.5) 167 (84.3)
Race
 American Indian/Alaskan Native 5 (1.3) 3 (0.4) 3 (0.4) 4 (1.0) 4 (0.6) 0 5 (1.3) 3 (0.3) 1 (0.4) 4 (1.6) 1 (0.5)
 Asian 6 (1.6) 10 (1.5) 25 (2.9) 15 (3.8) 10 (1.6) 5 (1.1) 23 (5.9) 13 (1.4) 33 (12.3) 0 2 (1.0)
 Black or African-American 0 285 (41.7) 245 (28.8) 19 (4.8) 128 (20.2) 242 (51.7) 11 (2.8) 104 (10.8) 20 (7.5) 35 (13.7) 65 (32.8)
 Native Hawaiian or other Pacific Islander 6 (1.6) 2 (0.3) 0 0 1 (0.2) 0 3 (0.8) 1 (0.1) 0 1 (0.4) 0
 White 349 (90.6) 371 (54.2) 570 (66.9) 353 (89.4) 335 (52.8) 204 (43.6) 339 (87.2) 837 (87.2) 205 (76.5) 213 (83.2) 110 (55.6)
 Other 19 (4.9) 13 (1.9) 8 (0.9) 4 (1.0) 156 (24.6) 17 (3.6) 8 (2.0) 2 (0.2) 9 (3.4) 3 (1.2) 20 (10.1)
Number of prior sPTDs
 1 or more 39 (14.8) 117 (24.5) 71 (11.5) 54 (16.4) 98 (21.0) 120 (33.3) 36 (13.2) 103 (13.9) 35 (18.0) 42 (22.2) 23 (18.9)
 None 224 (85.2) 360 (75.5) 547 (88.5) 275 (83.6) 369 (79.0) 240 (66.7) 236 (86.8) 638 (86.1) 160 (82.0) 147 (77.8) 99 (81.1)

N, number of subjects; sPTD, spontaneous preterm delivery.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Supplementary Table 3 Forty-four proteins meeting analytical filters that were up- or down-regulated in sPTD vs term controls

Uniprot ID Short name Protein name
A2GL_HUMAN LRG1 Leucine-rich alpha-2-glycoprotein
AFAM_HUMAN AFM Afamin
ANGT_HUMAN AGT Angiotensinogen
APOC3_HUMAN APOC3 Apolipoprotein C-III
APOH_HUMAN APOH Beta-2-glycoprotein 1
B2MG_HUMAN B2M Beta-2-microglobulin
BGH3_HUMAN TGFBI Transforming growth factor-beta-induced protein ig-h3
CATD_HUMAN CTSD Cathepsin D
CBPN_HUMAN CPN1 Carboxypeptidase N catalytic chain
CD14_HUMAN CD14 Monocyte differentiation antigen CD14
CFAB_HUMAN CFB Complement factor B
CHL1_HUMAN CHL1 Neural cell adhesion molecule L1-like protein
CO5_HUMAN C5 Complement C5
CO6_HUMAN C6 Complement component C6
CO8A_HUMAN C8A Complement component C8 alpha chain
CRIS3_HUMAN CRISP3 Cysteine-rich secretory protein 3
ENPP2_HUMAN ENPP2 Ectonucleotide pyrophosphatase/phosphodiesterase family member 2
F13B_HUMAN F13B Coagulation factor XIII B chain
FBLN3_HUMAN EFEMP1 EGF-containing fibulin-like extracellular matrix protein 1
FETUA_HUMAN AHSG Alpha-2-HS-glycoprotein
HABP2_HUMAN HABP2 Hyaluronan-binding protein 2
HEMO_HUMAN HPX Hemopexin
HLAG_HUMAN HLA-Ga HLA class I histocompatibility antigen, alpha chain G
IBP2_HUMAN IGFBP2 Insulin-like growth factor-binding protein 2
IBP3_HUMAN IGFBP3 Insulin-like growth factor-binding protein 3
IBP4_HUMAN IGFBP4 Insulin-like growth factor-binding protein 4
INHBC_HUMAN INHBC Inhibin beta C chain
ITIH3_HUMAN ITIH3 Inter-alpha-trypsin inhibitor heavy chain H3
ITIH4_HUMAN ITIH4 Inter-alpha-trypsin inhibitor heavy chain H4 N-term
ITIH4_HUMAN ITIH4 Inter-alpha-trypsin inhibitor heavy chain H4 C-term
KNG1_HUMAN KNG1 Kininogen-1
LBP_HUMAN LBP Lipopolysaccharide-binding protein
LYAM3_HUMAN SELP P-selectin
PAPP1_HUMAN PAPPA Pappalysin-1
PEDF_HUMAN SERPINF1 Pigment epithelium–derived factor
PGRP2_HUMAN PGLYRP2 N-acetylmuramoyl-L-alanine amidase
PRG2_HUMAN PRG2 Bone marrow proteoglycan
PSG11_HUMAN PSG11 Pregnancy-specific beta-1-glycoprotein 11
PSG2_HUMAN PSG2 Pregnancy-specific beta-1-glycoprotein 2
PSG9_HUMAN PSG9 Pregnancy-specific beta-1-glycoprotein 9
SHBG_HUMAN SHBG Sex hormone–binding globulin
TENX_HUMAN TNXB Tenascin-X
TIE1_HUMAN TIE1 Tyrosine-protein kinase receptor Tie-1
VTNC_HUMAN VTN Vitronectin

a Peptide surrogate for HLA-G was not unique to this protein.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Supplementary Table 4 Performance of the ln(IBP4/SHBG) classifier by study and BMI restriction

Discovery Verification Validation
Sample # (17–28 wk)a 86, 172 50, 100 81, 162
Sample # (all BMI)a 22, 44b 9, 18b 18, 36c
AUC (all BMI) (95% CI) 0.74 (0.59–0.87) 0.77 (0.56–0.94)b 0.67 (0.52–0.81)
Sample # (BMI <35)a 17, 33b 6, 17b 15, 29c
AUC (BMI <35) (95% CI) 0.79 (0.65–0.92)b 0.79 (0.58–0.95)b 0.70 (0.53–0.86)c

a Number of cases, number of controls

b Gestational age at blood draw weeks 190/7 - 216/7

c Optimal gestational age at blood draw interval from fixed sequence validation (191/7 - 206/7).

95% CI calculated based on 5000 bootstrap iterations.

AUC, area under the curve; BMI, body mass index; CI, confidence interval.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

Supplementary Table 5 Performance of the ln(IBP4/SHBG) classifier with different BMI restrictions

BMI (kg/m2) AUROC (95% CI) Cases Controls
All BMI 0.67 (0.52–0.82) 18 36
BMI ≤45 0.67 (0.52–0.82) 18 35
BMI ≤40 0.68 (0.52–0.83) 17 34
BMI ≤37 0.71 (0.53–0.86) 16 31
BMI >18 0.67 (0.52–0.82) 18 35
BMI >20 0.65 (0.51–0.82) 17 32
BMI >22 0.69 (0.53–0.86) 14 27
22 < BMI ≤ 37 0.75 (0.56–0.91) 12 23

95% CI calculated based on 5000 bootstrap iterations.

Gestational age at blood draw 191/7 - 206/7 weeks.

AUROC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval.

Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

fx1

Supplementary Figure 1 Filtering of candidate proteins prior to predictor discoveryShown is the starting number of proteins in the discovery dataset. Candidate proteins were reduced by analytic criteria that included presence of SIS peptide, lack of targeting by MARS-14 depletion column, good detectability, precision, lack of processing order effects, good preanalytical stability, lack of effect of serum storage age, and evidence of regulation. Forty-four proteins that were either up- or down-regulated in overlapping 3-week GA intervals remained for predictor development. Interbatch coefficients of variability—calculated using pHGS specimens across multiple batches, processing days, and instrumentation—are reported for the predictor proteins IBP4 and SHBG using the discovery assay.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

fx2

Supplementary Figure 2 Reversal predictive performance across gestationShown is the ROC predictive performance (AUROC) of reversals formed from the ratio of up-regulated proteins over down-regulated proteins using samples in overlapping 3-week intervals across GABD. Predictor performance was both analyte and GABD dependent, with spikes in performance occurring in relatively narrow GABD ranges. Examples are given for specific reversals that demonstrated the phenotypic properties observed (eg, waves of performance that were high early in gestation, late in gestation, or of consistent but moderate level).APOH, beta-2-glycoprotein 1; ITIH4, inter-alpha-trypsin inhibitor heavy chain family, member 4; BGH3, transforming growth factor-beta-induced protein ig-h3; PSG2, pregnancy-specific beta-1-glycoprotein 2.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

fx3

Supplementary Figure 3 Histogram of IBP4/SHBG predictive performanceShown is the frequency of AUROC values obtained by application of a bootstrapping procedure to the combined discovery and verification datasets. The total number of samples selected with replacement in each of the 2000 bootstrap iterations was equivalent to the number of samples in the combined sample set. The mean AUROC value, shown in red, was 0.76, with 95% confidence intervals shown in blue.Saade et al. Validation of a preterm delivery predictor. Am J Obstet Gynecol 2016.

References

  • 1 J.P. Vogel, A.C.C. Lee, J.P. Souza. Maternal morbidity and preterm birth in 22 low- and middle-income countries: a secondary analysis of the WHO Global Survey dataset. BMC Pregnancy Childbirth. 2014;14(1):56 Crossref
  • 2 R.E. Behrman, A.S. Butler (Eds.) Preterm Birth: Causes, Consequences, and Prevention (National Academies Press, Washington, D.C, 2007) 1-765
  • 3 H. Blencowe, S. Cousens, M.Z. Oestergaard, et al. National, regional, and worldwide estimates of preterm birth rates in the year 2010 with time trends since 1990 for selected countries: a systematic analysis and implications. Lancet. 2012;379(9832):2162-2172 Crossref
  • 4 E.S. Reich. Pre-term births on the rise. Nature. 2012;485(7396):20
  • 5 J. McManemy, E. Cooke, E. Amon, T. Leet. Recurrence risk for preterm delivery. Am J Obstet Gynecol. 2007;196(6):576.e1 Crossref
  • 6 S.K. Laughon, P.S. Albert, K. Leishear, P. Mendola. The NICHD Consecutive Pregnancies Study: recurrent preterm delivery by subtype. Am J Obstet Gynecol. 2013;210(2):131.e1-131.e8
  • 7 P.J. Meis, M. Klebanoff, E. Thom, et al. Prevention of recurrent preterm delivery by 17 alpha-hydroxyprogesterone caproate. N Engl J Med. 2003;348(24):2379-2385 Crossref
  • 8 R.L. Goldenberg, J.F. Culhane, J.D. Iams, R. Romero. Epidemiology and causes of preterm birth. Lancet. 2008;371(9606):75-84 Crossref
  • 9 J.D. Iams, R.L. Goldenberg, P.J. Meis, et al. The length of the cervix and the risk of spontaneous premature delivery. National Institute of Child Health and Human Development Maternal Fetal Medicine Unit Network. N Engl J Med. 1996;334(9):567-572
  • 10 M.G. Gravett, A. Thomas, K.A. Schneider, et al. Proteomic analysis of cervical-vaginal fluid: identification of novel biomarkers for detection of intra-amniotic infection. J Proteome Res. 2007;6(1):89-96 Crossref
  • 11 R. Romero, J. Espinoza, F. Gotsch, et al. The use of high-dimensional biology (genomics, transcriptomics, proteomics, and metabolomics) to understand the preterm parturition syndrome. BJOG. 2006;113(Suppl3):118-135 Crossref
  • 12 R.L. Goldenberg, J.D. Iams, B.M. Mercer, et al. The preterm prediction study: the value of new vs standard risk factors in predicting early and all spontaneous preterm births. NICHD MFMU Network. Am J Public Health. 1998;88(2):233-238 Crossref
  • 13 M.G. Gravett, M.J. Novy, R.G. Rosenfeld, et al. Diagnosis of intra-amniotic infection by proteomic profiling and identification of novel biomarkers. JAMA. 2004;292(4):462-469 Crossref
  • 14 G.A. Dekker, S.Y. Lee, R.A. North, L.M. McCowan, N.A.B. Simpson, C.T. Roberts. Risk factors for preterm birth in an international prospective cohort of nulliparous women. PLoS One. 2012;7(7):e39154 Crossref
  • 15 J.A. Bastek, M.A. Elovitz. The role and challenges of biomarkers in spontaneous preterm birth and preeclampsia. Fertil Steril. 2013;99(4):1117-1123 Crossref
  • 16 B.M. Mercer, R.L. Goldenberg, A. Das, et al. The preterm prediction study: a clinical risk assessment system. Am J Obstet Gynecol. 1996;174(6):1885-1893
  • 17 J.M. Schaaf, A.C.J. Ravelli, B.W.J. Mol, A. Abu-Hanna. Development of a prognostic model for predicting spontaneous singleton preterm birth. Eur J Obstet Gynecol Reprod Biol. 2012;164(2):150-155 Crossref
  • 18 Committee on the Review of Omics-Based Tests for Predicting Patient Outcomes in Clinical Trials. Board on Health Care Services, Board on Health Sciences Policy, Institute of Medicine. C.M. Micheel, S.J. Nass, G.S. Omenn (Eds.) Evolution of Translation Omics: Lessons Learned and the Path Forward (The National Academies Press, Washington, DC, 2012) 1-355
  • 19 Microsoft. Microsoft Excel. 2013.
  • 20 R Development Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, Vienna, Austria, 2011) http://www.R-project.org/ Available at:Accessed October 30, 2015
  • 21 Tillé Y, Matei A. Sampling: Survey Sampling. R package version 2.7. Available at: https://CRAN.R-project.org/package=sampling. Accessed October 30, 2015.
  • 22 J.A. Martin, B.E. Hamilton, M.J. Osterman, S.C. Curtin, T.J. Mathews. Births: final data for 2012. Natl Vital Stat Rep. 2014;63(09):1-86 Crossref
  • 23 T. Sing, O. Sander, N. Beerenwinkel, T. Lengauer. ROCR: visualizing classifier performance in R. Bioinformatics. 2005;21(20):7881
  • 24 A. Dmitrienko, A.C. Tamhane, F. Bretz (Eds.) Multiple Testing Problems in Pharmaceutical Statistics (CRC Press, Boca Raton, Florida, 2009) 1-320
  • 25 A. Dmitrienko, R.B. D'Agostino, M.F. Huque. Key multiplicity issues in clinical drug development. Stat Med. 2012;32(7):1079-1111
  • 26 D. Bamber. The area above the ordinal dominance graph and the area below the receiver operating characteristic graph. J Math Psychol. 1975;12(4):387-415 Crossref
  • 27 S.J. Mason, N.E. Graham. Areas beneath the relative operating characteristics (ROC) and relative operating levels (ROL) curves: statistical significance and interpretation. QJR Meteorol Soc. 2002;128(584):2145-2166 Crossref
  • 28 J.A. Martin, B.E. Hamilton, M.J. Osterman, S.C. Curtin, T.J. Matthews. Births: final data for 2013. Natl Vital Stat Rep. 2015;64(1):1-65 Crossref
  • 29 J.R. Petrini, W.M. Callaghan, M. Klebanoff, et al. Estimated effect of 17 alpha-hydroxyprogesterone caproate on preterm birth in the United States. Obstet Gynecol. 2005;105(2):267-272 Crossref
  • 30 S.S. Hassan, R. Romero, D. Vidyadhari, et al. Vaginal progesterone reduces the rate of preterm birth in women with a sonographic short cervix: a multicenter, randomized, double-blind, placebo-controlled trial. Ultrasound Obstet Gynecol. 2011;38(1):18-31 Crossref
  • 31 L. Pereira, A.P. Reddy, A.L. Alexander, et al. Insights into the multifactorial nature of preterm birth: proteomic profiling of the maternal serum glycoproteome and maternal serum peptidome among women in preterm labor. Am J Obstet Gynecol. 2010;202(6):555.e1-555.e10 Crossref
  • 32 L. Pereira, A.P. Reddy, T. Jacob, et al. Identification of novel protein biomarkers of preterm birth in human cervical-vaginal fluid. J Proteome Res. 2007;6(4):1269-1276 Crossref
  • 33 S. Dasari, L. Pereira, A.P. Reddy, et al. Comprehensive proteomic analysis of human cervical-vaginal fluid. J Proteome Res. 2007;6(4):1258-1268 Crossref
  • 34 M.S. Esplin, K. Merrell, R. Goldenberg, et al. Proteomic identification of serum peptides predicting subsequent spontaneous preterm birth. Am J Obstet Gynecol. 2010;204(5):391.e1-391.e8
  • 35 C.P. Weiner, C.W. Mason, Y. Dong, I.A. Buhimschi, P.W. Swaan, C.S. Buhimschi. Human effector/initiator gene sets that regulate myometrial contractility during term and preterm labor. Am J Obstet Gynecol. 2010;202(5):474.e1-474.e20 Crossref
  • 36 S.S.C. Chim, W.S. Lee, Y.H. Ting, O.K. Chan, S.W.Y. Lee, T.Y. Leung. Systematic identification of spontaneous preterm birth-associated RNA transcripts in maternal plasma. PLoS One. 2012;7(4):e34328 Crossref
  • 37 D.A. Enquobahrie, M.A. Williams, C. Qiu, et al. Early pregnancy peripheral blood gene expression and risk of preterm delivery: a nested case control study. BMC Pregnancy Childbirth. 2009;9(1):56 Crossref
  • 38 K.Y. Bezold, M.K. Karjalainen, M. Hallman, K. Teramo, L.J. Muglia. The genomics of preterm birth: from animal models to human studies. Genome Med. 2013;5(4):34 Crossref
  • 39 R. Romero, D.R.V. Edwards, J.P. Kusanovic, et al. Identification of fetal and maternal single nucleotide polymorphisms in candidate genes that predispose to spontaneous preterm labor with intact membranes. Am J Obstet Gynecol. 2010;202(5):431.e1-431.e34 Crossref
  • 40 K.A. Swaggart, M. Pavlicev, L.J. Muglia. Genomics of preterm birth. Cold Spring Harb Perspect Med. 2015;5(2):a023127 Crossref
  • 41 R. Haataja, M.K. Karjalainen, A. Luukkonen, et al. Mapping a new spontaneous preterm birth susceptibility gene, IGF1R, using linkage, haplotype sharing, and association analysis. PLoS Genet. 2011;7(2):e1001293 Crossref
  • 42 J.J. McElroy, C.E. Gutman, C.M. Shaffer, et al. Maternal coding variants in complement receptor 1 and spontaneous idiopathic preterm birth. Hum Genet. 2013;132(8):935-942 Crossref
  • 43 R. Menon, J. Jones, P.R. Gunst, et al. Amniotic fluid metabolomic analysis in spontaneous preterm birth. Reprod Sci. 2014;21(6):791-803 Crossref
  • 44 K. Forbes, M. Westwood, P.N. Baker, J.D. Aplin. Insulin-like growth factor I and II regulate the life cycle of trophoblast in the developing human placenta. Am J Physiol Cell Physiol. 2008;294(6):C1313-C1322 Crossref
  • 45 E.J. Crosley, C.E. Dunk, A.G. Beristain, J.K. Christians. IGFBP-4 and -5 are expressed in first-trimester villi and differentially regulate the migration of HTR-8/SVneo cells. Reprod Biol Endocrinol. 2014;12(1):123 Crossref
  • 46 Q. Qiu, M. Bell, X. Lu, et al. Significance of IGFBP-4 in the development of fetal growth restriction. J Clin Endocrinol Metab. 2012;97(8):E1429-E1439 Crossref
  • 47 G.L. Hammond. Diverse roles for sex hormone-binding globulin in reproduction. Biol Reprod. 2011;85(3):431-441 Crossref
  • 48 D.C. Anderson. Sex-hormone-binding globulin. Clin Endocrinol (Oxf). 1974;3(1):69-96 Crossref
  • 49 F. Larrea, L. Díaz, C. Cariño, et al. Evidence that human placenta is a site of sex hormone-binding globulin gene expression. J Steroid Biochem Mol Biol. 1993;46(4):497-505 Crossref
  • 50 R. Simó, C. Sáez-López, A. Barbosa-Desongles, C. Hernández, D.M. Selva. Novel insights in SHBG regulation and clinical implications. Trends Endocrinol Metab. 2015;26(7):376-383
  • 51 C.R. Mendelson. Minireview: fetal-maternal hormonal signaling in pregnancy and labor. Mol Endocrinol. 2009;23(7):947-954 Crossref
  • 52 N. Gomez-Lopez, D. StLouis, M.A. Lehr, E.N. Sanchez-Rodriguez, M. Arenas-Hernandez. Immune cells in term and preterm labor. Cell Mol Immunol. 2014;11(6):571-581 Crossref
  • 53 T.M. Lindström, P.R. Bennett. The role of nuclear factor kappa B in human labour. Reproduction. 2005;130(5):569-581
  • 54 L.D. Brown, T.T. Cai, A. DasGupta. Interval estimation for a binomial proportion. Statistical Science. 2001;16(2):101-133

Supplementary References

  • 1 Committee on the Review of Omics-Based Tests for Predicting Patient Outcomes in Clinical Trials. Board on Health Care Services, Board on Health Sciences Policy, Institute of Medicine. C.M. Micheel, S.J. Nass, G.S. Omenn (Eds.) Evolution of Translation Omics: Lessons Learned and the Path Forward (The National Academies Press, Washington, DC, 2012) 1-355
  • 2 Microsoft. Microsoft Excel. 2013.
  • 3 R Development Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, Vienna, Austria, 2011) http://www.R-project.org/ Avaiable at:Accessed October 30, 2015
  • 4 A. Matei, Y. Tillé. The R “sampling” package. European Conference on Quality in Survey Statistics. 2006;
  • 5 J.A. Martin, B.E. Hamilton, M.J. Osterman, S.C. Curtin, T.J. Mathews. Births: final data for 2012. Natl Vital Stat Rep. 2014;63(09):1-86 Crossref
  • 6 L.D. Brown, T.T. Cai, A. DasGupta. Interval estimation for a binomial proportion. Stat Sci. 2001;16(2):101-133
  • 7 D. Geman. d'Avignon C, Naiman DQ, Winslow RL. Classifying gene expression profiles from pairwise mRNA comparisons. Stat Appl Genet Mol Biol. 2004;3(1) Article 19
  • 8 N.D. Price, J. Trent, A.K. El-Naggar, et al. Highly accurate two-gene classifier for differentiating gastrointestinal stromal tumors and leiomyosarcomas. Proc Natl Acad Sci U S A. 2007;104(9):3414-3419 Crossref
  • 9 X.-J. Li, L.W. Lee, C. Hayward, et al. An integrated quantification method to increase the precision, robustness, and resolution of protein measurement in human plasma samples. Clin Proteomics. 2015;12(1):3
  • 10 X.-J. Li, C. Hayward, P.-Y. Fong, et al. A blood-based proteomic classifier for the molecular characterization of pulmonary nodules. Sci Transl Med. 2013;5(207):207ra142 Crossref
  • 11 B. Efron, R.J. Tibshirani. An Introduction to the Bootstrap. (Chapman and Hall/CRC Press, Boca Raton, Florida, 1994)

Footnotes

a Department of Obstetrics & Gynecology, The University of Texas Medical Branch, Galveston, Texas

b Department of Obstetrics and Gynecology, Division of Maternal-Fetal Medicine, University of North Carolina, Chapel Hill, North Carolina

c Department of Obstetrics and Gynecology, Medical University of South Carolina, Charleston, South Carolina

d Maternal Fetal Medicine, Baystate Medical Center, Springfield, Massachusetts

e Department of Obstetrics & Gynecology, The Ohio State University, Columbus, Ohio

f Department of Obstetrics and Gynecology, Maricopa Integrated Health System/District Medical Group, Phoenix, Arizona

g Division of Maternal-Fetal Medicine, Oregon Health & Science University, Portland, Oregon

h Division of Maternal Fetal Medicine, Intermountain Healthcare, Murray, Utah

i San Diego Perinatal Center, Maternal Fetal Medicine Division, Rady Children’s Specialists of San Diego, San Diego, California

j Regional Obstetrical Consultants, Chattanooga, Tennessee

k Department of Obstetrics & Gynecology, Christiana Care Health System, Newark, Delaware

l Sera Prognostics, Inc, Salt Lake City, Utah

m Dana Farber Cancer Institute, Department of Biostatistics and Computational Biology, Boston, Massachusetts

n Integrated Diagnostics, Inc, Seattle, Washington

Corresponding author: George R. Saade, MD.

Dr Markenson is currently affiliated with Maternal Fetal Medicine, Boston Medical Center, Boston, Massachusetts. Dr Coonrod is currently affiliated with the Department of Obstetrics and Gynecology, University of Arizona College of Medicine, Tucson, Arizona. Dr Esplin is currently affiliated with the University of Utah, Maternal Fetal Medicine, Salt Lake City, Utah. Dr Lam is currently affiliated with the Department of Obstetrics and Gynecology, UT College of Medicine Chattanooga, Chattanooga, Tennessee.

The following authors of this manuscript disclose that their institutions received money from Sera Prognostics to cover the costs of the study: G.R.S., K.A.B., S.A.S., G.R.M., J.D.I., D.V.C., L.M.P., M.S.E., L.M.C., G.K.L., M.K.H. The remaining authors (R.D.S., T.P., J.S.F., A.C.F., A.J.L., S.R.R., C.T.H., M.T.D., C.L.B., M.S.E., I.E.I., T.C.F., A.D.P., G.C.C., J.J.B., D.E.H., E.M., P.E.K.) are either employees or consultants for the study sponsor, Sera Prognostics.

The study described in this manuscript was supported in full by Sera Prognostics, Inc. The sponsor collaborated with site principal investigators and outside consultants regarding study design, data analysis and interpretation, manuscript writing, and the decision to submit the paper for publication. The sponsor was not involved with specimen collection, specimen processing, or data collection at the clinical sites. The sponsor conducted proteomic analysis of the specimens.

Cite this article as: Saade GR, Boggess KA, Sullivan SA, et al. Development and validation of a spontaneous preterm delivery predictor in asymptomatic women. Am J Obstet Gynecol 2016;214:633.e1-24.