Cluster analysis of spontaneous preterm birth phenotypes identifies potential associations among preterm birth mechanisms

Abstract

Objective

We sought to employ an innovative tool based on common biological pathways to identify specific phenotypes among women with spontaneous preterm birth (SPTB), in order to enhance investigators' ability to identify to highlight common mechanisms and underlying genetic factors responsible for SPTB.

Study Design

A secondary analysis of a prospective case-control multicenter study of SPTB. All cases delivered a preterm singleton at SPTB ≤34.0 weeks gestation. Each woman was assessed for the presence of underlying SPTB etiologies. A hierarchical cluster analysis was used to identify groups of women with homogeneous phenotypic profiles. One of the phenotypic clusters was selected for candidate gene association analysis using VEGAS software.

Results

1028 women with SPTB were assigned phenotypes. Hierarchical clustering of the phenotypes revealed five major clusters. Cluster 1 (N=445) was characterized by maternal stress, cluster 2 (N=294) by premature membrane rupture, cluster 3 (N=120) by familial factors, and cluster 4 (N=63) by maternal comorbidities. Cluster 5 (N=106) was multifactorial, characterized by infection (INF), decidual hemorrhage (DH) and placental dysfunction (PD). These three phenotypes were highly correlated by Chi-square analysis [PD and DH (p<2.2e-6); PD and INF (p=6.2e-10); INF and DH (p=0.0036)]. Gene-based testing identified the INS (insulin) gene as significantly associated with cluster 3 of SPTB.

Conclusion

We identified 5 major clusters of SPTB based on a phenotype tool and hierarchal clustering. There was significant correlation between several of the phenotypes. The INS gene was associated with familial factors underlying SPTB.

Introduction

Spontaneous preterm birth (SPTB) remains the leading cause of morbidity and mortality¹ in non-anomalous newborns, yet our understanding of the causes of SPTB is limited. This is, in part, because SPTB is a multifactorial condition with multiple etiologies and likely results from specific interactions between the environment and genetic factors.^2-6 There is support of a genetic component to SPTB suggested by the presence of racial disparities that persist despite controlling for multiple risk factors.⁷ In addition, there is a strong risk for recurrence of SPTB in women with a personal history of SPTB in a previous pregnancy.^8,9 In addition, a clear familial predisposition has been demonstrated.¹⁰ Finally, twin studies support the role of genetic risk factors in preterm birth by estimating the heritability at 20 to 40 percent.¹¹

Efforts to identify the genetic causes of SPTB have produced overall disappointing results. A recent large genome wide association (GWA) study of SPTB identified specific single nucleotide polymorphisms (SNPs) that were associated with SPTB but these could not be subsequently validated.¹² One attempt to summarize the genetic contribution to SPTB concluded that no robustly validated genetic variants contributing to this complex disease process have been identified.¹³ This lack of success is likely due, at least in part, to inadequate phenotyping of SPTB cases, the heterogeneity of the disease process, differences amongst patient populations, or a combination of these factors.

The Genomic and Proteomic Network for Preterm Birth Research (“GPN-PBR”, abbreviated GPN) was established by the Eunice Kennedy Shriver National Institute for Child Health and Human Development to study the genetic and environmental etiologies, and with a goal of deciphering mechanisms underlying SPTB. Accurate and precise phenotypes were needed to accomplish this goal. We have previously created a unique phenotyping tool using clinical features present at the time of delivery to define nine phenotypes suggestive of underlying etiologies of SPTB. We applied the phenotype tool to more than 1000 women with SPTB, were able to classify over 95% of women into one or more phenotype categories, and demonstrated that most cases of SPTB have evidence of two or more phenotypes present and that phenotypes vary by gestational age at delivery and by race.¹⁴ Assigning a phenotype that suggests similar underlying etiology or etiologies for SPTB among a group of women will likely result in an enhanced ability to identify genes or pathways associated with that phenotype.

We hypothesized that associations exist between SPTB phenotypes that highlight common mechanisms responsible for SPTB will enhance our ability to identify the underlying genetic factors responsible for this complication. We further hypothesized that cluster analysis using sub-categories within phenotypes might identify subsets of women with a similar genetic risk for SPTB. We sought to test this by evaluating candidate genes that might be associated with SPTB among one of the subsets identified.

Materials and Methods

This is a secondary analysis of a multicenter, prospectively cohort women enrolled in the GPN case-control study.

Patient Recruitment

Women with SPTB and matched uncomplicated term controls were prospectively recruited from November 2007 through January 2011 across eight clinical sites including the University of Utah / Intermountain Healthcare, University of Texas Medical Branch – Galveston, University of Alabama at Birmingham, Columbia University, Northwestern University, University of Texas – Houston, University of North Carolina – Chapel Hill, and Brown University. This study was approved by the Institutional Review Board at each center, and a written informed consent was obtained from all participants.

Women were included in the study if they experienced a preterm birth of a singleton pregnancy between 20 0/7 and 33 6/7 weeks gestation following spontaneous labor. The inclusion criteria for the study have been published previously.¹²

Women were excluded from the study if they were diagnosed with a stillbirth prior to presentation to labor and delivery or if they needed an indicated delivery for maternal or fetal complications. Women who experienced an intrapartum stillbirth or who had spontaneous labor in addition to maternal or fetal complications were not excluded.

A control group was also collected consisting of women who experienced a singleton live birth after spontaneous labor at 39 weeks or greater. Controls were excluded if they had a history of a prior pregnancy complicated by SPTB. Controls were used only for the analysis of candidate genes.

Data Collection

Clinical and demographic data were collected for cases and controls by trained research nurses using in-person interviews prior to hospital discharge whenever possible. All interviews and abstraction of medical records were performed within 14 days of delivery. Data collected included demographics, medical, social, family, and obstetric history, obstetric course and complications during the current pregnancy. Patients also completed validated questionnaires to assess factors such as anxiety (Beck anxiety index), depression (Beck depression inventory), perceived stress (Perceived stress scale), and attitude of the subject and partner with respect to pregnancy.

Cluster Analysis

A phenotyping tool was designed by the authors (MSE, TAM, MWV) that grouped maternal social, demographic, family history, and obstetric factors into SPTB categories.¹⁴ (See Table 1) Category clinical factors were classified into levels of evidence as providing “strong”, “moderate”, and “possible” evidence of the phenotype. Cluster analysis was used to classify 1028 unique SPTB cases. The data included binary indicator variables for several phenotypes relating to SPTB including Infection/inflammation, maternal stress (Hypothalmus-Pituitary Axis (HPA) activation), decidual hemorrhage, uterine distension, cervical insufficiency, preterm premature rupture of membranes (PPROM), placental dysfunction, maternal comorbidities, and familial phenotypes. There were two or three levels of evidence for each of the phenotypes. Identification of one level of evidence of a specific phenotype was not mutually exclusive for the other levels of evidence for the same phenotype. For example, one subject might have strong, moderate and possible evidence for one or more phenotypes. It is possible that the true presence of a phenotype may be more likely in women who had more than one indicator of the phenotype. Thus, this information was used to calculate a “weighted” score for each factor. Three points were given for “strong” evidence, 2 points for “moderate” evidence, and 1 point for “possible” evidence for each phenotype. Thus a subject with evidence from each of the categories “strong”, “moderate” and “possible” for a particular phenotype would receive six points for that phenotype. The maximum score any individual could receive for each phenotype was therefore 6 points. There was no limit to the number of phenotypes or levels of evidence that each woman could be assigned, provided she met criteria.

Table 1.

Phenotyping tool used for analysis. Characteristics are labeled as “Strong”, “Moderate” and “Possible” evidence of each phenotype. The tool was applied to all women with SPTB and each subject could thus have Strong, Moderate and/or Possible evidence of more than one phenotype.

Phenotype	Strong Evidence	Moderate Evidence	Possible Evidence
Infection / Inflammationa	- Histologic chorioamnionitis or funisitis - Positive placental culture or presence of placental viral inclusions	- Clinical chorioamnionitis requiring intrapartum antibiotic treatment - Placental pathology positive for deciduitis villitis, microabscess, arteritis, and/or phlebitis	- Clinical endometritis requiring postpartum antibiotic treatment - Major antenatal maternal systemic infection (pneumonia, pyelonephritis, pancreatitis, hepatitis) - Symptomatic urinary tract infection - Sexually transmitted disease diagnosed at any time during pregnancy (chlamydia, gonorrhea, trichomoniasis, HIV)
Decidual Hemorrhagea	- Hemosiderin deposits or tightly adherent clot on placental pathology - At least 25% hemorrhage on fetal or maternal interface on placental pathology	- Placental pathology demonstrating 1-25% or unspecified percentage of hemorrhage on fetal or maternal interface - Active vaginal bleeding plus at least one ol the following - non-reassuring fetal heart tones, uterine tenderness, or uterine tachysystole - Clinical diagnosis of abruption requiring delivery	- Trauma to abdomen or motor vehicle accident during pregnancy - Vaginal bleeding during pregnancy, not otherwise specified - Placenta previa
Maternal Stress	- Moderate to severe depression/anxiety requiring medication treatment during pregnancy	- Beck Depression Index score indicates severe depression - Perceived stress score = ‘very high’ or life stressors questionnaire indicated ‘severe distress’	- Mild to moderate depression/anxiety not requiring medication treatment - Illicit drug use or current binge alcohol use during pregnancy - High risk socioeconomic risk factor: income less than poverty level, less than a high school degree
Cervical Insufficiency	- Cervical dilation ≥2 cm prior to 28 weeks gestation in the absence of labor - Cervical length <0.5 cm prior to 28 weeks in the absence of labor At least one pregnancy loss prior to 24 weeks gestation due to painless cervical dilation	- Cervical length <1.50 cm prior to 28 weeks gestation in the absence of labor - Cervical length 1.50-2.5cm prior to 28 weeks gestation AND hourglassing membranes/marked funneling	- Cervical length 1.50-2.50 cm prior to 28 weeks gestation in the absence of labor - History of cervical conization procedure or loop electro-excision procedure
Uterine Distensiona	n/a	- Polyhydramnios (4-quadrant AFI >25cm or single deepest pocket >8cm - Birthweight >90% for gestational age	- Sonographically confirmed presence of uterine fibroids - Placental weight >90% for gestational age
Placental Dysfunctiona	- Birthweight <3% for gestational age and gender - Placental weight <3% for gestational age - At least 25% placental infarction on pathology - Reverse end diastolic flow on cord Doppler prior to delivery - Pre-eclampsia with severe features or eclampsia	- Birthweight <10% for gestational age and gender - Placental weight <10% for gestational age - Absent end diastolic flow on cord Doppler prior to delivery - Any placental infarction with no percentage listed or <25% on placental pathology - Four quadrant amniotic fluid index <5cm or single deepest pocket <2cm on ultrasound - Pre-eclampsia without severe features	- Placental calcifications on pathology - Umbilical artery cord Doppler S/D ratio > 4cm/sec but no evidence of absent- or reversed- end diastolic flow - Meconium staining on placental pathology - Velamentous cord insertion on placental pathology
Preterm premature rupture of membranes	- Preterm, premature rupture of membranes diagnosed with sterile speculum examination, dye test, or amnisure at least 48 hours prior to the onset of labor	- Preterm, premature rupture of membranes diagnosed with sterile speculum examination, dye test, or amnisure 12-48 hours prior to the onset of labor	- History of PPROM and delivery less than 37 weeks in a prior pregnancy
Maternal Comorbidities	- Class B or higher diabetes mellitus - Chronic hypertension - Systemic lupus erythematosus - Antiphospholipid antibody syndrome - Chronic renal failure or insufficiency	- Gestational diabetes in the current gestation - Other medical condition affecting a major organ system, not otherwise specified -i.e. pulmonary disease, renal disease, autoimmune disease, history of seizures	n/a
Familial	- At least one first degree relative with history o spontaneous preterm birth	- At least one first degree relative with history of medically indicated preterm birth - At least one second degree relative with history of spontaneous preterm birth	- At least one second degree relative with history of medically indicated preterm birth

Cluster analysis incorporated demographic variables (including maternal age, race, Hispanic ethnicity, educational attainment, marital status and nulliparity), binary indicators for each level of phenotypic evidence and the weighted score for each phenotype category. Using these variables as input, a sample dissimilarity matrix was generated using the “Daisy” method in the R “Cluster” package. Chi-square analysis was performed to evaluate potential correlation among specific phenotypes.^15,16 Figure 1. Illustrates the clustering of each individual included in the analysis.

Figure 1.

Hierarchical clustering of 1028 women with SPTB. The samples were divided into five main clusters for further analysis, as indicated by colors in the figure.

Candidate Gene Analysis

Once cluster analysis was complete, one cluster (cluster 3) was selected to use for gene-based analysis. We chose this cluster because it contained women with a strong familial phenotype and we thought it likely that they might have a genetic contribution to their SPTB. The women within the sample cluster were compared to 717 term controls.

All cluster cases and term controls had biologic samples collected at the time of their delivery and DNA was subsequently extracted for all study subjects. Genotypes for 905,682 SNPs were generated with the Affymetrix SNP 6.0 genotyping array as previously described.¹² For the present study, genotype data were downloaded from dbGaP in binary PLINK format. The files contained genotypes for 1419 individual mothers including 702 with at least one SPTB and 717 women with no history of PTB. Quality assurance testing was done to identify an appropriate set of samples and SNPs for use in association testing. The samples were screened for sex discrepancies, sample duplications, and high Mendelian error rates. Principal components analysis (PCA) was performed to ascertain population stratification within the data and to confirm the reported ancestry of individuals in the study. Identity-by-descent (IBD) estimates were calculated to assess relationships between all pairs of samples. Samples were removed from the analysis if the mean pair-wise IBD value compared to all other samples was greater than 0.04. Samples with autosomal SNP call rates less than 0.95 were also excluded from further analysis. SNPs with call rates less than 0.95, minor allele frequency less than 0.005, or significant departure from Hardy-Weinberg equilibrium (p<5 × 10^-8 in non-Hispanic Caucasian controls) were removed from analysis. A total of 841,350 SNPs on chromosomes 1-22 and the X chromosome passed all criteria used for association testing.

Gene-based testing

Genotype association tests were performed to compare SPTB cases in cluster 3 to the non-SPTB controls. All tests were performed with Golden Helix SNP and Variation Suite software version 8.1. Significance was tested with logistic regression, assuming an additive genetic model. Results were adjusted for three principal components to account for population structure and ethnic stratification within the data. The output of the SNP tests was processed with the VEGAS program to generate gene-level association test results.¹⁷ The VEGAS program compiles the significance of all SNPs in or near each gene to determine the significance of the entire gene region using a simulation procedure. The test for each gene includes all SNPs within 50kb of the gene, thereby capturing most cis regulatory regions and other important features in the region of the gene. VEGAS combines the significance of individual SNPs using a linkage disequilibrium (LD) model to determine the expected correlation patterns within the gene. Several independent SNP associations within a gene may thus be combined to assess the overall significance of the gene region.

966 genes from previously identified inflammatory pathways were selected for evaluation in this candidate gene analysis.¹⁸ We chose to use genes from inflammatory pathways because inflammation is a common underlying mechanism for multiple etiologies of SPTB. Based on the number of tests required to evaluate 966 genes, a p-value of about 5e-5 was required to declare significance in the analysis.

Results

We applied the phenotyping tool to 1,028 women with SPTB. Hierarchical clustering of the dissimilarity matrix using R revealed five major data clusters. Clusters can be visualized in Figure 1. Cluster 1 (n=445) is characterized by “maternal stress” (HPA activation), cluster 2 (n=294) by premature membrane rupture, cluster 3 (n=120) by familial factors, and cluster 4 (n=63) by maternal comorbidities. Cluster 5 (n=106) is multifactorial, characterized by infection, decidual hemorrhage and placental dysfunction. Significant co-occurrence was observed between these three phenotypic categories. Chi-square analysis shows correlation between placental dysfunction and decidual hemorrhage (p<2.2e-6), placental dysfunction and inflammation/infection (p=6.2e-10), and between inflammation/infection and decidual hemorrhage (p=0.0036).

We chose to use cluster 3 for additional gene based testing since this cluster was characterized primarily by women with a family history of SPTB. The seventy-eight women with genotyping data available from dbGaP from cluster 3 were compared to 717 term controls. Demographic information for the cases and controls is found in Table 2. There are notable differences in maternal race, education level and marital status likely due to the fact that these variables were part of the hierarchical clustering. Gene test results were adjusted for race using PCA.

Table 2.

Comparison of demographic information between cases from Cluster 3 and control subjects. There is a significant difference in races between the two groups. All gene association tests are adjusted for principal components in order to account for racial differences.

	Cases in Cluster 3	Controls	p-value
N	78	717
Age (mean, SD)	25.6, 5.5	25.5, 5.7	0.897
Caucasians	66 (84.6%)	486 (67.9%)	0.003
African Americans	8 (10.3%)	169 (23.6%)	0.011
Hispanic (any race)	10 (12.8%)	133 (18.6%)	0.273
Education=4 (13-16 years)	41 (52.6%)	313 (43.7%)	0.166
Education=3 (9-12 years)	34 (43.6%)	347 (48.5%)	0.492
Married/living with partner	47 (60.3%)	400 (55.9%)	0.525
Never married, living with partner	14 (17.9%)	118 (16.5%)	0.860
Never married, not living with partner	12 (15.4%)	174 (24.3%)	0.105

The insulin (INS) gene (p=3.8e-5) was significant in the cluster-3 analysis. The result is based on the combined evidence from 34 SNPs located in or near the INS gene. None of the SNPs were individually significant after multiple test adjustment.

Comment

We have identified five major clusters of women with SPTB using precise phenotyping and clustering tools. There was significant correlation between several of the phenotypes indicating the possibility of common mechanisms underlying these groups. We hypothesized that use of this approach would identify a SPTB cluster that may be associated with a genotype and indeed we found a significant difference in the INS gene, which had a group of 34 SNPs that were significantly different between cluster cases in which women either had recurrent SPTB or a strong family history of PTB and controls.

We found significant co-occurrence among placental dysfunction, decidual hemorrhage and the inflammation/infection pathways. The decidual hemorrhage and inflammation/infection pathways have previously been identified in the same population of Caucasian women with SPTB.¹⁹ Thrombosis within the decidua and placenta is often associated with an acute inflammatory process, and therefore it is not surprising that thrombosis is seen in conjunction with inflammation more frequently in SPTB.²⁰ In addition, one study reported that findings of inflammation and/or hemorrhage in placental pathology at the time of SPTB are associated with an increased risk of recurrent SPTB in subsequent pregnancies.²¹ It is therefore logical to find a cluster that includes all three clinical presentations.

Previous attempts to identify the genetic causes of SPTB have met with only modest success.¹³ In fact, a recent GWA study performed on the women included in this analysis identified only a few maternal and fetal SNPs associated with SPTB, despite relatively large numbers of prospectively collected cases and controls.¹² This may be due to insufficient numbers of cases and controls and thus inadequate power to detect the more subtle differences that exist between the groups. An alternative hypothesis is that the phenotypes that are currently used to define cases and controls are too broad, do not identify groups of women who share similar etiologies for their SPTB, and thus are not likely to identify common genetic causes. Several authors have recently proposed more sophisticated approaches to improve the assignment of phenotype to cases of preterm birth.^22,23 We have previously described a phenotyping system that was used to evaluate the SPTBs that occurred in the GPN case control study.¹⁴ Our phenotyping tool improves on other previously proposed PTB classification systems by providing more specific classification yet was applied with the use of readily available data. Use of this system demonstrated that most women had evidence of 2 or more phenotypes confirming the complex nature of SPTB. Previous reports of increased risk of recurrent SPTB even after a prior indicated PTB highlight the potential for overlap in the pathophysiology responsible different type of PTB.^9,24

In order to further refine the phenotype of SPTB, in the current study we used hierarchal clustering to identify women with common clinical characteristics that are not limited to the pre-defined phenotypes of the original tool. The hierarchal cluster analysis found five separate groups of women with one or more common clinical features and thus provides another opportunity to identify genetic factors associated with specific subsets of SPTB.

We chose the cluster with family history of SPTB as a predominant feature (cluster 3) for genetic analyses. We hypothesized that women with a strong personal history (recurrent SPTB) or family history of SPTB would be more likely to have inherited a genetic factor that would increase their risk of SPTB.^8,25 We chose a previously reported comprehensive set of inflammatory genes as the candidates for our candidate gene analysis.¹⁸ The INS gene, which is found on chromosome 11 and is known to be involved in the MAPK and in the NF-kB signaling pathways,¹⁸ was significantly associated with SPTB among women in cluster 3. Both the MAPK signaling and NFkB signaling pathways play important roles in the inflammatory response and have been previously implicated in the pathogenesis of SPTB.^26,27 However, the exact contribution of the INS gene to SPTB remains to be evaluated.

There are several limitations to this study. First, some of the clinical information used by the phenotyping tool was absent in some of the women with SPTB. For example, placental pathology was not available for every subject. Placental pathology after a SPTB is a powerful tool for elucidating the underlying etiology of the delivery. There is a well-characterized association between SPTB and the histologic presence of chorioamnionitis, with funisitis and vasculitis being particularly associated with deliveries at increasingly early gestational ages.²⁸ In addition, the identification of phenotypes of decidual hemorrhage and placental dysfunction would be enhanced by routine pathologic evaluation of the placenta in cases of SPTB and these findings may be associated with the risk of recurrent SPTB.²¹ It is clear that this phenotyping approach would be enhanced if all pertinent data were available for all subjects, however, the absence of some data would lead to under-classification of subjects and would make the identification of genetic contributors more difficult. There are likely some number of confounders, such as environmental exposures, that were not assessed. Separation of women into clusters was a second limitation in that it resulted in a reduction of the number of subjects available for the candidate gene analysis and therefore reduced the power of this study to identify genes with less robust association. Our study was also limited to the maternal genetic contribution to SPTB. Finally, we have chosen to focus on inflammatory genes and it is possible that other genes from other pathways are associated with SPTB. Future studies should assess the role of fetal genotype, as well as other pathways and other clusters of SPTB phenotypes in the genetic causes of SPTB.

Despite the limitations, this study has taken a novel, granular approach to the phenotyping of women who experience SPTB and who are thus more likely to share common etiologies and common genetic predispositions. We have identified one gene that was associated with women who have SPTB with a strong familial history. We recognize that the interaction between a pregnant woman's environment and her personal biological responses is complex and that advances in our understanding of these relationships is an ongoing, iterative process. However, we believe that the use of hierarchical clustering generates unbiased phenotyping information, which may enhance our understanding of the pathways leading to SPTB. Future studies may address the genetic similarities of the other clusters identified in this study providing new insight into the phenotypes that cluster together based on clinical characteristics. In addition, this type of evaluation may be used to assess recurrence risks in future pregnancies or neonatal outcomes among women with similar clinical phenotypes.

Figure 2.

Graphic representation of phenotype distributions for samples in each cluster. The coloring of each cell in the figure indicates the proportion of samples in the cluster that are positive for the specified variable. For example, the mean of “maternal stress_any” for Cluster 1 is 0.81, indicating that 81% of samples in the cluster are positive for some level of HPA activation. The color gradient is defined in the legend on the top edge of the figure. The labels on the left side represent the different levels of evidence from each of the phenotype categories that were included in the final cluster analysis. The first word indicates the phenotype category and the number represents the level of evidence within that category (1=Strong, 2= Moderate, and 3=Possible).

Table 3.

This table demonstrates the raw values for each of the variables assessed in each cluster. The number in each space represents the percentage of women in each cluster that were found to have each specific characteristic. For example, 40.2% of women in cluster 1 were married and living with their partner (See Column one, Row 1)

	Cluster 1	Cluster 2	Cluster 3	Cluster 4	Cluster 5
Married, live together	40.2%	48.6%	62.5%	54.0%	78.3%
Never married, live tog.	26.3%	19.0%	16.7%	25.4%	7.5%
Never mar'd, not live tog.	28.1%	29.9%	14.2%	15.9%	10.4%
9-12 yrs education	54.8%	53.1%	40.8%	50.8%	34.0%
13-16 yrs education	30.3%	43.2%	55.8%	30.2%	60.4%
Race=Black	19.1%	44.9%	9.2%	9.5%	0.9%
Race=Caucasian	68.8%	49.3%	85.0%	74.6%	92.5%
Hispanic (any race)	31.7%	5.8%	11.7%	46.0%	9.4%
Nulliparity	45.4%	51.0%	49.2%	36.5%	44.3%
Infection_1	3.8%	2.0%	6.7%	0.0%	50.9%
Infection_2	6.3%	5.8%	10.0%	9.5%	32.1%
Infection_3	27.2%	25.5%	25.0%	20.6%	30.2%
Infection_any	33.3%	31.0%	38.3%	28.6%	78.3%
Maternal Stress_2	44.7%	5.1%	47.5%	34.9%	32.1%
Maternal Stress_3	63.1%	25.9%	25.0%	42.9%	15.1%
Maternal Stress_any	80.7%	29.9%	49.2%	65.1%	41.5%
Hemorrhage_1	0.4%	0.0%	0.0%	0.0%	0.9%
Hemorrhage_2	12.8%	6.1%	11.7%	6.3%	53.8%
Henorrhage_3	20.0%	11.2%	29.2%	17.5%	32.1%
Hemorrhage_any	30.3%	15.3%	37.5%	23.8%	71.7%
Distension_2	26.1%	3.4%	16.7%	22.2%	14.2%
Distension_3	1.6%	4.4%	5.8%	4.8%	13.2%
Distension_any	26.7%	7.8%	22.5%	25.4%	26.4%
Cervical_1	4.0%	10.2%	7.5%	4.8%	6.6%
Cervical_2	2.9%	7.1%	6.7%	9.5%	2.8%
Cervical_3	1.6%	1.4%	3.3%	3.2%	4.7%
Cervical_any	7.4%	16.0%	15.0%	14.3%	10.4%
Placental Dysf_1	0.7%	0.3%	4.2%	3.2%	23.6%
Placental Dysf_2	2.5%	1.4%	7.5%	4.8%	42.5%
Placental Dysf_3	3.1%	0.7%	3.3%	0.0%	33.0%
Placental Dysf_any	4.9%	2.0%	9.2%	4.8%	61.3%
Mat Comorbid_1	1.1%	7.1%	4.2%	81.0%	5.7%
Mat Comorbid_2	10.3%	13.9%	21.7%	76.2%	20.8%
Mat Comorbid_any	10.6%	18.4%	23.3%	100.0%	24.5%
Familial_1	13.9%	21.4%	60.8%	17.5%	17.0%
Familial_2	10.8%	13.9%	35.8%	4.8%	7.5%
Familial_any	23.4%	29.9%	84.2%	22.2%	22.6%
PPROM_1	12.1%	34.4%	8.3%	30.2%	25.5%
PPROM_2	13.3%	19.0%	12.5%	7.9%	6.6%
PPROM_3	3.6%	6.8%	3.3%	3.2%	6.6%
PPROM_any	26.5%	53.7%	22.5%	38.1%	34.0%