Original Article

A Bayesian Stepwise Discriminant Model for Predicting

Risk Factors of Preterm Premature Rupture of Membranes:
A Case-control Study
Li‑Xia Zhang1,2, Yang Sun3, Hai Zhao2, Na Zhu2, Xing‑De Sun4, Xing Jin5, Ai‑Min Zou6, Yang Mi7, Ji‑Ru Xu1
1
Department of Microbiology and Immunology, Health Science Center, Xi’an Jiaotong University, Xi’an, Shaanxi 710061, China
2
Deparment of Clinical Laboratory, Shaanxi Provincial People’s Hospital, Xi’an, Shaanxi 710068, China
3
Department of Medical Statistics, Shaanxi Provincial People’s Hospital, Xi’an, Shaanxi 710068, China
4
Deparment of Clinical Laboratory, Xi’an Fourth Hospital, Xi’an, Shaanxi 710004, China
5
Deparment of Clinical Laboratory, Xi’an Gaoxin Hospital, Xi’an, Shaanxi 710075, China
6
Deparment of Clinical Laboratory, Chang’an Hospital, Xi’an, Shaanxi 710018, China
7
Department of Obstetrics and Gynecology, The Northwest Women and Children Hospital, Xi’an, Shaanxi 710061, China

Abstract
Background: Preterm premature rupture of membrane (PPROM) can lead to serious consequences such as intrauterine infection, prolapse
of the umbilical cord, and neonatal respiratory distress syndrome. Genital infection is a very important risk which closely related with
PPROM. The preliminary study only made qualitative research on genital infection, but there was no deep and clear judgment about the
effects of pathogenic bacteria. This study was to analyze the association of infections with PPROM in pregnant women in Shaanxi, China,
and to establish Bayesian stepwise discriminant analysis to predict the incidence of PPROM.
Methods: In training group, the 112 pregnant women with PPROM were enrolled in the case subgroup, and 108 normal pregnant women
in the control subgroup using an unmatched case-control method. The sociodemographic characteristics of these participants were collected
by face‑to‑face interviews. Vaginal excretions from each participant were sampled at 28–36+6 weeks of pregnancy using a sterile swab.
DNA corresponding to Chlamydia trachomatis (CT), Ureaplasma urealyticum (UU), Candida albicans, group B streptococci (GBS),
herpes simplex virus‑1 (HSV‑1), and HSV‑2 were detected in each participant by real‑time polymerase chain reaction. A model of Bayesian
discriminant analysis was established and then verified by a multicenter validation group that included 500 participants in the case subgroup
and 500 participants in the control subgroup from five different hospitals in the Shaanxi province, respectively.
Results: The sociological characteristics were not significantly different between the case and control subgroups in both training
and validation groups (all P > 0.05). In training group, the infection rates of UU (11.6% vs. 3.7%), CT (17.0% vs. 5.6%), and GBS
(22.3% vs. 6.5%) showed statistically different between the case and control subgroups (all P < 0.05), log‑transformed quantification of
UU, CT, GBS, and HSV‑2 showed statistically different between the case and control subgroups (P < 0.05). All etiological agents were
introduced into the Bayesian stepwise discriminant model showed that UU, CT, and GBS infections were the main contributors to PPROM,
with coefficients of 0.441, 3.347, and 4.126, respectively. The accuracy rates of the Bayesian stepwise discriminant analysis between the
case and control subgroup were 84.1% and 86.8% in the training and validation groups, respectively.
Conclusions: This study established a Bayesian stepwise discriminant model to predict the incidence of PPROM. The UU, CT, and
GBS infections were discriminant factors for PPROM according to a Bayesian stepwise discriminant analysis. This model could
provide a new method for the early predicting of PPROM in pregnant women.

Key words: Bayesian Stepwise Discriminant Analysis; Etiological

Factors; Infection; Preterm Premature Rupture of Membranes Address for correspondence: Prof. Ji‑Ru Xu,
Department of Microbiology and Immunology, Health Science Center,
Xi’an Jiaotong University, Xi’an, Shaanxi 710061, China
E‑Mail: [email protected]
Introduction
Preterm premature rupture of membrane (PPROM) is This is an open access article distributed under the terms of the Creative Commons
Attribution‑NonCommercial‑ShareAlike 3.0 License, which allows others to remix,
a common perinatal complication in pregnant women. tweak, and build upon the work non‑commercially, as long as the author is credited
and the new creations are licensed under the identical terms.
Access this article online For reprints contact: [email protected]
Quick Response Code: © 2017 Chinese Medical Journal  ¦  Produced by Wolters Kluwer ‑ Medknow
Website:
www.cmj.org
Received: 18‑07‑2017 Edited by: Xin Chen
How to cite this article: Zhang LX, Sun Y, Zhao H, Zhu N, Sun XD,
DOI: Jin X, Zou AM, Mi Y, Xu JR. A Bayesian Stepwise Discriminant Model
10.4103/0366-6999.216396 for Predicting Risk Factors of Preterm Premature Rupture of Membranes:
A Case-control Study. Chin Med J 2017;130:2416-22.

2416 Chinese Medical Journal  ¦  October 20, 2017  ¦  Volume 130  ¦  Issue 20
It is responsible for one‑third of all preterm births. The Study design and participants
worldwide prevalence of PPROM ranges from 2% to 10%. Abnormal vaginal discharge was examined in each of
PPROM might occur among women of the reproductive the participants. The quantitative levels of CT, UU, NG,
age group, especially during the period of pregnancy C. albicans, GBS, HSV‑1, and HSV‑2 were detected in each
before 37 weeks of gestation. PPROM can cause maternal of the participants. Based on the etiological detection, a type
and fetal infection in pregnant women and their unborn of linear discriminant analysis was used to discriminate
children, a lower Apgar score, pulmonary hypoplasia, between normal pregnant women and those with PPROM.
preterm delivery, and a low birth weight. However, the The accuracy of the Bayesian stepwise discriminant model
etiology of PPROM is unclear. PPROM may be caused was validated by both a training group (including 112 cases
by cervical incompetence, genital infections, and uterine in case subgroup and 108 cases in control subgroup) and a
abnormality. Some studies have shown that a history multicenter validation group (including 500 cases in case
of PPROM, race, smoking status, poor nutrition, and subgroup and 500 cases in control subgroup). An unmatched
genital infection are risk factors for PPROM. This case-control design was used in this study. Inclusion
study developed a model to explore genital infections criteria for normal pregnant women were as follows:
that might activate inflammatory cells and then induce women with 28–36+6 weeks of gestation, no use of any
PPROM. The etiologies of genital infection include antibiotics within 2 months, and no history of any chronic
Chlamydia trachomatis (CT), Ureaplasma urealyticum (UU), diseases (such as diabetes, cardiovascular disease, and
Candida albicans, syphilis, Neisseria gonorrhoea (NG), hypertension). Inclusion criteria for the PPROM patients
group B streptococci (GBS), herpes simplex virus (HSV), were as follows: pregnant women with 28–36+6 weeks of
and bacterial vaginosis (BV).[1,2] gestation, membrane rupture within 12 h, no use of any
antibiotics within 2 months, and no history of any basic
Genital infections might cause a release of cytokines
diseases (such as diabetes, cardiovascular disease, and
and other inflammatory mediators that may weaken the
hypertension). The PPROM is defined as the onset of
membrane and cause PPROM. Studies by Chow and Blas
amniotic fluid leakage from the vagina before the onset
showed that CT infection was associated with the occurrence of uterine contractions at less than 37 weeks’ gestational
of PPROM.[3,4] Pregnant women with BV more readily age.[8] The PPROM includes having a history of drainage
developed PPROM than women without BV.[4‑6] Candidiasis of clear fluid that wets the perineum and runs along the
infection in pregnant women with PPROM is controversial, thighs and legs as well as a sterile speculum examination
and a recent study showed that the treatments for candidiasis showing fluid pooling in the posterior vaginal fornix or fluid
might reduce the incidence of PPROM.[7] Pregnant women freely flowing from the cervix. The laboratory definition of
who were infected with NG had a six‑time higher risk of PPROM is positivity for insulin‑like growth factor‑binding
developing PPROM than women without NG infection. protein 1 in the vaginal discharge.[9]
GBS might cause the activation of inflammatory cells in
fetal membranes, which could lead to PPROM.[5,8] In the preliminary study, 112 pregnant women with PPROM
and 108 normal pregnant women were randomly recruited
Although some studies have reported that PPROM was from Department of Obstetrics and Gynecology, Shaanxi
related to genital infections, the proportions of women Provincial People’s Hospital between June 2011 and May
with confirmed genital infections with or without PPROM 2012. Bayesian stepwise discriminant analysis was used to
in China are unknown.[3‑5,7‑9] Studies of the relationship analyze the etiological infections of CT, UU, C. albicans,
between genital infection and PPROM are still rare. GBS, HSV‑1, and HSV‑2. A multicenter validation group
This study aimed to determine the association between included 500 pregnant women with PPROM (case subgroup)
etiological infection and PPROM. Discriminant analysis and 500 normal pregnant women (control subgroup) from
is a multivariate statistical method that can distinguish five different hospitals in the Shaanxi province between June
newly acquired samples according to the quantitative 2012 and January 2013, respectively. These five hospitals
characteristics of the existing observational sample. In were Northwest Women and Children Hospital, Xi’an Fourth
this study, a Bayesian stepwise discriminant model was Hospital, Xi’an Gaoxin Hospital, Chang’an Hospital, and
established, and a corresponding linear discriminant Xianyang 215 Hospital in Shaanxi province.
function was built. This model could predict and reduce
the occurrence of PPROM. Data and specimen collection
Face‑to‑face questionnaires were used to collect the
sociodemographic characteristics (including age, gravidity,
Methods parity, marital status, and occupation) and gynecological
Ethical approval histories (including obstetric history, past history of
The study was conducted in accordance with the PPROM, and history of trauma to the cervix). A vaginal
Declaration of Helsinki and was approved by Institutional swab and a cervical swab were collected within 12 h of
Review Board of Shaanxi Provincial People’s Hospital. membrane rupture of PPROM cases, and the control group
Informed written consent was obtained from all the were collected at 28–36+6 weeks of gestation during routine
participants before their enrolment in this study. examination.

Chinese Medical Journal ¦ October 20, 2017 ¦ Volume 130 ¦ Issue 20 2417

Nucleic acid extraction parity, marital status, and occupation between the
Each swab was suspended in the 1.5 ml of sterile saline normal pregnant women (control subgroup) and those
(0.85%). Nucleic acid was extracted from the swab with PPROM (case subgroup) in the training group [all
specimens using QIAamp MiniStool kit (QIAGEN, Hilden, P > 0.05, Table 1].
Germany) following the manufacturer’s instruction, and the Univariate analysis of the etiologic agents in the training
DNA was eluted in the 45 µl of elution buffer.
group
Quantitation of etiological agents by real‑time All participants were negative NG, HIV, and syphilis. In
polymerase chain reaction the training group, there were significant differences in
Real‑time polymerase chain reaction (PCR; Triplex the positive rates of abnormal vaginal discharge, UU, CT,
International Biosciences Co., LTD., China), following and GBS between the normal pregnant women and those
the manufacturer’s instructions, was used to detect NG, with PPROM [all P < 0.05; Table 2]. To study the effects
UU, CT, GBS, and C. albicans in the vaginal swabs and of different etiological agents on PPROM, the quantitative
HSV‑1 and HSV‑2 in the cervical swabs. The threshold levels of UU, CT, HSV‑2, and GBS were converted into
of detection of the PCR was equal to or greater than 103 log‑normal distribution data. The quantitative levels of
copies/ml. UU, CT, GBS, and HSV‑2 showed significant differences
between the normal pregnant women and those with
Antibody against HIV and syphilis detection PPROM [all P < 0.05; Table 3]. However, the C. albicans
All participants’ serum was collected to detect antibody and HSV‑1 distributions were not significantly different
against HIV and syphilis using enzyme linked immunosorbent between the normal pregnant women and those with
assay (ELISA) kits (Shanghai Kehua Bioengineering Co., PPROM [all P > 0.05, Table 3]. Positive rates of each
Ltd., China). etiological agent were analyzed using Chi‑square test. The
translated data were analyzed using Wilcoxon two‑sample
Statistical analysis test method. Then, the translated data were separately
The data were analyzed using SPSS version 19.0 analyzed by a forward selection method, and significant
software (SPSS Inc., Chicago, IL, USA). Data were variables were selected for the Bayesian stepwise
considered to have a normal distribution if the P value given discriminant analysis.
by the Shapiro-Wilk test was more than 0.05. If the test data
set did not show a normal distribution, the data could be Bayesian stepwise discriminant analysis
normalized by logarithmic transformation. The mean levels The Bayesian stepwise discriminant analysis is described
of infectious agents were compared using the Wilcoxon in statistical language as follows: Assume g populations
two‑sample test method. The Chi‑square test was used to follow g multivariate normal distributions. Probability of
analyze the differences between categorical data. The value misclassifying a subject in class i into class j, P(i|j); Loss due
of P < 0.05 was considered to be statistically significant. to misclassification, a(j|i). The Bayesian criterion: minimize
the expected misclassification loss.
Furthermore, the original data for the etiological factors
of PPROM were log‑transformed and then translated as
discriminant functions. The translated data were analyzed by Table 1: Baseline characteristics of all participates in
a forward selection method (sle = 0.1, sls = 0.1). Significant training group of this study
variables were identified by Bayesian stepwise discriminant Characteristics Case Control χ2 P
analysis. subgroup subgroup
(n = 112) (n = 108)
The quantitative levels of the etiological agents for each Marital status, n (%) 0.303 0.860
of the pregnant women were skewed. Hence, these data Single 2 (1.8) 1 (0.9)
were converted into a log‑normal distribution. Linear Married/cohabiting 109 (97.3) 106 (98.2)
combinations of data were used to form discriminant Divorced/separated 1 (0.9) 1 (0.9)
functions for the separation of categories by minimization Age, n (%) 0.897 0.085
of the within‑class and between‑class ratios of the sum of <20 years 1 (0.9) 1 (0.9)
squares. Bayesian stepwise discriminant analysis was used ≥20 years and <35 years 89 (79.5) 91 (84.3)
to distinguish normal pregnant women from those with ≥35 years 22 (19.6) 16 (14.8)
PPROM. Forward stepwise analysis was used to select Gravidity, n (%) 0.598 0.439
significant variables for the discriminant analysis. An Primegravidae 85 (75.9) 77 (71.3)
obvious difference in the selected variables was observed Gravida ≥2 27 (24.1) 31 (28.7)
Parity, n (%) 0.429 0.513
when the translated variables were used.
Primipara 94 (83.9) 87 (80.6)
Secondary ≥2 18 (16.1) 21 (19.4)
Results Occupation, n (%) 0.001 0.984
Baseline characteristics of the training group House wife 52 (46.4) 50 (46.3)
There were no significant differences in age, gravidity, Employee/business 60 (53.6) 58 (53.7)

2418 Chinese Medical Journal  ¦  October 20, 2017  ¦  Volume 130  ¦  Issue 20
 individual observation indices were introduced into the
Table 2: Infectious status of training group in this study
equation, according to the results of the individual, to infer
Variables Case Control χ2 P the type of a statistical method.
subgroup subgroup
(n = 112) (n = 108) Two Bayesian function equations were established
Abnormal vaginal 17.774 0.001 based on the discriminant coefficients. To investigate the
discharge, n (%) contribution of the etiological factors, the tests of equality
Yes 53 (47.3) 22 (20.4) of three groups (UU, CT, and GBS) were statistically
No 59 (52.7) 86 (79.6) different (P < 0.05), then Bayesian discriminant method
Candida albicans, n (%) 2.711 0.100 could be carried out. The significance test of the discriminant
Positive 7 (6.3) 2 (1.9) function are shown in Table 4, Wilks’ λ value was 0.530,
Negative 105 (93.7) 106 (98.1) Chi‑square value was 137.535, so the discriminant result
HSV‑2, n (%) 0.376 0.617 was proved to be effective. The classification function of
Positive 1 (0.9) 2 (1.9) Bayesian model was established as follows:
Negative 111 (99.1) 106 (98.1)
UU, n (%) 4.817 0.028 f1 = − 95.383 + 0.441x1 + 3.347 x2 + 4.126 x3
Positive 13 (11.6) 4 (3.7)
Negative 99 (88.4) 104 (96.3) f 2 = − 71.580 + 0.381x1 + 3.263 x2 + 2.642 x3
CT, n (%) 7.105 0.008
Positive 19 (17.0) 6 (5.6) X1 is the distribution of UU, X2 is the distribution of CT, and
Negative 93 (83.0) 102 (94.4) X3 is the distribution of GBS. f1 is the function for the PPROM
GBS, n (%) 11.098 0.001 group, and f2 is the function for the non‑PPROM group.
Positive 25 (22.3) 7 (6.5)
The results of the Bayesian stepwise discriminant analysis
Negative 87 (77.7) 101 (93.5)
showed that UU, CT, and GBS infections were key factors
HSV‑1, n (%) 1.218 0.270
that could affect the occurrence of PPROM, with coefficients
Positive 5 (4.5) 2 (1.9)
of 0.441, 3.347, and 4.126, respectively [Table 4]. According
Negative 107 (95.5) 106 (98.1)
UU: Ureaplasma urealyticum; CT: Chlamydia trachomatis;
to the Bayesian stepwise discriminant model, associations
GBS: Group B Streptococci; HSV‑1: Herpes simplex virus type 1; were observed among UU, CT, and GBS infections, and
HSV‑2: Herpes simplex virus type 2. PPROM. No associations were found among HSV‑1,
HSV‑2, C. albicans, and PPROM. The Bayesian stepwise
discriminant analysis was used to differentiate normal
Table 3: Distribution of etiological agents in training
pregnant women from those with PPROM. The results
group of this study
showed that the accuracy of this method was 84.1%.
Etiological Case Control Z P
agents subgroup* subgroup* Validation of Bayesian stepwise discriminant analysis
(n = 112) (n = 108) All classification rules developed through Bayesian stepwise
UU 6.16 5.03 3.534 0.002 discriminant analysis should be prospectively validated before
CT 5.75 3.72 11.521 0.001 their use in clinical practice; therefore, we designed a prospective
GBS 7.93 6.51 3.249 0.004 validation group. There were no significant differences in age,
Candida albicans 4.66 4.66 0.168 0.569 gravidity, parity, marital status, and occupation between the
HSV‑1 3.59 3.61 –0.264 0.795 normal pregnant women (control subgroup, 500 cases) and
HSV‑2 4.12 3.72 3.586 0.002 those with PPROM (case subgroup, 500 cases) in the validation
*The quantitative levels of UU, CT, HSV‑2, and GBS were converted group [all P > 0.05, Table 5]. The distributions of abnormal
into log‑normal distribution data. UU: Ureaplasma urealyticum; CT:
vaginal discharge, UU, CT, GBS, and C. albicans showed
Chlamydia trachomatis; GBS: Group B streptococci; HSV‑1: Herpes
simplex virus type 1; HSV‑2: Herpes simplex virus type 2. significant differences between two subgroups [all P < 0.05,
Table 6]. However, the distributions of HSV‑2 and HSV‑1
Classification function: were not significantly different [all P > 0.05, Table 6]. The
log‑transformed quantification of quantitative levels of UU,
f1 = a10 + a11 x1 + … + a1 p x p CT, HSV‑2, and GBS showed statistical differences between
f 2 = a20 + a21 x1 + … + a2 p x p the case and control groups but C. albicans did not show
statistical difference [Table 7]. After the bias discriminant
f g = ag 0 + ag1 x1 + … + agp x p function cross‑validation, the accuracy of this method was
86.8% to separate normal pregnant women and PPROM
aj0, aj1., ajp (j = 1,2.... g): the parameters to be estimated; fji women in validation group [Table 8].
represents positively related to the probability of being in
the jth population. Discussion
Bayesian stepwise discriminant analysis was used to The mechanisms of PPROM are unclear. The presence
establish a function using retrospective data, and then the of infections may cause PPROM through the release of

Chinese Medical Journal ¦ October 20, 2017 ¦ Volume 130 ¦ Issue 20 2419

Table 4: Results of Bayesian stepwise discriminant function
Variables F P Coefficients Wilks’ λ χ2 P
PPROM Non‑PPROM
UU 52.999 <0.001 0.441 0.381 0.530 137.535 <0.001
CT 31.247 <0.001 3.347 3.263
GBS 125.065 <0.001 4.126 2.642
UU: Ureaplasma urealyticum; CT: Chlamydia trachomatis; GBS: Group B streptococci; PPROM: Preterm premature rupture of membranes.

Table 5: Baseline characteristics of all participants in validation group of this study

Characteristics Case subgroup (n = 500) Control subgroup (n = 500) χ2 P
Marital status, n (%) 1.001 0.606
Single 4 (0.8) 2 (0.4)
Married/cohabiting 495 (99.0) 496 (99.2)
Divorce/separated 1 (0.2) 2 (0.4)
Age, n (%) 1.450 0.484
<20 years 6 (1.2) 6 (1.2)
≥20 years and <35 years 432 (86.4) 444 (88.8)
≥35 years 62 (12.4) 50 (10.0)
Gravidity, n (%) 2.067 0.150
Primegravidae 324 (64.8) 302 (60.4)
Gravida ≥2 176 (35.2) 198 (39.6)
Parity, n (%) 2.569 0.109
Primipara 417 (83.4) 435 (87.0)
Secondary ≥2 83 (16.6) 65 (13.0)
Occupation, n (%) 0.065 0.799
Home maker 219 (43.8) 215 (43.0)
Employed/business 281 (56.2) 285 (57.0)

Table 6: Infectious status of validation group in this study

Variables Case subgroup (n = 500) Control subgroup (n = 500) χ2 P
Abnormal vaginal discharge, n (%) 9.019 0.003
Yes 178 (35.6) 134 (26.8)
No 322 (64.4) 366 (73.2)
Candida albicans, n (%) 4.384 0.036
Positive 23 (4.6) 11 (2.2)
Negative 477 (95.4) 489 (97.8)
HSV‑2, n (%) 2.016 0.156
Positive 6 (1.2) 2 (0.4)
Negative 494 (98.8) 498 (99.6)
UU, n (%) 22.353 0.001
Positive 67 (13.4) 24 (4.8)
Negative 433 (86.6) 476 (95.2)
CT, n (%) 23.349 0.001
Positive 57 (11.4) 17 (3.4)
Negative 443 (88.6) 483 (96.6)
GBS, n (%) 36.424 0.001
Positive 78 (15.6) 21 (4.2)
Negative 422 (84.4) 479 (95.8)
HSV‑1, n (%) 0.504 0.478
Positive 5 (1.0) 3 (0.6)
Negative 495 (99.0) 497 (99.4)
UU: Ureaplasma urealyticum; CT: Chlamydia trachomatis; GBS: Group B streptococci; HSV‑2: Herpes simplex virus type 2; HSV‑1: Herpes simplex
virus type 1.

inflammatory cytokines and proteases.[6,9] Indeed, genital Infection may impair the antimicrobial effect of the pregnant
infection has been identified as a risk factor for PPROM. cervix, making it more susceptible to other microbes.[10,11] This

2420 Chinese Medical Journal  ¦  October 20, 2017  ¦  Volume 130  ¦  Issue 20
 Some studies have shown that infections with HIV, syphilis,
Table 7: Distribution of etiological agents in validation
and NG might be risk factors for PPROM in pregnant
group in this study
women.[8,22] In this study, all participants were HIV, syphilis
Etiological Case Control Z P and NG negative, so these three pathogens were not included
agents subgroup* subgroup* in the analysis of this study.
(n = 500) (n = 500)
UU 7.70 5.55 7.280 0.001 This study showed that CT infection was associated with
CT 6.12 3.95 11.183 0.001 PPROM. Some studies have shown that CT infection of
GBS 8.44 7.55 4.453 0.001 pregnant women could cause release of inflammatory mediators
Candida albicans 4.50 4.70 −2.291 0.091 that could be implicated in membrane rupture.[15,23,24] Some
HSV‑1 3.17 3.22 −0.626 0.539 studies have shown that infection with GBS might release
HSV‑2 3.78 3.37 4.684 0.001 cytokines and other inflammatory modulators which could
*The quantitative levels of UU, CT, HSV‑2, and GBS were converted cause membrane rupture.[23,25] This study found an association
into log‑normal distribution data. UU: Ureaplasma urealyticum; CT:
between PPROM and GBS. In this study, the prevalence rates of
Chlamydia trachomatis; GBS: Group B streptococci; HSV‑1: Herpes
simplex virus type 1; HSV‑2: Herpes simplex virus type 2. GBS in the women with PPROM ranged from 4.2% to 22.3%,
similar with the results of other studies.[14,26,27]

Table 8: Cross validation of training and validation The associations between etiological factors and PPROM are
groups in this study still unclear, and no tool is available to evaluate the association
between quantitative levels of etiological agents and
Groups PPROM Non‑PPROM Total PPROM.[24,28] In this study, we established a Bayesian stepwise
Training group*, n (%) discriminant model to identify normal pregnant women
Case subgroup 97 (81.5) 15 (18.5) 112 (100.0)
and those with PPROM. We found that CT, UU and GBS
Control subgroup 20 (18.4) 88 (86.6) 108 (100.0)
infections were associated with PPROM. Using this method,
Validation group†, n (%)
84.1% and 86.8% of the pregnant women with PPROM
Case subgroup 426 (85.2) 74 (14.8) 500 (100.0)
could be distinguished from the normal pregnant women in
Control subgroup 66 (13.2) 434 (86.8) 500 (100.0)
*The accuracy was 84.1% to separate normal pregnant women and
the training and validation groups, respectively. However, the
PPROM women in training group; †The accuracy was 86.8% to separate cause of PPROM is complicated, only main etiological agents
normal pregnant women and PPROM women in validation group. were involved in this study, but noninfectious factors were
PPROM: Preterm premature rupture of membranes. not included, so some pregnant women with PPROM could
not be distinguished from normal pregnant women using this
study investigated the associations between the selected genital Bayesian stepwise discriminant analysis.
infections (abnormal vaginal discharge, UU, CT, GBS, NG,
C. albicans, HSV‑1, HSV‑2, HIV, and syphilis) and PPROM in In this study, a Bayesian stepwise discriminant model was
Shaanxi province, China. The distributions of abnormal vaginal established to predict the incidence of PPROM. The UU, CT,
discharge, UU, CT and GBS were significantly different and GBS infections were discriminant factors for PPROM
between normal pregnant women and those with PPROM. according to a Bayesian stepwise discriminant analysis. This
However, the distributions of C. albicans, HSV‑1, HSV‑2, model could provide a new method for the early predicting
were not significantly different between normal pregnant of PPROM in pregnant women to hopefully reduce the
women and those with PPROM. The relationship between incidence of PPROM.
abnormal vaginal discharge and PPROM has been reported in Financial support and sponsorship
other studies.[12‑14] some studies have shown that C. albicans This work was supported by grants from the Natural
was protective against PPROM.[11,15] Pregnant women with Science Found of Shaanxi Province (No. 2014JM2‑8200
C. albicans were 73% less likely to develop PPROM than and No. 2010JM4031).
pregnant women without C. albicans. The possible reason
for this finding might be that the amniotic fluid washes out Conflicts of interest
the yeast cells, which could lead to negative results.[16‑18] We There are no conflicts of interest.
found that the positive rate of C. albicans had no statistical
difference between two subgroups in the training group, but had References
significant different between two subgroups in the validation 1. Luo X, Pan J, Wang L, Wang P, Zhang M, Liu M, et al. Epigenetic
group, and bias might be caused by the sample scale. Different regulation of lncRNA connects ubiquitin‑proteasome system with
infection‑inflammation in preterm births and preterm premature
outcomes were reported about the relationship between
rupture of membranes. BMC Pregnancy Childbirth 2015;15:35. doi:
HSV‑1 and HSV‑2 with PPROM.[19] In the training group, the 10.1186/s12884‑015‑0460‑0.
positive rate of HSV‑2 showed no statistical difference, but the 2. Ferrazzi E, Muggiasca ML, Fabbri E, Fontana P, Castoldi F, Lista G,
quantitative level of HSV‑2 was significantly different between et al. Assessment of fetal inflammatory syndrome by “classical”
markers in the management of preterm labor: A possible lesson from
normal pregnant women and those with PPROM. This result metabolomics and system biology. J Matern Fetal Neonatal Med
might be caused by the use of different detection methods for 2012;25:54‑61. doi: 10.3109/14767058.2012.716984.
HSV‑1 and HSV‑2 in pregnant women.[20‑22] 3. Chow JM, Kang MS, Samuel MC, Bolan G. Assessment of the

Chinese Medical Journal ¦ October 20, 2017 ¦ Volume 130 ¦ Issue 20 2421

 association of Chlamydia trachomatis infection and adverse diagnosis of premature rupture of membranes: Comparison of
perinatal outcomes with the use of population‑based chlamydia performance indexes. BMC Pregnancy Childbirth 2014;14:183. doi:
case report registries and birth records. Public Health Rep 10.1186/1471‑2393‑14‑183.
2009;124 Suppl 2:24‑30. doi: 10.1177/00333549091240S205. 17. Lorthe E, Ancel PY, Torchin H, Kaminski M, Langer B, Subtil D,
4. Blas MM, Canchihuaman FA, Alva IE, Hawes SE. Pregnancy et al. Impact of latency duration on the prognosis of preterm infants
outcomes in women infected with Chlamydia trachomatis: after preterm premature rupture of membranes at 24 to 32 weeks’
A population‑based cohort study in Washington State. Sex Transm gestation: A National population‑based cohort study. J Pediatr
Infect 2007;83:314‑8. doi: 10.1136/sti.2006.022665. 2017;182:47‑52.e2. doi: 10.1016/j.jpeds.2016.11.07.
5. Li W, Han L, Yu P, Ma C, Wu X, Moore JE, et al. Molecular 18. Surve MV, Anil A, Kamath KG, Bhutda S, Sthanam LK,
characterization of skin microbiota between cancer cachexia patients Pradhan A, et al. Membrane vesicles of group B streptococcus
and healthy volunteers. Microb Ecol 2014;67:679‑89. doi: 10.1007/ disrupt feto‑maternal barrier leading to preterm birth. PLoS Pathog
s00248‑013‑0345‑6. 2016;12:e1005816. doi: 10.1371/journal.ppat.1005816.
6. Ganor‑Paz Y, Kailer D, Shechter‑Maor G, Regev R, Fejgin MD, 19. Cordeiro CN, Althaus J, Burke A, Argani C. Herpes simplex virus
Biron‑Shental T, et al. Obstetric and neonatal outcomes after preterm cervicitis mimicking preterm premature rupture of membranes. Obstet
premature rupture of membranes among women carrying group B Gynecol 2015;126:378‑80. doi: 10.1097/AOG.0000000000000700.
streptococcus. Int J Gynaecol Obstet 2015;129:13‑6. doi: 10.1016/j. 20. Furman B, Shoham‑Vardi I, Bashiri A, Erez O, Mazor M. Clinical
ijgo.2014.10.024. significance and outcome of preterm prelabor rupture of membranes:
7. Roberts CL, Rickard K, Kotsiou G, Morris JM. Treatment of Population‑based study. Eur J Obstet Gynecol Reprod Biol
asymptomatic vaginal candidiasis in pregnancy to prevent preterm 2000;92:209‑16. doi: 10.1016/S0301‑2115(99)00257‑2.
birth: An open‑label pilot randomized controlled trial. BMC 21. van der Ham DP, van Kuijk S, Opmeer BC, Willekes C, van Beek JJ,
Pregnancy Childbirth 2011;11:18. doi: 10.1186/1471‑2393‑11‑18. Mulder AL, et al. Can neonatal sepsis be predicted in late preterm
8. Mercer BM. Preterm premature rupture of the membranes: Diagnosis premature rupture of membranes? Development of a prediction
and management. Clin Perinatol 2004;31:765‑82, vi. doi: 10.1016/j. model. Eur J Obstet Gynecol Reprod Biol 2014;176:90‑5. doi:
clp.2004.06.004. 10.1016/j.ejogrb.2014.02.003.
9. Xing X, Wenli G. Obsterics and Gynecology. 8th ed. Beijing: People’s 22. Waters TP, Mercer B. Preterm PROM: Prediction, prevention,
Health Press; 2013. p. 133‑4. principles. Clin Obstet Gynecol 2011;54:307‑12. doi: 10.1097/
10. Sweeney EL, Kallapur SG, Gisslen T, Lambers DS, Chougnet CA, GRF.0b013e318217d4d3.
Stephenson SA, et al. Placental infection with Ureaplasma species 23. Alger LS, Lovchik JC, Hebel JR, Blackmon LR, Crenshaw MC. The
is associated with histologic chorioamnionitis and adverse outcomes association of Chlamydia trachomatis, Neisseria gonorrhoeae, and
in moderately preterm and late‑preterm infants. J Infect Dis group B streptococci with preterm rupture of the membranes and
2016;213:1340‑7. doi: 10.1093/infdis/jiv587. pregnancy outcome. Am J Obstet Gynecol 1988;159:397‑404. doi:
11. Lorthe E, Quere M, Kayem G. Prolonged latency after preterm 10.1016/S0002‑9378(88)80093‑0.
premature rupture of membranes: An independent risk factor for 24. Yan SF, Liu XY, Cheng YF, Li ZY, Ou J, Wang W, et al.
neonatal sepsis? Am J Obstet Gynecol 2017;216:84. doi: 10.1016/j. Relationship between intrauterine bacterial infection and early
ajog.2016.08.022. embryonic developmental arrest. Chin Med J 2016;129:1455‑8. doi:
12. Nakubulwa S, Kaye DK, Bwanga F, Tumwesigye NM, Mirembe FM. 10.4103/0366‑6999.183411.
Genital infections and risk of premature rupture of membranes in 25. Eleje GU, Adinma JI, Ugwuanyi DC, Ikechebelu JI, Okafor CI,
Mulago Hospital, Uganda: A case control study. BMC Res Notes Ezeama CO, et al. Genital tract microbial isolate in women with
2015;8:573. doi: 10.1186/s13104‑015‑1545‑6. preterm pre‑labour rupture of membranes in resource‑constrained
13. Qin L, Li XX, Zhang LX, Zhang LL, Xi FY, Yan L, et al. Preterm community setting. J Obstet Gynaecol 2015;35:465‑8. doi:
premature rupture of membranes vaginal abnormal effects on maternal 10.3109/01443615.2014.970145.
and infant outcomes (in Chinese). Shaanxi Med J 2013;42:1220‑1. 26. Mikamo H, Sato Y, Hayasaki Y, Kawazoe K, Hua YX, Tamaya T,
doi: 10.3969/j.issn.1000‑7377.2013.09.051. et al. Bacterial isolates from patients with preterm labor with and
14. French JI, McGregor JA, Draper D, Parker R, McFee J. Gestational without preterm rupture of the fetal membranes. Infect Dis Obstet
bleeding, bacterial vaginosis, and common reproductive tract Gynecol 1999;7:190‑4. doi: 10.1155/S1064744999000320.
infections: Risk for preterm birth and benefit of treatment. Obstet 27. Dechen TC, Sumit K, Ranabir P. Correlates of vaginal colonization
Gynecol 1999;93:715‑24. doi: 10.1016/S0029‑7844(98)00557‑2. with group B streptococci among pregnant women. J Glob Infect Dis
15. Dadkhah F, Kashanian M, Eliasi G. A comparison between the pregnancy 2010;2:236‑41. doi: 10.4103/0974‑777X.68536.
outcome in women both with or without threatened abortion. Early 28. Ye F, Chen ZH, Chen J, Liu F, Zhang Y, Fan QY, et al. Chi‑squared
Hum Dev 2010;86:193‑6. doi: 10.1016/j.earlhumdev.2010.02.005. automatic interaction detection decision tree analysis of risk factors
16. Palacio M, Kühnert M, Berger R, Larios CL, Marcellin L. for infant anemia in Beijing, China. Chin Med J 2016;129:1193‑9.
Meta‑analysis of studies on biochemical marker tests for the doi: 10.4103/0366‑6999.181955.

2422 Chinese Medical Journal  ¦  October 20, 2017  ¦  Volume 130  ¦  Issue 20