We collected the clinical data of 228 pregnant women diagnosed with second-trimester threatened miscarriage in the development cohort retrospectively who were admitted to our hospital from January 2016 to August 2018, as well as the known pregnancy outcomes, including 33 patients with late miscarriage and 195 patients without late miscarriage. A total of 24 patients with second-trimester threatened miscarriage admitted between September 2018 and November 2018 were included in the validation group, of whom 2 patients suffered from late miscarriage and 22 patients didn’t experience late miscarriage.
Inclusion and exclusion criteria
There are different definitions of ‘threatened miscarriage’ and ‘second trimester’ among the various articles and guidelines. Threatened miscarriage is the diagnosis for the pregnant patient who presents with vaginal bleeding and no cervical dilation or effacement before 20 weeks of gestation. These patients may present with spotting to severe vaginal bleeding of several hours to days in duration . The time span is up to 24 weeks of gestation . In the present study, second-trimester threatened miscarriage occurs during the second trimester (14 to 28 weeks of gestation).
The exclusion criteria were as follows:
(1) Bleeding in the first and third trimester only.
(2) Women who did not participate in follow-up or delivered outside the hospital.
(3) Women who opted for termination.
(4) Known genetic/acquired thrombophilia.
(5) Large leiomyomata distorting the uterine cavity which transvaginal ultrasound shows.
(6) Women admitted to the hospital with fibroid degeneration, torsion of an ovarian cyst pedicle or an acute abdomen.
(7) Women with bleeding originating from an extragenital region or with other causes of vaginal bleeding (e.g., postcoital bleeding, cancer, cervical polyp, vaginitis) and with unknown bleeding histories.
(8) Hydatidiform moles.
Successful treatment criteria included the following:
(1) After proper treatment, vaginal bleeding stopped, there was no abdominal discomfort, and there was no sense of uterine contractions.
(2) Ultrasound reexamination indicated that fetal development was in line with the gestational weeks and fetal heart rate, there was no cervical dilation or effacement.
This study was approved by the Ethics Committee of Peking University First Hospital (ethics approval number: 2019 − 139) prior to the commencement of the study, and written informed consent was provided by the participants. The study was conducted according to the principles of the Declaration of Helsinki and its amendments.
We enrolled the baseline clinical data of the patients with threatened miscarriage. More than one hundred variables were obtained, including general information (gestational weeks, race/nationality, maternal age, admission temperature, admission pulse, maternal height, pregestational weight, pregestational body mass index (BMI), use of assisted reproductive technology, singleton or multiple pregnancy, prenatal diagnosis, admission treatment, pregnancy complications, uterine malformation, polycystic ovarian syndrome (PCOS), leiomyoma, uterine adenomyosis or adenomyoma, cervical surgical history, cervical lesions (including cervical incompetence), previous obstetric history (including history of miscarriage and induced abortion, previous preterm delivery, previous term delivery, history of obstetric abnormality and previous ectopic pregnancy), type of delivery and pregnancy outcome. Specific information was also collected, to include characteristics of threatened miscarriage, such as several admissions for second-trimester threatened miscarriage, abnormal sonographic features, specific symptoms (e.g., vaginal bleeding or/and abdominal pain), and vaginal blood loss, Other data included laboratory test results (routine blood examination, urine culture, vaginal secretion culture, cervical secretion culture, microflora assessment of the vaginal sample, procalcitonin and placental pathology). All baseline data were recorded in a Microsoft Excel.
The measurement of vaginal blood loss during pregnancy after admission is objective. Vaginal blood loss was measured after admission using the weighing method (multiply the wet weight minus the dry weight by 0.95). Since the subjective estimation of vaginal blood loss before admission is always incorrect and objective assessment may be unrealistic. We quantified the amount of vaginal blood loss at the initial visit.
Gestational weeks were calculated from the first day of the last menstrual period (LMP) and confirmed according to the first two ultrasound reports. If the embryonic bud length and crown-rump length (CRL) dating differed by more than 7 days or multiples of 7 days from the LMP dating, the gestational week was checked based on the embryonic bud length and CRL dating. [30, 31] If the patient received assisted reproductive technology, the gestational week was estimated based on the transplanted date, LMP and ultrasound scan.
C-reactive protein (CRP) is sharply elevated in the plasma when the body is infected or damaged by tissue. It activates and strengthens the phagocytosis of phagocytic cells to regulate and remove the invading organism or damaged, necrotic or apoptotic tissue cells.
Vaginal blood loss is a term referring to the amount of vaginal bleeding since the initial hospital visit.
Premature rupture of membranes (PROM) is a term referring to the natural rupture of membranes before delivery.
Uterine adenomyosis is defined as a benign lesion of the endometrium that invades the muscular layer and can be diagnosed by ultrasound, clinical examination and the patient's previous medical history. A few adenomyosis lesions show localized growth to form nodules or clumps, similar to intermural myoma, called adenomyoma.
We came to a conclusion that the baseline data in the research were not normally distributed by statistics analysis. Thus, these measurement data were presented as medians (25th-75th% quartiles), and the count data were presented as frequencies (percentages). We applied the Mann-Whitney U test to analyze the measurement data and the chi-square test to analyze the count data respectively. In addition,we assessed the associations of the variables in regard to late miscarriage in the development group by univariate and multivariate logistic regression analyses. Variables which were shown to be statistically significant in the univariate analysis were incorporated into the multivariate logistic regression analysis, and then the independent risk factors were selected through a forward stepwise process for model fitting. The most significant independent risk factors, a total of five, were screened and applied to establish a model for predicting the occurrence of miscarriage. Odds ratios (ORs) were calculated to estimate the relative risk of adverse consequences. It is believed that probability value (P-value) < .05 was statistically significant. Since the OR indicates the magnitude of the effect, any choice of OR cut-off ≥ 2.0 represents clinically significant risks and enhances the relevance of the research.
We used the ROC curve calculation cutoff value corresponding to the largest value on the Youden index to group the continuous variables, such as CRP. We eventually divided CRP into two groups, CRP ≤ 8 mg/L and CRP > 8 mg/L. Moreover, we used the optimal scale regression grouping to segment the continuous variables, such as gestational weeks, and divided the data into three groups according to the closed scores. We eventually divided the gestational weeks into three groups: ≤18 weeks, between 19 and 23 weeks and ≥ 24 weeks.
We developed an individualized predicted model of subsequent miscarriage associated with second-trimester threatened miscarriage on the basis of regression coefficients corresponding to five predictors screened from multivariate analysis. We assessed the predictive model according to its discrimination and calibration. The discrimination of the predictive model is the ability to differentiate between women with second-trimester threatened miscarriage that developed into miscarriage from those didn’t develop into miscarriage. The receiver operating characteristic (ROC) analysis was applied for the assessment of the diagnostic performance. The ROC curve was drawn by drafting the sensitivity against 1-specificity of various cut-off values. The area under the ROC curves (AUC) refers to the probability of the trial statistically, thus differentiating normal condition from abnormal condition. The AUC value ranges from 0.5 to 1.0, when the AUC value is 1 which means the trial is perfect, while the value of 0.5 represents the trial is not ideal. The closer the AUC value is to 1, the better the discriminative ability of the predictive model. Generally speaking, it is believed that the AUC value of the predictive model between 0.5 and 0.75 is acceptable, and the AUC value > 0.75 suggests that the model exhibits remarkable differentiation.  The AUC value of the development group was 0.882, indicating that the discrimination ability of predictive model is good enough.
The degree of calibration represents the consistency between the predicted probabilities and actual probabilities. The small P-value (P < .05) suggests that the calibration of the predictive model is not good enough. The large P-value on the Hosmer-Lemeshow test indicates that there is no compelling proof that the predictive model lacks fitting degree. The P-value of Hosmer-Lemeshow test in the development group was 0.923; therefore, the calculated probability derived by predictive model was highly in accordance with the observed probability, indicating that the calibration of the predictive model in the development group was perfect and the predictive model had a strong coordination. A validation group of 24 patients was applied for further verification. The diagnostic performance was determined based on accuracy, sensitivity and specificity.
IBM SPSS statistics 23 and R software (Version 3.5.0, USA) were applied to analysis statistics. P ≤ .05 implied that the differences were obviously statistically significant.