Patients
Patients with moderate or severe IUAs were selected who had been treated for infertility or recurrent abortion, from January 2013 to January 2017, at the Gynecology Minimally Invasive Center of Beijing Maternity Hospital Affiliated to Capital Medical University in China. This is a tertiary medical institution and hysteroscopic diagnosis and treatment center with nearly 1000 hysteroscopic surgeries performed annually.
For inclusion, each patient was aged 20 to 40 years; with an outpatient hysteroscopic diagnosis of moderate or severe IUAs (AFS ≥ 5) [18]; met fertility requirements; and with ovulation. Patients with any of the following were excluded: the shape of the uterine cavity did not recover; no second look was performed ≤3 months after the operation; there were uterine malformations, endometrial lesions, or tuberculosis of the reproductive system; serious adenomyosis; or abnormal semen of the male partner (World Health Organization version 5 standard).
Hysteroscopic adhesiolysis
The cervix was softened by 400 µg of misoprostol, 12 hours before the operation. Tracheal intubation and intravenous general anesthesia were administered. Olympus S70 surgical hysteroscopy series equipment was used, and the perfusion medium was normal saline. Under the direct view of the hysteroscope, the shape of uterine cavity, and the position and degree of adhesion were observed. Adhesion tissue was separated via needle electrode and scar tissue by ring electrode, expanding the uterine cavity volume and removing scar tissue with care to protect the residual endometrium [19]. Successful separation of IUAs was defined as restoration of normal IU anatomy, display of bilateral uterine angle, no adhesions [20].
The operator was a senior doctor with a surgical team and extensive experience in the separation of IUAs. A physical barrier (intrauterine suitable balloon or Foley balloon or heart-shaped copper intrauterine device) was used to prevent the recurrence of IUAs [21]. All patients were given postoperative routine prevention of infection and hormone cycle therapy to promote endometrial growth [21]. Hormone therapy was started on the second day after the operation and consisted of estradiol valerate at 4 mg/d for 21 days, with the addition of dydrogesterone at 20 mg/d for the final 10 days of estrogen therapy. This was followed by a period of 7 days of no hormone therapy. The hormone therapy regimen was then repeated for an additional 2 months.
Research variables
Live birth is defined as delivery occurring at any time between 28 and 42 completed weeks of gestation. In this study, the diagnosis and score of IUAs referred to the American Fertility Society (AFS) standard of 1988. According to the nature, scope, and menstrual pattern of adhesions, the quantitative scores were: mild, 1-4 points; moderate, 5-8 points; or severe, 9-12 points.
For measurement of uterine volume and endometrial thickness, all patients were examined by vaginal color ultrasound (model: GE E8). Patients with menstruation were examined during the latter period of menstruation proliferation. Patients with amenorrhea were examined at no specific time. The length, width, and thickness of the uterine body were measured. The length was considered the distance from the bottom of the uterus to the internal orifice of cervix. The width and thickness of the uterine body were the transverse diameters of the coronal and sagittal sections, respectively. The volume of the uterus was calculated as length × width × thickness × 0.523, in cm3[22]. The depth of the uterine cavity is the distance from the bottom of the endometrial cavity to the external orifice of the cervix.
According to the modified vas method of Osada et al. [23], the patients’ previous menstruation was judged as either normal or amenorrhea. For those with normal previous menstruation, the changes in menstrual volume before and after surgery were compared. There were three types of postoperative menstrual patterns: normal menstrual volume, improvement of menstruation, and no improvement of menstruation. The causes of IUAs were divided into three categories: early pregnancy termination; middle and late pregnancy termination; or non-pregnancy related factors.
Follow-up observations
Data was collected from the electronic database and telephone return visits of inpatients, including demographic and clinical characteristics, surgical records and hysteroscopic pictures, pregnancy outcome, and related complications after treatment. The pregnancy outcome was followed up by telephone in February 2018.
Three months after the operation, menstruation was evaluated for improvement in the outpatient department, and hysteroscopy was performed [20]. A normal anatomy of the uterine cavity and no cause of infertility outside the uterine cavity suggested that natural conception was possible. If there were other infertility factors or natural pregnancy failure, patients were recommended to receive assisted reproductive technology.
Obstetric delivery mode and obstetric complications
A clinical pregnancy was defined as within the IU gestational sac. Live birth was ≥28 weeks of gestation, with a live birth obtained. Full term birth was considered ≥37 weeks of gestation, and premature birth <37 weeks of gestation. Loss of pregnancy was recorded as spontaneous abortion, fetal arrest, or stillbirth.
Statistical analysis
The final analysis comprised 374 patients (i.e., samples), including 23 variables, with live birth as the outcome variable and 22 other related factors as the predictive variables (Table 1). Because the positive and negative samples of the original data were not uniform (live production = 111, inactive production = 263, total sample size = 374), we used the ROSE (Randomly Over Sampling Examples) package in R language [24, 25] to deal with class imbalance. We under-sampled multiple data and oversampled less data, while keeping the new sample size the same as the original sample size. Moreover, we set the parameter p = 0.5 of the resample function in ROSE to keep the proportion of positive sample size approximatively equal to 0.5. Consequently, the number of samples with live and non-live production is more or less balanced (live production = 173, non-live production = 201, total sample size 374). The new data was used for further analyses.
Table 1
Attribute variables or categories of each influencing factor
|
Data type
|
Code
|
Live birth *
|
2 categories
|
No = no live birth, yes = live birth
|
Age
|
Continuity
|
|
Times of pregnancy
|
Frequency
|
|
Childbirth
|
2 categories
|
No = no, yes = yes
|
Chief complaint
|
3 categories
|
a = primary infertility, b = secondary infertility, c = recurrent abortion
|
Menstrual pattern
|
3 categories
|
a = less than 1/4, b = 1/4 ~ 1/2, c = more than 1/2
|
Etiology
|
3 categories
|
a = primary infertility, b = secondary infertility, c = recurrent pregnancy
|
Pregnancies loss
|
3 categories
|
z = 0 times, a = 1 time, b = more than or equal to 2 times
|
IU operations
|
Frequency
|
|
History of TCRA
|
2 categories
|
No = yes, yes = no
|
Endometrial thickness
|
3 categories
|
a = ≤3 mm, b = 4-6 mm, c = ≥7 mm
|
Uterine volume
|
Frequency
|
|
Preoperative AFS
|
Frequency
|
|
Degree of adhesion
|
2 categories
|
a = moderate, b = severe
|
Adhesion type
|
2 categories
|
a = mixed type, b = peripheral type
|
IU depth preop, mm
|
Frequency
|
|
Uterine horn closure preop, n
|
3 categories
|
z = 0 side, a = 1 side, b = 2 side
|
Tubal ostia preop
|
3 categories
|
z = 0 side, a = 1 side, b = 2 side
|
Isolation barrier
|
3 categories
|
a = balloon, b = uterine suitable balloon, c = IUD
|
Postoperative AFS
|
Frequency
|
|
Tubal ostia postoperative
|
3 categories
|
z = 0 side, a = 1 side, b = 2 side
|
Menstrual patterns postoperative
|
3 categories
|
z = improvement, a = no improvement, b = normal amount
|
IU depth postoperative
|
Continuity
|
|
* The only dependent variable; the remaining are independent.
Preop, preoperative; TCRA, transcervical resection of adhesions
|
The R language (www.r-project. ORG) random Forest package was used to rank the features of the independent variables (predictive variables) according to the mean decrease accuracy index. The first 10 most important feature variables were selected. To get the best predictive efficiency, various data partition schemes were tried. We finally choose 90% of the original data to build the decision tree model, a classification and regression tree (CART) algorithm, and 10% of the data to verify the model. The decision tree model was constructed using the R language rpart (Recursive Partitioning and Regression Trees) software package. Using the rpart package, the decision tree was constructed by CART algorithm based on gini index splitting criteria. The minimum sample size of the parent node was 30, the minimum sample size of the leaf node was 10, and the depth was set to 5.
The logistic regression model was constructed with the top 10 variables using the enter method (SPSS 23.0). P < 0.05 was statistically significant.