Random Forest Model for Labor Induction in Pregnant Women With Hypertensive Disorders Using a Cervical Double Balloon.

CONTEXT

Women with hypertensive disorders of pregnancy often need to have labor induced. The use of cervical double balloons to trigger cervical ripening, combined with the use of oxytocin, has been widely used for labor induction in recent years. In the evaluation of factors affecting the success rate of labor induction, previous predictive models have been limited to use of linear correlation, which simplifies the complex relationship between a large number of variables.

OBJECTIVE

The study intended to retrospectively analyze the factors influencing the outcomes of cervical dilatation using a cervical double balloon in the induction of labor for pregnant women with hypertensive disorders and to establish a predictive model based on the random forest (RF) method that is able to manage multifeatured data, provide fast training speeds, offer high predictive accuracy, and analyze the impact of various features.

DESIGN

The research team performed a retrospective analysis of data.

SETTING

The study took place at the Fujian Provincial Maternity and Child Health Hospital at the Affiliated Hospital of Fujian Medical University in Fuzhou, China.

PARTICIPANTS

Participants were 201 women in late pregnancy who came to the hospital for delivery between January 2014 and December 2018, who had hypertensive disorders of pregnancy, and for whom doctors induced labor using a cervical double balloon.

INTERVENTION

The research team divided participants into an intervention group, who had a successful induced labor, and a control group, who had a failed induced labor.

OUTCOME MEASURES

The research team analyzed the medical records of the groups using the RF method of ensemble learning and the multifactor logical regression method. The team used the receiver operating characteristic curve (ROC) to evaluate the working efficiency of the two models. The RF prediction model examined the factors influencing induced labor: the pregnancy method, the ultrasound EFW, the amniotic fluid index (AFI), the serum LDH level of the pregnant women, the placental volume, the cervical Bishop score before use of the balloon, the duration of the balloon's use, and the hours of use of oxytocin after balloon removal.

RESULTS

The success rate for induced labor with use of a cervical double balloon for women with hypertensive disorders during pregnancy was 77.18%. The incidence of postpartum hemorrhage was 4.7% and of fetal distress was 12.7%. The most important 10 features were: (1) hours of oxytocin use, (2) fetal weight, (3) placental volume, (4) AFI, (5) LDH, (6) BMI, (7) the Bishop score before use of the COOK balloon, (8) duration of the balloon's use, (9) pregnancy method, and (10) weight gain during pregnancy. The area under the ROC curve for successful induction for the RF model was 0.983. The multivariate logistic regression model based on RF showed that multiple births, high cervical Bishop scores before labor induction, less time for use of oxytocin after balloon removal, and a small placental volume were independent risk factors, with the area under the ROC curve for successful induction being 0.918.

CONCLUSIONS

Medical practitioners can use the cervical double balloon effectively for the induction of labor for women with hypertensive disorders during the third trimester of pregnancy, and the prediction model for induction of labor based on RF had a good working efficiency.