Skip to main content
Download PDF

Accurate prediction of mediolateral episiotomy risk during labor: development and verification of an artificial intelligence model

BMC Pregnancy and Childbirth volume 25, Article number: 370 (2025) Cite this article

Abstract

Objective

The study developed an intelligent online evaluation system for mediolateral episiotomy, which incorporated machine learning algorithms and integrated maternal physiological data collected during delivery.

Methods

In this study, a predictive model for mediolateral episiotomy was constructed first, and based on this, an early warning system using open-source R software was established. The physiological data of 1191 parturients who delivered at Deyang People's Hospital in western China from January 2022 to December 2022 were collected and divided into training set and test set according to a ratio of 8:2. The factors affecting mediolateral episiotomy were determined based on the expert consultation method. Six machine learning models, namely Logistic regression(LR), Support Vector Machine(SVM), K-Nearest Neighb(KNN), Random Forest (RF), Light Gradient Boosting Machine(LightGBM), and eXtreme Gradient Boosting‌(XGBoost) were constructed on this basis. The models’ performance was evaluated using accuracy, precision, recall, F1 value, and area under the receiver operating characteristic curve (AUC) measures. Additionally, a confusion matrix was employed to assess their performance across different categories. SHapley Additive exPlanation (SHAP) provided interpretability. The clinical external verification process focused on data collected from January to March 2023, and an intelligent online evaluation system for mediolateral episiotomy was developed.

Results

Twenty eight factors influencing mediolateral episiotomy were screened. The model evaluation results showed that the SVM model has the best prediction ability among the six models, with an accuracy of 0.793, a recall rate of 0.981, a precision rate of 0.790, and a F1 value of 0.875. The area under ROC curve of SVM was 0.882, The verification results showed that the prediction accuracy was 74% for undergoing mediolateral episiotomy and 93% for not undergoing it. SHAP analysis identified perineal elasticity, number of pregnancies, BMI, perineum edema, and age as top predictors. An early warning system for mediolateral episiotomy was successfully constructed, which can assist the clinical medical staff in decision-making by inputting the maternal data.

Conclusion

The early warning system for the risk of mediolateral episiotomy constructed in this study can accurately and rapidly utilize the physiological data of parturients during labor to predict the risk of mediolateral episiotomy in the third stage of labor.

Peer Review reports

Introduction

Episiotomy, a surgical incision of the perineum, is performed during the second stage of labor to make the vaginal opening larger for childbirth. There are 2 types of episiotomy incision: midline incision and mediolateral incision [1]. A timely and appropriate episiotomy can reduce obstruction of the perineum to the fetal head, enlarge the vaginal opening, accelerate the fetal delivery, and lessen perineal trauma in situations of fetal distress, shoulder dystocia, or forceps delivery [2]. However, routine episiotomy to prevent possible injury will increase the risk of adverse outcomes, including sever tears, edema, hematoma, local pain and infection at the incision site [3]. Long-term adverse consequences may include skin papilloma, vaginal prolapse, vaginal stenosis, pain or discomfort during sexual intercourse [4, 5]. In severe cases, it even causes pelvic floor dysfunction and have negative impacts on various aspects of the woman's life, including physical, psychological, and social adaptation.

With the changes in modern obstetric service models, the continuous updating in midwifery practices, and the deepening of people's understanding of the complications associated with episiotomy, many scholars have begun to question the rationality of routine episiotomy. In recent years, developed countries have reduced the use of episiotomy, with the rates of 2.3% in the United States, 1.3% in France, and around 10% in the United Kingdom and Australia [6, 7]. While developing countries like India and the Philippines still maintain episiotomy rates of over 50% [8]. In China, due to the different acceptance of the concept of restrictive episiotomy in recent years, the rate of episiotomy varies. Some small and medium-sized hospitals still perform routine episiotomy, while others have seen a decrease in rates. However, the majority of them fluctuate around 20–40% [9], which is significantly higher than the World Health Organization (WHO)'s recommendation to limit episiotomy rates within 10% [10]. Therefore, strict adherence to indications for episiotomy and reducing the episiotomy rate are crucial for improving the quality of obstetric services and the maternal experience of labor and delivery.

At present, there is a lack of clear and specific criteria and guidelines for episiotomy in clinical practice. As decision-makers for episiotomy, midwives often rely on their experience and subjective judgment to determine whether to perform the procedure. The indication for episiotomy mentioned in obstetrics and gynecology textbooks and guidelines is as follows [11]: "Tight perineum or fetal macrosomia with an anticipated unavoidable tear or urgent need for delivery due to maternal or fetal pathology." However, terms such as "rigid perineum," "tight perineum," and "anticipated unavoidable tear" introduce subjectivity into the decision-making process for episiotomy. Moreover, the indications are too general and lack specific quantitative index. Due to variations in clinical experience and knowledge levels, it is often challenging for junior midwives and basic medical units to determine the indications for episiotomy, leading to significant differences in episiotomy rates among different regions, hospitals, and even different healthcare providers within the same hospital. This further increases the risks associated with episiotomy.

Related work

Currently, there are limited prospective and systematic researches on the assessment system for the risk of mediolateral episiotomy, which mainly falls into two types. The first type is an assessment scale developed through expert discussions [12], which includes factors such as perineal conditions, maternal age, severe pregnancy complications, fetal macrosomia, fetal distress, and amniotic fluid contamination. However, in clinical practice, assignment of values to these items and the critical values for judgment are limited and subjective. And there are issues such as unreasonable distribution of evaluation scores and incomplete item settings. Moreover the paper version of the assessment scale has poor clinical practicality in the critical moment of maternal labor.

The second type involves using statistical methods to build a logistic regression model. For example, Guo Lin et al. built a logistic regression model by exploring factors influencing the decision for episiotomy, including 7 independent predictive factors such as perineal length, perineal elasticity and perineal edema [13]. The validation of the model shows that the area under the ROC curve is 0.932, the sensitivity is 0.883, and the specificity is 0.80, demonstrating good clinical applicability. However, this study had a small sample size and lacked prospective clinical validation, so the actual effectiveness of the model is unknown. The complex etiological mechanism and interplay of various risk factors affecting episiotomy make disease prediction challenging. The establishment of a risk prediction model must consider the integration, complexity, and dynamics of childbirth, as well as the nonlinear synergistic interactions among various risk factors. Traditional statistical methods are insufficient to meet these requirements, and there is a lack of intelligent and information-based early risk assessment systems to assist midwives in clinical decision-making.

At present, there is no AI-based risk decision system for episiotomy in clinical practice. The aim of this study is to utilize the powerful classification and prediction capabilities of artificial intelligence technology, specifically machine learning, to develop an early risk prediction model for episiotomy during childbirth. Based on this model, an intelligent and convenient clinical decision support system for episiotomy in obstetrics will be developed. The goal is to achieve precise prediction and early intervention, providing accurate and objective decision-making support for midwives in performing episiotomy.

Materials and methods

Participants

This study was designed as a prospective cohort study. Clinical medical records of women who underwent vaginal delivery in the Obstetrics Department of Deyang People's Hospital in Sichuan Province, China, from January 2022 to December 2022 were collected. Clinical trial number: not applicable.

The inclusion criteria were as follows

  1. (1)

    Full-term pregnancy

  2. (2)

    Vaginal delivery

  3. (3)

    Vertex presentation

  4. (4)

    Singleton pregnancy

  5. (5)

    In-hospital delivery

The exclusion criteria were as follows

  1. (1)

    Presence of contraindications to vaginal delivery.

  2. (2)

    Cesarean section.

  3. (3)

    Women with mental illness or disturbance of consciousness who were unable to communicate effectively.

Outcome definition

The women who underwent mediolateral episiotomy during the second stage of vaginal delivery will be assigned to the experimental group. On the other hand, the women who did not undergo mediolateral episiotomy during the second stage of vaginal delivery will be assigned to the control group.

Influencing factors in the study

Preliminary determination of factors influencing mediolateral episiotomy

Now all obstetrics and gynecology textbooks provide only a general framework for the indications of episiotomy, lacking specific quantitative index. Therefore, a comprehensive analysis of factors related to mediolateral episiotomy during the delivery process is of great significance in midwifery. The researchers conducted a literature review and preliminarily determined the influencing factors based on expert group meetings and clinical experience, resulting in the initial draft of the Questionnaire for Factors Influencing Mediolateral Episiotomy, which includes 41 indicators:

  1. (1)

    General information: age, gestational age, parity, history of stillbirth, obesity, uterus scarring, pregnancy complications, twin or multiple pregnancy, pendulous abdomen.

  2. (2)

    Conditions during delivery: perineal length, perineal elasticity, perineal color, perineal thickness, perineal edema, perineal skin tear, hymenal laceration, vaginal laceration, uterine contractions, intrapartum fever, shoulder dystocia, assisted breech delivery, instrumental delivery, use of labor analgesia, duration of first stage of labor, duration of second stage of labor, duration of third stage of labor, total duration of labor, characteristics of amniotic fluid, estimated fetal weight, positive across pubic sign, premature infants, post-term infants, late deceleration, severe variable deceleration, abnormal fetal position, fetal malformation.

  3. (3)

    Other conditions: maternal cooperation, working years of midwives, abdominal pressure, waiting for labor in a free position, delivering in a free position.

Determination of factors influencing mediolateral episiotomy using the Delphi expert consultation method

The Delphi expert consultation method was used to revise the preliminary version of the Questionnaire on Factors Influencing Mediolateral Episiotomy. The final version of the questionnaire was developed based on this process.

  1. (1)

    Selection of Experts

The inclusion criteria for consultant experts:

  • ➀ Bachelor's degree or above;

  • ➁ Engaged in clinical and nursing work related to obstetrics;

  • ➂ Having a minimum of 15 years of experience in the field;

  • ➃ Holding a vice senior title or above;

  • ➄ Voluntary participation in this study.

  1. (2)

    Delphi consultation process

In the first round, experts were introduced to the research background, objectives, tasks, and instructions for completing the questionnaire. Their informed consent was obtained, and they were requested to provide their responses within 1 week. After the questionnaires were collected, the opinions of the experts were organized and summarized, and the indicator scores were statistically analyzed. The research team then screened and revised the indicators based on the indicator selection criteria" and prepared another questionnaire for the second round of expert consultation. The results and modified parts from the first round were explained to the experts. Finally, based on the results of the second round of consultation, the Questionnaire on Factors Influencing Mediolateral Episiotomy was finalized.

  1. (3)

    Evaluation of expert consultation results

The evaluation of expert consultation results mainly included expert positive coefficient, expert authority level, the level of coordination and concentration of experts’ opinions, and validity of the scale.

  1. (4)

    Indicator selection criteria

During the correspondence process, experts were asked to provide their opinions regarding the modification, addition, or deletion of indicators. After collecting the questionnaires, the research team combined the statistical results and expert opinions to screen the indicators at all levels. Indicators that met the criteria of mean value > 3.50 and CV < 0.25 were retained to determine the influencing factors in this study.

Data acquisition and cleaning

A data collection team consisting of four members was formed, and they underwent unified training. The Questionnaire on Factors Influencing Mediolateral Episiotomy determined after expert consultation, was used to collect relevant data from pregnant women who delivered in the obstetrics department from January 2022 to December 2022.

Data preprocessing: among the variables, 17 variables such as age and height were numerical variables. The cervical maturity was assessed by Bishop score and recorded as numerical variables. Variables such as labor analgesia, fetal membrane status, and amniotic fluid contamination are binary variables and were entered as 1/0 assignments.

Missing datas handling: there were missing or incomplete data for the first stage of labor, estimated fetal weight, and weight. The missing data for the first stage of labor were due to incomplete formatting, which was addressed by standardizing the format. Median imputation was used to fill in the missing data for estimated fetal weight and weight.

Establishing the prediction model based on machine learning algorithms

In this study, the data were divided into training set and test set in a 8:2 ratio. The training set had 952 cases and the cross-validation set had 239 cases. Logistic regression (LR), Support Vector Machine (SVM), K-Nearest Neighb (KNN), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and eXtreme Gradient Boosting‌ (XGBoost) were used to build a risk prediction model for episiotomy during childbirth. The parameters of the model are set as shown in Table 1:

Table 1 The six model parameter settings
Full size table

Evaluation of the prediction model

The performance of the prediction model was evaluated by common evaluation metrics used in machine learning such as accuracy, precision, recall, and F1 score. The confusion matrix is used to observe the performance of the model across different class. It is a standard format for accuracy evaluation in machine learning, providing a statistical representation of the prediction and actual conditions, as shown in Table 2.

Table 2 Confusion matrix
Full size table

Clinical prospective verification of predictive model

After comparing the six models, the best-performing predictive model was selected for clinical validation. The data for validation was obtained from the same hospital, involving relevant information of pregnant women who underwent vaginal delivery between January and February 2023. The differences between the actual outcomes of episiotomy in clinical practice and the decision results of the predictive model were compared and expressed as accuracy (%).

Statistical analysis

SPSS22.0 was used for data analysis. Continuous variables were described as "mean ± standard deviation" or "median (interquartile range)" depending on distribution, while categorical variables were presented as percentages. Group comparisons were performed using t-tests, one-way ANOVA, rank-sum tests, or chi-square tests, as appropriate. Additionally, different types of machine learning algorithms were established and validated using MATLAB.

Result

Baseline characteristics

From January to December 2022, there were a total of 3,141 delivery cases in the Department of Obstetrics of Deyang People's Hospital. Among them, 1,482 cases were vaginal deliveries, and 1,949 cases did not meet the inclusion criteria (among which 1,679 were delivered via cesarean section, 270 cases had missing data, and 1 case was delivered out of hospital). A total of 1,191 cases were ultimately included in this study. There were 300 cases in the mediolateral episiotomy group and 891 cases in the non-mediolateral episiotomy group. The flow chart of inclusion process is shown in Fig. 1, all patients’ baseline clinical features analysis shown in Table 3.

Fig. 1
figure 1

A flow chart for study population selection

Full size image
Table 3 All patients’ baseline clinical features analysis
Full size table

Results of expert consultation

Basic information of the experts

The 15 selected experts all come from tertiary hospitals or universities, specializing in obstetric nursing, obstetrics clinical medicine, and obstetric education. There are 10 experts in obstetric nursing, 4 experts in obstetrics clinical medicine, and 1 expert in obstetric education. All experts hold vice senior titles or above, with working experience ranging from 17 to 38 years and an average of 30.70 years of experience.

Expert positive coefficients

In the first round of expert consultation, 15 experts were contacted, and 15 valid questionnaires were received, with a positive coefficient of 100%. In the second round, 15 experts were contacted again, and 15 valid questionnaires were received, with a positive coefficient of 100%.

Expert authority coefficients

In this study, Ca = 0.909 and Cs = 0.853, therefore Cr = 0.881. It is generally believed that when Cr ≥ 0.7, the expert authority is considered high, and the consultation results are more reliable.

Coordination level of experts’ opinions

The Kendall’s ω of the first and the second rounds of expert consultation were 0.312 and 0.487 respectively. The Kendall’s ω was significant increased after two rounds of expert consultation. The CV in the first round ranged from 0.00 to 0.35, while in the second round, it ranged from 0.00 to 0.21. This indicates that after expert consultation, there was greater consensus and better coordination of expert opinions.

Concentration level of experts’ opinions

In the first round of consultation, the mean scores of all indicators ranged from 2.87 to 5.00, with a full score ratio of 0.07 to 1.00. In the second round, the mean scores of all indicators ranged from 3.73 to 5.00, with a full score ratio of 0.18 to 1.00. There is an improvement in the scores.

Content validity

In the first round of expert consultation, the scale-level content validity index based on the average method (S-CVI/Ave) was 0.87, the scale-level content validity index based on the universal agreement method (S-CVI/UA) was 0.44, and the item-level content validity index (I-CVI) ranged from 0.20 to 1.00. In the second round, the S-CVI/Ave was 0.87, the S-CVI/UA was 0.56, and the I-CVI ranged from 0.55 to 1.00, indicating that the questionnaire has good content validity.

Results of the factors influencing mediolateral episiotomy

The criteria for item selection in this study were to retain indicators with mean > 3.50 and CV < 0.25. The changes in items after the first round of consultation are as follows:

  • Deleted Indicators:

    • Secondary indicators: The mode of delivery was deleted due to the I-CVI not meeting the standard. Experts suggested that the gestational age be deleted as it overlapped with the tertiary indicator of gestational age.

    • Tertiary indicators:

      1. 1)

        In the case history data, indicators such as uterus scarring and pendulous abdomen were deleted as their mean and CV did not meet the criteria.

      2. 2)

        In the perineal condition data, indicators such as perineal color, hymenal laceration, and vaginal laceration were deleted as their mean and CV did not meet the criteria.

      3. 3)

        In the labor process data, indicators such as epidural analgesia and intrapartum fever were deleted as their mean and CV did not meet the criteria.

      4. 4)

        In the fetal factors, indicators such as positive across pubic sign, preterm infants, post-term infants, and fetal malformation were deleted as their mean and CV did not meet the criteria.

      5. 5)

        In the midwife factors, indicators such as more than 5 years of midwifery experience, experience judgement, in demonstration, operation by internship students, abdominal pressure, waiting for labor in a free position, and delivering in a free position were deleted as their mean and CV did not meet the criteria. A total of 18 indicators were deleted.

  • Modified Indicators:

    1. 1)

      Based on expert opinions and group discussion, the secondary indicators of teaching and working years were merged into the operator's working years. Additionally, the indicator of obesity was modified to BMI. After the second round of consultation, all items met the inclusion criteria, and expert opinions tended to be consistent, with no further modification suggestions raised. Finally, the Questionnaire on Factors Influencing Mediolateral Episiotomy was formed, consisting of 28 influencing factors, as shown in Table 4.

Table 4 The questionnaire on factors influencing mediolateral episiotomy
Full size table

Ranking of variable characteristics and weights

All the factors mentioned above were included in the model. The distribution of the variables incorporated in the SVM prediction model is shown in Fig. 2. Perineal elasticity, perineal elasticity, number of pregnancies, BMI, and perineal edema were identified as the top five predictors of episiotomy risk.

Fig. 2
figure 2

Ranking of Variable Importance

Full size image

SHAP analysis

SHAP analysis was used to identify the key factors influencing the risk of mediolateral episiotomy and to clarify their specific contributions. As shown in Fig. 3, perineal elasticity was the most important predictor of episiotomy risk. Poor perineal elasticity is strongly associated with an increased risk of episiotomy. Following perineal elasticity, number of pregnancies, BMI, and perineal edema were also identified as key predictors, indicating that the patient's obstetric history and perineal conditions play an important role in the model's risk prediction. In addition, patient age had a significant effect on the risk of episiotomy, with a higher likelihood of episiotomy in women with advanced maternal age.

Fig. 3
figure 3

Feature dependency and interaction effects

Full size image

The waterfall plot can visually show the influence of different features on the predicted value, which helps us understand which factors are most important for the prediction results. Figure 4 presents an individual SHAP analysis for a specific patient. As an example, one patient was assigned a predicted risk score of −2.40, suggesting a minimal likelihood of undergoing an episiotomy. A more thorough analysis revealed that perineal elasticity (0 = poor elasticity) played the most significant role in decreasing the risk score by 1.63, emphasizing that poor perineal elasticity is associated with a higher risk of lateral episiotomy. Furthermore, the number of pregnancies (1 time) and the thickness of the perineum (2 cm) also contributed positively to lowering the risk, with contributions of 0.707 and 0.356 respectively. This indicates that a higher number of pregnancies and a thicker perineum are associated with a decreased risk of lateral episiotomy. Conversely, fetal birth weight (3000 g) and the presence of prenatal perineal skin laceration (1 = perineal skin laceration) incremented the risk factors by −0.585 and −0.301 respectively. This indicates that patients with higher fetal birth weight and those who have prenatal perineal skin tears are at greater risk of episiotomy.

Fig. 4
figure 4

SHAP waterfall plot of a single sample

Full size image

Model performance

SVM model has the best prediction ability among the six models, with an accuracy of 0.793, a recall rate of 0.981, a precision rate of 0.790, and a F1 value of 0.875. Similarly, the RF model showed comparable results, with an accuracy of 0.793, the recall of 0.950, an F1 score of 0.872, as shown in Table 5.

Table 5 The results were predicted by six models
Full size table

External verification results of the prediction model

Clinical data of pregnant women who underwent vaginal delivery in the same hospital from January to February 2023 were collected for external clinical verification of the model. Among them, 32 cases underwent mediolateral episiotomy, while 125 cases did not. The verification results showed that the prediction accuracy was 74% for undergoing mediolateral episiotomy and 93% for not undergoing it. The results of the confusion matrix are shown in Table 6. The area under ROC curve of SVM was 0.882, which showed great prediction performance, as shown in Fig. 5.

Fig. 5
figure 5

The area under ROC curve of SVM model

Full size image
Table 6 External verification results of prediction model
Full size table

System development results

The system is built on the BS architecture. When it is deployed on a server, users can directly access the system through the browser. The system framework is developed based on Spring boot. The data analysis, data modeling, and prediction parts are developed using the Hadoop + Spark distributed system.

An example of using the tool is depicted in Figs. 6 and 7. A user needs to enter the variable values when weaning, leaving missing values blank and clicking the predict button. and the result will be shown to the user. According to the warning prompts of the system, the doctor made the decision of whether to perform episiotomy or not, combined with the actual situation of the patient and the progress of labor.

Fig. 6
figure 6

Interface diagram of the early warning system of mediolateral episiotomy

Full size image
Fig. 7
figure 7

Interface diagram of the early warning system of mediolateral episiotomy Note: This figure shows the prediction and results page of the early warning system of mediolateral episiotomy. Step 1- System check and start: Click on the computer to start the episiotomy risk warning system and enter the main interface. Step 2-Patient information input: Input the patient's basic information, such as name, age, gestational age, etc., on the system interface. Step 3- Data entry and analysis: The system analyzed and processed the data according to preset algorithms and models to evaluate the individual risk of episiotomy. Step 4-Warning prompt: After the data input is completed, click the "prediction" button, and the system generates the warning prompt of episiotomy according to the analysis results

Full size image

Discussion

Currently, the development of clinical decision support systems based on artificial intelligence (AI) technology has been widely implemented in the clinic, such as assisting in diagnosis and treatment decision-making. For example, the Development, Validation and Application Value of the AI Clinical Decision Support System for Liver Cancer [14] and the Study of Clinical Decision Support System for Pressure Injury [15]. These studies all input the patients’ clinical examination data and imaging data into the system, and the diagnosis and treatment plan will be given immediately after calculation.

In this study, an early risk prediction model for mediolateral episiotomy during labor was developed based on machine learning algorithms, and on this basis, an intelligent and convenient clinical decision support system for mediolateral episiotomy was developed. Clinical medical staff can input relevant data of pregnant women into this system before childbirth. The system can then provide direct recommendations on whether to perform a mediolateral episiotomy. Clinical validation has shown that the system performs well with high prediction accuracy, providing accurate and objective decision-making support for midwives to perform episiotomy. The design of the early warning system of mediolateral episiotomy has fully considered the limitation of computing resources in different clinical Settings. The system uses a lightweight front and back-end architecture and a responsive web design on the front end to ensure smooth performance across a variety of devices and browsers. The back end uses an efficient cloud computing platform, which can dynamically adjust computing resources according to the number of users and the number of requests to ensure the stability and response speed of the system. In addition, data security is one of the core considerations in the design of the episiotomy warning system. The system adopts multi-level data security measures to ensure patient privacy and data security. First, the system encrypts the transmitted and stored data to prevent data leakage. In addition, the system has established a strict user authentication and authorization mechanism, and regularly carries out security audit and vulnerability scanning to find and repair security risks in time. Through multiple measures, data security is guaranteed to meet the needs of different medical units.

The pathogenesis of perineotomy and the interactions among its risk factors are complex, thereby complicating disease prediction. When establishing a population-based disease risk warning model for this condition, it is essential to consider the holistic, intricate, and dynamic nature of maternal delivery, as well as the nonlinear synergistic effects among the risk factors. Traditional logistic regression models analyze the potential impacts of single or multiple variables on lateral perineal incision by constructing equations based on regression coefficients. In contrast, machine learning employs more advanced algorithms to train classifiers, offering superior data processing capabilities and enabling precise analysis of large and complex clinical datasets. In this study, the SVM algorithm demonstrated the best performance. By harnessing the sophisticated kernel trick, SVM transform input data into a high-dimensional feature space through a nonlinear mapping, facilitating the identification of an optimal separating hyperplane within this transformed domain. This distinctive capability allows SVM to adeptly address nonlinear patterns that might escape detection by traditional logistic regression models, which are inherently limited to linear relationships. This nonlinear mapping capability enables SVM to deal with the nonlinear relationships among the factors affecting the complex episiotomy. Moreover, SVM adhere to the fundamental principle of maximizing the margin, striving to identify a separating hyperplane that maximizes the distance between data points belonging to different classes. This margin maximization not only enhances the model's robustness but also improves its generalization capability, ensuring that the learned decision boundary generalizes well to unseen data.

Considering the significant regional variations in lateral episiotomy rates, which are influenced by both cultural factors and medical practices, the early warning model was developed through expert consultation to incorporate a broad spectrum of variables. This approach ensures the model's adaptability and generalizability. By implementing standardized data processing and correction strategies, the model can be fine-tuned for specific contexts, whether in regions with high or low episiotomy rates. Consequently, it effectively accommodates differences in medical practices across diverse cultural settings, providing a robust scientific foundation and decision support for global episiotomy management.

Among the many variables used to assess the risk of episiotomy, perineal elasticity was identified as the most influential factor. Good perineal elasticity plays a key role in the process of delivery, which can effectively reduce labor resistance and pain, and reduce the risk of perineal tears. Studies have pointed out that insufficient perineal elasticity significantly increases the possibility of episiotomy: when the maternal perineum is not flexible, the fetal head through the vaginal opening may cause perineal laceration. To avoid more severe tears (e.g., grade III or IV), physicians or midwives often perform episiotomy on high-risk women. Therefore, perineal malelasticity is seen as a key factor that increases the risk of episiotomy. In the present study, perineal elasticity contributed the most to the warning model and was consistent with the clinical reality, further confirming the strong link between perineal elasticity and lateral episiotomy. This finding suggests that health care providers need to continuously monitor maternal perineal elasticity and take appropriate measures to enhance its elasticity during labor.

This finding has important guiding implications for clinical practice. Firstly, perineal elasticity assessment should become a routine item in prenatal care, using standardized tools and methods to ensure the accuracy and consistency of the assessment results. Second, medical staff should tailor the delivery plan for the mother based on the results of the perineal elasticity assessment. For parturients with poor perineal elasticity, more aggressive interventions can be considered to avoid severe perineal tears. In addition, pregnant women should be educated about the importance of perineal muscle exercise and encouraged to perform perineal muscle exercise during pregnancy and postpartum to improve the flexibility and toughness of the perineum.

The limitation of this study lies in the fact that the model was developed based on data from a single center, which may limit its applicability in a broader context. Given the diversity in statistical characteristics of patient populations and medical conditions across different regions, it is necessary for future research to conduct multicenter trials to validate and establish the applicability of the model in various settings.

Conclusion

In this study, the data are modeled based on machine learning algorithm, in which the SVM model has the best performance, high prediction accuracy, and stability, and could be utilized to predict the outcome of mediolateral episiotomy.

Research prospect

Selective mediolateral episiotomy is closely related to subjective and objective factors such as clinical experience, judgment, and midwifery techniques of midwives. These factors may introduce certain biases when constructing the mediolateral episiotomy prediction model. So, it is necessary to widely apply the model in clinical practice to gather a larger sample of clinical data, and to further test its rationality and accuracy through practical operation.

Data availability

We have uploaded the raw data to the submission system. In addition, the datasets and system access right of the study available from the corresponding author on reasonable request.

References

  1. Ghulmiyyah L, Sinno S, Mirza F, Finianos E, Nassar AH. Episiotomy: history, present and future - a review. J Matern Fetal Neonatal Med. 2022;35(7):1386–91.

    Article  CAS  PubMed  Google Scholar 

  2. Jiang H, Qian X, Carroli G, Garner P. Selective versus routine use of episiotomy for vaginal birth. Cochrane Database Syst Rev. 2017;2(2):CD000081.

    PubMed  Google Scholar 

  3. Jiang H, Qian X, Carroli G, et al. Selective versus routinr use of episiotomy for vaginal bith[J]. Cochrane Database Syst Rev. 2017;2:CD000081.

    PubMed  Google Scholar 

  4. Zang Y, Lu H, Zhang H, et al. Benefits and risks of upright positions during the second stage of labour: An overview of systematic reviews. Int J Nurs Stud. 2021;114: 103812.

    Article  PubMed  Google Scholar 

  5. Seijmonsbergen SA, Thompson S, Feijen DJ, et al. Understanding the perspectives and values of midwives, obstetricians and obstetric registrars regarding episiotomy: qualitative interview study. BMJ Open. 2021;11(1): e037536.

    Article  Google Scholar 

  6. Brismar Wendel S, Liu C, Stephansson O. The association between episiotomy or OASIS at vacuum extraction in nulliparous women and subsequent prelabor cesarean delivery: A nationwide observational study. Acta Obstet Gynecol Scand. 2023;102(3):378–88.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Levaillant M, Loury C, Venara A, Hamel-Broza JF, Legendre G. Is there still an indication for episiotomy? Results from a French national database analysis. Int J Gynaecol Obstet. 2023;160(3):880–5.

    Article  PubMed  Google Scholar 

  8. Baruch Y, Gold R, Eisenberg H, Amir H, Reicher L, Yogev Y, Groutz A. High Incidence of Obstetric Anal Sphincter Injuries among Immigrant Women of Asian Ethnicity. J Clin Med. 2023;12(3):1044.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Li L. Clinical application status and research progress of episiotomy [J]. Shanghai Nursing. 2016;16(01):64–7.

    Google Scholar 

  10. Marty N, Verspyck E. Déchirures périnéales obstétricales et épisiotomie: aspects techniques. RPC prévention et protection périnéale en obstétrique CNGOF [Perineal tears and episiotomy: Surgical procedure - CNGOF perineal prevention and protection in obstetrics guidelines]. Gynecol Obstet Fertil Senol. 2018;46(12):948–67 French.

    CAS  PubMed  Google Scholar 

  11. Xiuxia Z. Obstetrics and Gynecology Nursing [M]. 5th ed. Beijing: People’s Medical Publishing House; 2014. p. 212.

    Google Scholar 

  12. Liu X, Wan L. Observation of the effect of perineal evaluation scale on guidance of perineal incision in parturient women [J]. J Nurs. 2020;27(12):60–4.

    CAS  Google Scholar 

  13. Guo L, Ding Y, Zhang Z, et al. Construction of influencing factors prediction model for perineotomy decision during childbirth [J]. Chin J Nurs. 2019;54(10):1469–74.

    Google Scholar 

  14. Choi GH, Yun J, Choi J, Lee D, Shim JH, Lee HC, Chung YH, Lee YS, Park B, Kim N, Kim KM. Development of machine learning-based clinical decision support system for hepatocellular carcinoma. Sci Rep. 2020;10(1):14855.

  15. Tschannen D, Anderson C. The pressure injury predictive model: A framework for hospital-acquired pressure injuries. J Clin Nurs. 2020;29(7–8):1398–1421.

Download references

Funding

The study were supported by National Natural Science Foundation of China (52403043, U22A20334), Sichuan Science and Technology Programs (2025YFHZ0246), Sichuan University Innovative Research Project (2023SCUH0032), and Sichuan Health Information Society(2023026).

Author information

Authors and Affiliations

  1. Sichuan University-The Hong Kong Polytechnic University Institute for Disaster Management and Reconstruction, Chengdu, 610041, China

    Tingting Hu

  2. People’s Hospital of Deyang City, Deyang, 618000, China

    Tingting Hu, Lin He, Xiaoli Zhong & Junjie Chen

  3. Medicine and Engineering Interdisciplinary Research Laboratory of Nursing & Materials, West China Hospital, Sichuan University/West China School of Nursing, Sichuan University, Chengdu, 610041, China

    Tingting Hu, Yanting Han & Ka Li

  4. Chengdu Jincheng College, Chengdu, Sichuan, 610000, China

    Liheng Zhao

  5. Chengdu University of Technology, Chengdu, Sichuan, 610000, China

    Xueling Zhao

  6. Department of Gastroenterology, Affiliated Tumor Hospital of Xinjiang Medical University, Urumqi, 830000, China

    Zhe Yin

Authors
  1. Tingting Hu
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Liheng Zhao
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Xueling Zhao
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Lin He
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Xiaoli Zhong
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Zhe Yin
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Junjie Chen
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Yanting Han
    View author publications

    You can also search for this author inPubMed Google Scholar

  9. Ka Li
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

H.T.T. contributed to the data collection, wrote the first draft, and revised the manuscript. Z.L.H. contributed to performing the data analysis. Z.X.L. contributed to the language polishing. H.L. and Z.X.L contributed to the revised manuscript. Y.Z. contributed to the data collection and coding C.J.J. H.Y.T.and L.K.is the principal investigator, who contributed to the study idea and design and the subsequent drafts.

Corresponding authors

Correspondence to Junjie Chen, Yanting Han or Ka Li.

Ethics declarations

Ethics approval and consent to participate

The experimental protocols were approved by the Ethics Committee of People's Hospital of Deyang City (2021–04-065-K1). We confirmed that all methods were carried out in accordance with the relevant guidelines, and written informed consent was obtained from all participants included in the study. This study was conducted by the Declaration of Helsinki.

Consent for publication

Consent for publication was obtained the participants.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Hu, T., Zhao, L., Zhao, X. et al. Accurate prediction of mediolateral episiotomy risk during labor: development and verification of an artificial intelligence model. BMC Pregnancy Childbirth 25, 370 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12884-025-07441-2

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12884-025-07441-2

Keywords

BMC Pregnancy and Childbirth

ISSN: 1471-2393

Contact us