Abstract
Background
Women who are pregnant again after a prior cesarean section are faced with the choice between a vaginal trial and another cesarean section. Vaginal delivery is safer for mothers and babies, but face the risk of trial labor failure. Predictive models can evaluate the success rate of vaginal trial labor after cesarean section, which will help obstetricians and pregnant women choose the appropriate delivery method.
Objective
To review the existing prediction models of vaginal delivery after cesaean.
Methods
Seven databases, including CNKI, Wanfang Data, Chinese Science and Technology Periodical Database, China Biomedical Literature Database, PubMed, Embase, and Web of Science, were searched for studies on the predictive model of VBAC from inception to July 20, 2022. Two researchers independently screened the literature and extracted the data. The risk of bias and applicability of the included researches was evaluated using the Prediction model Risk of Bias Assessment Tool.
Results
Twenty-six studies that covered 26 models were included. The overall property of the included models was good, but validation of the included models was insufficient. The methodological quality of the included studies was generally low, with 3 studies rated as having a low risk of bias and 23 studies rated as having a high risk of bias. The main predictors in the models were the Bishop score, history of vaginal delivery, neonatal weight, maternal age, and BMI.
Conclusions
Although a variety of prediction models have been developed globally, the methodology of these studies has limitations and the models have not been adequately validated. In the future, more prospective and high-quality research is needed to develop visual models to serve clinical work more effectively and conveniently. Obstetricians or midwives could use predictive models to help a woman choose the right delivery method.
Keywords: Prediction model; Scoping review; Trial of labor after cesarean delivery; Vaginal birth after cesarean, VBAC; TOLAC
Introduction
The cesarean section rate has been increasing worldwide over the current decade [4]. Data showed that the cesarean delivery rate increased from 5% to 30–32% over the last 10 years in America [2]. A large Chinese study showed that the cesarean section rate rose from 28.8% in 2008 to 34.9% in 2014 [22]. The latest available data shows that 21.1% of women gave birth by cesarean worldwide [4]. The increasing cesarean section rates have led to an increase in maternal mortality and morbidity [2]. Affected by the dictum ”once a cesarean always a cesarean”, women with a prior cesarean section often choose to have another cesarean section in the next delivery, which in turn contributes to the high cesarean section rates.
Women with a prior cesarean section once become pregnant again face a choice between a trial of labor after cesarean (TOLAC) and an elective repeat cesarean delivery (ERCD). The vaginal birth after cesarean (VBAC) guidelines of the American College of Obstetricians and Gynecologists recommend a TOLAC for women with a prior cesarean section who become pregnant again [16]. Outcomes of a TOLAC include VBAC and emergency cesarean delivery after a failed trial of labor. A successful TOLAC helps mothers avoid abdominal surgery, leading to a lower incidence of bleeding, thromboembolism, and infection and a shorter recovery period than those undergoing an ERCD [16]. However, a TOLAC has the risk of failure. There are more complications and possibly uterine rupture when the mother undergoes emergency cesarean delivery after a failed TOLAC. The inability to predict the probability of success for a trial of labor has become a barrier for obstetricians and mothers in choosing a TOLAC. If the probability of VBAC can be accurately predicted, it will help obstetricians screen suitable candidates for a TOLAC, help mothers establish confidence in a vaginal trial of labor, avoid complications caused by a second cesarean section, and reduce the overall cesarean section rate. VBAC predictive models have been researched in many developed countries and various predictive models have been developed and validated in different populations [7, 14, 31]. China implemented the two-child policy in 2016, allowing a couple to have two children. In 2021, the three-child policy was introduced, permitting a couple to have three children. Prior to these changes, China had been adhering to a one-child policy for several decades, encouraging couples to have only one child as a response to rapid population growth. In recent years, with the launch of the “Two-Child Policy” and “Three-Child Policy”, there has been increasing research on developing VBAC predictive models in China [8, 10, 33]. However, the factors and model performance of the predictive models of various studies are quite different, and there is a lack of external validation. In this review, the construction, factors, and performance of relevant predictive models worldwide were reviewed to guide the selection of appropriate predictive models in a clinical setting and related research in the future.
Methods
Search strategy
We searched CNKI, Wanfang Data, Chinese Science and Technology Journal Database, China Biomedical Literature Database, PubMed, Embase, and Web of Science for publication from inception to July 20, 2022. We used a search method that involved combining MeSH Terms and free-language terms. The search formula was as follows: (“vaginal birth after cesarean” [MeSH Terms] OR “trial of labor after cesarean” [Title/Abstract] OR “VBAC” [Title/Abstract] OR “TOLAC” [Title /Abstract]) AND (“prediction” [Title/Abstract]). Subsequently, a manual search of references of the included research papers was performed.
Inclusion and exclusion criteria for the literature
The inclusion criteria were as follows: 1) the research subjects underwent a TOLAC, the gestational age was more than 37 weeks, there was a single live fetus; the subjects had a previous history of cesarean section, and the re-pregnancy was more than 2 years postoperatively; 2) the study was to construct or to validate a predictive model of VBAC section or to test the predictive power of a model on VBAC section; and 3) the article was the original study for model construction or validation, such as a cohort study, case-control studies, and cross-sectional studies.
The exclusion criteria were as follows: 1) non-Chinese or non-English literature; 2) unavailable full text; 3) low-quality literature; and 4) reviews.
Literature screening
The citations obtained from the search were imported into the Endnote X9 software to check for bibliographic duplicates. After eliminating the duplicates, 2 post-graduate researchers trained in evidence-based nursing practice independently conducted preliminary screening by reading titles and abstracts according to the research topic. After initial screening, they read the full text and re-screened the research according to the inclusion and exclusion criteria. If there was any disagreement during the screening process, a third researcher was consulted, and these disagreements were resolved by consensus. Finally, the research that met the criteria was determined.
Data extraction and analysis
Two researchers independently extracted data from the chosen literature. In the event of a disagreement, a third researcher was consulted, and these disagreements were resolved by consensus. Data extracted are as follows: author, year, country, study site, sample size, model construction, validation method, model predictors, presentation format, performance, and others.
Evaluation of methodological quality
Two researchers independently assessed the methodological quality of the included researches according to the Prediction Model Risk of Bias Assessment Tool (PROBAST) [38], including the risk of bias and applicability assessments. The PROBAST list includes four evaluation areas, including research objects, predictors, results, and analyses, with 20 questions. In this study, all areas of the included researches were evaluated for risk of bias, and the first 3 were evaluated for applicability. The risk of bias evaluation questions were answered with either “yes”/ “maybe,” “no”/”maybe,” or “no information;” the applicability evaluation questions were evaluated by “low applicability risk,” “high applicability risk,” or “unclear.” In disagreement during the evaluation process, a third researcher was consulted, and these disagreements were resolved by consensus.
Results
Literature search and screening results
A total of 970 articles were obtained from the preliminary search, including 563 English and 407 Chinese articles. After a series of screening processes of duplicate checking, reading the title and abstract, and reading the complete text, 26 articles were finally included. All studies were model-constructing studies to construct predictive models for VBAC section to validate their properties. The flow chart of literature screening is shown in Fig. 1.
Fig. 1.
Literature screening flowchart
Methodological quality evaluation of the included researches
The quality of the included researches was assessed using PROBAST. Among the 26 chosen studies, 3 had a low risk of bias, and 23 had a high risk of bias; the overall methodological quality needs improvement. The risk of bias assessment results are shown in Table 1.
Table 1.
Risk of bias assessment of the included studies
| Chosen study | Research subject | Predictor | Result | Statistical analysis | Overall risk of bias assessment | ||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ① | ② | Risk of bias | ③ | ④ | ⑤ | Risk of bias | ⑥ | ⑦ | ⑧ | ⑨ | ⑩ | ⑪ | Risk of bias | ⑫ | ⑬ | ⑭ | ⑮ | ⑯ | ⑰ | ⑱ | ⑲ | ⑳ | Risk of bias | ||
| Grobman [15] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Y | Y | Y | Low | Low |
| Kiran [19] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | Higha |
| Lakra [21] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | Higha |
| Li [23] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Liao [25] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | Higha |
| Lin [26] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | N | Y | High | High |
| Wollmann [27] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Y | Y | Y | Low | Low |
| Meyer [28] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Mi [29] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Mizrachi [30] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Y | Y | Y | Low | Higha |
| Schoorel [32] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Y | Y | Y | Low | Higha |
| Zhang [42] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Chen [8] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Fang [10] | N | N | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | N | Y | Y | Y | Y | Y | Y | Y | High | High |
| Hu [17] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Hua [18] | Y | N | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | N | N | Y | Y | N | Y | Y | Y | Y | High | High |
| Lai [20] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Li [24] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | Low | High |
| Shui [33] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Sun [35] | N | N | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | N | Y | Y | Y | Y | High | High |
| Yu [41] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | N | Y | Y | N | Y | Y | Y | Y | High | High |
| Bi [5] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | N | Y | Y | N | Y | Y | Y | Y | High | High |
| Carlsson [7] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | N | Y | Y | N | Y | Y | Y | Y | High | High |
| Gerhardy [11] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | N | Y | Y | Y | Y | Y | Y | Y | High | High |
| Gonen [12] | N | Y | High | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | N | Y | Y | Y | Y | NI | N | Y | High | High |
| Grobman [14] | Y | Y | Low | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Low | Y | Y | Y | Y | Y | Y | Y | Y | Y | Low | Low |
aIndicates that no external verification has been carried out; all are evaluated as high risk
Basic features of the chosen studies
Among the 26 included studies, the design types were primarily retrospective cohort studies, retrospective case-control studies, and prospective cohort studies. Seventeen studies were single-center studies, and 9 were multi-center studies. The included studies came from 7 countries: 15 from China, 2 from the United States, 2 from India, 2 from Sweden, 3 from Israel, 1 from the Netherlands, and 1 from Australia. The construction of the model for the included studies is shown in Table 2.
Table 2.
Building of chosen study models
| Author | Year | Country | Study site | Study design | Model construction sample size | Model validation sample size | Model building method | Model predictors | Model mode of presentation | Model validation method | Model performance | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Discrimination | Calibration | |||||||||||
| Grobman [15] | 2021 | USA | Multi-center | Prospective cohort study | 5741 | 5946 | Logistic regression | Age, pre-pregnancy BMI, height, symptom stop, vaginal delivery before cesarean section, history of VBAC, chronic hypertension history | Web-based calculator | External (ROC) | 0.75 | Unclear |
| Kiran [19] | 2020 | India | Single-center | Prospective cohort study | 194 | Not reported | Logistic regression | Age, BMI, Bishop score, history of previous vaginal delivery | Scoring system | Unclear | 0.853 | 0.804 |
| Lakra [21] | 2020 | India | Single-center | Observational study | 150 | Not reported | Logistic regression | Number of previous deliveries, natural childbirth, Bishop score, gestational age, BMI | Scoring system | Unclear | 0.77 | Unclear |
| Li [23] | 2019 | China | Multi-center | Retrospective cohort study | 1491 | 373 | Logistic regression | Gestational age, history of vaginal delivery, estimated birth weight, BMI, onset of labor, Bishop score, premature rupture of membrane | Nomogram | External (ROC, Youden index) | 0.77 | 0.82 |
| Liao [25] | 2020 | China | Single-center | Retrospective descriptive study | 1062 | 1062 | Logistic regression | Age, history of vaginal delivery, birth interval, cesarean section before trial labor, dystocia as an indication for previous cesarean section, internodular diameter, maternal prenatal BMI, gestational age, estimated fetal weight, hypertensive disorders | Scoring System | Unclear | 0.777 | 0.71 |
| Lin [26] | 2019 | China | Single-Center | Prospective Study | 162 | 162 | Logistic Regression | Bishop’s score, natural childbirth | Nomogram | Unclear | 0.953 | 0.948 |
| Wollmann [27] | 2021 | Sweden | Multi-Center | Cohort Study | 1558 | 1558 | Conditional Inference Tree, Conditional Random Forest, LASSO binary regression | Indications for first cesarean section | Machine learning model | External | 0.61–0.69 | 0.684–0.704 |
| Meyer [28] | 2020 | Israel | Single-center | Cohort study | 792 | 197 | Random Forest, Generalized Linear Model, eXtremeGradient-Boosted Decision Trees | History of vaginal delivery, vaginal delivery after cesarean section, height, increased cervical disappearance and dilation, low fetal position, induction of labor, cessation of previous descent | Machine learning model | External | 0.336–0.351 | Unclear |
| Mi [29] | 2021 | China | Single-center | Retrospective cohort study | 551 | 227 | Logistic regression | Number of previous deliveries, BMI, Bishop score, history of vaginal delivery, newborn birth weight | Nomogram | External | 0.730 | Unclear |
| Mizrachi [30] | 2017 | Israel | Single-center | Retrospective cohort study | 231 | 231 | Logistic regression |
Model 1: History of VBAC, cervical disappearance at TOLAC, head position at previous Cesarean section, neonatal weight at previous cesarean section Model 2: Cervical position at previous cesarean section, cervical disappearance at TOLAC, neonatal weight from previous cesarean section |
Regression equation | Unclear |
Model 1: 0.80 Model 2: 0.76 |
Unclear |
| Schoorel [32] | 2014 | Holland | Multi-center | Retrospective cohort study | 515 | 515 | Logistic regression | Estimated fetal weight, previous discontinuous deliveries, history of vaginal delivery, induction of labor, pre-pregnancy BMI, ethnicity | Equation | Internal validation (bootstrapping) | 0.708 | 0.13 |
| Zhang [42] | 2020 | China | Single-center | Retrospective cohort study | 483 | 225 | LASSO regression | Maternal height, maternal BMI at delivery, fundus height, cervical Bishop Score, maternal delivery age, gestational age, history of vaginal delivery | Nomogram | Internal (ROC, bootstrap resampling, DCA), External | 0.89 | Unclear |
| Chen [8] | 2021 | China | Single-center | Retrospective case-control study | 149 | 75 | Logistic regression | Low maternal BMI on admission, high cervical Bishop score, spontaneous labor, thick myometrium in the lower uterine segment, and neonatal weight < 3500g | Regression equation | Unclear | 0.908 | Unclear |
| Fang [10] | 2019 | China | Single-center | Retrospective case-control study | 430 | Not reported | ANN, Logistic regression | Maternal age, weight gain during pregnancy, whether there is an indication for cesarean section in this pregnancy, whether this pregnancy has an artificial rupture of membranes or planned delivery, and Bishop score | Regression equation | Unclear |
ROC curve OF ANN predictive models: 0.850 ROC curve for logistic regression prediction models: 0.810 |
Unclear |
| Hu [17] | 2019 | China | Single-center | Prospective cohort study | 290 | Not reported | Logistic regression | Age, height, weight gain during pregnancy (kg), history of vaginal delivery, cervical disappearance on admission (%), induction of labor | Regression equation | Unclear | 0.778 | Unclear |
| Hua [18] | 2009 | China | Single-center | Prospective cohort study | 94 | Not reported | Logistic regression | Age, time since the last cesarean section, thickness of the lower uterine segment, combined with high blood pressure during pregnancy | Regression equation | Unclear | 0.828 | Unclear |
| Lai [20] | 2018 | China | Single-center | Prospective cohort study | 532 | Not reported | Logistic regression | Neonatal birth weight, cervical Bishop score on admission, rupture of membranes | Regression equation | Unclear | 0.926 | 0.416 |
| Li [24] | 2020 | China | Single-center | Retrospective case-control study | 975 | 244 | Logistic regression | Gestational age, number of abortions, history of vaginal delivery, BMI (kg/m2), high blood pressure during pregnancy, labor induction, cervical Bishop score | Regression equation | External (ROC) |
AUC = 0.83 (Internal) AUC = 0.81 (Assessment sample) |
Unclear |
| Shui [33] | 2022 | China | Multi-center | Retrospective case-control study | 294 | 83 | Logistic regression | Age, prenatal BMI, fetal weight, lower uterine segment myometrial thickness, Bishop score, number of prenatal training | Nomogram predictive model | Internal (Bootstrap) | 0.927 | Unclear |
| Sun [35] | 2022 | China | Single-center | Retrospective case-control study | 184 | Not reported | Logistic regression | Maternal labor admission, The Bishop score is high at admission (protective factor); Excessive pregnancy weight gain, high neonatal weight (risk factor) | Regression equation | Unclear | 0.814 | 0.127 |
| Yu [41] | 2021 | China | Single-center | Retrospective case-control study | 324 | Not reported | Logistic regression | Whether they are close to giving labor, premature labor, premature rupture of membrane, height, birth interval, neonatal weight, and indications for previous pregnancy surgery | Constructed scoring system | Unclear | 0.918 | 0.826 |
| Bi [5] | 2020 | China | Multi-center | Retrospective case-control study | 1013 | Not reported | Logistic regression | Gravity, parity, number of previous deliveries, source of referral, premature rupture of membrane, prenatal hemorrhage, placental hyperplasia profile, preeclampsia, neonatal weight, oxytocin induction | Nomogram | Unclear |
AUC in the first trimester model = 0.661 Prenatal model AUC = 0.743 |
Unclear |
| Carlsson [7] | 2019 | Sweden | Multi-center | Retrospective cohort study | 19343 | 19343 | Logistic regression | Previous VBAC and previous non-cephalic cesarean delivery | Regression equation | Unclear | 0.67 | Unclear |
| Gerhardy [11] | 2022 | Australia | Multi-center | Retrospective cohort study | 13237 | 4412 | Logistic regression | Maternal place of birth, BMI, parity, if last birth was CS, any previous vaginal births, number of previous CS(s), maternal age, diabetes (pre-existing or gestational diabetes), hypertensive disorders (pre-existing, gestational or preeclampsia) and fetal abnormalities | Regression equation | Unclear |
Full model AUC = 0.7887 Prenatal model AUC = 0.7384 |
Unclear |
| Gonen [12] | 2004 | Israel | Single-center | Retrospective cohort study | 475 | Not reported | Logistic regression | Abnormal presentation as an indication for primary CS, previous VBAC, cervical dilation, gestational age < 41 weeks, low gestational age at primary CS | Scoring system | No information | No information | Unclear |
| Grobman [14] | 2007 | USA | Multi-center | Prospective cohort study | 7660 | Not reported | Logistic regression | Age, BMI, ethnicity, prior vaginal delivery, the occurrence of a VBAC, a potentially recurrent indication for the cesarean delivery |
Regression equation, nomogram |
Internal | 0.75 | Unclear |
Construction and validation of the predictive model for VBAC section
Model construction method and presentation
Twenty-six VBAC predictive models were built from the 26 included studies. 23 studies used logistic regression to construct the predictive models, 1 used LASSO regression, and 2 used machine learning algorithms. 11 models are presented in the form of regression equations, which have a clear mathematical form, making them easy to understand and interpret. 7 models in the form of nomograms which can visually display data distribution and trends. 5 models as scoring systems which simplify complex information into one or more scores, facilitating quick assessment and decision-making. 2 models as machine learning models which can handle large amounts of complex data with high predictive accuracy. can handle large amounts of complex data with high predictive accuracy. 1 model as a web-based calculator which is interactive, allowing users to input data in real-time and receive immediate feedback.
Model predictors
Each of the chosen studies involved 1–10 predictors; the predictors involved in each predictive model were different due to demographic and child-birth-related factors. Among them, the predictors with the highest frequency were the Bishop score, vaginal childbirth history, neonatal weight, maternal age, and BMI.
Model’s property
The property evaluation of the predictive models includes discrimination and calibration [39]. Among the 26 studies included, 25 used the area under the receiver operating characteristic curve to evaluate the discriminative degree of the model. Among them, 3 models had a discriminatory capacity of less than 0.7, 17 had a discriminatory capacity between 0.7 and 0.9, and 5 had a discriminatory capacity of more than 0.9. This indicates that most of the prediction models had good discrimination, and only 1 study did not report the model’s discriminatory capacity [12]. The Hosmer-Lemeshow test was used in 9 studies to verify the models’ calibration; the models in 6 studies showed a reasonable degree of calibration.
Model validation
Model validation is either internal or external [39]. Of the 26 models included, 4 were validated internally, 6 were validated externally, and only 1 was validated internally and externally. Most models (11 researches) are presented in the form of regression equations; other modes include nomograms (7 researches), machine learning models (2 researches), scoring systems (5 researches), and web-based calculators (1 research).
Discussion
In this review, we summarized 26 predictive models for vaginal birth after cesarean (VBAC). Due to differences in medical and cultural backgrounds across regions, there is a wide variety of VBAC predictive models. However, most of these models are presented in the form of regression equations and have not been visualized, which increases the complexity of clinical use. Regarding model validation, the majority of models lack external validation, indicating deficiencies in the verification aspects of existing VBAC predictive model studies. In terms of study populations, 15 of these predictive models originate from China. The predominance of Chinese literature can be attributed to the surge in second-child pregnancies in China following the implementation of the two-child policy in 2016. This demographic shift included a significant number of women with a history of cesarean delivery, who were faced with the decision to attempt a trial of labor after cesarean (TOLAC). This may explain the abundance of related Chinese literature. Since most of the included studies are concentrated in China, these predictive models may primarily be applicable to the medical environment and population characteristics in China. Differences in medical resources, cultural backgrounds, and maternal health status across regions suggest that the applicability of these models in other countries and regions needs further validation and adjustment to ensure their effectiveness and reliability. Regarding research methods, these models primarily employed various statistical techniques, including logistic regression, LASSO regression, and machine learning algorithms. Logistic regression is a commonly used statistical method that establishes a model by estimating the relationship between independent variables and a binary dependent variable. It is suitable for handling data with binary outcomes and can provide probability estimates. LASSO regression is a regularized linear regression method that introduces a penalty term to constrain regression coefficients, thereby achieving feature selection. This method can effectively address issues of multiple collinearity, enhancing the interpretability and predictive power of the model. Additionally, some studies employed machine learning algorithms such as random forests and support vector machines (SVM), which can handle complex nonlinear relationships and perform well on large datasets.
Existing models were not fully validated and had low methodological quality
In this review, we summarized the property of 26 VBAC prediction models. Among which 15 [5, 8, 10, 17, 18, 20, 23–26, 29, 33, 35, 41, 42] models were developed in China. The main reason was that after the launch of the “two-child policy” in 2016, a large number of women chose to give birth to a second child, including those women who had prior cesarean sections. Because of this special reason, there have been many studies on the prediction model of vaginal delivery after cesarean section in China. In terms of model validation, except for one study [12] that did not report the discrimination (AUROC), all other studies reported the discrimination (AUROC = 0.336 ~ 0.927), and most of the models showed good discrimination(AUROC > 0.7). Only 9 [19, 20, 23, 25–27, 32, 35, 41] of the models reported Calibration. Reporting on model performance was incomplete and may affect the use of the model. Our study also found that the vast majority of the researches lacked external validation. These all show that existing studies on VBAC prediction model still have deficiencies in model validation. External validation is an important step in evaluating the generalizability of a model, helping to determine its applicability across different populations and settings. The lack of external validation may result in poor performance of the models on new datasets, limiting their practical application value. Thus, subsequent studies need improvement in model validation.Methodologically, of the 26 researches included, 23 had methodological limitations. Only 3 researches [14, 15, 27] were at low risk of bias after quality assessment. The main reason for the risk of methodological bias was that most of the researches used univariate factor analysis to screen the predictors. The significant variables in the univariate analysis were included in the regression model. For example, Liao's [25] research adopted Chi-square test, T-test or fisher's exact test to conduct univariate factor analysis, and the variables with significant differences were incorporated into the logistic regression model. Although this is a commonly used method to screen variables, it may miss some key variables and lead to bias, especially when the sample size is small. The second main reason for bias was the data came from inappropriate sources. Data from randomized controlled studies, prospective cohort studies, nested case-control studies, or case cohort studies are appropriate [38]. In the researches we included, 15 [5, 7, 8, 10, 11, 23, 25, 29, 30, 32, 33, 35, 41, 42] were retrospective, and retrospective researches were more likely to have bias. The third major reason was for the bias was the processing of independent variables. Continuous variables were not suitable for conversion to categorical variables [38], which were found in 7 researches [5, 7, 10–12, 18, 41]. All these evidence indicate that the methodological quality of the current researches on VBAC prediction model is poor, and the follow-up research needs to improve methodological quality to reduce the risk of research bias.
Various types of VBAC prediction models exist, and there is an urgent need for model unification and visualization
Various types of predictive models were included in this study. Due to the research subjects' different races, backgrounds, and nationalities, different studies involve various predictors with different weights, which does not facilitate the scaling up of clinical studies. Furthermore, some models are presented as regression equations [7, 8, 10, 11, 17, 18, 20, 24, 30, 32, 35], increasing the complexity of clinical use. Model visualization such as nomogram and machine learning models are more intuitive and can reduce the tedious and complicated steps in use. Machine learning technique can provide diagnosis and analytical amenities so that it can be used in disease prediction to making effective decisions [36]. Future research can consider visualizing the predictive model using a nomogram, scoring system, and machine learning model.
Usage status of the VBAC prediction model
Among the 26 models included in this study, the model developed by Grobman [14] in 2007 is the most widely used. This predictive nomogram incorporates six variables: maternal age, body mass index,ethnicity, prior vaginal delivery, the occurrence of a VBAC and a potentially indiction for the cesarean delivery. This model was validated in external populations in the US [9], in Japan [40], in Spanish [3],in Italy [1] and so on. The model was validated in different populations, and the area under ROC curve was 0.67–0.80. Consideration of race may influence the clinician's choice of TOLAC. Grobman developed a model in 2021 that did not involve race and ethnicity [15]. This model has demonstrated good predictive accuracy when applied to the Japanese population [34]. Schoorel's [32] model is another one that is used more frequently. The property of the model has been validated in populations in Western Europe [32] and Dutch [37]. The predictors included in Grobman 's model were mainly indicators early in pregnancy, such as maternal age, prepregnancy weight, height, indication for previous cesarean delivery, obstetrical history, and treated chronic hypertension [14]. Schoorel's model, by contrast, included indicators both in early pregnancy and pre-labor predictors [32]. Pre-labor predictor, such as estimated fetal weight, can only be obtained in late pregnancy, before delivery. This means that obstetricians can use Grobman 's model to tell pregnant women the success rate of vaginal trial labor after cesarean section early in pregnancy, so that pregnant women have plenty of time to consider which method of delivery to choose. However, Schoorel's model can only be used before delivery, which may not be conducive to full consideration by pregnant women.
Recommendations for future research and practice
Evidence suggests that women who try a TOLAC may have fewer complications than those who choose an ERCD if they are more than 60–70% more likely to predict VBAC [6, 13]. High-quality predictive models could help pregnant women choose appropriate delivery methods and minimize risks. Various models have been developed worldwide, In addition to Grobman's model low methodological bias, and is widely used, the methodological quality of most models in the development process is low; in addition, there is a high risk of bias and a lack of validation. The model’s quality is not high and is complicated to use. In the future, it is necessary to conduct multi-center prospective cohort studies based on the domestic population to develop a high-quality predictive model VBAC, to improve the model evaluation and validation system, and to build a predictive model suitable for countries with different cultural backgrounds. For pregnant women who meet the clinical conditions of a TOLAC, the obstetrician or midwife should fully inform them of the benefits and possible risks of choosing a TOLAC so that pregnant women could choose a delivery method that is suits their vital interests.
Limitations
This study reviewed existing available studies on VBAC prediction models, however, the scope of literature searched was limited. Only Chinese databases and part of English databases were searched, and grey literature was not retrieved, so some relevant literature might be missed.
Conclusions
The extensive clinical use of the VBAC predictive model significantly reduces maternal complications and the overall cesarean section rate. This study systematically reviewed the characteristics of the construction, predictors, performance, and validation of VBAC prediction models of current researches. Evidence indicates that although a variety of prediction models have been developed globally, the methodology of these studies has limitations and the models have not been adequately validated. In the future, more prospective and high-quality research is needed to develop visual models to serve clinical work more effectively and conveniently. Obstetricians or midwifes could use predictive models to help a woman choose the right delivery method.
Acknowledgements
None
Abbreviations
- VBAC
vaginal birth after cesarean
- TOLAC
trial of labor after cesarean
- ERCD
elective repeat cesarean delivery
Authors’ contributions
Made substantial contributions to conception and design, or acquisition of data, or analysis and interpretation of data: Hong CUI, Tong LIU Involved in drafting the manuscript or revising it critically for important intellectual content: Hong CUI, Wenhui SHAN Given final approval of the version to be published. Each author should have participated sufficiently in the work to take public responsibility for appropriate portions of the content: Hong CUI, Wenhui SHAN, Quan NA, Tong LIU Agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved:
Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Data availability
All data generated or analyzed during this study are included in this published article.
Declarations
Ethics approval and consent to participate
Not Applicable.
Consent for publication
All authors have been informed and agree to publication.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Annessi E, Del Giovane C, Magnani L, Carossino E, Baldoni G, Battagliarin G, Fabio F. A modified prediction model for VBAC, in a European population. J Matern Fetal Neonatal Med. 2016;29(3):435–9. 10.3109/14767058.2014.1002767. [DOI] [PubMed] [Google Scholar]
- 2.Antoine C, Young BK. Cesarean section one hundred years 1920–2020: the Good, the Bad and the Ugly. J Perinat Med. 2020;49(1):5–16. 10.1515/jpm-2020-0305. [DOI] [PubMed] [Google Scholar]
- 3.Baranov A, Gratacos E, Vikhareva O, Figueras F. Validation of the prediction model for success of vaginal birth after cesarean delivery at the university hospital in Barcelona. J Matern Fetal Neonatal Med. 2017;30(24):2998–3003. 10.1080/14767058.2016.1271407. [DOI] [PubMed] [Google Scholar]
- 4.Betran AP, Ye J, Moller AB, Souza JP, Zhang J. Trends and projections of caesarean section rates: global and regional estimates. BMJ Glob Health. 2021;6(6). 10.1136/bmjgh-2021-005671. [DOI] [PMC free article] [PubMed]
- 5.Bi S, Zhang L, Chen J, Huang L, Zeng S, Jia J, Chen D. Development and Validation of Predictive Models for Vaginal Birth After Cesarean Delivery in China. Med Sci Monit. 2020;26:e927681. 10.12659/MSM.927681. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Cahill AG, Stamilio DM, Odibo AO, Peipert JF, Ratcliffe SJ, Stevens EJ, Macones GA. Is vaginal birth after cesarean (VBAC) or elective repeat cesarean safer in women with a prior vaginal delivery? Am J Obstet Gynecol. 2006;195(4):1143–7. 10.1016/j.ajog.2006.06.045. [DOI] [PubMed] [Google Scholar]
- 7.Carlsson Fagerberg M, Kallen K. Third-trimester prediction of successful vaginal birth after one cesarean delivery-A Swedish model. Acta Obstet Gynecol Scand. 2020;99(5):660–8. 10.1111/aogs.13783. [DOI] [PubMed] [Google Scholar]
- 8.Chen XM, Chen ZY, Sun JL, Zhong W, Jin JX. Establishment and validation of prediction model for transvaginal delivery of re-pregnancy after cesarean section. Adv Mod Obstet Gynecol. 2021;30(8):601–5. 10.13283/j.cnki.xdfckjz.2021.08.037. [Google Scholar]
- 9.Costantine MM, Fox K, Byers BD, Mateus J, Ghulmiyyah LM, Blackwell S, Saade G. Validation of the prediction model for success of vaginal birth after cesarean delivery. Obstet Gynecol. 2009;114(5):1029–33. 10.1097/AOG.0b013e3181bb0dde. [DOI] [PubMed] [Google Scholar]
- 10.Fang JH, Liang JL, Zheng JL, Yang Y, Xu M, Song AP. A prediction study of re-pregnancy vaginal delivery after cesarean section based on artificial neural network. Maternal child health care China. 2019;34(12):2680–3. 10.7620/zgfybj.j.issn.1001-441I.2019.12.07. [Google Scholar]
- 11.Gerhardy L. A predictive tool for vaginal birth after caesarean success in an Australian cohort. Aust N Z J Obstet Gynaecol. 2022;62(3):383–8. 10.1111/ajo.13473. [DOI] [PubMed] [Google Scholar]
- 12.Gonen R, Tamir A, Degani S, Ohel G. Variables associated with successful vaginal birth after one cesarean section: a proposed vaginal birth after cesarean section score. Am J Perinatol. 2004;21(8):447–53. 10.1055/s-2004-835961. [DOI] [PubMed] [Google Scholar]
- 13.Grobman WA, Lai Y, Landon MB, Spong CY, Leveno KJ, Human Rouse DJ., Development Maternal-Fetal Medicine Units, N. Can a prediction model for vaginal birth after cesarean also predict the probability of morbidity related to a trial of labor? Am J Obstet Gynecol. 2009;2009(1):56 e51-56. 10.1016/j.ajog.2008.06.039. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Grobman WA, Lai Y, Landon MB, Spong CY, Leveno KJ, Rouse DJ, Human Development Maternal-Fetal Medicine Units N. Development of a nomogram for prediction of vaginal birth after cesarean delivery. Obstet Gynecol. 2007;109(4):806–12. 10.1097/01.AOG.0000259312.36053.02. [DOI] [PubMed] [Google Scholar]
- 15.Grobman WA, Sandoval G, Rice MM, Bailit JL, Chauhan SP, Costantine MM, Human Development Maternal-Fetal Medicine Units N. Prediction of vaginal birth after cesarean delivery in term gestations: a calculator without race and ethnicity. Am J Obstet Gynecol. 2021;225(6):664 e661-664 e667. 10.1016/j.ajog.2021.05.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Gynecologists ACoOa. ACOG Practice Bulletin 205. Vaginal birth after sesarean delivery. Obstet Gynecol. 2019;133(2):e110–27. [DOI] [PubMed] [Google Scholar]
- 17.Hu HY, Zhong M, Luo ML, Yin Q, Liu CD. Establishment of a Prediction Model for Vaginal Delivery after Cesarean Section. J Practical Obstet Gynecol. 2019;35(3):195–8. [Google Scholar]
- 18.Hua XY, Li ZC, Wang ZP, Li YX, Zhang JF, Hu YG, Shang JW. The correlative factors and prediction model of vaginal trial outcome of second pregnancy after cesarean section. Prog Obstet Gynecol. 2009;18(9):696–8. [Google Scholar]
- 19.Kiran P, Patil KP, Metgud MC, et al. Prediction of Vaginal Birth after Cesarean Section using Scoring System at the time of Admission for Trial of Labor: A One-year Prospective Cohort Study. J South Asian Feder Obst Gynae. 2020;12(4):224–9. [Google Scholar]
- 20.Lai BL, Wang CH, Zhang QF, Yuan HL, Chen L. Establishment and validation of predictive model for trial of labor after cesarean section. J New Med. 2018;49(5):350–4. [Google Scholar]
- 21.Lakra P, Patil B, Siwach S, Upadhyay M, Shivani S, Sangwan V, Mahendru R. A prospective study of a new prediction model of vaginal birth after cesarean section at a tertiary care centre. Turk J Obstet Gynecol. 2020;17(4):278–84. 10.4274/tjod.galenos.2020.82205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Li HT, Luo S, Trasande L, Hellerstein S, Kang C, Li JX, Blustein J. Geographic Variations and Temporal Trends in Cesarean Delivery Rates in China, 2008–2014. JAMA. 2017;317(1):69–76. 10.1001/jama.2016.18663. [DOI] [PubMed] [Google Scholar]
- 23.Li YX, Bai Z, Long DJ, Wang HB, Wu YF, Reilly KH, Ji YJ. Predicting the success of vaginal birth after caesarean delivery: a retrospective cohort study in China. BMJ Open. 2019;9(5):e027807. 10.1136/bmjopen-2018-027807. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Li YX, Yin HT, Zhou LQ, Chen XY. Discussion on the prediction model of vaginal delivery after cesarean section. Chin J Family Plann Gynecotokology. 2020;12(7):88–92. [Google Scholar]
- 25.Liao Q, Luo J, Zheng L, Han Q, Liu Z, Qi W, Yan J. Establishment of an antepartum predictive scoring model to identify candidates for vaginal birth after cesarean. BMC Pregnancy Childbirth. 2020;20(1):639. 10.1186/s12884-020-03231-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Lin J, Hou Y, Ke Y, Zeng W, Gu W. Establishment and validation of a prediction model for vaginal delivery after cesarean and its pregnancy outcomes-Based on a prospective study. Eur J Obstet Gynecol Reprod Biol. 2019;242:114–21. 10.1016/j.ejogrb.2019.09.015. [DOI] [PubMed] [Google Scholar]
- 27.Lindblad Wollmann C, Hart KD, Liu C, Caughey AB, Stephansson O, Snowden JM. Predicting vaginal birth after previous cesarean: Using machine-learning models and a population-based cohort in Sweden. Acta Obstet Gynecol Scand. 2021;100(3):513–20. 10.1111/aogs.14020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Meyer R, Hendin N, Zamir M, Mor N, Levin G, Sivan E, Tsur A. Implementation of machine learning models for the prediction of vaginal birth after cesarean delivery. J Matern Fetal Neonatal Med. 2022;35(19):3677–83. 10.1080/14767058.2020.1837769. [DOI] [PubMed] [Google Scholar]
- 29.Mi Y, Qu P, Guo N, Bai R, Gao J, Ma Z, Luo X. Evaluation of factors that predict the success rate of trial of labor after the cesarean section. BMC Pregnancy Childbirth. 2021;21(1):527. 10.1186/s12884-021-04004-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Mizrachi Y, Barber E, Kovo M, Bar J, Lurie S. Prediction of vaginal birth after one ceasarean delivery for non-progressive labor. Arch Gynecol Obstet. 2018;297(1):85–91. 10.1007/s00404-017-4569-4. [DOI] [PubMed] [Google Scholar]
- 31.Mone F, Harrity C, Mackie A, Segurado R, Toner B, McCormick TR, McAuliffe FM. Vaginal birth after caesarean section prediction models: a UK comparative observational study. Eur J Obstet Gynecol Reprod Biol. 2015;193:136–9. 10.1016/j.ejogrb.2015.07.024. [DOI] [PubMed] [Google Scholar]
- 32.Schoorel EN, van Kuijk SM, Melman S, Nijhuis JG, Smits LJ, Aardenburg R, Scheepers HC. Vaginal birth after a caesarean section: the development of a Western European population-based prediction model for deliveries at term. BJOG. 2014;121(2):194–201. 10.1111/1471-0528.12539. discussion 201. [DOI] [PubMed] [Google Scholar]
- 33.Shui J, Gu SY, Chen X, Wang YJ. Construction of a predictive model for the feasibility of vaginal delivery in re-pregnancy of scarred uterus after cesarean section. Chin J Family Plann Gynecotokology. 2022;14(3):74–8. [Google Scholar]
- 34.Sugai S, Nishijima K. Validating a calculator without race and ethnicity to predict vaginal birth after cesarean delivery. Am J Obstet Gynecol. 2022;227(3):537–8. 10.1016/j.ajog.2022.05.017. [DOI] [PubMed] [Google Scholar]
- 35.Sun X, Sun CZ, Wei J, Liu JS. Construction of prediction model of vaginal birth in pregnancy again after cesarean section. Jiangsu Med J. 2022;48(6):566–70. 10.19460/j.cnki.0253-3685.2022.06.006. [Google Scholar]
- 36.Ullah Z, Saleem F, Jamjoom M, Fakieh B. Reliable Prediction Models Based on Enriched Data for Identifying the Mode of Childbirth by Using Machine Learning Methods: Development Study. J Med Internet Res. 2021;23(6):e28856. 10.2196/28856. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Vankan E, van Kuijk SMJ, Nijhuis JG, Aardenburg R, Delemarre FMC, Dirksen CD, Scheepers HC. External validation of a prediction model on vaginal birth after caesarean in a The Netherlands: a prospective cohort study. J Perinat Med. 2021;49(3):357–63. 10.1515/jpm-2020-0308. [DOI] [PubMed] [Google Scholar]
- 38.Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, Groupdagger P. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med. 2019;170(1):51–8. 10.7326/M18-1376. [DOI] [PubMed] [Google Scholar]
- 39.Xu Y, Zhu LY, Wang Y, Sun JH, Wang XJ, Deng HB, Ma YF. The status and implications of prediction model studies conducted by Chinese nursing scholars: a scope review. Chin Nurs Manage. 2022;22(5):744–9. 10.3969/j.issn.1672-1756.2022.05.021. [Google Scholar]
- 40.Yokoi A, Ishikawa K, Miyazaki K, Yoshida K, Furuhashi M, Tamakoshi K. Validation of the prediction model for success of vaginal birth after cesarean delivery in Japanese women. Int J Med Sci. 2012;9(6):488–91. 10.7150/ijms.4682. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Yu JZ, Liu CX, Wang ZW, Xiong XS, Cui H. The build and assessment of the predictive model of successful TOLAC. Chin J Practical Gynecol Obstet. 2021;37(7):782–5. 10.19538/j.fk2021070122. [Google Scholar]
- 42.Zhang HL, Zheng LH, Cheng LC, Liu ZD, Yu L, Han Q, Yan JY. Prediction of vaginal birth after cesarean delivery in Southeast China: a retrospective cohort study. BMC Pregnancy Childbirth. 2020;20(1):538. 10.1186/s12884-020-03233-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
All data generated or analyzed during this study are included in this published article.