Prediction of intrapartum caesarean section in vaginal breech birth: development of models for nulliparous and multiparous women

Introduction

The optimal birth mode for breech presentation at term is still debated 1], [2], [3. Therefore, women need to be counselled about the risks and benefits of vaginal breech birth and planned cesarean section 4], [5], [6 in order to choose their preferred mode of birth. The individual risk for intrapartum cesarean section, a major point in this decision, should also be addressed based on evidence-based predictions.

Various relevant studies have shown that a narrow obstetric conjugate (OC) 7], [8], [9], [10 or interspinous distance (ISD) [11], high birthweight [7, 12], wide head circumference [7], birth after estimated due date [13] and high maternal age [7, 14] are associated with the need for intrapartum cesarean section in vaginal breech birth. In order to predict the risk of this surgical outcome accurately, these variables of interest need to be integrated into a prediction model using multivariable regression (i.e. model-based approach) [15, 16]. However, the models developed in previous studies show significant weaknesses by lacking either statistical adjustment [10, 13], analysis of model performance [12] or internal model validation [7, 10], which are essential to demonstrate the clinical value of these predictors.

It is still unclear if OC can predict intrapartum cesarean section in vaginal breech births, as studies showed conflicting results 7], [8], [9, 11, 17], [18], [19. One explanation for this ambiguity could be that the effect of the OC on this outcome depends on the birthweight [8, 19]. Fittingly, a statistical interaction (also called effect modification) characterizes how additional factors (such as birthweight) can influence the effect of another factor (such as OC) on an outcome (such as intrapartum caesarean section). However, previous studies in this field did not explore such statistical interactions [7, 10], [11], [12, 18], even though this would clarify which women could actually profit from OC measurement prior to vaginal breech birth.

To the best of our knowledge, no study so far has analysed multiparous women separately focusing on the factors associated with intrapartum cesarean section in vaginal breech birth. In our view, this represents a shortcoming of the current research in this field. Factors predicting intrapartum cesarean section in multiparous women could be different from those observed in nulliparous women, which has been observed in the case of epidural analgesia [20].

Patient counselling and clinical decision making need to be based on accurate prediction. Poorly developed prediction models will inevitably make erroneous assumptions about the individual probability of an outcome and may thus lead to unnecessary intervention. Specifically, a poor model could overestimate the risk of intrapartum cesarean section and vaginal breech birth may consequently be wrongly discouraged.

This study aims to provide further evidence concerning which parameters have value in predicting intrapartum cesarean section both in nulliparous and multiparous women. A special focus will be put on how birthweight could influence the effect of OC on the risk of this surgical intervention.

Subjects and methods

This retrospective single-centre study included a total of 262 nulliparous women (1/2017–11/2021) and 230 multiparous women (1/2013–11/2021) who attempted vaginal breech birth at the University Hospital of Leipzig. If a woman had more than one breech delivery at our institution in the time span of this study, they were counted as two separate cases.

The primary outcome was the occurrence of intrapartum caesarean section and the secondary outcomes were the short-term neonatal outcomes (5 min Apgar-Score, arterial pH-value and base excess at birth).

We collected the data of possible predictors, which have been associated with the rate of intrapartum caesarean section in vaginal breech birth: OC 7], [8], [9, 19], ISD [11], fetal birthweight (cut-off at 3.8 kg) [7, 12], fetal size [7], fetal head circumference [7], maternal age [7, 11], delivery after estimated due date [13] and epidural anesthesia [7, 20]. Additionally, we included data of other variables of interest, which have not yet been significantly associated with the primary outcome: all other parameters of MR-pelvimetry, prelabor rupture of membranes (PROM), sonographic fetometric parameters (biparietal diameter, abdominal circumference, femur length), cervical dilation at first examination, induction of labor, maternal pre-pregnancy BMI and height. Also, an interaction term (OC * birthweight ≥3.8 kg) was added, as previous research suggested that one factor could modify the effect of the other [8, 19].

Estimated birthweight was not used as a predictor for several reasons: (1) fetal weight estimates may be very heterogenous due to the variety of estimation formulas used, occasional missingness of certain fetometric parameters (e.g. head circumference) and variable timespan between ultrasound and birth; (2) relevant selection bias must be suspected, as vaginal breech birth is usually discouraged if the estimated weight is below 2.5 kg or above 4 kg. This could mask the true effect of birthweight, since newborn weighting more than allowed would necessarily have lower estimated birthweights. In fact, a previous study showed that the difference between estimated and real birthweight was higher in newborns delivered by intrapartum cesarean section than in those born vaginally [11]; (3) to demonstrate a causal effect between birthweight and the risk of intrapartum cesarean section, this predictor had to be determined as precisely as possible.

Cases with known serious fetal anomalies were excluded from the study, as these may have impacted the decision to perform cesarean section.

Results

A total of 262 nulliparous and 230 multiparous women were included in the analysis. Of these, 78 (29.8 %) nulliparous and 23 (10 %) multiparous women needed intrapartum cesarean section.

In nulliparous women, newborns after vaginal breech birth had significantly lower arterial pH (p<0.0001) and base excess (p<0.0001) compared with those born through intrapartum caesarean section. However, Apgar-Scores at 5 min did not differ significantly (p=0.13). In multiparous women no significant differences in blood gases or Apgar-Scores could be observed between birth modes. The characteristics of the cohort are summarised in Table 1. No characteristic had more than 2 % missing values Supplementary Material (List S1/S2). Model equations are shown in the Supplementary Material (Eqs. S3–S5).

Nulliparous women

The following variables showed a possible association (p<0.25) with intrapartum cesarean section in nulliparous women: birthweight above 3.8 kg, length of the newborn, BMI, maternal height, OC, pelvic width, ISD and epidural analgesia during birth. These were integrated in the multivariable analysis together with an interaction term between OC and birthweight above 3.8 kg (OC * birthweight ≥3.8 kg; Table 2A), as OC did not show significancy in all birthweight-subgroups (Table 1). In the resulting multivariable model, the variables retaining significance were birthweight above 3.8 kg, use of epidural analgesia and the interaction term between OC and birthweight above 3.8 kg (i.e. if birth weight was above 3.8 kg then increasing OC was associated with an increasing rate of successful vaginal deliveries; Figure 1A and B). Backward elimination additionally selected maternal height into the reduced model. However, after internal validation, this model showed signs of overfitting (calibration slope: 0.85, heuristic shrinkage factor: 0.84) and penalized estimation was used to refit the model. A penalization factor of 3 was chosen, as this factor optimized calibration (calibration slope after penalization: 1.02). The adjusted AUC of the model was 0.67.

Table 2A:

Multivariable logistic models for prediction of intrapartum cesarean section in nulliparous women.

Predictor	Full model		Reduced (penalized) model			Predictor	Reduced (penalized) model
	coef.	p-Value	coef.	p-Value	aOR (95 % CI)		coef.	p-Value	aOR (95 % CI)
OC (per 1 cm)	−0.08	0.75	−0.18	0.41	0.83 (0.54–1.29)	OC (per 1 cm)	−0.09	0.69	0.91 (0.57–1.45)
OC * BW ≥3.8 kg (per 1 cm)a	−1.72	0.04	−1.45	0.04	0.24 (0.06–0.97)	OC * AC ≥34 cm (per 1 cm)a	−1.22	0.02	0.29 (0.11–0.82)
BW ≥3.8 kg	1.07	0.04	0.9	0.03	2.45 (1.07–5.62)	AC ≥34 cm	0.66	0.04	1.93 (1.03–4.75)
Maternal height (per 10 cm)	−0.37	0.22	−0.44	0.08	0.64 (0.38–1.07)	Maternal height (per 10 cm)	−0.45	0.08	0.64 (0.38–1.06)
Epidural analgesia	0.82	0.01	0.76	0.01	2.14 (1.23–3.72)	Epidural analgesia	0.79	0.01	2.2 (1.26–3.81)
BMI, kg/m2	0.08	0.04
Interspinous distance (per 1 cm)	−0.2	0.21
Pelvic width (per 1 cm)	−0.16	0.45
Length of the newborn (per 1 cm)	0.08	0.2

1. The reduced model was built using backwards elimination and the model was re-fitted with penalized estimation due to overfitting. The adjusted odds ratios represent the change of the odds for intrapartum cesarean section for every additional unit of the variable. For example, the odds intrapartum cesarean section changes by the odds ratio of 0.64 if maternal height increases by 10 cm. ^aInteraction term representing the statistical interaction: if birthweight is ≥3.8 kg, OC, has an adjusted OR, of 0.2 (0.83 × 0.24) for every additional 1 cm in wideness (if birthweight is ≤3.8 kg, the aOR, remains 0.83). Similarly, every additional 1 cm in OC-wideness has an adjusted OR, of 0.26 (0.91 × 0.29) and 0.93 if AC, is ≥34 cm and ≤34 cm, respectively. OC, obstetric conjugate; BW, birthweight; AC, abdominal circumference; aOR, adjusted odds ratio; coef., coefficient.

Figure 1:

Graphical representation of the statistical interaction between OC and birthweight/AC. (A) and (C): Rate of intrapartum cesarean section depending on the obstetric conjugate. The two regression lines represent how this relationship differs depending on birthweight or fetal abdominal circumference. The rate of intrapartum cesarean section depend on OC predominantly in the subgroup of heavy newborn (blue regression lines). (B) and (D): The rate of intrapartum cesarean section depending on birthweight and fetal abdominal circumference. When OC is ≥13 cm, the change in this outcome is less dependent on birthweight or AC. The dots represent the individual cases in the cohort and are located on the top if intrapartum cesarean section occured and on the bottom if not (small random vertical variation has been added to the dots in order to avoid overplotting and improve visualisation). OC, obstetric conjugate; AC, abdominal circumference.

As birthweight can be accurately measured only after the birth, we built an alternative model by replacing birthweight above 3.8 kg with the abdominal circumference (AC) measured by ultrasound before labor. AC was chosen as it was the only fetometric parameter significantly associated with intrapartum cesarean section (Table 1). This alternative model containing AC (dichotomized at 34 cm, Table 2A) with the corresponding interaction term (OC * AC ≥34 cm) instead of birthweight showed an adjusted AUC of 0.68 and a good calibration (calibration slope after penalization: 0.97).

Discussion

This study suggests that the effect of the OC on the rate of intrapartum cesarean section is dependent on birthweight when a nulliparous woman attempts vaginal breech birth. In our cohort, the OC did not influence the rate of intrapartum cesarean section if the newborn weighted less than 3.8 kg (or sonographically estimated abdominal circumference was below 34 cm). However, the OC seemed to be crucially important when birthweight or estimated abdominal circumference were above those thresholds, as every additional centimeter in the OC significantly decreased the rate of intrapartum cesarean section (Figure 1A and C). This effect modification can also be visualised in the following way: increasing birthweight (or abdominal circumference) would not lead to an increase in intrapartum cesarean sections if the OC is above 13 cm (Figure 1B and D, Table 1).

This effect modification is taken into account by incorporating an interaction term (Table 2) into the prediction models for nulliparous women. For practical purposes, the prediction model with the abdominal circumference may be preferable, as birthweight can only be precisely measured after birth. However, as practicioners only have to decide whether the fetus weights more or less than 3.8 kg (instead of estimating a specific birthweight), we think that also the model based on the birthweight as a predictor has clinical value. To provide practical interpretation of the models, several hypothetical scenarios were tested and their predicted probabilities are shown in Table 3.

A birthweight-OC-ratio has been suggested previously to predict intrapartum cesarean section in vaginal breech birth [10]. In our opinion, such a ratio overestimates the risk of intrapartum cesarean section in those women with a heavy newborn but also a large OC. For example, a hypothetical nulliparous woman with an OC of 13.5 cm would be considered high-risk for intrapartum cesarean section if birthweight is 3.5 kg or above (since this woman would have a birthweight-OC-ratio above the cut-off of 257.8 g/cm [10]). However, this could not be observed in our cohort, as our model predicted that rates of intrapartum cesarean section were unaffected by birthweight when the OC was above 13 cm (Table 3, graphically shown in Figure 1B and D). Consequently, it is our opinion that a model accounting for statistical interaction is more appropriate than a birthweight-OC-ratio to predict successful vaginal breech birth in nulliparous women.

In summary, our results clarify why studies reported conflicting results on the role of the OC in vaginal breech births 7], [8], [9, 11, 17], [18], [19. Since the OC seems to be mostly relevant in high birtweight, a study with a low number of heavy newborns would probably not detect significant effects if statistical interaction was not applied. Importantly, not accounting for statistical interaction would lead to the false conclusion that the OC is equally important for all women regardless of the birthweight. The analysis of our cohort rejects this assumption showing that OC is not useful in predicting intrapartum cesarean section when birthweight is below 3.8 kg or the AC is less than 34 cm (while it is crucially important when birthweight or AC are higher).

These results show that the use of epidural analgesia is associated with intrapartum cesarean section. This is in line with another study [34] while it contradicts the result in the FRABAT-cohort [7, 20]. However, we cannot draw the conclusion that epidural analgesia per se leads to higher rates of intrapartum cesarean section, as this form of analgesia may be used more frequently when labor was already prolonged or obstructed.

In our cohort, ISD did not predict intrapartum cesarean section contrarily to what a previous study suggested [11]. Of note, its multivariable regression included ISD as a variable but omitted the OC. In fact, we observed that ISD did not have any statistically significant relationship when analysed together with OC (Table 2A), indicating that ISD lacks an independent predictive value. Similarly, birth after estimated due date and length of the newborn, which have been linked to intrapartum cesarean section [7, 13], lost significance when analysed together with birthweight (Table 2A).

To the best of our knowledge, this is the first study building an (internally) validated prediction model focusing specifically on vaginal breech birth in multiparous women. It identifies several risk factors (e.g. maternal BMI, epidural analgesia) but also shows that PROM is associated with a lower rate of intrapartum cesarean section. Contrarily to nulliparous women, birthweight was not independently associated with intrapartum cesarean section in multiparous women. So far, research on vaginal breech birth has analysed women together [12, 13, 35], possibly missing some risk factors, which are specific for multiparous women. This study suggests that multiparous women in fact have different risk factors for intrapartum caesarean section compared to nulliparous women.

Our study has some limitations. Firstly, this is a retrospective study which in itself presents a weakness. Especially the parameter of abdominal circumference might be inaccurate due to the fact that time between measurements and birth varied (Table 1). Also, practitioners caring for the women in this cohort were not blinded for several characteristics (e.g. OC) and the decision to perform intrapartum cesarean section may have been influenced by this information. Secondly, these models have only been validated internally and need to be validated externally in order to confirm their clinical value and generalisability. Thirdly, an important limitation is the sample size. Clearly a cohort of just above 200 cases is too small to build a prediction model that is perfectly accurate. However, our study used a statistically robust approach (penalisation and bootstrapping) [29, 30] to limit the shortcomings of the small sample sizes. Nevertheless, this model may have missed useful predictive variables that have been described in other larger studies [7]. Fourthly, while backward elimination is popular, it remains a potentially unprecise method of variable selection [15]. Consequently, we can’t be sure that our prediction models contain all “authentic” predictors nor can we exclude that some “noise” variables have been erroneously selected. Lastly, predicting intrapartum cesarean section is particularly challenging, due to the multitude of reasons for this surgical intervention (abnormal fetal heart rate, arrest of labor, ect.). For example, a model may be more accurate in predicting a cesarean section due to arrest of labor than due to abnormal fetal heart rate, as risk factors for these two outcomes are different.

This work has some clinical implications. Measuring the OC may have limited value for the prediction of intrapartum cesarean section when a fetus in breech presentation is of avarage weight. Conversingly, the OC (ideally measured by MR-pelvimetry) could be especially usefull when birthweight is above 3.8 kg. Specifically, an OC above 13 cm could compensate the additional risk for intrapartum cesarean section caused by a high birthweight. Additionally, women with an OC of less than 13 cm may benefit from induction of labor if the practitioner suspects that the fetus could reach a birthweight above 3.8 kg. However, such an intervention would need to be studied prospectively.

Conclusions

Our results show the most relevant risk factors associated with intrapartum cesarean section during vaginal breech birth. As a novel finding, this study clarifies the role of the OC by showing that it is only associated with intrapartum cesarean section if birthweight is high (above 3.8 kg). Contrastingly, the OC did not predict this surgical outcome when the newborn weighted less than 3.8 kg. This study further provides internally validated models for the prediction of intrapartum caesarean section for nulliparous and multiparous women, respectively. Risk stratification according to this models could be beneficial for the counselling of women who wish a vaginal breech birth.