Costantine, Maged M. MD1; Fox, Karin MD1; Byers, Benjamin D. DO1; Mateus, Julio MD1; Ghulmiyyah, Labib M. MD1; Blackwell, Sean MD2; Hankins, Gary D. V. MD1; Grobman, William A. MD, MBA3; Saade, George MD1
In 2006, the rate of cesarean delivery in the United States reached a new record of 31.1% (Centers for Disease Control and Prevention, final birth data for 2006). This represents a 50% increase during the past decade.1 This sharp increase in the total cesarean delivery rate is the result of multiple factors, including a rise in the rate of primary cesarean delivery, a decline in the rate of vaginal birth after cesarean (VBAC) to 8.5% in 2006 (from a maximum of 28.3% in 1996), and a progressive decline in the rate of operative vaginal delivery.1
In an effort to decrease the rate of cesarean delivery, the American College of Obstetricians and Gynecologists recommends that most pregnant women with a single previous low transverse cesarean delivery be counseled about VBAC and offered a trial of labor.2 A key aspect in the counseling process involves providing the patient with her individual chance of success as well as a discussion about the maternal and neonatal risks and benefits associated with a trial of labor. Although much progress has been made in identifying the population-level variables that are associated with a successful trial of labor and the complications from a failed one,3,4 providing the patient with an individual risk has remained challenging.
Between 1999 and 2002, the National Institute of Child Health and Human Development Maternal–Fetal Medicine Units (MFMU) Network conducted a prospective multicenter observational study of women with prior cesarean delivery undergoing either trial of labor or elective repeat cesarean delivery (more than 45,000 patients).3 Grobman et al, using data from this registry on term pregnant women with one prior cesarean delivery, developed a model that predicts the probability of successful VBAC.5 Published in 2007, the final model included six maternal variables that can be obtained at the first prenatal visit, thus allowing the obstetrician to provide the patient desiring a trial of labor with her individual chance of success early in the course of pregnancy. More than 8,000 patients with successful VBAC were included in that study; data from half of them were used in the training phase to develop the multivariable logistic regression model.5 Although the model was subsequently tested using data from the other half of the registry’s patients, no validation has been conducted on a completely different population. Thus, the objective of this study was to validate the VBAC prediction model and test its accuracy in a patient cohort independent from that of the National Institute of Child Health and Human Development MFMU Network’s cesarean delivery registry.
MATERIALS AND METHODS
The University of Texas Medical Branch in Galveston, Texas, was not a member in the MFMU Network at the time when the cesarean delivery prospective observational study was being conducted, and, therefore, none of its patients were included in the development or internal validation of the prediction model.
After obtaining approval by the institutional review board at the University of Texas Medical Branch, all pregnant women with a vertex singleton gestation at term (more than 36 6/7 weeks) and a single prior low transverse cesarean delivery who attempted a trial of labor between January 2002 and August 2007 were identified. The previously published VBAC prediction model was not in use at our site during the study period. Women with antepartum intrauterine fetal demise were excluded. Medical records were reviewed and pertinent data extracted. As reported by Grobman et al, the prediction model includes maternal characteristics that can be obtained at the first prenatal visit: maternal age, prepregnancy body mass index (kg/m2), ethnicity, any prior vaginal delivery, prior VBAC, and indication of prior cesarean delivery.5 The data from individual patients were incorporated into the reported regression formula, and the predicted rate for VBAC success for each woman was calculated.
As was performed during the original development of the prediction model, the predicted VBAC scores at the level of the study population then were partitioned into 10 decile groups (eg, 0–10%, 10–20%). In each of these deciles, the proportion (and the 95% confidence interval [CI]) of women with a successful trial of labor was calculated. These proportions represent the actual or observed rates of VBAC. The predictive ability of the model was assessed with both a receiver operating characteristic (ROC) and calibration curves.6,7 For the former analysis, the area under the curve (AUC) was determined by the trapezoidal rule (empirical or nonparametric method).7,8 For the latter, a curve was generated by plotting the predicted and observed rates, represented by the midpoint of each decile category, on a scatter gram and then smoothly connecting the points to form a curve (with its 95% confidence margin). In addition, the best-fit line was determined and found to be linear. Its intercept then was compared with zero and its slope to 1; these correspond to an ideal curve represented by a 45° straight line. In addition, the concordance between the predicted and observed VBAC rates in each decile was calculated.
Normality was assessed using the Kolmogorov-Smirnov test,9 and data that were not normally distributed were compared using the Mann-Whitney rank sum test. Also, correlation analyses used the Pearson correlation. P<0.05 was used to indicate statistical significance, and all tests were two-tailed. Statistical analyses and graphs were performed using SigmaPlot 10.0 (Systat Software Inc, Chicago, IL).
During the study period, 545 women met the inclusion criteria and were selected for data extraction. Complete data were available for 502 patients. A total of 262 patients had VBAC, corresponding to a rate of 52.2%. The patients’ characteristics are summarized in Table 1.
The predicted probability of VBAC, as calculated by the regression equation, was significantly higher in those who had a successful trial of labor (median 78.4%, interquartile range 62.1–88.2) than in those who did not (median 59.7%, interquartile range 50.8–75.3, P<.001). The model’s prediction of trial of labor success, as represented by the ROC curve, is presented in Figure 1. The AUC of this curve is 0.70 (95% CI 0.65–0.74, P<.001), which is consistent with the AUC reported in the development of the prediction model.
The calculated predicted VBAC success rates then were partitioned into deciles, as previously described, and compared with the observed VBAC rates (Table 2, Figs. 2 and 3). The actual and predicted VBAC rates did not differ when the predicted chance of success was less than 50%. The point estimates of the observed VBAC rates were approximately 10–20% lower than predicted when the predicted rate was at least 50%, although, in most cases, the 95% CI either approached or included the observed rate (Fig. 2). In addition, the best-fit line was determined and found to be linear. Its intercept and slope were 5.04 and 0.72, and neither was significantly different from an ideal intercept of zero (P=.09) and slope of 1 (P=.59). The correlation between the observed and predicted VBAC rates was high, with r=0.90 (P=.002) (Fig. 3).
Using an independent patient cohort, we have confirmed the ability of the previously published model to predict the probability of VBAC. Its overall test characteristics, such as that exemplified by the ROC curve, among women undergoing a trial of labor in our cohort were consistent with the characteristics observed during the model’s development.
The prediction model developed by Grobman et al5 provides important information that can be used to help provide individual and patient-specific success rates from a trial of labor. Its major benefit is that it provides success rates using data available at entry to care. This gives both the patient and the counseling obstetrician ample time to discuss the risks and benefits of a trial of labor and thus allows the patient to make an informed decision weeks if not months ahead of the onset of labor.
The strength of this study is that it was performed on an independent cohort of patients who were not included in the MFMU Network’s cesarean delivery registry from which the prediction model was developed. Despite our relatively small sample size as compared with the sample size during model development, the essential characteristics of the prediction model have been validated in this external population. Specifically, those women who are predicted to have a VBAC are significantly more likely to do so than those who are not, the AUC of the ROC curve is similar to that originally described, and the actual probability of VBAC increased in conjunction with an increase in the predicted probability. Also, as illustrated by the calibration curve, the predicted probabilities were similar to the observed probabilities.
In some cases, the point estimate of the observed was less than the predicted value, but this does not negate the general validation. First, because of the sample size, the width of the 95% CIs was wide and often incorporated or came close to the observed rate. Also, in a small, single-center population, this observed line can deviate from the predicted owing to differences in local practice. However, the published model remains more useful than giving all patients an average likelihood of VBAC. Although different obstetrical units across the country have different attitudes and practices regarding trial of labor, each one of them does not need to develop a new model or validate the current one. The model in question has been developed using a large number of patients from different parts of the country, then validated internally in the same cohort and, in this study, externally in an entirely different patient cohort. Moreover, our observed values correlated particularly well with the predicted values for those women with a low likelihood of VBAC success. This is important because these patients are ones most likely to decline VBAC once given a low chance of a successful vaginal delivery because they are ones the most likely to suffer the consequences of failed trial of labor. Maternal morbidities, such as uterine rupture/dehiscence, blood transfusion, and need for hysterectomy, are more common in women who attempt and fail a trial of labor as compared with those with successful VBAC or elective repeat cesarean delivery.10
As with the original development study, there are limitations in this study (and the prediction model) that should be noted. The model applies exclusively to pregnant women desiring a trial of labor when they reach term, thus no information can be deduced on the VBAC chance at less than 37 weeks. There is also a potential selection bias because the only patients in the model were those who attempted a trial of labor, and thus we cannot know how well the model would apply to women who were VBAC candidates but elected elective repeat cesarean delivery.
The primary intent of this model, as previously mentioned, is to give patients contemplating trial of labor individual chances of success to help them reach a more informed decision. Although it is tempting for clinicians to select a cutoff above or below which they will or will not offer a trial of labor, this model is not intended for this role because, ultimately, it is the patient’s decision to undergo a trial of labor and each patient will consider the pros and cons of the decision differently.
The presently analyzed prediction model is the last in a series of prediction models11–15 that have been developed to help in decision making regarding a trial of labor. Some involved an additive point-scoring system11,15 whereas others produced more complex logistic regression equations.12–14 Some were developed from a small number of patients,11,13 whereas, in others, the patient cohort had a limited range of success probabilities.15 In the majority of these models, data were collected retrospectively,11,13–15 sometimes from state or national registries,14,15 and then incorporated into the development of the model. Moreover, one major drawback of previously mentioned models is their inclusion of variables that will not be known until late in pregnancy (maternal anemia)15 or until the patient is admitted for delivery (such as nonreassuring fetal heart tones, cervical examination, labor induction, sex of the infant, and gestational age at delivery).11–14 Compared with all of these, the latest model (Grobman et al),5 validated in this study, was developed in a concurrent registry from a large number of patients who had a wide distribution of VBAC success scores, was validated internally, and includes only those variables that can be obtained from the first prenatal visit, thus allowing ample time for the physician and the patient to discuss options regarding a trial of labor.
In conclusion, the prediction model developed by Grobman et al5 provides individualized and patient-specific chances of VBAC success and has been validated in our study. This instrument may help pregnant women contemplating a trial of labor reach a more informed decision. An easy-to-use calculator using the previously published model can be found at http://www.bsc.gwu.edu/mfmu/vagbirth.html.
2. American College of Obstetricians and Gynecologists. Vaginal birth after previous cesarean. ACOG Practice Bulletin No. 54. Obstet Gynecol 2004;104:203–12.
3. Landon MB, Hauth JC, Leveno KJ, Spong CY, Leindecker S, Varner MW, et al. Maternal and perinatal outcomes associated with a trial of labor after prior cesarean delivery. N Engl J Med 2004;351:2581–9.
4. Macones GA, Peipert J, Nelson DB, Odibo A, Stevens EJ, Stamilio DM, et al. Maternal complications with vaginal birth after cesarean delivery: a multicenter study. Am J Obstet Gynecol 2005;193:1656–62.
5. Grobman WA, Lai Y, Landon MB, Spong CY, Leveno KJ, Rouse DJ, et al. Development of a nomogram for prediction of a vaginal birth after cesarean delivery. Obstet Gynecol 2007;109:806–12.
6. Zweig MH, Campbell G. Receiver-operating characteristic (roc) plots: a fundamental evaluation tool in clinical medicine [published erratum appears in Clin Chem 1993;39:1589]. Clin Chem 1993;39:561–77.
7. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29–36.
8. Lasko TA, Bhagwat JG, Zou KH, Ohno-Machado L. The use of receiver operating characteristic curves in biomedical informatics. J Biomed Inform 2005;38:404–15.
9. Lilliefors HW. On the Kolmogorov-Smirnov test for normality with mean and variance unknown. J Am Statist Assoc 1967;62:399–402.
10. Rossi AC, D’Addario V. Maternal morbidity following a trial of labor after cesarean section vs elective repeat cesarean delivery: a systematic review with metaanalysis. Am J Obstet Gynecol 2008;199:224–31.
11. Troyer LR, Parisi VM. Obstetric parameters affecting success in a trial of labor: designation of a scoring system. Am J Obstet Gynecol 1992;167:1099–104.
12. Flamm BL, Geiger AM. Vaginal birth after cesarean delivery: an admission scoring system. Obstet Gynecol 1997;90:907–10.
13. 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;21:447–53.
14. Smith GC, White IR, Pell JP, Dobbie R. Predicting cesarean section and uterine rupture among women attempting vaginal birth after prior cesarean section. PLoS Med 2005;2:e252.
15. Hashima JN, Guise JM. Vaginal birth after cesarean: a prenatal scoring tool Am J Obstet Gynecol 2007;196:e22–3.