Pregnant women who have had a prior cesarean delivery often are confronted with the decision of whether to attempt a trial of labor. One important component in this decision-making process is the likelihood that a trial of labor will result in a vaginal delivery. Correspondingly, investigators have attempted to elucidate the factors that are associated with successful vaginal birth after cesarean delivery (VBAC). Some of the factors that have been repeatedly demonstrated to improve the chance of a VBAC include a history of a vaginal birth, a nonrecurrent indication for cesarean delivery, and a lower body mass index.1 Nevertheless, although the identification of these factors allows the physician to provide a patient with some general guidance regarding the likelihood of achieving a VBAC, knowledge of these factors does not allow the physician to provide an individual woman who is considering a trial of labor with her own unique risk assessment.
Several investigators have attempted to develop models that could provide physicians with the ability to use multiple clinical factors to more accurately predict the chance of a VBAC for a given woman. For example, after identifying four variables negatively associated with VBAC, Troyer and Parisi2 accorded each of these variables one point if it was present in a woman undergoing a trial of labor after a prior cesarean. The total number of points accumulated by a woman was inversely correlated with her chance of VBAC. Other investigators have developed similar models.3–5 These predictive models, however, have not entered into widespread clinical use for a variety of reasons. First, these models have been based on retrospectively obtained data, thereby limiting incorporation of important but difficult to ascertain variables. Although these models have allowed some refinement in predictive ability, they still result in relatively crude categorizations, thereby limiting application of the model to any given individual. Also, all these models incorporate intrapartum factors. Ideally, the counseling regarding VBAC should take place before labor and delivery.
Predictive nomograms, although not yet widely used in obstetrics, have been successfully used in other medical disciplines to aid in medical decision making.6 These nomograms allow the independent contribution of multiple factors to be incorporated to provide a continuous probability prediction for a given outcome. Moreover, the graphic depiction of the nomogram facilitates its accessibility and clinical use. In this study, we have investigated the use of a nomogram to help in the prediction of achieving a VBAC for those women who undertake a trial of labor.
PARTICIPANTS AND METHODS
Nineteen academic medical centers belonging to the National Institute of Child Health and Human Development Maternal–Fetal Medicine Units Network participated in a study of pregnant women with prior cesareans between 1999 and 2002. Eight centers participated throughout the study: five participated only during the first 2 years, and six participated for part of the last 2 years. At each center, trained and certified research nurses concurrently identified women who were admitted for delivery and who had a history of cesarean delivery. Those women who were identified had their charts abstracted for demographic data, medical and obstetric history, and intrapartum and postpartum events. All data were transmitted weekly to the data-coordinating center and were edited for missing, out of range, and inconsistent values. Approval for the study was obtained from the institutional review board of each institution. Further details of the methodology of this study have been previously described.7
This analysis concerns those women in the registry with a vertex singleton gestation and one prior low-transverse cesarean delivery who underwent a trial of labor after 36 6/7 weeks of gestation. Women with an antepartum intrauterine fetal demise were excluded. Of the information that had been abstracted from the patient's charts, those factors that were considered for use in the predictive model were only those that could be ascertained at a first prenatal visit. These factors included demographic variables (maternal age, ethnicity, prepregnancy body mass index), variables related to the prior cesarean delivery (recurrent indication, length of time since cesarean), variables related to obstetric history (any prior vaginal delivery, number of prior vaginal deliveries, any vaginal delivery subsequent to cesarean, prior preterm vaginal deliveries only, maximal birth weight of a prior child), and variables reflective of pre-existing medical conditions. Recurrent indication for a cesarean delivery was defined as arrest of dilation or descent as the indication for the prior cesarean. A pre-existing medical condition was judged to exist if a patient had either asthma, chronic hypertension, renal disease, heart disease, or a connective tissue disorder. The association of pregestational diabetes mellitus with VBAC was analyzed separately from these other medical conditions.
In the initial phase of nomogram development, different classification techniques, including classification tree, logistic regression, random forest, support vector machine, and boosting analysis, were compared according to their classification errors for their ability to identify the most optimally predictive model.8–11 Marginal exploratory analysis was performed on the maternal age and body mass index (BMI) values to analyze whether these were best represented in the model as continuous or categorical variables. Based on this analysis, maternal age and BMI were used as continuous variables in classification analysis. The classification errors were estimated through a cross-validation procedure. Thus, the original data set was randomly and equally divided into a training set and a test set. The training set was used to identify predictive factors and build predictive models. Then the independent test set was used to estimate the classification errors. After analyzing the different classification techniques, logistic regression was chosen as the technique of choice because of its competitively low classification error.
The logistic regression model developed from the training set was validated as follows. After applying the regression model to the test set, the predicted probabilities of successful VBAC were partitioned into 10 groups (eg, 0–10%, 11–20%, etc). The midpoints of these probability ranges (eg, 5%, 15%, etc) were used to represent these groups. In each group, the proportion of women with successful VBAC was calculated to estimate the empirical probability of VBAC success. The scatter plots of the predicted and empirical VBAC probabilities were smoothly connected to form a curve. The ideal validation would generate a 45º straight line. Corresponding 95% confidence intervals for the curve were calculated from the normal approximation.
A graphic nomogram was generated to represent the logistic regression model. Using this nomogram, the predicted chance of VBAC for women who underwent trial of labor was calculated for different hypothetical patients. Ninety-five percent confidence intervals for the predicted chances were provided as well. Statistical analysis was performed with SAS 8.2 (SAS Institute, Cary, NC).
Of the 11,856 women who met inclusion criteria, 8,659 (73.0%) had successful VBACs. Descriptive characteristics of the population are presented in Table 1. This population was diverse with respect to both demographic factors (such as ethnicity) and obstetric factors (such as history of vaginal delivery). Seven thousand six hundred sixty women had complete sets of values for all variables under consideration, and this was the final population used for analysis.
The multivariable logistic regression model that was built to predict the probability of a successful VBAC revealed that multiple patient factors were associated with vaginal delivery. The full logistic regression equation can be found in the box, “Logistic Regression Equation for Prediction of Achieving VBAC After a Trial of Labor.” In this regression, VBAC was significantly more likely among women who were younger, had a lower prepregnancy BMI, and were of white race. Other factors that were significantly associated with a greater chance of VBAC were a nonrecurrent indication for the cesarean delivery and a history of vaginal delivery, particularly if this vaginal delivery was subsequent to the prior cesarean. The results of the multivariable logistic regression, presented with odds ratios and corresponding 95% confidence intervals, are presented in Table 2. The corresponding receiver operating characteristics curve, with an area under the curve of 0.754 (95% confidence interval 0.742–0.766), is presented in Figure 1.
The graphic nomogram derived from the logistic regression is presented in Figure 2. This nomogram is used by locating each patient characteristic and finding the number of points, on the uppermost point scale, to which that characteristic corresponds. For example, a maternal age of 26 years corresponds to approximately 10 points. Once the number of points generated from all of a patient's characteristics are added together, that sum is found on the “total points” scale, and the predicted probability of VBAC is the probability on the lowermost scale that is obtained by drawing a vertical line from the “total points” to the “probability” scale. Thus, a patient whose characteristics result in 60 total points has an approximately 78% chance of having a VBAC.
The cross-validation procedure showed that the performance of the nomogram on the test set was similar to that originally determined from the training set, with an area under the curve that remained 0.75. Figure 3 compares the predicted rates of VBAC with the empirical probabilities of VBAC for women in the test set. The estimated curve and its 95% confidence band confirm the overall consistency between the predicted and empirical probabilities and the adequate calibration of the nomogram.
Lastly, examples of the nomogram's predictive capability are illustrated by the calculation of VBAC success for four hypothetical pregnant women. Table 3 presents the characteristics of these women, the total number of points generated by these characteristics from the nomogram, and the final predicted chance (with the corresponding 95% confidence interval) of VBAC for each individual woman.
To this point in time, physicians have had several potential strategies that could be used to counsel a woman about her probability of having a VBAC if she undergoes a trial of labor. The simplest strategy uses the reported success rate, or range of rates, for the entire population of women who attempt a VBAC. When counseled with this strategy, a woman would be told her chances of a VBAC are approximately 60–80%.12 Although clearly simple, this approach does not make any attempt to individualize prediction, and women of vastly different risk status are placed into one group of similar risk.
A more individualized approach makes use of the many factors that have been associated with VBAC. Some examples of these factors include maternal age, body mass index, and history of vaginal delivery.1 Using these factors, physicians can try to assess how a woman's chance of VBAC may differ from the average of the overall population. Although this approach acknowledges the differences among women, it is essentially qualitative and does not allow the condensation of multiple patient characteristics into a specific and accurate assessment of the likelihood of success.
In an effort to incorporate a woman's individual characteristics and also provide a quantitatively determined point estimate of VBAC probability, some investigators have proposed the use of prediction models.2–5 In all of these models, patient characteristics associated with VBAC generate a certain number of points, and the total number of points accumulated indicates a woman's chance of VBAC. Yet, the models that have been put forth so far have several methodological limitations. In these scoring systems, the points that contribute to the final score are either not based or are only loosely based upon the actual magnitude of the association between the patient characteristic and VBAC outcome. Also, the range of points is limited, such that patients of very different risk may still appear to have an equivalent probability of VBAC. Lastly, these models use either intrapartum factors, which preclude counseling during the antepartum period, or factors such as birth weight that are not in fact measurable before delivery.
This predictive nomogram overcomes many of these limitations. This model includes only factors that are available at a first prenatal visit, allowing caregivers the possibility of providing accurate counseling early during the prenatal course. The number of points accorded to each patient characteristic is mathematically determined, and this type of determination optimizes accuracy of prediction. Moreover, because the probability outcomes in the nomogram are continuous, women can be given specific estimates of VBAC probability that reflect the many different combinations of characteristics that different patients may have. Indeed, this graphic nomogram can provide physicians and patients alike with a reasonably accurate assessment of a woman's chance of achieving a VBAC if she has a trial of labor. As evidenced in Figure 3, the chance of VBAC predicted by the model is extremely similar to the empirical chance of VBAC. The 95% confidence interval gives further reassurance of the model's precision. Only when the empirical chance of VBAC is quite unlikely (less than 35%) does the precision of the predictive model deteriorate. Yet, from a clinical standpoint, this imprecision is unlikely to be of great importance because many physicians and their patients would consider any estimate in this range to be a disincentive to attempting a VBAC.
This model cannot be used to predict outcome for all women considering a trial of labor. It has been generated from women at term with one prior cesarean and thus cannot be applied to women who are preterm or have undergone more than one cesarean. The inclusion criteria were chosen because they describe the large majority of those who attempt VBAC.7 Also, the model has been developed and internally validated from data derived from a consortium of academic medical centers. Future analyses will need to ascertain if its results are equally valid among patient populations in different hospital settings. Finally, predicted probabilities derived from this model may need to be modified based on circumstances that can arise near the end of gestation, such as the need for an induction of labor.
There is not one correct answer as to the probability of VBAC at which women should attempt a trial of labor. Even women with similar chances of a VBAC may pursue different choices. This situation is similar to that of prenatal diagnosis for aneuploidy. In that setting, women with similar risks may choose very different strategies of care depending on the strength of desire for wanted outcomes and risk aversion for adverse outcomes. Just as we enhance a woman's decision-making ability through the provision of an accurate aneuploidy risk estimate, we can similarly enhance her decision to proceed with a trial of labor through the provision of specific VBAC success rates. This prediction model, which provides a format designed to facilitate clinical application is available at http://www.bsc.gwu.edu/mfmu/vagbirth.html.
1. Landon MB, Leindecker S, Spong C, Hauth JC, Bloom S, Varner MW, et al. The MFMU Cesarean Registry: factors affecting the success of trial of labor after previous cesarean delivery. Am J Obstet Gynecol 2005;193:1016–23.
2. Troyer AR, Parisi VM. Obstetric parameters affecting success in a trial of labor: designation of a scoring system. Am J Obstet Gynecol 1992;167:1099–104.
3. Pickhardt MG, Martin JN Jr, Meydrech EF, Blake PG, Martin RW, Perry KG Jr, et al. Vaginal birth after cesarean delivery: are there useful and valid predictors or success or failure? Am J Obstet Gynecol 1992;166:1811–9.
4. Flamm BL, Geiger AM. Vaginal birth after cesarean delivery: an admission scoring system. Obstet Gynecol 1997;90:907–10.
5. Weinstein D, Benshushan A, Tanos V, Zilberstein R, Rojansky N. Predictive score for vaginal birth after cesarean section. Am J Obstet Gynecol 1996;174:192–8.
6. Eastham JA, Kattan MW, Scardino PT. Nomograms as predictive models. Semin Urol Oncol 2002;20:108–15.
7. 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.
8. Lemon SC, Roy J, Clark MA, Friedmann PD, Rakowski W. Classification and regression tree analysis in public health: methodological review and comparison with logistic regression, Ann Behav Med 2003;26:172–81.
9. Breiman L. Random forests. Machine Learning 2001;45:5–32.
10. Vapnik V. The nature of statistical learning theory. New York (NY): Springer-Verlag; 1995.
11. Freund Y, Schapire RE. A decision-theoretic generalization of on-line learning and an application to boosting. J Comput System Sci 1997;55:119–39.
12. Guise JM, Merlin M, McDonagh M, Osterweil P, Chan B, Helfand M. Safety of vaginal birth after cesarean: a systematic review. Obstet Gynecol 2004;103:420–9.
In addition to the authors, other members of the National Institute of Child Health and Human Development Maternal–Fetal Medicine Units Network are as follows:
- Ohio State University: J. Iams, S. Meadows, H. Walker
- University of Alabama at Birmingham: J. Hauth, A. Northen, S. Tate
- University of Texas Southwestern Medical Center: S. Bloom, D. Bradford
- University of Utah: M. Belfort, F. Porter, B. Oshiro, K. Anderson, A. Guzman
- University of Chicago: J. Hibbard, P. Jones, M. Ramos-Brinson, M. Moran, D. Scott
- University of Pittsburgh: K. Lain, M. Cotroneo, D. Fischer, M. Luce
- Wake Forest University: M. Swain, C. Moorefield, K. Lanier, L. Steele
- Thomas Jefferson University: A. Sciscione, M. DiVito, M. Talucci, M. Pollock
- Wayne State University: M. Dombrowski, G. Norman, A. Millinder, C. Sudz, B. Steffy
- University of Cincinnati: T. Siddiqi, H. How, N. Elder
- Columbia University: F. Malone, M. D‘Alton, V. Pemberton, V. Carmona, H. Husami
- Brown University: H. Silver, J. Tillinghast, D. Catlow, D. Allard
- Northwestern University: M. Socol, D. Gradishar, G. Mallett
- University of Miami, Miami, FL: G. Burkett, J. Gilles, J. Potter, F. Doyle, S. Chandler
- University of Tennessee: W. Mabie, R. Ramsey
- University of Texas at San Antonio: S. Barker, M. Rodriguez
- University of North Carolina: K. Moise, K. Dorman, S. Brody, J. Mitchell
- University of Texas at Houston: L. Gilstrap, M. Day, M. Kerr, E. Gildersleeve
- Case Western Reserve University: P. Catalano, C. Milluzzi, B. Slivers, C. Santori
- The George Washington University Biostatistics Center: E. Thom, S. Leindecker, H. Juliussen-Stevenson, M. Fischer
- National Institute of Child Health and Human Development: D. McNellis, K. Howell, S. Pagliaro