Objective
The purpose of this study was to determine outcomes, after the use of propensity score techniques, to create balanced groups according to whether a woman undergoes elective repeat cesarean delivery (ERCD) or trial of labor (TOL).
Study Design
Women who were eligible for a TOL with 1 previous low transverse incision were categorized according to whether they underwent an ERCD or TOL. A propensity score technique was used to develop ERCD and TOL groups with comparable baseline characteristics. Outcomes were assessed with conditional logistic regression.
Results
The rates of endometritis, operative injury, respiratory distress syndrome, and newborn infant infection were lower and the rates of hysterectomy and wound complication were higher in the ERCD group.
Conclusion
Propensity score techniques can be used to generate comparable ERCD and TOL groups. Some types of maternal morbidity (such as hysterectomy) are higher; other types (such as operative injury) are lower in the ERCD group. Although the absolute risk is low, neonatal morbidity appears to be lower in the ERCD group.
In March 2010, the Agency for Healthcare Research and Quality published the results of an evidence report and technology assessment about women who undergo either a trial of labor (TOL) after previous cesarean delivery or an elective repeat cesarean delivery (ERCD). The evidence report was based almost entirely on observational studies; the lack of randomized trials to compare these 2 approaches to delivery was cited as a critical research gap. It also was noted that “several papers indicate that such a trial is unlikely in the United States,” because of the perception that pregnant women would not be willing to be randomly assigned to a particular delivery approach and the large sample size that would be required to power such a study adequately.
A randomized trial is considered the gold standard of clinical research, given that, among other reasons, it is a design that optimizes the balance of observed and unobserved baseline covariates. Observational studies, conversely, are more prone to selection bias and often result in unbalanced baseline group differences. Various methods (such as stratification, matching, and regression) have been used to minimize this potential bias; however, these methods have limitations with regard to the number of covariates that can be incorporated. This limitation to control for confounding factors was mentioned specifically in the Agency for Healthcare Research and Quality Vaginal Birth After Cesarean evidence report.
One statistical method, the propensity score, which was introduced by Rosenbaum and Rubin, potentially circumvents these covariate restrictions. The propensity score is a single variable that represents the collection of baseline covariates for each individual or patient and is defined as the probability of being treated on the condition of the individual’s covariate values. This score therefore can be used to develop groups that differ according to the exposure of interest but that have otherwise similar baseline characteristics; its use creates “the observational study analog of randomization in experiments.” Although several methods have been proposed to optimize intergroup similarity, Rosenbaum and Rubin found the nearest Mahalanobis metric matching within calipers that were defined by the propensity score to produce the best balance of baseline covariates between groups. Further discussion on propensity score methods can be found in several papers and in textbooks.
Given the past and projected absence of randomized trials that compare ERCD with TOL, propensity scores offer an alternate method to compare the outcomes that are related to the approach to delivery after a previous cesarean delivery. This relatively new statistical technique has been used rarely in obstetrics and gynecology and does not appear to have been used in the context of a TOL. Correspondingly, this analysis was designed to assess, with the use of propensity score techniques, the maternal and neonatal outcomes that are associated with an ERCD compared with a TOL.
Materials and Methods
From 1999 through 2002, data were collected in a registry (the Cesarean Registry) by the Eunice Kennedy Shriver National Institute of Child Health and Human Development Maternal-Fetal Medicine Units Network. This observational study that included data on all women with a previous cesarean delivery who underwent delivery was conducted at 19 clinical centers throughout the United States and was approved by the institutional review board of each participating center. Study personnel at the participating medical centers abstracted data from patient charts under a waiver of informed consent. Further detail on the Cesarean Registry can be obtained from previously published articles.
This secondary analysis of the Cesarean Registry included women with a live, singleton, term, and vertex gestation and 1 previous low transverse incision (LTCS) who had an ERCD or a TOL whose labor began spontaneously. Term was defined as >36 weeks 6 days of gestation. An ERCD was defined as a cesarean delivery without any indication other than the previous cesarean delivery. This group included women whose cesarean delivery was scheduled and those women who chose cesarean delivery, in the absence of any other indication, after arriving in labor or with rupture of membranes. Correspondingly, women who had a repeat cesarean delivery for indications such as placenta previa or active herpes were excluded. Also, women were ineligible for the cohort if they had an ERCD at <39 weeks of gestation without spontaneous labor or premature rupture of membranes. Women with an ERCD and spontaneous labor or rupture of membranes at <39 weeks of gestation were included because they were eligible for, and would need to choose between, the 2 approaches to delivery. Women with induced labor also were excluded because this intervention has been associated with a lower probability of success and a higher probability of uterine rupture and is not purely a probabilistic possibility but a choice that a woman and her provider can make. A woman who is considering undergoing a spontaneous TOL or proceeding with an ERCD can be counseled only about the 2 approaches, either before or at the time of her labor, if the outcomes that are related to each alternative alone are known. This would be true, as well, if she were faced with the choice of undergoing a labor induction or an ERCD (in which case outcomes of induction vs an ERCD would be studied). The suitability of analyzing the cases of women whose labor is induced separately from those in spontaneous labor, in the appropriate circumstance, has been recognized in the literature.
Propensity scores were used to control for potential covariate imbalances and to create maximally nonbiased groups. The propensity scores were generated by logistic regression and represent the probability of being delivered by an ERCD given baseline covariate values. Propensity scores were based on the ERCD group to ensure as many matches as possible could be made because the number of patients in this group was less than that in the TOL group. All covariates that were known at 37 weeks of gestation that existed before labor were included (n = 43) in the logistic regression without a stepwise procedure. This included 18 dichotomous variables that were used to represent the different centers from which patients were enrolled. One-to-1 matching without replacement was conducted with the nearest Mahalanobis distance within calipers defined by the logit of the propensity score. The Mahalanobis distance was calculated by identifying, a priori, 3 variables that highly influenced the probability of cesarean delivery: maternal age, no previous vaginal delivery, and recurrent cesarean delivery indication (ie, cephalopelvic disproportion or failure to progress as the indication for the previous cesarean delivery). This matching method randomly ordered the treated subjects and chose the first subject in the list. All patients who underwent a TOL who were within 0.25 standard deviation of the logit of the propensity score (or the caliper) of the identified subject in the ERCD group were selected. The pair with the smallest Mahalanobis distance was then removed and saved for use in analysis. The process was repeated until as many matches as possible could be made for the women in the ERCD group.
The 2-sample Kolmogorov-Smirnov test was used to determine whether the distributions of the propensity scores were similar between the 2 groups. The standardized difference was used to assess the balance of the covariates because it is a property of the sample and, unlike significance testing, does not depend on the size of the sample. A standardized difference greater than the absolute value of 10 was used to indicate that the samples were meaningfully different. For the matched cohort, outcomes were compared with the use of the McNemar’s test, and odds ratios and 95% CIs were generated with the use of conditional logistic regression. The CIs for the number that was needed to treat were calculated by the Wilson score method.
The evaluated maternal outcomes through delivery were endometritis (clinical diagnosis of puerperal uterine infection in the absence of findings that suggested another source), wound complication (seroma, hematoma, or infection), uterine rupture (disruption or tear of the uterine muscle and visceral peritoneum or a uterine muscle separation with extension to adjacent structures), operative injury (broad ligament hematoma, cystotomy, or bowel or ureteral injury), peripartum hysterectomy, thromboembolism (deep vein thrombosis or pulmonary embolus), and maternal death. Evaluated neonatal outcomes that were collected up to 120 days after delivery or at the time of hospital discharge were transient tachypnea of the newborn infant, respiratory distress syndrome, infection that included suspected or confirmed sepsis, acidemia (arterial cord pH <7.0), hypoxic ischemic encephalopathy, and neonatal death. Cases of uterine rupture, hypoxic ischemic encephalopathy, and maternal death were confirmed by central review. For mothers and neonates, a separate composite adverse outcome variable that represented ≥1 of these outcomes also was generated. An additional composite outcome that included only major maternal morbidities was developed and was defined as the occurrence of hysterectomy, uterine rupture, or operative injury.
Statistical analysis was conducted with SAS software (SAS Institute Inc, Cary, NC); for the matched analysis, the R and the R MatchIt library was used ( http://www.r-project.org/ ). A 2-tailed nominal probability value of < .05 was considered significant. No adjustments were made for multiple comparisons.
Results
Within the Cesarean Registry, 22,068 women with a live, singleton, vertex birth with 1 previous LTCS who delivered at term after undergoing a TOL or an ERCD were available for analysis ( Figure 1 ) . Seven-hundred fifty-nine women who had indications for cesarean delivery that would normally preclude a TOL (such as nonreassuring antepartum testing, malpresentation, placenta previa, myomectomy and active genital herpes) were excluded. Similarly, to ensure that women who underwent repeat cesarean delivery truly had no indication for the cesarean delivery other than their choice, women who were reported to have had a cesarean delivery that was elective but who had a reported indication that implied that this was not the case (ie, cephalopelvic disproportion, failure to progress, cord prolapse, nonreassuring tracing, or abruption) were excluded (n = 262). An additional 3188 women who had an ERCD at <39 weeks of gestation and 3235 women who underwent labor induction were not included in this analysis. Patients who carried fetuses with aneuploidy or congenital malformations (which included trisomy 18 and 21, clubbed foot, cleft palate, and neural tube defect; n = 120) were removed as well. These women were excluded because the perinatal outcome of interest could be biased by their inclusion. This process left 14,504 women for analysis, of whom 8297 women had a TOL, and 6207 women had an ERCD. Of the 8297 women who attempted a TOL, 6421 women (77.4%) had a successful vaginal birth after a previous cesarean delivery. Of the 6207 women who had an ERCD, 4972 women (80.1%) delivered ≥39 weeks of gestation without labor, and 1235 women (19.9%) delivered between 37.0 and 38.6 weeks of gestation because of the onset of labor or premature rupture of membranes.
