Prostaglandins and cesarean delivery for non-reassuring fetal status in patients delivering small for gestational age neonates at term

Abstract

Objective:

Growth restricted fetuses have been excluded from many randomized trials of prostaglandin for labor induction. As prostaglandins, particularly misoprostol, are associated with increased rates of cesarean delivery for non-reassuring fetal heart tracing, it is important to assess their safety in pregnancies at higher risk of this complication. The objective of this study was to estimate the association between use of prostaglandins for labor induction in term singleton pregnancies complicated by delivery of small for gestational age (SGA) neonates and the risk of cesarean delivery for non-reassuring fetal status (NRFS).

Materials and Methods:

Retrospective cohort study of singleton deliveries ≥37 weeks following induction of labor in patients with SGA (birthweight <10% percentile for gestational age). Patients with prior cesarean delivery or neonates with major congenital anomalies were excluded. Patients were categorized by exposure to prostaglandins. The primary outcome was cesarean delivery for NRFS. Secondary outcomes were any cesarean delivery, a composite of 5-minute Apgar score <7, admission to the neonatal intensive care unit, or neonatal death, and a composite of maternal morbidity (transfusion, postpartum hemorrhage, wound infection, endometritis, fever). Propensity scores for exposure were estimated using a logistic regression model including parity, comorbidities, and Bishop score. Stabilized weights from inverse probability of treatment weighting were used. Outcomes were compared with relative risks (RR) and 95% confidence intervals (CI).

Results:

There were 1097 patients: 587 (53.5%) exposed to prostaglandins and 510 (46.5%) unexposed. Covariates were balanced in the stabilized sample. Overall 166 (15.1%) patients had cesarean deliveries for NRFS. In unadjusted analysis, prostaglandin use was associated with an increased relative risk of cesarean for NRFS (18.3% vs 11.0%, RR 1.71, 95% CI 1.27-2.30. In propensity-score weighted analysis, the relative risk for cesarean for NRFS was 1.22 (95% CI 0.93-1.59). There was no significant association between prostaglandin exposure and all-cause cesarean delivery, maternal morbidity or neonatal morbidity.

Conclusion:

In propensity-score analysis there was no association between the use of prostaglandins for labor induction at term and cesarean for NRFS in pregnancies complicated by SGA. However, given the retrospective nature of the study these results should be interpreted with caution.

INTRODUCTION

In 2015, 23.8% of all labors in the United States were induced [1]. Given the ubiquity of this procedure, defining the safest and most effective induction agents is essential. However, patients delivering small for gestational age (SGA) neonates have been excluded from many randomized trials of prostaglandins for labor induction and have been underrepresented in others, despite undergoing induction of labor at rates approaching 50%.[2-5] As prostaglandins, particularly misoprostol, are associated with increased rates of cesarean delivery for non-reassuring fetal heart tracing, although not with neonatal morbidity, it is important to assess their safety in fetuses at higher risk, including those with growth restriction.[6, 7]

To date there are only a few studies, including two small randomized controlled trials, specifically focusing on induction agents in pregnancies complicated by SGA.[7-12] In the absence of larger trial data, other analytic approaches are required to better define medication risks. One of the most significant limitations of observational studies for this question is confounding by indication, where clinicians choose treatment based on the clinical scenario at hand [13]. Propensity score analysis is one tool to reduce confounding by indication [14]. A propensity score is the “probability of treatment assignment conditional on observed baseline characteristics.”[15] When specified correctly, propensity scores balance the distribution of covariates in the treated and untreated groups, which reduces confounding by indication [15]. Propensity scores can also allow for positivity, where each patient has a defined probability of being treated or not treated [16]. Thus, the use of propensity scores can allow for direct estimation of risk and risk ratios in observational studies, avoiding the need for logistic regression and the use of less intuitive odds ratios.[14, 17]

The objective of this study was to use propensity score analysis to estimate the association between use of prostaglandins for labor induction at term in patients whose pregnancies are complicated by SGA and the risk of cesarean delivery for non-reassuring fetal status (NRFS). We hypothesized that exposure to prostaglandins would be associated with an increased risk of cesarean delivery for NRFS compared to other labor induction methods.

MATERIALS AND METHODS

This is was a secondary analysis of a cohort study of nonanomalous singleton deliveries at ≥37 weeks gestational age at Barnes-Jewish Hospital in St. Louis from 2004–2014. The study was approved by the Washington University in St. Louis Human Research Protection Office. Full details of the study have been published previously [18]. For this analysis, patients who presented for labor induction for any indication with Bishop score ≤6 and delivered liveborn SGA singletons were included. SGA was defined as birthweight at <10% percentile for gestational age based on recent national birthweight data [19]. Gestational age at delivery was based on best obstetrical estimate. Patients with prior cesarean delivery or delivering neonates with major congenital anomalies were excluded.

Demographic, clinical, and delivery data were abstracted from the medical record by trained obstetrical research staff. Patients were categorized by exposure to prostaglandins (misoprostol or dinoprostone). The primary outcome was cesarean delivery for NRFS. Secondary outcomes were cesarean delivery for any reason, composite neonatal morbidity (5-minute Apgar score <7, neonatal intensive care unit admission, or death) and maternal morbidity (transfusion, postpartum hemorrhage, postpartum wound infection, postpartum endometritis, postpartum fever).

Baseline characteristics were compared between the exposed and unexposed groups. Categorical variables were compared using the chi-square or Fisher exact test, as appropriate. Continuous variables were checked for normality using the Kolmogorov–Smirnov test and then compared using the Student t-test or the Mann-Whitney U test.

Propensity scores for exposure to prostaglandin were estimated in an iterative process using a logistic regression model. Variables were selected based on a theoretical basis as known or potential confounders. The variables included maternal age (as a continuous variable), use of regional anesthesia, Bishop score at admission, obesity (body mass index ≥30 kg/m²), nulliparity, chronic hypertension, hypertensive disorder of pregnancy (gestational hypertension or preeclampsia), gestational diabetes, tobacco smoking, race (black, white, other), oligohydramnios as indication for induction, and nonreassuring fetal test as indication for induction (due to nonreactive nonstress test, decelerations, or abnormal biophysical profile score). Higher order variables were added as necessary until the standardized difference between the treated and untreated group was <0.1 for all tested covariates after weighting (as described below). The final model included the above variables as well as Bishop score on admission squared and cubed. No interaction terms were included.

The inverse probability of treatment weight (IPTW) was calculated as the inverse of the probability of receiving the treatment that was actually received [20]. Stabilized weights were calculated as the proportion of patients receiving each treatment divided by the individual probability of receiving the treatment actually received. Stabilized weights were truncated at the 1^st and 99^th percentiles with all weights below or above those percentiles replaced with the 1^st or 99^th percentile value respectively [21]. The distribution of the weights was examined graphically.[15, 16, 20, 22] The balance of the covariates between exposed and unexposed patients was assessed in two ways. First, the standardized differences in each parameter between the exposed and unexposed group were calculated [17]. For continuous variables we also assessed standardized differences of higher-order moments (squares and cubes) as recommended by Austin [20]. There is no clear cutoff of standardized difference to determine balance, because a better balance (i.e. smaller standardized difference) may be more relevant for “prognostically important covariates” than for less important ones [17]. However, in general a standardized difference of less than 0.1 is considered optimal [23]. Second, box plots were used to visually compare the distribution of the continuous covariates.

After finalization of the propensity score model, the relative risks and 95% confidence intervals for the primary and secondary outcomes were calculated using log-linear regression using stabilized weights [24]. Although when using IPTW it is recommended that robust variance estimation or bootstrapping be used for confidence interval estimates, this is not necessary when using stabilized weights.[20, 25]

Because exposure to oxytocin could not be balanced with propensity scores, as a sensitivity analysis we included oxytocin exposure as a covariate in the regression models estimating the association between prostaglandin exposure and the outcomes. Additionally, we examined the association between prostaglandin use and the primary outcome in the subgroup of patients with suspected SGA prior to delivery. Since the diagnosis of SGA cannot be made until after delivery, this last analysis represents the situation facing the provider. Finally, the propensity score analysis does not permit the examination of the individual association between the potential confounders and the outcome (since propensity score is used instead of each individual variable in the final model). Therefore, we performed a post-hoc logistic regression to estimate the association between each of the covariates and the primary outcome.

Data analysis was performed in SAS 9.4 (SAS Institute, Cary, NC). Tests with two-sided p-value <0.05 were considered statistically significant. Due to the fixed sample size, an a priori power analysis was not done.

RESULTS

The study flow diagram is in Figure 1. Overall there were 1097 patients meeting inclusion criteria, of whom 587 (53.5%) were exposed to prostaglandins and 510 (46.5%) were not. In patients receiving prostaglandins, misoprostol was the most common type of prostaglandin used (432 patients, 73.6%). Dinoprostone was used in 121 (20.6%) patients, 32 (5.5%) patients had both, and 2 patients (0.3%) did not have the type of prostaglandin reported. Prostaglandin use was associated with nulliparity and low Bishop score (p<0.001) (Table 1). The primary reason for induction was maternal comorbidity in 278 (25.3%) patients, non-reassuring fetal status is 151 (13.8%) patients, obstetric (rupture of membranes, late-term, etc.) in 80 (7.3%) patients, suspected growth restriction in 375 (34.2%) patients, and elective in 213 (19.4%) patients.

Figure 1:

Study flow diagram for enrollment and inclusion in analysis.

Table 1:

Baseline Characteristics (n=1097)

Characteristic	Prostaglandin N=587	No prostaglandin N=510	P*	Standardized Difference before Weighting	Standardized Difference After Weighting
Maternal age, y	23 (20,28)	23 (20,28)	0.52	−0.023	−0.0040
Gestational age at delivery, wk	38 (37,40)	39 (37,40)	<0.001	0.21	−0.0072
Race			0.87
Black	442 (75.3)	391 (76.6)		−0.032	0.022
White	105 (17.9)	86 (16.9)		−0.027	0.036
Other	40 (6.8)	33 (6.5)		0.014	−0.078
BMI > 30 kg/m2	308 (52.4)	230 (45.1)	0.015	0.15	0.047
Nulliparity	426 (72.6)	278 (54.5)	<0.001	0.38	0.045
Chronic hypertension	28 (4.8)	16 (3.1)	0.17	0.084	−0.03
Preeclampsia	31 (5.3)	19 (3.7)	0.22	0.075	0.031
Gestational diabetes mellitus	21 (3.6)	8 (1.6)	0.039	0.13	−0.0053
Tobacco use	112 (19.1)	117 (22.9)	0.12	−0.095	−0.018
Regional anesthesia	555 (94.6)	477 (95.5)	0.48	0.043	−0.039
Oligohydramnios	53 (9.0)	44 (8.6)	0.81	0.014	−0.045
Nonreassuring fetal status onadmission	74 (12.6)	77 (15.0)	0.23	−0.072	−0.042
Suspected SGA on admission	267 (45.5)	200 (39.2)	0.036	0.16	0.014
Bishop score	1 (1,2)	3 (2,4)	<0.001	−1.22	−0.066
Oxytocin use	490 (83.5)	509 (99.8)	<0.001	−0.62	−0.65
Foley bulb use	221 (37.7)	90 (17.7)	<0.001	0.46	0.039

Data are n (%) or median (interquartile range)

BMI: body mass index, SGA: small for gestational age

^*P from chi-square or Mann-Whitney-U, prior to weighting

The range of the IPTW weights was 1.02–22.9 with a mean of 2.00, and for the stabilized weights the mean was 1.00 with a range from 0.48–10.65. The 1^st and 99^th percentiles of the stabilized weights (for truncation) were 0.48 and 5.95 respectively. There was good overlap between stabilized weights for treated and untreated patients. Covariates were balanced in the stabilized sample with standardized differences of all covariates except oxytocin of <0.1. (Table 1).

Overall 166 (15.1%) patients had cesarean deliveries for NRFS. In unadjusted analysis, prostaglandin use was associated with an increased relative risk of cesarean for NRFS (18.3% vs 11.0%, RR 1.71, 95% CI 1.27–2.30. There was also an increased risk of cesarean for any indication, but no significant increase in risk of adverse neonatal outcome or maternal morbidity (Table 2). In propensity-score weighted analysis, the relative risk for cesarean for NRFS was 1.22 (95%CI 0.93–1.59). The addition of oxytocin exposure to the regression model did not substantially change the estimated relative risk (1.14, 95% CI 0.86–1.53).

Table 2:

Outcomes

Outcome	Prostaglandin N=587 N(%)	No Prostaglandin N=510 N(%)	Unadjusted RR (95% CI)	RR from Stabilized IPTW (95% CI)
Cesarean for non-reassuring fetal status	110 (18.7)	56 (11.0)	1.71 (1.27-2.30)	1.22 (0.93-1.59)
Cesarean for any indication	153 (26.1)	76 (14.9)	1.75 (1.37-2.24)	1.07 (0.85-1.33)
Adverse neonatal outcome*	14 (2.4)	15 (2.9)	0.81 (0.40-1.66)	0.43 (0.21-0.88)
Adverse maternal outcome†	31 (5.3)	15 (3.0)	1.80 (0.98-3.29)	1.20 (0.68-2.12)

RR relative risk; CI confidence interval, IPTW inverse probability of treatment weight

^*Composite of 5-minute Apgar <7, admission to neonatal intensive care unit, or death

^†Composite of maternal fever, wound infection, endometritis, postpartum hemorrhage or blood transfusion

There was no significant association between prostaglandin exposure and maternal morbidity. The number of adverse neonatal outcomes was small: there were no deaths and there were 11 neonates in each group with 5-minute Apgar <7. The relative risk for adverse neonatal outcome for prostaglandin exposed patients was 0.43 (95%CI 0.21–0.88). Finally, in the group of patients with suspected SGA prior to admission (n=467), there was no association between exposure to prostaglandins and cesarean delivery for NRFS (RR 1.04, 95%CI 0.19–5.58).

In a post-hoc analysis we found that nulliparity and obesity were significantly associated with cesarean for NRFS in a fully adjusted model of prostaglandin exposure and this outcome, with an odds ratio of 3.57 (95%CI 2.20, 5.81) for nulliparity and 1.78 (95%CI 1.23, 2.59) for obesity.

DISCUSSION

We found that in patients undergoing induction of labor at term and delivering singleton SGA neonates, exposure to prostaglandins was not associated with an increased risk for cesarean delivery for NRFS after adjusting for confounders in propensity score modeling. Furthermore, there was no increased risk of neonatal or maternal morbidity.

There is a relative paucity of data on the safety of prostaglandin use for induction of labor in pregnancies complicated by SGA. In an analysis of the National Institute of Child Health and Human Development sponsored Consortium on Safe Labor database, there was no difference in cesarean delivery or neonatal morbidity rates in patients induced with misoprostol versus other induction agents [9]. Although a retrospective cohort study of 14,294 patients from the Netherlands showed an increased risk of emergency cesarean delivery with exposure to prostaglandins, there was no attempt to adjust for confounding as all the relative risks presented were unadjusted. Similarly, a retrospective cohort study of 222 growth restricted fetuses showed over 4-times higher risk of cesarean delivery for NRFS in fetuses exposed to prostaglandins compared to those induced without prostaglandins [11]. Finally, in a retrospective cohort of 99 IUGR fetuses comparing dinoprostone, misoprostol, and Cook balloon, the rate of cesarean for NRFS was 7.3% in the combined prostaglandin group compared to 8.6% in the Cook balloon group [8]. To our knowledge there has only been one randomized clinical trial specifically addressing prostaglandin use in pregnancies complicated by growth restriction. In this trial of 25 mcg of vaginal misoprostol compared to Foley bulb for induction of labor in 100 patients in India, there was no statistical significance in the difference in cesarean delivery rate, and there was no difference in tachysystole [10]. The rates of cesarean delivery for NRFS were similar between the two groups [10]. Thus, overall our results are concordant with the published literature supporting a lack of association between prostaglandin use and cesarean for NRFS in this population.

The findings of our study add to the body of literature that suggests that prostaglandin use is not associated with cesarean for NRFS in patients delivering SGA neonates at term. The primary mechanism thought to lead from prostaglandins to cesarean for NRFS is tachysystole [6]. However, although prostaglandins may be associated with an increased rate of tachysystole, this may be resolved with conservative measures and not uniformly require cesarean delivery [6]. Indeed, induction of labor with vaginal prostaglandins, compared to other methods, has the highest rate of vaginal delivery within 24 hours, even though the rates of tachysystole are also increased [26]. Randomized controlled trials are needed to fully elucidate the optimal induction methods in this patient population. However, based on our findings and the bulk of the literature it is reasonable to use prostaglandins for induction in these patients.

The sharp decrease in the RR for the primary outcome from 1.71 to 1.22 after propensity score analysis deserves comment. In our post-hoc exploratory analysis, we found that relationship between prostaglandin use and cesarean for NRFS was confounded most strongly by nulliparity and obesity. Since multiparous patients with no prior cesarean are more likely to have a vaginal delivery than nulliparous patients, and also less likely to need cervical ripening, this finding is consistent with clinical experience.

Our study has a number of strengths. First, we utilized a large study sample with comprehensive clinical data. Second, we used best practices for propensity score analysis and IPTW [27]. Propensity score analysis reduces confounding by indication while allowing for the estimation of risk ratios. After IPTW weighting we achieved acceptable standardized differences between the two groups. Additionally, the use of stabilized weights has been shown to preserve the true sample size of the original data and thus produce appropriate estimates of variance and type I error rate [25].

At the same time, there are a number of limitations to consider. First, the diagnosis of SGA is only known after delivery. However, we did examine the association between prostaglandin use and cesarean delivery for NRFS in the subgroup of patients with suspected growth restriction prior to delivery. We also did not have information on umbilical artery Doppler findings, although in our institution inductions of labor at term for IUGR generally have elevated or normal umbilical artery Doppler pulsatility index. Additionally, we only examined singleton term inductions of labor. The effect of prostaglandins on preterm SGA infants may be different. We were unable to balance oxytocin exposure between the two groups, likely because patients who receive prostaglandins may not receive oxytocin, while those who do not receive prostaglandins almost invariably receive oxytocin for cervical ripening and well as induction. However, we included oxytocin in the model estimating the relative risk of cesarean delivery for NRFS and found no substantial change in the relative risk estimates. Additionally, propensity scores cannot control for unmeasured confounding and the possibility of residual confounding must be considered. Finally, the issue of power must be considered. It is possible that the study was underpowered to detect the small difference in the primary outcome that was observed after propensity score analysis. A larger sample size may have made the results statistically significant. However, the sharp decrease in the RR after propensity score adjustment suggests that the association between prostaglandin use and cesarean delivery from NRFS is highly confounded, and we were able to adjust for the measured confounders. Furthermore, an RR of 1.22 from the propensity score analysis translates to an increase in the risk of cesarean delivery for NRFS from 11.0% in the unexposed group to 13.4% in the exposed group, a small absolute difference that is not clinically meaningful.

In conclusion, exposure to prostaglandins in patients undergoing induction of labor and delivering term SGA neonates was not associated with an increased risk for cesarean delivery for NRFS, cesarean for any reason, or neonatal morbidity after adjustment for confounding with a propensity score model. We have previously demonstrated that patients with SGA undergoing induction of labor at term have the same rate of vaginal delivery as those whose pregnancies are not complicated by SGA [18]. Our current results suggest that prostaglandins may be a safe choice for induction in these patients, assuming appropriate monitoring systems and clinical expertise is readily available.