This is a national retrospective cohort analysis using data abstracted from the NCHS and Centers for Disease Control and Prevention’s (CDC) Division of Vital Statistics database from 2015 to 2018 22. We chose the most recent 4 years of livebirth and fetal death data available. These years reflect outcomes prior to the formal recommendations regarding elective IOL at 39 weeks 19. The data is publicly available and de-identified, therefore institutional review board approval was not required.
The intervention group consisted of all women undergoing IOL at 39 weeks of gestation without an identifiable medical indication, irrespective of their final mode of delivery. The expectant management group consisted of women delivered between 40 and 42 weeks of gestation. Weeks were stated as completed weeks of gestation, which is how this variable is reported in the fetal death or live birth databases.
We excluded all women at less than 39 weeks or greater than 42 weeks of gestation, multifetal gestations, known fetal congenital anomalies or aneuploidy, with previous cesarean delivery, and infant deaths at greater than 28 days. These were excluded from analysis because of their association with postnatal complications often unrelated to the birth process 23,24. Deliveries > 42 weeks gestation were excluded primarily because it is no longer common practice to continue expectant management at this gestational age given the inherent neonatal morbidities associated with post-date delivery 25,26. Women with any form of diabetes or hypertension were excluded from the intervention group to isolate the effects of induction of labor at 39 weeks in a low risk cohort compared with expectant management in an otherwise low-risk population. These conditions are considered high-risk conditions with delivery recommended by 39 weeks 23–26. Gestational hypertensive disorders can present after 39 weeks but were excluded if diagnosed in the 39th week, as delivery is recommended by 37 weeks or at time of diagnosis 24. As such, expectant management would not be a reasonable option if the diagnosis was made in the 39th week, but would have been an option had they been diagnosed after 39 weeks, which is possible due to its incidence with advancing gestational age 17,27.
The fetal death database was merged with the livebirth database. There are variables that are included in the livebirth but not in the fetal death certificates. These include fetal congenital anomalies or aneuploidies, maternal puerperal infection, blood transfusion and cesarean hysterectomy. No imputations were performed and their absence is expected to produce an underestimate in the incidence of those outcomes.
We included all fetal deaths from 40 to 42 weeks, but only intrapartum at 39 weeks 22. We were able to identify this group by verifying when the diagnosis was made, whether it was pre-labor, intrapartum, or unknown. This variable is reported in all except for the following jurisdictions: District of Columbia, Hawaii, Kansas, Missouri, Montana, Nevada, and New York.
Maternal demographic information was compared between the two management groups using the appropriate univariate statistical test. The maternal outcomes of interest included: cesarean delivery, intra-amniotic infection or inflammation (triple I), blood transfusion, intensive care unit (ICU) admission, uterine rupture, and cesarean hysterectomy. Triple I, or chorioamnionitis as it was previously known. The neonatal outcomes of interest included: fetal death, 5-minute Apgar score ≤ 3, assisted ventilation for > 6 hours, NICU admission, seizure, and neonatal death < 28 days of life.
Multivariable log-binomial regression analysis was performed to calculate adjusted relative risk (aRR) to control for potential confounding variables based on historic significance and univariate analysis. These variables had to be included in both the certificate of live birth and fetal death and included maternal age, race, parity, education, prenatal care, cigarette use, and body mass index (BMI). Backward stepwise elimination method was performed to arrive at the final regression model, which included maternal education, ethnicity, parity, BMI and cigarette use. A secondary analysis reported frequencies for cesarean delivery, fetal and neonatal death, and seizures, in addition to calculating the relative risk for each subsequent week.
We chose to calculate the relative risk (RR), which is usually the parameter of interest in cohort studies especially since odds ratios can overestimate risk for more common outcomes and overestimation of the importance of a risk factor may lead to intervention errors 28. Statistical significance was defined as p-value < 0.01. Power analysis was not performed as the sample size included the entire population. All analyses were performed using Stata 14 statistical software (College Station, TX) 28 and we adhered to “Strengthening the Reporting of Observational Studies” (STROBE) guidelines for reporting cohort studies 29.