Materials and Methods
Data were derived from the national linked Perinatal Mortality data files assembled by the National Center for Health Statistics. Between 1995 and 1997, there were 11,436,020 singleton live births in the United States. Among 242,392 singleton live births delivered between 23 and 32 completed weeks' gestation, 39,251 (16.2%) had PROM, defined as rupture of membranes at least 12 hours before onset of labor. Labor was considered present if the birth record contained the indications tocolysis; cephalopelvic disproportion; labor induction or stimulation; precipitated, prolonged, or dysfunctional labor; vaginal birth after cesarean; or vaginal delivery that was spontaneous or by vacuum or forceps.
The primary outcome was death up to age 1 year, which was further classified as early neonatal (0–6 days), late neonatal (7–27 days), or postneonatal (28–365 days). We also examined respiratory distress syndrome (RDS), assisted ventilation, and neonatal seizures as secondary outcomes. Information on RDS and assisted ventilation was limited to conditions that developed any time before the mother's hospital discharge. Data available on neonatal seizures were limited to those within the first 12 hours of life. Available clinical information was abstracted from data on birth and infant death certificates. Conditions are recorded on vital statistics using a check-box format (yes or no).8
In the vital statistics database, gestational age assignment was predominantly based on last menstrual period (LMP) reported by women. However, in less than 5% of births the LMP date was missing and therefore was imputed.9 Inconsistent data on gestational age for birth weight was replaced by clinical estimates of gestational age reported on vital statistics data.8 The algorithm for deriving gestational ages in vital statistics data was done by the National Center for Health Statistics.
We excluded from analysis infants with congenital or chromosomal abnormalities (n = 1245), cases with implausible gestational ages (n = 3325), and pregnancies with missing data on labor (n = 87). Pregnancies were judged to have implausible gestational ages if infants' birth weights were outside gestational age–specific birth weight cutoffs proposed by Kiely.10 After the exclusions, the population for the study included 34,594 singleton live births delivered between 23 and 32 weeks' gestation and complicated by PROM.
A birth weight nomogram that comprised all live births in the United States between 1995 and 1997 was constructed. Small for gestational age infants were identified as those with birth weights below the tenth percentile for gestational age. Using that methodology, 10.3% (4049 of 34,594) of infants were classified as SGA. The risks for infant morbidity and mortality from labor were examined separately for SGA and non–SGA infants.
Rates of early neonatal, late neonatal, postneonatal, and infant death were calculated for women with and without labor, and unadjusted relative risks (RR) with 95% confidence intervals (CI) were derived as the measure of effect. Adjusted odds ratios (OR) were derived from multivariable logistic regression models after adjusting for potential confounders, which included gestational age, birth weight, maternal age, gravidity (gravida 1, gravida 2, or gravida ≥3), marital status (married or unmarried), maternal education (number of years of schooling), race or ethnicity (white, black or other), prenatal care use (number of prenatal care visits and month of prenatal care initiation), cigarette smoking and alcohol use (yes or no), and medical risk factors (diabetes, chronic hypertension, cardiac and renal disease, respiratory and circulatory disorders). Potential obstetric confounding variables included pregnancy history (including preterm deliveries or SGA infants), intrapartum fever, placental abruption, placenta previa, uterine bleeding of unknown origin, pregnancy-induced hypertension, eclampsia, oligohydramnios and polyhydramnios, cord prolapse, fetal heart rate abnormalities, birth injury, meconium staining, and electronic fetal monitoring during labor. Route of delivery also was considered a potential confounder. All those conditions are recorded on current United States birth and infant death certificates using the check-box format.8
We also evaluated the data for an interaction between presence or absence of labor and “fetal distress” for all outcomes examined. If the presence of a confounder changed the adjusted OR for labor by at least 10% in the logistic regression model, the confounder was retained in the model for adjustment. Adjusted ORs derived from the regression models were transformed to adjusted relative risks (RR) based on the methods described by Zhang and Yu.11 The fit of logistic regression models was evaluated based on the Hosmer-Lemeshow goodness-of-fit tests12 and by examining the distribution of residuals.
Several covariates in the study, including maternal age, education, gestational age, and birth weight were continuous and therefore were spline transformed.13 Cubic spline is a nonparametric smoothing procedure that does not impose restriction on the form of relation between continuous variables and outcomes. The shape of the distribution of gestational age and birth weight was nonlinear, so we applied a four-piece cubic spline transformation for gestational age in SGA infants with knots (connection points) assigned at 23, 25, 28, and 32 weeks and at 369, 504, 624, and 1247 g for birth weight. Among non–SGA infants, knots for gestational age were assigned at 24, 28, 31, and 32 weeks and at 624, 1095, 1615, and 2268 g for birth weight. Locations of knots for gestational age and birth weight among SGA infants were slightly lower than those for non–SGA infants because, by definition, a greater proportion of SGA infants were likely to be delivered earlier and be of lower weight. Several models with quadratic and cubic splines with 3, 4, 5, and 6 knots were fit and were compared using the likelihood-ratio test13; the spline transformations with four knots provided the best (and most parsimonious) fit. The relationships between maternal age and education and outcomes examined were essentially linear and so were retained in logistic regression models with a linear term. Statistical analyses were done with the SAS system (SAS Institute, Cary, NC; 8.0 for UNIX). To account for the possible effect of fetal presentation at delivery, a separate analysis was done only with infants that had cephalic presentation.
Among 34,594 women who met study criteria, 73.4% (n = 25,382) were in labor before delivery, and 79.8% (n = 27,591) delivered infants in cephalic presentation. The demographic, obstetric, and infant characteristics of women who did and did not labor are presented in Table 1. The rates of infant death, RDS, assisted ventilation, and seizures were 11.6%, 15.1%, 25.9%, and 0.2%, respectively.
Risks and RR for infant death, RDS, assisted ventilation, and seizures for labored compared with nonlabored women are shown in Table 2. Among SGA infants, labor was associated with a 23% increased risk of infant death within the first year of life, after adjusting for confounding factors (RR 1.23, 95% CI 1.02, 1.49). That association was caused specifically by more early neonatal deaths (RR 1.24, 95% CI 1.11, 1.38). Labor was associated with lower incidence of assisted ventilation but had no effect on incidence of RDS. In the adjusted analysis of non-SGA infants, labor was associated with higher rates of RDS and assisted ventilation but had no effect on infant death or neonatal seizures.
Small for gestational age infants in cephalic presentation had a 33% higher risk of early neonatal death with exposure to labor, whereas non-SGA infants had a 16% higher risk (Table 3). However, labor was not associated with higher rate of infant death in SGA or non-SGA infants. For non-SGA cephalic infants, labor was associated with higher rates of RDS and assisted ventilation. There was no effect of labor on those outcomes in SGA infants.
The entire study was replicated when association between presence and absence of labor and the outcomes were evaluated by each week of gestation between 23 and 32 weeks. The results (not presented) were consistent with the overall analysis.
This study suggests that labor has an adverse effect on outcomes in preterm infants delivered by women with PROM. Infants with low or borderline reserve, (SGA) appear particularly vulnerable to effects of labor, manifested as greater risk of death particularly in the early neonatal period. The higher mortality rate in those infants might explain their lower risk of pulmonary complications, because fewer survive the early neonatal period. However, additional studies are needed to clarify the exact causes of death in those infants. Based on our findings, infants with normal reserve (non-SGA) who are exposed to labor have higher risk of morbidity with no effect on mortality.
The large size of the study, including nearly all deliveries in the United States over a recent 3-year period, offers credibility to our findings. A post-hoc power analysis indicated that for a given study size (n = 34,594) and type 1 error of 5%, our study had power in excess of 90% for analysis of infant death, respiratory distress, and assisted ventilation and a power of 80% for analysis of neonatal seizures. Previous studies based on state birth certificate information by necessity analyzed data collected over longer periods of time. With rapid advances in obstetric and neonatal care leading to improved neonatal outcomes at early gestational ages, interpretation of data based on extended periods of observation might be difficult. Another strength of our methodology is that the study population was confined to women who delivered preterm with the same high-risk condition, PROM. Outcomes for infants delivered by women with PROM might not be comparable to those for infants delivered preterm for other indications, such as maternal hypertensive disease.7 Combining etiologically distinct or heterogeneous groups in analysis leads to attenuated estimates of relative risks.
Recent evidence suggests that magnitude of error in gestational age assessment based on LMP might be substantial, as much as 4 weeks.14 Kramer et al15 reported that ultrasound dating shifts the gestational age distribution to the left, and therefore is likely to overstate the proportion classified as preterm. Although it is virtually impossible to accurately assess gestational age in population-based data, we excluded inconsistent gestational age for birth weight, which should have improved estimates of gestational age.
The limitation of the study design is inherent in birth certificate data. The reliability of some information extracted from birth certificate data can be questioned. Our main exposure, the presence or absence of labor, was determined indirectly; however, the proportion of laboring women in our study was as expected, 73.4%, based on current obstetric practice. One commonly accepted definition of PROM is rupture of the membranes any time before onset of labor; however, in this study, PROM was defined as rupture of membranes more than 12 hours before onset of labor. The duration of ruptured membranes before delivery is associated with higher rates of neonatal complications (eg, infection), so caution should be used when generalizing the results reported here.
Our findings are in general agreement with those published3,4,16; however, our method also allowed us to identify a specific group of infants who were at the highest risk of death from exposure to labor (SGA infants). That finding might have implications for treatment of women who present with preterm PROM. To assess significance of our findings from a public health perspective, the population attributable risk (percent) was computed. For early neonatal death with labor, the population attributable risk was 6.2%, which implies that among women who delivered singleton live births between 23 and 32 weeks with PROM, 6.2% of all early neonatal deaths were attributable to exposure to labor. We believe our findings justify the performance of randomized controlled trials to determine whether labor is an important variable in determining infant outcomes. Women with preterm PROM (before 33 weeks) who remain undelivered after steroid therapy might be randomly assigned to delivery by cesarean early after spontaneous onset of labor or to standard trial of labor. Opinions and recommendations on clinical management should await future studies.
1. Westgren M, Dolfin T, Halperin M. Mode of delivery in the low birthweight fetus: Delivery by cesarean section independent of fetal lie versus vaginal delivery in vertex presentation. Acta Obstet Gynecol Scand 1985;64:51–7.
2. Malloy MH, Onstad L, Wright E. The effect of cesarean delivery on birth outcome in very low birth weight infants. Obstet Gynecol 1991;77:498–503.
3. Anderson GD, Bada HS, Sibai BM. The relationship between labor and route of delivery in the preterm infant. Am J Obstet Gynecol 1988;158:1382–90.
4. Anderson GD, Bada HS, Shaver DC, Harvey CJ, Korones SB, Wong SP, et al. The effect of cesarean section on intraventricular hemorrhage in the preterm infant. Am J Obstet Gynecol 1992;166:1091–101.
5. Novy MJ, Thomas CL, Lees MH. Uterine contractility and regional blood flow responses to oxytocin and prostaglandin E2 in pregnant rhesus monkeys. Am J Obstet Gynecol 1975;122:419–33.
6. Jonas HA, Lumley JM. The effect of mode of delivery on neonatal mortality in very low birthweight infants born in Victoria, Australia: Caesarean section is associated with increased survival in breech-presenting, but not vertex-presenting infants. Paediatr Perinat Epidemiol 1997;11:181–99.
7. Wolf EJ, Vintzileos AM, Rosenkrantz TS, Rodis JF, Salafia CM, Pezzullo JG. Do survival and morbidity of very-low-birth-weight infants vary according to the primary pregnancy complication that results in preterm delivery? Am J Obstet Gynecol 1993;169:1233–9.
8. Taffel SM, Ventura SJ, Gay GA. Revised US certificate of birth-new opportunities for research on birth outcome. Birth 1989;16:188–93.
9. Taffel S, Johnson D, Heuse R. A method of imputing length of gestation on birth certificates. Vital Health Stat 1982;93:1–11.
10. Kiely JL. What is the population-based risk of preterm birth among twins and other multiples? Clin Obstet Gynecol 1998;41:3–11.
11. Zhang J, Yu KF. What's the relative risk? A method of correcting the odds ratio in cohort studies of common outcomes. JAMA 1998;280:1690–1.
12. Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley & Sons, 1991.
13. Durrelman S, Simon R. Flexible regression models with cubic splines. Stat Med 1989;8:551–61.
14. Gjessing HK, Skjaerven R, Wilcox AJ. Errors in gestational age: Evidence of bleeding early in pregnancy. Am J Pub Health 1999;89:213–8.
15. Kramer MS, McLean FH, Boyd ME, Usher PH. The validity of gestational age estimation by menstrual dating in term, preterm, and postterm gestation. JAMA 1988;250:3306–8.
16. Barrett JM, Boehm FH, Vaughn WK. The effect of type of delivery on neonatal outcome in singleton infants of birth weight of 1,000 g or less. JAMA 1983;250:625–9.