We conducted a retrospective cohort study of term births that occurred in California from 1997 to 2006. We obtained institutional review board approval from the Committee on Human Research at the University of California, San Francisco, the institutional review board at Oregon Health and Science University, and the California Office of Statewide Health Planning and Development and the Committee for the Protection of Human Subjects. Because the data are deidentified and part of the public record of vital statistics, informed consent was not required.
The data for these calculations come from the California Vital Statistics Birth Certificate Data, California Patient Discharge Data, Vital Statistics Death Certificate Data, and Vital Statistics Fetal Death File.10
The State of California maintains linked data sets that include maternal antepartum and postpartum hospital records for the 9 months before delivery and 1 year after delivery as well as birth records and all infant admissions occurring within the first year of life. Linkage is performed by the California Office of Statewide Health Planning and Development Healthcare Information Resource Center under the State of California Health and Human Services Agency using a unique “record linkage number” specific to the mother–infant pair.
The birth certificate data use last menstrual period as the basis for gestational age dating in days. This gestational age is converted to weeks and treated as an ordered categorical variable. If the last menstrual period was missing or nonsensical, the mother– infant pair was excluded for analysis. This study includes all births from 37 to 42 weeks of gestation; 37 weeks of gestational age included births ranging from 37 0/7 weeks to 37 6/7 weeks, and 42 weeks of gestational age included births from 42 0/7 weeks to 42 6/7 weeks. We excluded multiple gestations, pregnancies complicated by diabetes mellitus (pre-existing or gestational) and chronic hypertension, and infants with congenital anomalies or genetic causes of death based on the International Classification of Diseases (ICD), 9th and 10th Revision codes. Causes of infant death were taken from the ICD, 9th Revision (years 1997–1998) or ICD, 10th Revision (years 1999–2007) codes on death certificates and were grouped into large thematic categories.
The incidence of stillbirth at a given gestational age was calculated as the number of stillbirths (whether antepartum or intrapartum) at that gestational age per 10,000 ongoing pregnancies. Infant mortality at each gestational age was calculated as the number of infants born at this gestational age who die within 1 year of life per 10,000 live births at that same gestational age. For reference, a neonatal death is defined as death within the first 30 days of life, whereas early neonatal death, the metric included in estimates of perinatal mortality, is defined as death within 7 days of life.
The goal of this project was to compare the mortality risks between delivery at a certain gestational age with that of expectant management (ie, continuing the pregnancy for another week and then delivering 1 week later). More specifically, the mortality risk of delivery at a given week was defined as the rate among those infants born at that week of gestation. The mortality risk of 1 week of expectant management was defined as the risk of stillbirth over that week plus the mortality risk experienced by infants born in the subsequent week of gestation. Infant death, rather than neonatal death, was chosen as the preferred metric to examine because of its greater magnitude and persistent correlation with gestational age at delivery. As mentioned previously, infant mortality has been shown to vary with gestational age at term and shares many of the same risk factors as stillbirth.4,6
Although only early neonatal death rates have classically been included in estimates of perinatal risk, as neonatal intensive care improves, a larger proportion of children with complications resulting from gestational age or intrapartum events may be surviving beyond the neonatal period, contributing to the decrease in neonatal mortality rates over time.4,11
Also, recent data demonstrate that term infants who die within the first year of life are more likely to do so within the postneonatal period (age 29–365 days of life) than in the neonatal period.12
Any gestational age-related mortality effect in these children will be better captured by examining infant death rates.
Our calculations rely on the following assumptions: 1) the risks of stillbirth and infant death have a uniform distribution throughout a particular week of gestation; 2) when estimating the risk of delivering at a particular gestational age, the fetus is not at risk for stillbirth beyond that gestational age; therefore, their mortality risk in that week is equal only to the risk of infant death; and 3) all probabilities are conditional rather than cumulative; that is, the risk of stillbirth at 41 weeks of gestation includes the assumption that the pregnancy is viable at that gestational age and has not had a stillbirth in the weeks prior.
The composite risk of expectant management for 1 week represents the sum of the probability of stillbirth during a given week of gestation plus the probability of infant death when birth occurs the subsequent week. This composite risk of expectant management beyond each given week of gestation then was compared with the risk of infant death for children born in the given week of gestation. The “number needed to deliver” was calculated as an analogous measure to the “number needed to treat” by taking the reciprocal of the absolute risk difference between delivery and expectant management.
Statistical calculations were performed with Excel and Stata 12, including proportions, relative risks, and 95% confidence intervals (CIs). Exponential modeling was performed and goodness of fit was reported with the coefficient of determination, R2. Chi square tests were performed to compare proportions of independent variables and analysis of variance was performed to compare means. Statistical significance was reached with a P value of<.05 or if 95% CIs did not overlap. We assumed that the binomial probability distributions of both mortality risks approximated the normal distribution and derived the CI of the composite risk using the sum of the variances plus twice the covariance of the estimates of infant death and stillbirth.