Study population

All women who gave birth in a hospital to a live infant >500 g birthweight with a gestational age between 34 ^0/7 and 40 ^6/7 weeks in Ontario, Canada, between 2005 and 2012 were eligible for inclusion in this study. Women who gave birth between 41 ^0/7 and 41 ^6/7 weeks’ gestation were excluded to avoid capturing an increased risk of obstetric interventions in the 41st week of gestation.

Sources of data

Data were obtained from the BORN Information System (BIS), which is a provincial, internet-based maternal newborn surveillance system managed by the Better Outcomes Registry & Network (BORN Ontario). Maternal, fetal, and obstetric data were extracted for the study time period, April 1, 2004, to March 31, 2012. In a birth number validation exercise between the CIHI Discharge Abstract Database and the BIS, the BORN system captured 82% of hospital births in the 2004-2005 fiscal year and increased yearly until 100% of hospital births from 106 sites were captured by 2010-11. The BIS includes information on maternal demographic characteristics and health behaviors, preexisting maternal health problems, obstetric complications, intrapartum interventions, and birth outcomes. Various quality assurance mechanisms were used to ensure high data quality and consistency including data entry operating manuals, automated system checks, and training sessions for data entry staff. A validation study of the BORN birth record system found a high level of agreement with the original patient record for many of the variables used in this study including gestational age, induced delivery and cesarean section, which had 91%, 90%, and 99% agreement with the original patient record respectively. The variance for percentage agreement across all variables evaluated was 97.2% (interquartile range, 93.0–99.4%).

Exposure variables

Our primary exposure was ‘any obstetric intervention’ defined as prelabor cesarean section or induced delivery. Our secondary exposures were prelabor cesarean section and induced delivery separately. To avoid overlap between obstetric intervention categories and capture of the event that primarily influenced gestational age at birth, induced deliveries were defined as those with medical or surgical intervention to initiate uterine contractions before the onset of spontaneous labor and included those that subsequently resulted in cesarean section. Prelabor cesarean sections were defined as delivery by cesarean section without prior induction of labor or spontaneous onset of labor. Spontaneous onset of labor was defined as labor that was initiated without intervention, which may have been preceded by spontaneious rupture of membranes.

We also explored maternal and obstetric factors known or suspected to be associated with preterm birth. These included maternal characteristics such as maternal age at delivery, socioeconomic status, smoking during pregnancy, parity (nulliparous, primiparous, or multiparous), multifetal pregnancy, previous cesarean section, and number of previous preterm births; maternal health problems including preexisting diabetes (insulin and noninsulin dependent), gestational diabetes, chronic hypertension (hypertension that predates the pregnancy or was diagnosed before the 20th week of gestation), gestational hypertension, infection during pregnancy (composite of any periodontal, urinary tract, vaginal, or cervical infections), and obstetrical complications including preeclampsia, eclampsia, placenta previa, uterine bleeding, placental abruption (premature separation of a normally implanted placenta after the 20th week of gestation and before the fetus is delivered), cord prolapse, chorioamnionitis, breech presentation and small for gestational age (<10 percentile gender specific birthweight for gestational age relative to a Canadian standard reference population for any fetus). Socioeconomic status was measured using material and social deprivation from the Deprivation Index, a national index that has been used in many epidemiologic studies and by the Canadian Institute for Health Information as a proxy for socioeconomic status. The material deprivation component includes: (1) the proportion of persons aged 15 and over who have no high school diploma; (2) the average income of persons aged 15 and over; and (3) the employment to population ratio for persons aged 15 and over. The social deprivation component includes (1) the proportion of people who live alone; (2) the proportion of people separated, divorced, or widowed; and (3) the proportion of single parent families. A factor score is derived from a principal component analysis of these variables at the dissemination area (DA) level, a small, relatively stable geographic unit with a population of 400 to 700 people of relatively homogenous socioeconomic status. Each DA is divided into quintiles based on the corresponding factor scores. DAs in the highest deprivation quintile (Q5) represent the neighborhoods at the highest social disadvantage. We used automated geographic coding based on the Statistics Canada postal code conversion files (PCCF+ v5h) to link the maternal residential postal code in the BORN dataset to the corresponding quintile of deprivation.

Outcome variables

For our primary outcome, gestational age at birth was dichotomized as LP (34 ^0/7 to 36 ^6/7 weeks’ gestation) or term (37 ^0/7 to 40 ^6/7 weeks’ gestation) based on estimated gestational age at the time of delivery as reported on the birth record. Gestational age was estimated from the date of last menstrual period or dating ultrasound as indicated on the patient record.

Statistical methods

For the primary analysis, multivariable generalized estimating equation (GEE) regression was used to test the association between obstetric interventions and late preterm birth (vs term birth) adjusted for maternal and obstetric factors known or suspected to be associated with preterm birth (see above). We chose to use GEE regression to account for clustering within hospitals, which would allow for a more equitable representation of the relationship between risk factors and LP birth than traditional regression techniques. Variables considered for inclusion contained less than 10% missing data and records with missing information on variables included in the models were excluded in the final computation. All variables specified above were included in the final models regardless of traditional statistical significance as all were believed to be clinically important. Unadjusted and adjusted risk ratio (RR) estimates were computed using a log binomial regression model with hospital site used as the clustering unit for the GEE model to account for clustering of obstetrical interventions within hospitals. Colinearity between model parameters was assessed using the variance inflation factor with a cutoff value of 2.0. To explore a dose effect, the association between obstetric interventions and gestational age at birth was explored by week of gestation between 34 and 37 weeks relative to term.

For our secondary analysis, the primary analysis was repeated but using induced delivery and prelabor cesarean section as separate exposures. We also conducted an exploratory analysis to examine the independent association between maternal and obstetrical factors known or suspected to be associated with preterm birth and LP birth (vs term birth) adjusted for obstetric interventions.

To internally validate our model and minimize the risk of spurious associations from multiple hypothesis testing, for both the primary and secondary analyses, all births were randomly assigned to equally sized derivation and validation cohorts. The analysis was completed on both cohorts, however, the validation cohort was used to test the associations identified through the derivation cohort. All covariates were included in both derivation and validation models.

Statistical analyses were performed using SAS statistical software version 9.2 (SAS Institute, Cary NC) with statistical significance evaluated using 2-sided P values at the 5% level. The study protocol was approved by the Research Ethics Boards of the Children’s Hospital of Eastern Ontario, the St. Michael’s Hospital and the Mount Sinai Hospital in Toronto.