The Consortium on Safe Labor included 12 clinical centers (with 19 hospitals) across 9 American College of Obstetricians and Gynecologists (ACOG) U.S. districts. There were 8 university affiliated teaching hospitals, 9 teaching community hospitals and 2 non-teaching community hospitals. They were chosen because of the availability of electronic medical records at each institution and because their geographic distribution covers all ACOG US districts. A total of 228,668 deliveries with 233,844 newborns between 2002 and 2008 were included in the study. 87% of births occurred between 2005 and 2007. All births at 23 weeks or later in these institutions were included. 9.5% of women contributed more than one delivery to the database. To avoid intra-person correlation, we selected the first delivery from each subject in the study, leaving 206,969 deliveries for analysis. Participating institutions extracted detailed information from their electronic medical records on maternal demographic characteristics, medical history, reproductive and prenatal history, labor and delivery summary, postpartum and newborn information. Information from the neonatal intensive care unit (NICU) was linked to the newborn records. Data on labor progression were extracted from the electronic labor database. Information on hospital and physician characteristics was collected from surveys of the local investigators, and maternal and newborn discharge summaries (in ICD-9 codes) were linked to each delivery. This project was approved by the Institutional Review Boards of all participating institutions.
Data transferred from the clinical centers were mapped to pre-defined common codes for each variable at the data coordinating center. Data inquiries, cleaning, recoding and logic checking were performed. We also conducted validation studies for four key outcome diagnoses including cesarean for non-reassuring fetal heart rate tracing, asphyxia, NICU admission for respiratory conditions, and shoulder dystocia. To validate data, eligible charts were selected, and investigators were asked to recollect data with chart abstraction done by hand. We compared the information hand collected from the medical charts with that downloaded from the electronic medical records. indicates that most variables that were reviewed in this study are highly accurate. Although our records were not sampled randomly, the consistency among different records on the same variable (e.g., singleton, gestational age, attempting vaginal birth, live birth, vertex presentation) indicates that the information provided in the validation studies is reliable and likely to be generalizable to the entire database. Thus, the electronic medical records are reasonably accurate representation of the medical charts.
Validity of data from electronic medical records comparing to medical charts in selected variables
Approximately 5.9% of women in our study had missing information on fetal presentation. Given the importance of fetal presentation in our analysis, we performed multiple imputation.8
A logistic regression model imputed the likelihood of vertex/non-vertex presentation in a particular subject multiple times based on other obstetric characteristics, including maternal race, parity, previous uterine scar, number of fetus, external cephalic version, smoking, placenta previa, cephalopelvic disproportion, gestational age, reason for admission to labor/delivery, trial of labor, induction, fetal scalp electrode, operative vaginal delivery and mode of delivery. When the imputed data were analyzed, the uncertainty that was related to imputation was taken into account.
To make our study population reflect the overall U.S. obstetric population and to minimize the impact of the various number of births from different institutions, we assigned a weight to each subject based on ACOG district, maternal race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic and others), parity (nulliparas vs. multiparas) and plurality (singleton vs. multiple gestation). We first calculated the probability of each delivery with these four factors according to the 2004 National Natality data;9
publicly available National Natality data can no longer be separated by state after 2004. Then, based on the number of subjects each hospital contributed to the database, we assigned a weight to each subject. indicates that the weighted study population is close to the entire U.S. obstetric population. Therefore, we used the weighted sample throughout our analyses.
Description of the study population in comparison to the 2004 U.S. birth cohort
We defined “attempting vaginal delivery or a trial of labor” as all vaginal deliveries plus cesarean deliveries with at least two vaginal examination data in the labor progression (or labor curve) database. For indications for cesarean delivery, we first listed all major indications and the percent of cesarean deliveries with a specific indication (one woman may have more than one indication). “Elective cesarean delivery” was defined as cesarean for clinical indications of: 1) elective as denoted in the electronic medical record, 2) declining a trial of labor, and 3) a variety of factors that are not considered accepted indications for cesarean delivery such as elderly gravida, multiparity, remote from term, postterm/postdates, diabetes, chorioamnionitis, chronic or gestational hypertension without preeclampsia/eclampsia, premature rupture of the membranes, HPV infection, GBS positive, polyhydramnios, fetal demise, tubal ligation, and social/religion concerns. We then grouped all indications into three hierarchical, mutually exclusive categories: “clinically indicated”, “mixed”, and “truly elective”. The “mixed” group included cesarean deliveries where not enough detailed information (e.g., HIV with an unknown viral load or unknown presentation of twins) was available to judge the necessity or where the clinical indications were not that strong (e.g., preeclampsia).
Duration of labor arrest was calculated as the duration of no appreciable change of cervical dilation in the 1st stage and the time interval between the first 10 cm and delivery in the 2nd stage. “No appreciable change in cervical dilation” was defined as within 1 cm of change in dilation prior to delivery. All statistical analyses were performed using SAS version 9.1. Given that this is a descriptive analysis with a very large sample size, no statistical testing was performed; nor were confidence intervals provided.