This was a retrospective cohort study of women prescribed antidepressants before and during pregnancy. This study was approved by the Indiana University Institutional Review Board and Regenstrief Institute in Indianapolis. Deidentified data were obtained from electronic medical records (EMRs) through the Regenstrief Institute in Indianapolis, an honest data broker for EMR data from multiple large health systems in Indiana [8, 9]. Data related to pregnancies within Eskenazi Health or Indiana University Health Systems between January 1, 2010, and December 31, 2019 were collected. Medication orders (i.e. prescriptions) for SSRI or SNRI dated 100 days before the last menstrual period through the date of delivery were captured. Antidepressants of interest included the SSRIs citalopram, escitalopram, fluoxetine, and sertraline as well as the SNRI bupropion. Data on the SSRI paroxetine and SNRIs desvenlafaxine, duloxetine, and venlafaxine were collected and ultimately excluded from analyses due to the number of subjects with orders for these medications at timepoints of interest failing to reach n = 200, limiting statistical ability to effectively detect rates for the outcomes of interest. Tricyclic and tetracyclic antidepressants were also excluded due to being less commonly prescribed during pregnancy for our population. This was a planned secondary analysis focusing on timing of prescriptions of a larger antidepressant database [7].
The following variables were collected for eligible women: maternal age at time of delivery, race, ethnicity, insurance, estimated due date, history of prior preterm birth, any other drugs prescribed during pregnancy, history of diabetes (types I, II), development of gestational diabetes (GDM), and development of a hypertensive disorder of pregnancy. Infant outcomes collected were gestational age at birth, birth weight, birth length, birth head circumference, stillbirth, diagnosis of any adaptation syndrome, neonatal intensive care unit (NICU) admission, 5-minute APGAR score, jaundice requiring treatment, diagnosis of transient tachypnea of the newborn (TTN) or respiratory distress syndrome (RDS), persistent pulmonary hypertension of the newborn (PPHN), neonatal seizures, and cardiac malformations.
The primary outcomes of interest were newborn diagnosis of any adaptation syndrome and NICU admission. Due to a change in hospital coding for the diagnosis, coded diagnoses of “neonatal abstinence syndrome” (NAS) or “pediatric adaptation syndrome” (PAS) were combined for the diagnosis of any adaptation syndrome. Discharge summaries, delivery records, and administrative ICD codes were used to extract diagnoses. Appendix Table S1 lists the codes and sources used by the data extraction process. Extraction of clinical diagnoses was contingent upon use of standard clinical criteria documented by the clinician in the discharge summary of an infant. The Regenstrief Institute Data Core’s documented process based on ICD9/10 codes was utilized for capture of diagnoses [8, 9]. Through the linkage of maternal and infant medical record numbers across health systems, all diagnoses and diagnostic codes for the infant were extracted by the Data Core for both inpatient and outpatient encounters. Identification and verification for accuracy against the Cerner or Epic medical records of the respective health system was completed for approximately 1% of records [7]. Data for these subjects was then re-deidentified for the final analyses.
The data were aggregated and organized using SPSS 26 and Microsoft Excel and analyzed using SPSS 28. Data were analyzed using chi-square testing for discreet variables. Logistic regression models, adjusting for maternal age, race, ethnicity, and insurance status were used to determine the impact of prescription timing on the primary outcomes.
Because the objective of this study was to screen for any impact on maternal and infant outcomes based on timing of antidepressant exposure, women who were prescribed an antidepressant within 100 days before the last menstrual period, as well as in first (0–14 weeks gestation), second (15–28 weeks), and third trimesters (> 28 weeks) of their pregnancy were used as the control group (i.e. subjects with assumed continual medication exposure throughout pregnancy). Exposure timing was extrapolated based on the date the prescription was ordered. As we did not have data on whether the person actually took the medication, we assumed that the drug was taken and there was exposure.
Three separate comparisons were performed to explore the impact of timing and lack of exposure to each of the drugs. Consistent exposure throughout pregnancy was compared to 1) exposure up to the third trimester (no prescriptions in the third trimester), 2) exposure starting in the second trimester (no record of prescriptions before pregnancy or during first trimester), and 3) exposure occurring only during the second trimester (prescriptions dated during second trimester alone; potentially representing women without exposure in the first and third trimesters). All three exposure timing groups were assessed for each drug. Women who delivered before the third trimester were excluded from comparison 1. Logistic regression models, adjusting for maternal age, race, and primary insurance payor, were then constructed to examine the impact of timing of drug exposure on the primary outcomes. These results were reported in adjusted odds ratios (aOR) and 95% confidence intervals (CI).