Study population

The NICHD Consecutive Pregnancy Study collected data retrospectively from electronic medical records of 20 hospitals in Utah ( Appendix ). Women with at least 2 pregnancies delivered between 2002 through 2010 were included resulting in 114,679 pregnancies (live births or stillbirths at ≥20 weeks’ gestation) from 51,086 women. Extensive data on maternal demographics, reproductive and medical history, prenatal complications, labor and delivery information, and neonatal outcomes were extracted. Data on infants admitted to the neonatal intensive care unit were collected from birth to hospital discharge or death. International Classification of Diseases, Ninth Revision ( ICD-9 ) codes were collected from maternal and newborn discharge summaries and linked to each delivery. All participating sites obtained approval for the study and waiver of informed consent from their individual institutional review boards.

The study was limited to women delivering singleton births in their first 2 pregnancies at study entry (parity range, 0–14). If women had >2 pregnancies during the study period, their subsequent pregnancies were only included if they were also singletons. A total of 49,868 women (78.3% with 2 pregnancies, 19.1% with 3 pregnancies, and 2.6% with 4-6 pregnancies) were included. Seven categories according to diabetes status in the previous or current pregnancy were created resulting in the following pregnancy pairs: (1) women without diabetes in the previous and the current pregnancy; (2) women with GDM in the previous pregnancy only (and not in the current pregnancy); (3) women with GDM in the current pregnancy only (and not in the previous pregnancy); (4) women with recurrent GDM (in the previous and the current pregnancy); (5) women who had no GDM in the previous pregnancy but developed DM between their previous and current pregnancy (DM in current pregnancy only); (6) women with GDM in the previous pregnancy who transitioned to DM between their previous and current pregnancy; and (7) women with pregestational DM (type 1 or 2) prior to the first observed pregnancy in the dataset. Women with >2 pregnancies could have been included in >1 of the examined groups. For example, a woman with 3 pregnancies, of which GDM was diagnosed only in her first pregnancy and not in her subsequent pregnancies, was included in the previous GDM-only group for her first and second pregnancies (category 2) and again in the no diabetes group for her second and third pregnancies (category 1). Since women could have entered the study at any parity, we performed sensitivity analyses using only the first 2 singleton births among women who were nulliparous at study entry (n = 27,064).

Gestational diabetes or DM

Maternal diabetic status was ascertained from electronic medical records supplemented with ICD-9 codes. If the diagnosis was coded in either source, then women were considered to have the condition during that pregnancy. In the medical records, diabetic status was recorded as gestational or pregestational (the Supplementary Table lists the ICD-9 codes used to identify diabetes and other maternal complications). Women whose records indicated pregestational diabetes in 1 pregnancy were categorized as such for all subsequent pregnancies.

Neonatal outcomes

PTB was defined as <37 weeks’ gestation based on obstetrical estimate in the medical record. We further classified PTB into spontaneous, indicated, and elective using a previously published algorithm by our group. Spontaneous PTB was the result of preterm labor or preterm premature rupture of membranes. Indicated PTB was defined among women without preterm premature rupture of membranes or spontaneous labor but with potential maternal, fetal, and/or obstetrical pregnancy complications. The elective group included women with labor inductions or cesarean deliveries recorded as elective by the study site without any obstetrical, fetal, and/or maternal indications. Large for gestational age (LGA) and small for gestational age were defined based on sex-specific birthweight >90th percentile and <10th percentile for gestational age (by week), respectively. Macrosomia was defined as birthweight >4000 g. Respiratory distress syndrome (RDS) was based on medical records and discharge summaries. Hypoglycemia, congenital anomalies, and jaundice were based on the ICD-9 codes. Shoulder dystocia, documented in both the electronic medical records and discharge summaries, was defined among women with vaginal deliveries only. Stillbirth and neonatal mortality were recorded in the electronic medical records. For a complete list of the examined newborn outcomes and the ICD-9 codes, refer to the Supplementary Table .

Data exclusions

From the 49,868 women with at least 2 repeat singleton pregnancies, 24 women were excluded from analyses; 23 had ICD-9 code for “infant of a diabetic mother” with no diabetes recorded for the mother and 1 had diabetes controlled by insulin with no diabetes diagnosis. This resulted in a final sample size of 49,844 women with 111,857 singleton deliveries and 62,013 repeat singleton deliveries (2 deliveries equivalent to 1 repeat, 3 deliveries equivalent to 2 repeats) for the main analyses. Sensitivity analysis restricted to women nulliparous at study entry resulted in a sample size of 27,064 repeats. (the Supplementary Figure displays the distribution of the women in regard to their parity and the change in diabetic status between pregnancies).

Statistical analysis

To examine the relative risk (RR) of adverse newborn outcomes across different groups, we used Poisson regression models with robust variance estimators. This approach provides valid inference for consecutive pregnancies and allows comparison of disease risk across groups. This technique was used in both unadjusted and adjusted analyses. In unadjusted models, we first assessed whether GDM in the previous pregnancy influenced neonatal outcomes in the current pregnancy by testing the significance of 2 interaction terms. The first interaction term (ie, GDM in previous pregnancy × GDM in current pregnancy) tested whether recurrent GDM conveyed greater risks of neonatal outcomes than expected based on the independent (additive) risks of having a previous GDM and a current GDM pregnancy. Similarly, the second interaction term (ie, GDM in previous pregnancy × DM in current pregnancy) tested whether transitioning from GDM to DM between pregnancies led to greater risks of neonatal outcomes than expected based on the independent risks of having a previous GDM and a current DM pregnancy. We decided a priori to keep both interaction terms in the model if at least 1 had a P value < .05 based on the score test, and for these models we reported the risk estimates from the model with both interaction terms included. To be parsimonious, for models with nonsignificant interaction terms, we calculated risk estimates obtained from the additive effects model. Of all examined neonatal morbidities, only macrosomia had a significant interaction term (GDM in previous pregnancy × DM in current pregnancy, P value = .013) and as such, the reported risk estimates for macrosomia were based on the model with interaction terms. In subsequent models, we adjusted for study site, maternal age, race, parity, interpregnancy interval, current prepregnancy body mass index, and maternal smoking (noted in table footnotes). For some of the outcomes, certain study sites were excluded due to sample size limitations (noted in the table footnotes). No further risk estimates were reported for neonatal outcomes displayed in the descriptive table when there were no significant differences in the percentages between the GDM/DM groups based on the global score test. Sensitivity analyses among women nulliparous at study entry were conducted in similar fashion. Indicated PTB was not examined in sensitivity analysis due to limited sample size. For all comparisons, the same referent group of women with no GDM/DM in the previous and current pregnancy was used. Statistical analyses were conducted using software (SAS, version 9.3; SAS Institute, Cary, NC).