Study participants were from the Nurses’ Health Study II (NHSII) population, a cohort study of 116,678 female nurses who were aged 25 to 42 years at the start of the study in 1989. NHSII follows participants biennially by questionnaire to obtain both health-related behavior information and data on the occurrence of diseases, including diabetes. The follow-up rate has been approximately 90% for each biennial questionnaire. For purposes of this study, we followed women from their first pregnancy (the earliest pregnancy in our study population occurred in 1964) through 2005, the last reported follow-up for incident diabetes.

Our study population is based on a subset of the NHSII population who had detailed data on pregnancy history and T2DM. More specifically, in 2001, NHSII sent a supplemental questionnaire to study participants who were deemed good responders — those women who typically responded to the first or second mailing of the biennial questionnaires. This supplemental questionnaire included detailed data on history of pregnancies that lasted at least 12 weeks; 68,376 women (75% of those mailed the supplemental questionnaire and 58% of the original study population) returned the questionnaire (Figure 1). We excluded women with twins or triplets, women who had had more than 5 pregnancies (the questionnaire length precluded collection of detailed pregnancy data for more than 5 pregnancies), and women whose questionnaires were missing data on gestational age or birth weight or were missing covariate data (Figure 1). We also excluded women who were diagnosed with diabetes at or before age 30 to avoid including those with type 1 diabetes mellitus and women who were diagnosed with diabetes mellitus before their first pregnancy. This study was approved by The Partners Human Research Committee (institutional review board) of Brigham and Women’s Hospital.

Study participants indicated the length of each of their 5 most recent pregnancies lasting at least 12 weeks in the following categories: 12 to less than 20 weeks, 20 to less than 24 weeks, 24 to less than 28 weeks, 28 to less than 32 weeks, 32 to less than 37 weeks, 37 to 42 weeks, and 43 or more weeks. Pregnancies were categorized as very preterm (20 to ≤32 weeks), moderate preterm (33to ≤37 weeks), term (37 to ≤42 weeks, referent) and postterm (≥43 weeks).

Participants were asked the birth weight for each of their 5 most recent pregnancies (<5 lbs, 5–5.4 lbs, 5.5–6.9 lbs, 7.0–8.4 lbs, 8.5–9.9 lbs, ≥10 lbs). Low birth weight was defined as less than 5.5 pounds at term and macrosomia as 10 pounds or more at term.

In the NHSII 1989 baseline questionnaire and subsequent biennial questionnaires study participants indicated whether they had ever been diagnosed with GDM by a physician. A validation of GDM showed high validity of self-report of this condition compared with medical records (16).

Study participants responded to “Have you ever had toxemia/pre-eclampsia (raised blood pressure and proteinuria) with any pregnancy?” on the NHSII baseline questionnaire and biennially. A recent validation of pre-eclampsia compared with medical record review showed self-reported preeclampsia is a moderately good indicator of hypertensive disorders of pregnancy (HDOP) (positive predictive value, 73%) (17).

NHSII study participants received a supplemental questionnaire if they reported having been diagnosed with diabetes mellitus on the biennial questionnaires between 1991 and 2005. These supplemental questionnaires collected information to distinguish between type 1 and type 2 diabetes mellitus on the basis of diabetes diagnosis and treatment. The validation of this method of case confirmation has been reported previously (18). The 6% of respondents who reported having diabetes on the baseline 1989 questionnaire were not sent the supplemental questionnaire and, consequently, had details only on date of diagnosis, without information on whether the diabetes was type 1 or type 2. To restrict the analysis to cases likely to be T2DM, we included women who reported having diabetes mellitus on the 1989 questionnaire only if they were age 30 years or older at the time of diagnosis.

On the 1989 baseline questionnaire, study participants reported their current weight and height and their weight at age 18 (19). Participants were asked to update their weight on all biennial questionnaires. Body mass index (BMI, kg/m²) was calculated from reported height and weight data. BMI was derived for ages at which weight was not reported (ie, between age 18 and age at the baseline questionnaire) from a formula using weight at age 18 and weights reported on biennial questionnaires, as well as somatograms at ages 20, 30, and 40 to assign BMI at each age starting at 18 years and through the end of the study period.

Participants reported their race/ethnicity at baseline. Subsequent biennial questionnaires queried family history of diabetes mellitus, as well as a personal medical history (eg, hypertension, cancer, gestational diabetes, pre-eclampsia, toxemia). In addition, the 2001 supplemental questionnaire asked about smoking status during pregnancy (eg, ever, never). Participants were also asked at baseline to report their menstrual regularity at age 18 to 22 years, which was categorized as regular, irregular, or no menstrual periods.

We used Cox proportional hazards models to examine the association between preterm and postterm birth, as well as low birth weight and macrosomia with T2DM risk, modeling gestational age and birth weight separately. Very preterm birth, moderate preterm birth, and postterm birth were evaluated as indicator variables, with the reference group of term. For the association between birth weight and T2DM, we did not have enough information on birth weights under 5.5 pounds to create small-for-gestational-age categories among preterm births. Therefore, we restricted our analysis to women who had a term birth (n = 42,502). Models were constructed among all term births by using indicator variables for low birth weight, macrosomia, and normal birth weight (reference group).

Women were followed from their first birth through 2005 or up to 35 years after their first pregnancy; we were able to follow more than 95% of the cohort for up to 34 years after their first birth. First, we evaluated the association between gestational age, infant birth weight, and T2DM risk in first pregnancies for the entire 35-year follow-up period. Second, we explored time since first birth in 5-year intervals up to 35 years after the first pregnancy to determine at what time points the differences between gestational age or birth-weight groups were significant. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated. We adjusted for potential confounders, those factors that preceded pregnancy complications of interest and were associated with diabetes mellitus. Finally, we used the Breslow estimator to calculate the age-adjusted risk differences from the Cox proportional hazards models. All analyses were conducted using SAS statistical software (version 9.3; SAS Institute, Inc, Cary, North Carolina).

In age-adjusted analysis, we observed a graded association of very and moderate preterm birth with risk of T2DM, as well as elevated risks among women delivering their first child postterm (Table 2). However, adjustment for age at first birth, age in 1989, race/ethnicity, pre-pregnancy BMI, family history of diabetes mellitus, menstrual irregularity, smoking status, history of GDM, and history of HDOP dampened several of these associations; very preterm birth and postterm birth conferred an increased risk of T2DM, and moderate preterm birth no longer showed a significant increased risk of T2DM.

We explored the association between gestational age and T2DM in 5-year intervals following first pregnancy (Figure 2). In the first decade after very preterm birth, there was no increased risk of T2DM compared with term birth. However, there was an inconsistent change in risk in the second decade, which reached statistical significance. No significant associations were found after 21 years following first pregnancy among women with a very preterm birth. In contrast, women who had a moderate preterm birth had significant, roughly two-fold, increased risk of T2DM for the first 10 years after their first pregnancy, which thereafter returned to the baseline risk of women who had delivered at term. Postterm birth did not appear to be associated with T2DM at any time period.

We also explored the findings using risk differences. We found that during the 6 to 10 years following the first pregnancy, only 12 excess T2DM cases per 10,000 women who experienced a moderate preterm birth occurred compared with women who delivered at term. By 26 to 30 years after the first pregnancy, 298 excess T2DM cases per 10,000 women who delivered a very preterm birth occurred compared with those who delivered at term.

Although term low birth weight was not associated with T2DM in the overall population (fully adjusted HR, 1.26; 95% CI, 0.88–1.81), an increased risk for women who delivered a term low birth weight infant was seen among those without a history of GDM or HDOP (fully adjusted HR, 1.62; 95% CI, 1.01–2.59). Women who gave birth to a first infant that weighed 10 pounds or more had a 1.61 increased risk of T2DM (95% CI, 1.24–2.08) (Table 3). Restricting the results to women without a history of GDM or HDOP somewhat strengthened the association (fully adjusted HR, 2.22; 95% CI, 1.50–3.28).

In our exploratory analysis evaluating time trends in risk of T2DM by 5-year intervals since first birth, we found women who had a term, low birth weight infant had an inconsistently elevated risk of developing T2DM within the first 2 decades after the first pregnancy (Figure 3); this association reached significance only at 11 to 15 years after delivery (adjusted HR, 2.36; 95% CI, 1.11–5.03). Macrosomia also conferred a large but decreasing risk of T2DM. Associations between macrosomia and subsequent risk of T2DM were only significant between 6 and 20 years after the first pregnancy.

When evaluating absolute risk of T2DM for women who delivered an infant that weighed 10 pounds or more, we found an excess of 115 T2DM cases per 10,000 women in this group compared with women who delivered a normal weight infant in the 21 to 25 years following the first pregnancy. After 30 years, the elevated risk associated with having delivered a large infant had disappeared.

Because the prevalence of screening for GDM and HDOP during pregnancy has changed over the period of the births in this cohort (1964–2001), we repeated the analysis among first births occurring after the baseline 1989 questionnaire, because after that time period, screening was more widely practiced as standard care and gestational age-dating of pregnancies had probably improved with the advent of ultrasound. The hazard ratios estimated by the fully adjusted models (including GDM and HDOP diagnoses) were stronger for birth weight and T2DM, which mitigated the concern that macrosomia in the earlier pregnancies might merely be a marker for undiagnosed GDM. However, the association of very preterm delivery with T2DM was no longer detectable when limiting the analysis to pregnancies occurring after 1989.