Identifying cohort

KPNC is an integrated healthcare delivery system that provides medical care for approximately one-third of the population in the San Francisco Bay area. KPNC subscribers are representative of the region.

The source population consisted of women who were KPNC members who attended a voluntary MHC at the Kaiser Permanente Oakland Medical Centers between 1984 and 1995. KPNC members at these facilities were invited to participate on enrollment. The MHC consisted of a clinic visit for completion of questionnaires and clinical measurements of blood pressure, weight, height, and random serum glucose and cholesterol tests, with the goal of providing health maintenance through early diagnosis. Measurements of serum glucose level were assessed by the hexokinase method; total cholesterol was assessed with a Kodak Ektachem Chemistry analyzer (Eastman Kodak Co, Rochester, NY). Both tests were performed by the regional laboratory of KPNC, which is a laboratory that participates in the College of American Pathologists’ accreditation and monitoring program. BMI was calculated with standard methods from height that was measured with a stadiometer and weight that was measured with a balance beam scale. Information on age, sex, race/ethnicity, education level, cigarette smoking, alcohol consumption, coffee consumption, use of medications, and hours since last food ingestion was collected with a self-administered questionnaire that has been described previously.

Among women of reproductive age who participated in the MHC from 1985–1996 (n = 22,356), 4084 women who subsequently delivered an infant up to 2005 were identified by a search of the KPNC hospitalization database and the Pregnancy Glucose Tolerance Registry. This is an active surveillance registry that annually identifies, among KPNC members, all pregnancies that have resulted in a livebirth or stillbirth and includes all results from screening and diagnostic tests for GDM by including data from the electronic laboratory database (available since 1991).

GDM case definition

There were 186 women with GDM according to their glucose values that were obtained during a standard 100-g, 3-hour oral glucose tolerance test that met the American Diabetes Association (ADA) plasma glucose thresholds. An additional 47 women had a hospital discharge diagnosis of GDM on the electronic hospital discharge database before the electronic laboratory data were available, which was confirmed by a standardized medical chart review to confirm that the women had had a 100-g, 3-hour oral glucose tolerance test that met the ADA criteria for GDM. Cases were excluded if, at the time of the MHC examination, they had an indication of overt diabetes mellitus (ie, a physician diagnosis of diabetes mellitus [n = 5]), which left a total of 228 confirmed GDM cases.

Control selection and matching criteria

From the cohort of 3849 women described earlier who did not meet the ADA criteria for the diagnosis of GDM, control subjects were selected randomly. For each case, 2 control subjects were matched individually to cases by year of MHC serum measurement date (±3 months), age at MHC serum test (±2 years), and age at delivery of index pregnancy (±2 years).We matched age at serum measurement and age at pregnancy to account for the fact that there were varying time intervals between the MHC examination and the subsequent pregnancy. Had we not matched, our cases would have had a longer time interval between the MHC visit and pregnancy because GDM is more common in older women.

Exclusion criteria

Women were excluded if they had missing information on glucose or cholesterol level, BMI, or blood pressure (25 cases and 62 control subjects) at the MHC examination. To further ensure that we excluded women with possible preexisting diabetes mellitus and to narrow the range of pregravid glucose tolerance that we assessed, we first categorized women into a dichotomous variable based on whether their hours since last food ingestion was at or above the mean (≥6 vs <6 hours). We then excluded women with <6 hours since last food ingestion and a glucose value of >140 mg/dL (n = 2 cases) and women with ≥6 hours since last food ingestion and a glucose value of ≥110 mg/dL (n = 2 cases). For 17 cases, we were only able to identify 1 control subject who met these criteria. Our final analytic cohort consisted of 199 cases and 381 control subjects.

Pregravid cardiometabolic risk factors

When data from >1 MHC examination were available, data from the visit before but closest to the index pregnancy were used in the analysis. Serum glucose and total cholesterol levels and diastolic and systolic blood pressure were first categorized into tertiles according to the distribution in the control subjects. In addition, women were categorized according to clinically relevant adverse levels of cardiometabolic risk factors. Overweight/obesity was defined as a BMI of >25 kg/m ² . Prehypertension/hypertension was defined as a systolic blood pressure of >120 mm Hg and/or diastolic blood pressure of at least 80 mm Hg for diastolic or a reported use of antihypertensive treatment. Hypercholesterolemia was defined as a serum cholesterol of ≥200 mg/dL. Mild hyperglycemia was defined as a pregravid random serum glucose level of ≥100 mg/dL but ≤140 mg/dL, given that 100 mg/dL represented the 95th percentile for our study population.

Statistics

Conditional logistic regression was used to obtain odds ratios as estimates of the relative risk of GDM in relation to each individual cardiometabolic risk factor. We also examined a multivariate model that contained all the clinically relevant cardiometabolic risk factors defined earlier to identify independent cardiometabolic predictors of GDM. To assess confounding, we entered covariates into a logistic regression model one at a time and then compared the adjusted and unadjusted odds ratios. Final adjusted logistic regression models included covariates that altered unadjusted odds ratios by at least 10% for at least 1 cardiometabolic risk factor. Variables that were evaluated for confounding were race/ethnicity, pregravid BMI (kilograms per square meter, except for the model with obesity), parity, maternal education in years, alcohol consumption, cigarette smoking, caffeine consumption, and first-degree family history of diabetes mellitus. Sensitivity analyses were conducted only for cases who met the more stringent National Diabetes Data Group criteria for GDM to see whether the association varied by severity of disease or by hours since last food ingestion (<6 vs ≥6 hours) to see whether the associations varied by time since last food intake. In addition, we conducted a sensitivity analysis to examine the multivariate model of cardiometabolic risk factors and GDM that was restricted to nulliparous women to ensure that the association was similar among women without a previous GDM pregnancy, because we were unable to control for history of GDM.

To assess the potential modifying effects of prepregnancy BMI (overweight, ≥25 kg/m ² vs not overweight, <25 kg/m ² ) and time since MHC examination (>7 vs ≤7 years), we tested interaction terms in unconditional logistic regression models that had been adjusted for time between MHC examination and pregnancy. This study was approved by the human subjects committee of the Kaiser Foundation Research Institute.