Research design and methods
The HCHS/SOL is a probability sample and community-based cohort study of 16 415 self-identified Hispanic/Latino persons aged 18–74 years from randomly selected households in four US field centers (Chicago, Illinois; Miami, Florida; Bronx, New York; San Diego, California). Sample design and cohort selection have been described previously.14 Participants were enrolled in 2008–2011 (baseline) and a second clinic visit was conducted in 2014–2017, on average 6 years after baseline examination. Annual follow-up interviews were conducted by telephone to collect basic health and healthcare information, including a diagnosis of diabetes.
Study participants
At baseline, 9835 participants of the cohort were women of which 6661 (68%) had at least 1 pregnancy. We additionally excluded 272 because of missing GDM history information resulting in an analytic sample of 6389 women for the cross-sectional analyses. Sociodemographic characteristics and medical history of participants were obtained at baseline through an interviewer-administered questionnaire in their language of preference (www.cscc.unc.edu/hchs). Participants were asked to report their current age, number of pregnancies, and which of the following best describes their Hispanic/Latino heritage: Central American, Cuban, Dominican, Mexican, Puerto Rican, South American, other, or more than one heritage. Participants self-reported the following sociodemographic and access to care characteristics: language preference, age of immigration among those not born in the US mainland, highest level of education, household income, employment status, occupation, access to care (health insurance, number of physician visits in past year), years living in the USA, and smoking status (current, former, never).
Gestational diabetes
Self-reported history of GDM, defined as any degree of glucose intolerance with onset or first recognition during pregnancy, was determined by answering ‘yes’ to (1) a physician diagnosis of diabetes only during pregnancy at the baseline visit or (2) asked again at visit 2 for diabetes first diagnosed during pregnancy before the baseline visit (80% of women participated at both baseline and visit 2).
Diabetes
Prevalent diabetes at baseline was defined by self-report of a physician diagnosis of diabetes to the question “Has a doctor ever said that you have diabetes?” and no report of GDM to the follow-up question of ‘Was this during pregnancy only?’, fasting plasma glucose (FPG) ≥126 mg/dL, 2-hour postoral glucose tolerance test (OGTT) ≥200 mg/dL, or A1c ≥6.5%. Incident diabetes was defined by self-report at any annual follow-up telephone interview or by self-report or laboratory measures (FPG, OGTT, A1c) at visit 2. The methods for the OGTT, FPG, and A1c were the same for the baseline and follow-up visits. For the OGTT, participants were instructed to drink a serving of glucola within 5 min; a 2-hour blood sample was obtained 2 hours after the participants initiated with glucola drink.15 Venipunctures for the OGTT, FPG, and A1c were performed similarly with technicians applying a tourniquet, identifying a vein, cleansing the site, inserting the needle, and appropriating 10 tubes of blood.
Cardiometabolic risk factors
Cardiometabolic risk factors were measured at baseline. Height and weight were measured by trained examiners to determine body mass index (BMI (kg/m2)). Waist circumference was measured and a circumference of >88 cm was considered high risk for cardiovascular disease.16 Hypertension was defined as self-report of antihypertensive medication or a blood pressure reading of ≥140/90 mm Hg. Elevated low-density lipoprotein (LDL) cholesterol was defined as use of lipid-lowering medication or LDL cholesterol ≥100 mg/dL. Low high-density lipoprotein (HDL) was defined as HDL cholesterol <50 mg/dL and elevated triglycerides were defined as ≥150 mg/dL. Statin use was self-reported and verified by scanning medication bottles. Albuminuria was defined as an albumin-to-creatinine ratio ≥30 mg/g.
Statistical analysis
Women’s baseline demographic, behavioral, and health characteristics (per cent, SE) were estimated and stratified by self-reported GDM status. All baseline analyses included all women regardless of prevalent diabetes status. For cross-sectional analyses at baseline, we estimated the prevalence (per cent, SE) of a history of GDM overall and by heritage group and other characteristics of the HCHS/SOL study population. We used logistic regression to estimate the OR (95% CI) for the association between GDM and diabetes at baseline. Estimates were determined overall and stratified by participant characteristics (study center, current age, number of pregnancies, Hispanic/Latino heritage, sociodemographic characteristics, health insurance, number of physician visits, and cardiometabolic risk factors). For stratified analyses, no correction factors (eg, Bonferroni) were used. In addition, OR estimates from logistic regression models were (1) unadjusted, (2) adjusted for study center, age, and number of pregnancies, (3) additionally adjusted for Hispanic/Latino heritage, (4) additionally adjusted for sociodemographic characteristics and access to care, and (5) additionally adjusted for cardiometabolic risk factors.
For prospective analyses, among women without prevalent diabetes at baseline, the overall cumulative incidence of diabetes (per 100 persons) was determined by whether or not women reported history of GDM at baseline and additionally stratified by women’s characteristics. To do this, we used predictive marginals from logistic regression, which allows for inference for internal comparisons of subgroups (GDM vs no GDM) within the target population from which the sample is drawn. Second, bivariate interactions between GDM and women’s characteristics were assessed for incident diabetes. Third, we used logistic regression to estimate the OR for the association between history of GDM at baseline and incident diabetes at visit 2 (~6 years after baseline) overall and stratified by participant characteristics. In addition, OR estimates from logistic regression models were adjusted sequentially as was done for the prevalent models. Lastly, for incident diabetes analyses, manual backwards stepwise selection, starting with all variables included in the stratified analyses and any significant interactions, was used to define the most parsimonious model with variables having a statistical significance level of p<0.10 at each model selection step and p<0.05 in the final model.
All statistical analyses used sampling weights and accounted for clustering and stratification in the HCHS/SOL sampling design using SUDAAN (SUDAAN User’s Manual, Release 11, 2012; Research Triangle Institute). The HCHS/SOL baseline sampling weights are a product of a base weight (reciprocal of the probability of selection) and three adjustments: (1) non-response adjustments made relative to the sampling frame, (2) trimming to handle extreme values, and (3) calibration of weights to the 2010 US Census according to age, sex, and Hispanic heritage. Visit 2 sampling weights accounted for visit 2 non-response.