Study sample: mother–infant dyads
This analysis utilises data from the SWIFT study, which is a prospective, longitudinal study of 1035 women with recent GDM enrolled into in-person research examinations from September 2008 to December 2011. The study eligibility criteria were as follows: (1) Kaiser Permanente Northern California (KPNC) member delivered at a Kaiser Permanente hospital (2008–2011), (2) maternal age 20–45 years at delivery, (3) diagnosed with GDM according to the 3 h 100 g OGTT per Carpenter and Coustan criteria , (4) delivery of a singleton, live birth at ≥35 weeks gestation, (5) no history of overt diabetes or other serious medical conditions, (6) electronic health record (EHR) data for the GDM pregnancy, (7) not planning to become pregnant again or move out of the area in the next 2 years, (8) not taking any medications that affect blood glucose tolerance, (9) able to speak English or Spanish, and (10) exclusive or mostly BF, or exclusive or mostly FF and the intention to continue for ≥4 months as detailed in the SWIFT study design [25, 27].
Briefly, the study contacted women with GDM during late pregnancy and again at 4 weeks postpartum to evaluate BF intention and study eligibility. The SWIFT study enrolled 1035 women within a single integrated healthcare system who provided written informed consent for three in-person research visits starting from 6 to 9 weeks postpartum (study baseline) and annually thereafter for 2 years. In this analysis, 835 of 1031 SWIFT mother–infant dyads exposed to GDM and continued in study follow-up had complete information on BF and infant dietary intake variables (from birth to 1 year), sociodemographics, prenatal and newborn clinical outcomes, child weight and height measurements at 2–5 years of age, and mothers did not have overt diabetes at study baseline. Those excluded from the analytic sample (n = 196) had lower attained education, worse maternal glucose intolerance and shorter BF duration than women included (electronic supplementary material (ESM) Table 1).
At each SWIFT study in-person research visit for women (6–9 weeks postpartum, 1 year and 2 year post-baseline), trained research staff conducted a 2 h 75 g OGTT (glucose and insulin) and obtained measurements of maternal anthropometry (weight, height, waist circumference), and administered interviewer- and self-administered surveys to assess reproductive and health history, infant feeding/diet, child health, sociodemographic factors (race/ethnicity, education, income), depression and maternal lifestyle behaviours (diet, physical activity, sleep). Participants also completed monthly mailed surveys from delivery to 1 year later to report BF intensity and duration, health status, and detailed infant dietary intake .
This study also obtained clinical data from the Kaiser Permanente EHR to determine study eligibility and account for prenatal exposure to maternal hyperglycaemia, including GDM diagnosis and treatment (i.e. prenatal 3 h 100 g OGTT, gestational age at GDM diagnosis, and GDM treatment with diet, oral medication, and/or insulin therapy), maternal height, pre-pregnancy weight and last weight measured before delivery (gestational weight gain). Data on newborn outcomes (i.e. infant sex, gestational age, birthweight, weight and length for size-at-birth percentiles) and child weight and height measurement between the ages of 2 and 5 years were obtained from the EHR. The SWIFT Study protocols were approved by the Kaiser Permanente Northern California institutional review board and were carried out in accordance with the Declaration of Helsinki as revised in 2008.
Infant feeding and dietary intake measures
Telephone interviews were conducted during late pregnancy to assess BF intention and again at 4–6 weeks post-delivery to assess BF intensity and intention to assess eligibility for enrolment. Women completed detailed infant feeding surveys at in-person research visits up to 2 years, a telephone interview at 6 months, and monthly mailed surveys from birth to 12 months. Women reported the average frequency of BF and FF within the past 7 days for each month, including total number of all feedings, BF and daily amounts of formula fed, as well as the date of introduction of solids, fruit juice, sweetened beverages, water, sugar water, and other beverages, including types and amounts of each item. BF duration (total months of any BF) was evaluated as a continuous variable, and dichotomised as adequate BF (≥6 months) or inadequate BF (none or <6 months). EBF duration was dichotomised as adequate EBF (≥6 months) or inadequate EBF (none or <6 months). The SWIFT study utilised the method of Piper and Parks  to quantitatively assess BF intensity and duration . This method calculates a BF intensity score for each month by dividing the number of breast milk feeds by the total number of feeds (FF, BF, other liquids) in 24 h (on average during the past 7 days). The monthly score ranges from 0 for exclusive FF to 1 for EBF. The 12 month BF intensity/duration score is the sum of the 12 intensity ratios from birth to 12 months (range of scores: 0 to 12), as a single continuous measure across time.
The infant diet variables were assessed from birth to 12 months, including age at initiation of first complementary foods, intake of fruit juice (100% fruit juice or juice with added sugar), liquids with added sugar, and SSBs. Intake of beverages other than milk feeds during the first year of life were categorised as: (1) consuming SSBs (including any sugary drinks, Pedialyte, water or juices with added sugar), (2) 100% fruit juice intake (unsweetened/with no added sugar), or (3) no SSB/no 100% fruit juice from birth to 1 year of age. Other food items were assessed and included in the complementary food intake variable categorised by age of initiation.
Clinical anthropometric measurements from the EHR at birth and 2–5 years
Clinical measurements of neonatal weight and length in the supine position were obtained from the EHR to calculate birthweight percentiles based on the KPNC population [29, 30]. We also obtained child weight and height from the EHR clinical paediatric well check visits (measured height in a standing position without shoes and heels against a wall using a stadiometer, and measured weight using a calibrated digital scale) between 2 and 5 years of age [31, 32]. For this analysis, the weight and height measured at the oldest age between 2 and 5 years for each child was selected to calculate the standardised BMI percentiles for age and sex, and categorise each child as normal (BMI <85th percentile), overweight (BMI ≥85th to <95th percentile) or obese (BMI ≥95th percentile) based on the Centers for Disease Prevention and Control growth percentiles . ESM Table 2 displays per cent of individuals among the BMI percentile categories by child age at last weight and height measurements.
Descriptive statistics (i.e. mean, SD, range, median and quartiles, histograms and Q-Q plots) assessed the variable distributions. Four separate linear and multinomial logistic regression models estimated adjusted OR (aOR) for the following BF measures: (1) the continuous BF duration, (2) any BF duration categories, (3) EBF duration categories, and (4) 12 month BF intensity/duration score, as well as SSB intake for child BMI percentile categories at ages 2–5 years. For the current analysis, a priori covariates were selected for their known influence on child obesity, including maternal education level (years of formal schooling), gestational weight gain, parity (primiparous vs multiparous), Special Supplemental Nutrition Program for Women, Infants and Children (WIC) participation, infant sex, size for gestational age, race/ethnicity, and child age at BMI measurement. Prenatal GDM severity was measured according to three variables: (1) 3 h 100 g OGTT laboratory results calculated as the sum of four z scores for glucose, (2) GDM treatment type (diet only vs oral hypoglycaemic agents or insulin), and (3) gestational age at GDM diagnosis.
Finally, to evaluate the joint association between the infant dietary intake categories (SSB, and 100% fruit juice) and categorical BF variables, two separate multinomial regression models were constructed, each stratified by BF adequacy and EBF adequacy adjusted for the above covariates. All analyses were performed with SAS version 9.4 (SAS, North Carolina, USA). Significance was denoted at p < 0.05.