Study design
We performed a secondary and prospective cohort analysis of the Look AHEAD study, details of which have previously been published [18]. Briefly, Look AHEAD was a randomized clinical trial in which 5145 overweight or obese adults with T2DM aged 45 to 76 years were recruited from August 2001 to April 2004 across 16 centers in the USA and randomly assigned to receive either an intensive lifestyle intervention (ILI) or diabetes support and education (DSE). For the current analysis, we included participants with full data on body mass index (BMI) and waist circumference (WC) at the baseline, 12-month, 24-month, and 36-month visits. Participants with history of prevalent HF, those who developed HF or died during the initial 36-months (body weight variability assessment period) were excluded (n = 828); this was done in order to ensure that the exposure measurement preceded the development of the outcome. We also excluded participants in the Look AHEAD who had consent restrictions (n = 244). A total of 4073 participants were included in our final analyses. Additional file 1: Fig S1 summarizes the study exclusion process.
The research protocol was reviewed and approved by the Institutional Review Board at each participating clinical center and informed consent was obtained from each participant [18].
Assessment of variability in adiposity indices
The variability of adiposity indices was assessed during the first 36-months of follow-up in Look AHEAD (Additional file 1: Fig S1). For each participant, height and weight were measured at each visit twice using a stadiometer and digital scale, respectively; the average of the measurements was calculated and used for the analyses. BMI was computed as weight in kilograms divided by the square of height in meters. Waist circumference (WC) was measured using a non-metallic, constant tension tape placed at midlevel between the highest point of the iliac crest and lowest point of the costal border on the mid-axillary line [18].
For each adiposity index (BMI, WC or body weight), variability was defined by three metrics: the intraindividual standard deviation (SD), the coefficient of variation (CV) calculated as 100*SD/mean, and the variability independent of the mean (VIM) calculated as 100*SD/meanβ where β represents the regression coefficient based on the natural logarithm of SD on the natural logarithm of the obesity measure’s mean [19].
Ascertainment of incident heart failure events
Participants were followed from the end of the variability assessment period (36-month visit) through the occurrence of a HF event, death, or end of the study. Semiannual telephone calls and annual visits were conducted. Incident HF events were ascertained by an adjudication committee after reviewing relevant medical records using criteria adapted from the Women’s Health Initiative [20]. Potential cases were grouped into definite or possible acute decompensated HF, chronic stable HF, HF unlikely, or unclassifiable. Incident HF referred to the first hospitalization for definite or possible acute HF exacerbation [21]. Further details about the ascertainment of HF events are provided in Additional file 1: Method S1.
Covariates
The covariates assessed at baseline include: age, sex, race/ethnicity, randomization arm, duration of diabetes, history of prevalent ASCVD (history of coronary artery disease or stroke at baseline), current smoking, alcohol use, and estimated glomerular filtration rate (eGFR) calculated using the Chronic Kidney Disease Epidemiology Collaboration equation [22]. Additionally, we assessed variables collected during the variability assessment period which include average systolic blood pressure, use of antihypertensive medication, average ratio of total to high-density lipoprotein (HDL) cholesterol, average glycosylated hemoglobin (HbA1C) as well as average BMI, WC, and body weight [18]. We also evaluated the variability of other physiologic parameters including SD of blood pressure, heart rate and HbA1C.
Statistical analyses
We compared the characteristics of the participants across quartiles of the VIM of BMI using the χ2 test for categorical variables, and the Analysis of Variance or Kruskal–Wallis test for continuous variables.
Incidence rates were calculated as the ratio of the cumulative number of HF events to the total person-years. Person-years were calculated from the end of the variability assessment period to the earliest of HF event, death, or September 14, 2012 (date of trial’s termination).
We used Cox proportional hazards regression models to compute adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for incident HF. Each adiposity variability index was modelled both as a continuous variable and quartiles using the lowest quartile as reference group.
We built sequential regression models as follows: (1) a first model adjusting for age, sex, race/ethnicity, randomization arm (model 1); (2) a second model adjusted for model 1 variables plus additional adjustment for current smoking, alcohol drinking, use of antihypertensive medication, average systolic blood pressure, average ratio of total to high-density lipoprotein cholesterol, estimated glomerular filtration rate, duration of diabetes, average HbA1C, and history of cardiovascular disease (model 2). When variability was assessed using SD or CV, a third model (Model 3) was constructed as model 2 variables with further adjustment for average BMI (when assessing BMI variability), average WC (when evaluating WC variability) or average body weight (when assessing the variability of body weight). In supplementary analyses, we further adjusted for CAD as a time-varying covariate.
In order the assess the effects of the variability in other physiologic variables, we conducted sensitivity analyses adjusting for SD of SBP, heart rate and HbA1C.
A two-sided P-value of less than 0.05 was considered statistically significant and all analyses were conducted using STATA 14.2 (Stata, Inc, College Station, TX).