Data are presented as mean±s.d. or percentage, as appropriate. BMI was natural logarithmically transformed to best approximate-conditional normality and homogeneity of variance, conditional on cohort, disease status, age and gender. To account for potential gender and cohort differences in the distribution of BMI, subjects were stratified by gender in each cohort and by disease status (ESRD in cohort 4 and T2DM in cohort 6). Within each strata, an individual was considered an outlier if the data was outside of four s.d. A total of three outlier observations were removed from further analyses. The data were then adjusted for age in a linear model and residuals were standardized to a mean of zero and variance of one (that is, Z-score). These Z-scores are the primary unit of analysis for both within individual cohorts and the meta-analyses.
