Statistical methods

Characteristics of the study participants are summarized as descriptive statistics (Table ​(Table1).1). Every study participant has multiple outcome observations. Two DM and two NDM participants were measured twice. Due to repeated measurements of the outcome within the same study participant, a linear mixed model. was fitted to compare average elastic moduli between DM and NDM groups (Table ^S1). A linear mixed model captures the variability of measurements between individuals and the variability within measurements from an individual by including fixed and random effects⁴⁵. Specifically, DM status was a fixed effect and each participant was a random effect in the linear mixed model. Because residuals of the linear mixed model were not normally distributed with equal variance, we used a natural logarithm transformation of the outcome and performed the linear mixed model with the same analysis strategy. The distribution of the natural logarithm transformation of the outcome is illustrated in box plots irrespective of multiple measurements. Mean elastic moduli and standard deviations for each group are reported on the raw and natural logarithm scales (Table ​(Table22).

For hydroxyproline and Sirius Red data, datasets were tested for outliers and outliers removed, then tested for normality and independent t-test used to compare DM and NDM groups. For adipocyte size data, analysis was done using SPSS Statistics (IBM SPSS Statistics for Mac, Version 27.0. Armonk, NY: IBM Corp.). The mean adipocyte area for each image was calculated. A linear mixed model was run with average adipocyte area as the output; DM status, age, and BMI as fixed effects; and patient as a random effect. Analysis of adipocyte area frequency was done using a generalized linear mixed model that accommodates a negative binomial distribution used for count data. Sizing groups were determined based on the overall size distribution and corresponding size percentiles. The frequency within each sizing group for each patient was counted. A generalized linear mixed model was run for each sizing group with DM status, age, and BMI as fixed effects; patient as a random effect; and total number of cells per patient as the offset.