Data were analyzed with SAS, version 9.4 (SAS Institute, Cary, NC). Due to skewed distributions and small sample sizes, continuous data are summarized as median with interquartile range (IQR) (25th–75th percentiles). Categorical data are summarized as frequency counts and percentages. For continuous data, Kruskal-Wallis tests were used for comparisons between insulin groups. χ2 and Fisher exact tests were used, as appropriate, for group comparisons of categorical data. Ordinal (independent, low, high insulin) and binary (on/off insulin) logistic regression models were used to assess predictors at 3, 6, and 9 months post-TPIAT for insulin requirements at 1 year. The covariates examined in each model were age at TPIAT, BMI percentile at the particular visit, baseline BMI percentile, change in BMI percentile (from baseline to the visit time point), fasting glucose, HbA1c, AUCC-peptide, fasting C-peptide, and peak C-peptide. BMI percentile at the time of each visit and age at TPIAT were included in all logistic regression models as covariates to control for age differences and the potential influence BMI can have on metabolic measures. P values and Akaike information criterion were used to assess which variables were better predictors in the models. A P < 0.05 was considered significant.