Metabolic dysfunction-associated steatotic liver disease (MASLD) is the most common cause of chronic liver disease globally and is becoming the leading indication for liver transplantation in Western countries [1]. MASLD affects approximately 30 % of the general adult population [2], up to 70 % of individuals with type 2 diabetes (T2D) [3], and nearly 30–35 % of adults with type 1 diabetes (T1D) [4]. Notably, the presence of T2D is associated with a faster progression to more advanced forms of MASLD, including metabolic dysfunction-associated steatohepatitis (MASH), advanced fibrosis, and cirrhosis [5], as well as relevant cardiovascular and renal complications [6,7].

While T2D is typically characterized by insulin resistance and relative insulin deficiency [8], T1D is typically marked by absolute insulin deficiency [9]. Due to the increasing global prevalence of overweight and obesity even among adult individuals with T1D, insulin resistance has become a relatively common feature also in this patient group [9]. Persons with T1D are subject to substantial glycemic variability, often alternating between episodes of hyperglycemia and hypoglycemia. Notably, subcutaneous insulin injection bypasses the hepatic portal circulation, resulting in higher insulin levels in the periphery and reduced insulin levels in the liver compared to when insulin is released by the pancreas [9]. This mechanism might promote long-term hepatic fat accumulation via mechanisms involving insulin resistance and ectopic fat deposition, potentially explaining the non-negligible prevalence of MASLD in adult individuals with T1D [4]. However, the interplay among chronic hyperglycemia, glycemic fluctuations, and MASLD in T1D remains poorly understood.

In this context, we aimed to identify clinical and glycemic predictors of MASLD (hepatic steatosis) in adults with T1D by using a machine-learning approach. Specifically, we applied a random forest regression model to a multicenter cohort of 262 adults with T1D who underwent continuous glucose monitoring (CGM) and liver fat quantification using FibroScan-derived controlled attenuation parameter (CAP). The model incorporated specific CGM-derived metrics alongside clinical variables to assess their relative contributions to hepatic fat accumulation and to explore potential nonlinear associations through partial dependence analysis.