Deciphering metabolic dysfunction-associated steatotic liver disease: insights from predictive modeling and clustering analysis

Abstract

Background and aim: New nomenclature of steatotic liver disease (SLD) including metabolic dysfunction-associated SLD (MASLD), MASLD and increased alcohol intake (MetALD), and alcohol-associated liver disease (ALD) has recently been proposed. We investigated clustering analyses to decipher the complex landscape of SLD pathologies including the former nomenclature of nonalcoholic fatty liver disease (NAFLD) and metabolic dysfunction-associated fatty liver disease (MAFLD).

Methods: Japanese individuals who received annual health checkups including abdominal ultrasonography (n = 15 788, men/women: 10 250/5538, mean age: 49 years) were recruited.

Results: The numbers of individuals with SLD, MASLD, MetALD, ALD, NAFLD, and MAFLD were 5603 (35.5%), 4227 (26.8%), 795 (5.0%), 324 (2.1%), 3982 (25.8%), and 4946 (31.3%), respectively. Clustering analyses using t-distributed stochastic neighbor embedding and K-means to visually represent interconnections in SLDs uncovered five cluster formations. MASLD and NAFLD mainly shared three clusters including (i) low alcohol intake with relatively low-grade obesity; (ii) obesity with dyslipidemia; and (iii) dysfunction of glucose metabolism. Both MetALD and ALD displayed one distinct cluster intertwined with alcohol consumption. MAFLD widely shared all of the five clusters. In machine learning-based analyses using algorithms of random forest and extreme gradient boosting and receiver operating characteristic curve analyses, fatty liver index (FLI), calculated by body mass index, waist circumference, and levels of γ-glutamyl transferase and triglycerides, was selected as a useful feature for SLDs.

Conclusions: The new nomenclature of SLDs is useful for obtaining a better understanding of liver pathologies and for providing valuable insights into predictive factors and the dynamic interplay of diseases. FLI may be a noninvasive predictive marker for detection of SLDs.

Keywords: clustering; machine learning; metabolic dysfunction‐associated fatty liver disease (MAFLD); metabolic dysfunction‐associated steatotic liver disease (MASLD); nonalcoholic fatty liver disease (NAFLD).