Hepatic steatosis and incident diabetes share a non-glycemic cardiometabolic profile in Japanese adults with BMI below 25 kg/m²: a retrospective cohort study

Hepatic steatosis is a common liver manifestation of metabolic dysfunction and can occur in adults with body mass index (BMI) < 25 kg/m². Although hepatic steatosis is associated with incident diabetes, whether baseline steatosis and subsequent diabetes follow concordant gradients across a multidimensional non-glycemic cardiometabolic phenotype space remains unclear. This study assessed that relationship in Japanese adults. A secondary analysis of the longitudinal NAGALA cohort included 12,932 adults with BMI < 25 kg/m² and no diabetes at baseline. A continuous phenotype space was constructed using discriminative dimensionality reduction via trees from eight non-glycemic traits spanning adiposity, lipids, blood pressure, and gamma-glutamyl transferase (GGT). Glycemic measures, hepatic steatosis, and incident diabetes were excluded from phenotype construction. Logistic and Cox models mapped baseline ultrasound-detected hepatic steatosis and incident diabetes onto this space. Cosine similarity quantified concordance between the two outcome gradients, and a steatosis-aligned phenotype score was evaluated in sequentially adjusted Cox models. At baseline, 1,334 participants (10.3%) had hepatic steatosis. Over 78,629 person-years of follow-up, 201 participants developed diabetes. Both outcomes were concentrated in the region with lower Dimension 1 and higher Dimension 2, characterized by greater central adiposity, higher triglycerides and GGT, and lower high-density lipoprotein cholesterol. Per 1-SD increase, age- and sex-adjusted odds ratios for hepatic steatosis were 0.27 (95% CI, 0.24–0.30) for Dimension 1 and 2.09 (95% CI, 1.93–2.26) for Dimension 2; the corresponding hazard ratios for incident diabetes were 0.61 (95% CI, 0.53–0.71) and 1.27 (95% CI, 1.10–1.46), respectively. The two gradients showed high directional concordance (cosine similarity, 0.992). The steatosis-aligned phenotype score remained associated with incident diabetes after adjustment for lifestyle factors and baseline glycemia (hazard ratio per 1-SD increase, 1.33; 95% CI, 1.14–1.56). Baseline hepatic steatosis and incident diabetes showed concordant adverse gradients across a continuous non-glycemic cardiometabolic phenotype space in Japanese adults with BMI < 25 kg/m². The shared direction was characterized by central adiposity, higher triglycerides and GGT, and lower high-density lipoprotein cholesterol, although the two outcomes did not have identical risk profiles. Independent cohorts are needed to confirm the reproducibility of the phenotype structure and gradient concordance.