Background <p>Hepatic steatosis measured by imaging fails to capture the variation in cardiometabolic risk and intervention response, which may be better characterized by metabolomic profiles. We aimed to construct metabolomics-based indices to define this variation.</p> Methods <p>Using data from a three-arm lifestyle intervention randomized trial in adults with type 2 diabetes (T2D) and overweight/obesity, we constructed two novel indices from untargeted plasma metabolomics and MRI-measured liver fat: a metabolomics-based liver fat score (mliver fat), and the discordance between mliver fat and MRI-measured liver fat (Δliver fat). We examined their associations with cardiometabolic traits and intervention response.</p> Results <p>Both mliver fat and Δliver fat were associated with body composition, glucose indices, insulin sensitivity, and triglyceride, but only Δliver fat was independent of MRI-measured liver fat. Despite having comparable MRI-measured liver fat, compared with the participants with a high Δliver fat (mliver fat &gt; MRI-measured liver fat), those with a low Δliver fat (mliver fat &lt; MRI-measured liver fat) had a more favorable cardiometabolic profile and derived greater benefits and more sustained benefits from diet intervention, with more pronounced long-term improvements in weight, insulin sensitivity, and β-cell function.</p> Conclusions <p>Among individuals with T2D, a metabolomics-based liver fat score, particularly the discordance between metabolomic and imaging assessments, identifies systemic metabolic heterogeneity and differential responsiveness to lifestyle interventions. Future research is warranted to evaluate its performance in improving risk stratification and personalizing lifestyle intervention.</p> Trial registration <p>NCT03839667</p> Graphical abstract