Hepatorenal index and an exploratory combined model for identifying moderate-to-severe histologic steatosis in patients with liver injury of undetermined etiology

Background Liver biopsy is the reference standard for grading hepatic steatosis, but its invasiveness limits routine use. We evaluated the hepatorenal index (HRI), derived from routine two-dimensional grayscale ultrasound images, and an exploratory combined model incorporating HRI with clinical parameters for identifying moderate-to-severe histologic steatosis (S ≥ 2) in patients with liver injury of undetermined etiology. Methods This retrospective single-center cohort included 80 biopsy-referred patients who had undergone same-day grayscale ultrasound, liver stiffness measurement, and liver biopsy. HRI was calculated offline. The exploratory combined model used Firth bias-reduced logistic regression followed by intercept correction, with 1,000-resample bootstrap internal validation. Results The S0, S1, S2, and S3 groups comprised 46, 17, 12, and 5 patients, respectively. Median HRI differed among the S0–S1, S2, and S3 groups (1.02, 1.25, and 1.59; P < 0.001). Body mass index (BMI), prognostic nutritional index (PNI), and HRI were positively associated with S ≥ 2 steatosis, whereas the aspartate aminotransferase-to-platelet ratio index (APRI) was inversely associated. HRI yielded an apparent area under the receiver operating characteristic curve (AUROC) of 0.866 (95% CI, 0.782–0.950). At the data-derived cutoff of >1.05, HRI identified all 17 S ≥ 2 cases, with an apparent sensitivity of 100.00% (95% CI, 80.49%–100.00%) and an apparent specificity of 60.32%. The exploratory combined model yielded an apparent AUROC of 0.942 and a bootstrap optimism-corrected AUROC of 0.927. Conclusion HRI showed promising apparent discrimination and potential rule-out value requiring independent validation. The combined model retained high discrimination after bootstrap correction and may provide a low-cost adjunct where dedicated fat quantification is unavailable, but remains hypothesis-generating pending external validation.