The purpose of this investigation was to develop and validate a risk model based on natural killer (NK) cell-associated lncRNAs to predict outcomes in individuals with hepatocellular carcinoma (HCC). To further explore the role of NK cells in the HCC tumor microenvironment, we leveraged single-cell RNA sequencing data and the TCGA-LIHC dataset to identify NK cell-associated lncRNAs. Using Cox regression and LASSO techniques, we pinpointed four key lncRNAs as prognostic markers for the model. The model demonstrated robust predictive power across the training set, validation set, and entire dataset. Additionally, we identified a synergistic interaction between NK cells and other immune cells, particularly CD8 + T cells, in HCC. Moreover, we uncovered novel molecular subgroups of HCC and their associations with the immune microenvironment and drug sensitivity. To further validate these findings, we performed experimental validation of the expression and function of the model lncRNAs in HCC. These results suggest that the NK cell-associated lncRNA model not only serves as an effective prognostic tool for HCC patient outcomes but also offers potential molecular targets for immunotherapy and targeted therapies.

Background/Objectives: The controlled attenuation parameter (CAP) is widely used for non-invasive assessment of hepatic steatosis, but its discrimination for moderate-to-severe steatosis may be limited in patients with obesity. We developed and internally validated a model combining CAP, alanine aminotransferase (ALT), and glycated haemoglobin (HbA1c) to identify biopsy-confirmed S ≥ 2 steatosis in bariatric surgery candidates. Methods: This retrospective single-centre study included 320 bariatric surgery candidates with preoperative CAP and laboratory data and intraoperative wedge liver biopsy as the reference standard. Candidate predictors were evaluated using univariable logistic regression, least absolute shrinkage and selection operator (LASSO) screening, and multivariable logistic regression. Model performance was assessed using receiver operating characteristic analysis, paired DeLong testing, bootstrap internal validation, calibration, Brier score, and decision curve analysis. Results: S ≥ 2 steatosis was present in 182 patients (56.9%). The combined model showed higher discrimination than CAP alone (area under the curve [AUC] 0.768, 95% CI 0.717–0.819 vs. 0.671, 95% CI 0.612–0.731), with an AUC difference of 0.096 (95% CI 0.049–0.144; paired DeLong p < 0.001). At the internally derived Youden threshold, sensitivity was 53.8% and specificity was 86.2%. The combined model also showed a lower Brier score than CAP alone (0.1941 vs. 0.2234). Bootstrap internal validation yielded an optimism-corrected AUC of 0.760 and a mean absolute calibration error of 0.017. Decision curve analysis suggested modest incremental net benefit over CAP alone. Conclusions: The CAP–ALT–HbA1c model showed higher discrimination and lower overall prediction error than CAP alone. It may complement CAP as an adjunctive tool for preoperative risk stratification, but external validation and assessment of clinically appropriate decision thresholds are required before routine clinical implementation.