Letter to the Editor: Machine Learning Models for Screening Steatosis in MASLD : An International Validation Study

In their article, Ye et al. 1 present a rigorous, multi-database analysis of 34 diagnostic and prognostic approaches for metabolic dysfunction–associated steatotic liver disease (MASLD), including traditional scores and advanced machine-learning models. Their study, encompassing nearly 25 000 participants across five cohorts, represents an important step toward standardising and validating diagnostic strategies for this increasingly prevalent condition 1. The consistent performance of the extreme gradient boosting (XGB) model (AUROC > 0. 8) across all cohorts, together with the identification of the triglyceride–glucose index and waist circumference as a robust combination for lean individuals, offers clinically actionable insights 1. The observed reduction in overall survival among patients with MASLD and the prognostic strength of logistic regression further underscore the need for early detection and risk stratification 1. These findings complement and extend our prior work using FibroX, an XGBoost and explainable AI (SHAP), on NHANES data, where we achieved an AUC of 0. 95 for identifying high-risk metabolic dysfunction–associated steatohepatitis (MASH), outperforming FIB-4 (AUC = 0. 50) 2. In subsequent research, FibroX improved detection of advanced fibrosis (AUROC 0. 97 vs. 0. 62 for FIB-4) and predicted cardiovascular mortality (adjusted hazard ratio, 2. 76) 3, highlighting the broader clinical implications of machine-learning approaches. Importantly, we estimated potential U. S. healthcare savings of 3. 3 billion through reduced unnecessary procedures 3. Moving forward, AI for MASLD detection should focus on integrating these validated machine learning models into clinical practice. Of note, the emergence of agentic AI systems offers an exciting next step. These systems have the potential to autonomously streamline clinical workflows, from generating diagnostic protocols and documentation to integrating real-time multimodal data (imaging, genomics, biosignals) for highly personalised treatment planning and continuous monitoring 4. In addition, the inherent capability of multi-agent systems to manage intricate clinical trial workflows, from protocol optimization to automated patient recruitment and adaptive randomisation, will be crucial for accelerating the validation and deployment of these novel diagnostic tools. However, as Ye et al. 1 rightly acknowledge through the use of agentic systems, rigorous clinical validation, human oversight and the establishment of robust governance frameworks for accountability and safety must be prioritised to ensure these powerful tools translate into meaningful improvements in patient care rather than remaining confined to research settings. We congratulate the authors on this landmark contribution to advancing MASLD care. The authors have nothing to report. The authors have nothing to report. The authors have nothing to report. The authors declare no conflicts of interest. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.