Among validated serumbased prognostic tools, the Fibrosis-4 (FIB-4) index is the most widely used for routine practice^[14-16]. It is calculated as: FIB-4 = [age × aspartate aminotransferase (AST)]/[platelet count × √alanine aminotransferase (ALT)]^[17]. A large meta-analysis of 2,518 biopsy-proven MASLD patients confirmed the prognostic utility of FIB-4: patients with FIB-4 > 2.67 had a significantly higher risk of adverse outcomes (20.8%) than those with FIB-4 < 1.3 (1.3%). For 5-year outcome prediction, the time-dependent area under the receiver operating characteristic curve (AUC) of FIB-4 (0.74) was comparable to that of histological fibrosis stage on biopsy (0.72), suggesting that FIB-4 can substitute for biopsy in selected situations^[18]. MASLD prognosis is influenced by dynamic factors, including lifestyle modifications and pharmacotherapy, and longitudinal changes in FIB-4 can reflect intervention effects and prognostic trends. Specifically, dynamic elevation of FIB-4 is significantly associated with increased risks of all-cause mortality, CVD, and LREs in MASLD patients - those with high and persistently elevated FIB-4 levels have nearly a 7-fold higher risk of LREs than patients with low baseline FIB-4^[19]. In contrast to static risk stratification, monitoring FIB-4 changes guides clinical decision-making and allows timely adjustment of treatment strategies.

Emerging composite models integrating multi-dimensional indicators have been developed in recent years. These models combine multi-level data including clinical characteristics, routine biochemical parameters, and omics to achieve more refined prognostic risk stratification. A study by Chinese researchers developed a metabolomics-based MASH prediction score using machine learning algorithms. This score incorporates four indicators: body mass index (BMI), AST, tyrosine, and the ratio of phospholipids to total lipids in very low-density lipoprotein. It can not only identify MASH and stratify its risk, with AUCs of 0.87 [95% confidence interval (CI): 0.83-0.91] and 0.81 (95%CI: 0.75-0.88) in Chinese and European populations, respectively, but also identify individuals at high risk of liver-related mortality (AUC 0.83, 95%CI: 0.79-0.86). Its prognostic performance is significantly superior to the NAFLD Fibrosis Score (NFS)^[20]. This score was developed in a Chinese patient cohort and only externally validated in the Finnish population; its applicability in other ethnic groups requires further verification in multicenter, multi-ethnic cohorts. In addition to metabolomics-based models, proteomic profiling has also identified novel biomarkers for MASLD prognostic assessment. Another study using proteomic profiling found that Fibulin-3 is not only significantly associated with progressive liver fibrosis in MASLD, but also an independent predictor of LREs. The AUC of Fibulin-3 concentrations for predicting 5-year LREs was 0.761. Using a cutoff value of 6.0 μg/mL, individuals with high Fibulin-3 levels had a significantly increased risk of LREs. This association was validated in an independent cohort of 226 MASLD patients, supporting the value of extracellular vesicle proteins as liver microenvironment-specific biomarkers^[21]. The LiverRisk score was derived from international prospective cohort data and demonstrates excellent performance in predicting liver stiffness. The LiverRisk score can effectively identify high-risk populations for LREs, enabling precise risk stratification of liver-related outcomes. Stratified analysis showed that compared with the very low-risk group, the high-risk group had a hazard ratio (HR) of up to 471 for liver-related mortality. Its overall AUC for predicting 10-year liver-related death was 0.90, significantly higher than FIB-4’s 0.84^[22]. Another prospective cohort study developed and validated the LiverPRO tool in 6 independent cohorts from Denmark, Germany, and the United Kingdom. For liver fibrosis prediction, the AUC of LiverPRO was 0.81, comparable to the Enhanced Liver Fibrosis (ELF) score (0.78) and LiverRisk score (0.81), and significantly higher than FIB-4 (0.69) and NFS (0.74). Further validation in UK Biobank cohort showed that the LiverPRO tool had a C-statistic of 0.80 for predicting 2-year LREs, confirming its stable performance in short-term MASLD prognosis prediction^[23].

Models based on specific fibrosis-related biomarkers are also used for MASLD prognostic assessment. Among them, the ELF score is a well- validated proprietary assay^[24-26]. It combines three biomarkers: hyaluronic acid (HA), N-terminal propeptide of type III collagen (PIIINP), and tissue inhibitor of metalloproteinases 1 (TIMP-1), and performs well in fibrosis staging^[27]. A multicenter study from the UK confirmed its prognostic value: Cox proportional hazards model analysis showed that compared with the reference group with ELF score < 8.34, the HRs for LREs in the ELF score ranges of 8.34-10.425, 10.426-12.51, and 12.52-16.67 were 5.0, 20.0, and 75.0, respectively, indicating a clear dose-response relationship between ELF score and adverse outcome risk^[28]. However, as a proprietary assay, ELF has key limitations: high cost and interference from age and extrahepatic fibroinflammatory conditions, which restrict its widespread clinical use. Other models based on specific serum markers have also demonstrated potential clinical value: for example, the ADAPT score [age, diabetes status, procollagen III peptide (PRO-C3), platelet count] and the FIBROSpect test (α2-macroglobulin, HA, TIMP-1), both perform well in identifying advanced MASLD fibrosis^[29,30]. However, current research on the ability of these two models to predict long-term clinical outcomes such as LREs and all-cause mortality is relatively scarce, and their prognostic value requires further validation in large prospective multicenter cohorts.

With the continuous rise in the global prevalence of metabolic diseases, the incidence of MASLD-related HCC has increased significantly. Its pathogenesis involves interactions between metabolic disorders, chronic inflammation, and genetic susceptibility^[31]. Genetic variations in lipid metabolism genes are key drivers of MASLD progression to HCC^[32]. The patatin-like phospholipase domain-containing protein 3 (PNPLA3) variant is the best-established genetic risk factor, driving progression to cirrhosis and HCC^[33-35]. A meta-analysis of 109 studies including 118,302 MASLD patients showed that compared with non-carriers, carriers of PNPLA3 gene variations had more severe liver injury and histological changes. Meanwhile, the PNPLA3 GG genotype was significantly associated with higher mortality and LREs in MASLD patients^[36]. Grimaudo et al. followed 471 patients with histologically confirmed MASLD or clinically diagnosed compensated MASLD-related cirrhosis and found that PNPLA3 variants were independently associated with higher risks of decompensation (HR = 2.10), HCC (HR = 2.68), and liver-related death (HR = 3.64)^[37]. This association remained stable in the subgroup with F3 fibrosis or cirrhosis. To improve the accuracy of genetic prediction, several studies have moved beyond single-gene variants and integrated PNPLA3 with other genetic variations associated with progressive liver diseases (such as TM6SF2, GCKR, MBOAT7) to construct polygenic risk scores (PRS) for MASLD prognostic assessment^[38-40]. One study evaluated a PRS including these four genes in predicting HCC risk in MASLD patients. The results confirmed that this PRS can effectively predict HCC risk in MASLD patients, with prognostic performance significantly superior to single-gene variations^[41]. More recently, rare pathogenic variants identified by whole-exome sequencing have been integrated into a PNPLA3-based weighted PRS, further refining HCC risk stratification^[42]. However, this approach has practical limitations: whole-exome sequencing is costly, and data analysis requires specialized bioinformatics expertise, which restricts its popularization and application in primary care and routine clinical practice. Overall, PNPLA3 variants and derived PRS provide a new genetic perspective for noninvasive risk stratification of MASLD-related HCC. Future studies can improve PRS accuracy and generalizability by integrating additional newly identified liver-specific MASLD susceptibility variants. Given the central role of PNPLA3 in MASLD progression, clinical trials are needed to evaluate whether treatment response to emerging antifibrotic drugs differs by PNPLA3 genotype, thereby providing evidence for genetically guided individualized therapy and improving outcomes for MASLD-related HCC.

These emerging serum-based prognostic models, developed using omics, genetic, and machine learning approaches, remain largely in the early stages of investigation. Most generally lack large-scale, multicenter external validation, prospective evidence of clinical utility, and feasibility for widespread clinical implementation, and their translational value requires confirmation in additional high-quality studies.

Conventional ultrasonography is the first-line imaging modality for initial evaluation of MASLD, with well-recognized advantages of low cost, widespread equipment availability, and safety profile. It can detect mild or greater degrees of hepatic steatosis based on typical features such as hepatic parenchymal hyper-echogenicity and hepatorenal echo contrast^[43,44]. However, its diagnostic performance is limited by inter-device variability and operator-dependent subjectivity, which leads to suboptimal reproducibility. Semi-quantitative ultrasound indices, such as the hepatorenal ratio (HRR), have demonstrated considerable clinical utility for MASLD, particularly hepatic steatosis. One study has demonstrated a strong correlation between HRR and histological steatosis severity (r = 0.80, P < 0.01), with a cut-off of ≥ 1.24 for diagnosing steatosis (sensitivity 92.7%, specificity 92.5%)^[45]. HRR is the ratio of echogenic intensity between regions of interest placed in hepatic and renal parenchyma at the same depth, which may partially reduce operator-dependent bias^[46]. However, it cannot be reliably used in patients with acute or chronic kidney injury or those who have undergone right nephrectomy^[47]. In addition to assessing hepatic steatosis, conventional ultrasonography can reliably exclude focal hepatic lesions and identify indirect signs of portal hypertension, including splenomegaly, ascites, and portal vein dilatation, thus providing preliminary evidence for identifying patients at high risk of adverse hepatic outcomes^[48]. Despite its central role in first-line evaluation, conventional ultrasonography has limited ability to accurately quantify liver fibrosis and fails to enable precise risk stratification for long-term adverse outcomes. To overcome this limitation, elastography techniques represented by vibration-controlled transient elastography (VCTE) have been developed and widely used in clinical practice.

Among imaging modalities, VCTE is the most widely used noninvasive tool for assessing liver fibrosis^[49-53]. With technical advantages such as ease of use, short time consumption, high patient tolerance, and high repeatability, it avoids the invasive risks and sampling errors of liver biopsy, and has become a core auxiliary tool for the diagnosis, staging, and prognostic assessment of MASLD patients. Its core detection indicator - liver stiffness measurement (LSM) - is not only a key quantitative indicator for assessing liver fibrosis staging and identifying high-risk populations in MASLD patients but can also dynamically track changes in liver stiffness after treatment, providing real-time evidence for disease prognosis monitoring^[54,55].

Multiple high-quality studies have confirmed the predictive value of LSM for adverse outcomes in MASLD^[18]. A large US cohort study enrolled 30,414 MASLD participants with a total follow-up of 69,974 person-years. The results showed that for each 5 kPa increase in LSM, the risk of HCC increased by 18%; and the annual incidence of HCC increased gradually with increasing LSM intervals. Annual incidence rates were 0.34, 0.45, 0.78, and 0.94 per 100 person-years in the LSM 10-14.9, 15-19.9, 20-24.9, and ≥ 25 kPa groups, respectively^[56]. The composite scoring system derived from LSM has further improved the accuracy of prognostic prediction. A multicenter study involving 16 centers and 16,603 MASLD patients (median follow-up of 51.7 months) showed that the VCTE-based Agile score (integrating LSM with clinical parameters such as platelet count, transaminases, diabetes status, age, and sex) was significantly associated with the risk of LREs. Its time-dependent AUC was not only superior to traditional noninvasive models such as FIB-4, NFS, and LSM but also significantly better than histological fibrosis staging by liver biopsy. Moreover, patients with improved Agile scores from baseline had a significantly lower subsequent risk of LREs, providing a reliable basis for evaluating treatment response^[8]. Meanwhile, the study has also verified the diagnostic value of Agile3+ for at-risk MASH, with an AUC of 0.708 (0.672-0.744)^[57]. Recently, the VCTE-derived acFibroMASH index was proposed based on data from 3,004 biopsy-proven MASLD patients in 29 Chinese and 9 international cohorts. In terms of diagnostic performance, the index had an AUC of 0.808 (95%CI: 0.748-0.869) for identifying at-risk MASH with significant fibrosis, which was significantly higher than that of the conventional FibroScan-AST (FAST) score (0.764; 95%CI: 0.694-0.834, P = 0.040). For clinical use, the index uses dual cut-offs: < 0.15 rules out at-risk MASH with 90% sensitivity and 93% negative predictive value (NPV), while > 0.39 rules in at-risk MASH with 90% specificity and 60% positive predictive value (PPV). This addresses the dual clinical needs of screening and confirmatory diagnosis. Further validation of its prognostic value showed that, consistent with its diagnostic stratification performance, patients with an acFibroMASH index > 0.39 had an HR of 11.23 for LREs, compared with those with an index < 0.15. The index also achieved an AUC of 0.835 for predicting the 5-year risk of LREs, with prognostic performance that was again significantly superior to that of the FAST score (0.750). These findings confirm that the acFibroMASH index is a reliable tool for identifying at-risk MASH and predicting LREs risk^[58]. However, a core parameter of acFibroMASH is serum creatinine, which is affected by non-hepatic factors including renal function, dehydration, muscle mass, and drug use. Failure to adjust for these confounders may reduce model accuracy in specific populations, such as patients with CKD, emaciation, or obesity^[59,60]. Dynamic changes in LSM also have prognostic value in MASLD. The 2022 Baveno VII criteria first incorporated baseline LSM and longitudinal changes into clinical outcome prediction^[61]. A multicenter prospective study involving 1,403 MASLD patients showed that patients with LSM progressing to ≥ 10 kPa had a significantly higher cumulative incidence of LREs (16%) than non-progressors (4%), while patients with LSM improving to < 10 kPa had a significantly lower incidence of LREs (7%) than non-regressors (32%), confirming the close association between dynamic changes in LSM and LREs risk^[55]. However, recent evidence challenges this conclusion: a global multicenter cohort study across 5 countries found that both previous and current LSM values were associated with LREs, but current LSM better reflects real-time liver status. When the latest LSM data are available, the additional predictive value of baseline LSM and dynamic changes is negligible^[62]. This conclusion differs from previous studies, indicating that larger, longer longitudinal studies are needed to clarify the predictive performance of baseline and most recent LSM and the magnitude of dynamic changes for MASLD clinical outcomes. Such studies will inform the development of clinical monitoring strategies.

As a noninvasive assessment technology, magnetic resonance elastography (MRE) has shown high diagnostic accuracy for MASLD fibrosis staging^[63-65]. Liver stiffness measured by MRE is associated with adverse outcomes such as LREs and all-cause mortality^[66]. A recent multicenter individual-patient-data meta-analysis of 454 patients evaluated the prognostic value of MRE-derived tools. The results showed that positivity for the MEFIB index (combination of MRE and FIB-4) (HR = 49.22, 95%CI: 6.23-388.64, P < 0.001) and positivity for the MAST score [combination of AST, MRE, and magnetic resonance imaging proton density fat fraction (MRI-PDFF)] (HR = 3.86, 95%CI: 1.46-10.17, P < 0.001) were both independent predictors of new-onset liver decompensation, and the discriminative power of the MEFIB index was significantly higher than the MAST score^[67]. One study also validated the diagnostic value of the MAST score and MEFIB index for identifying at-risk MASH, with corresponding AUCs of 0.79 and 0.68, respectively^[68]. These MRI-based tools offer additional options for identifying high-risk patients with MASLD. However, available data are limited to single-time-point MRE assessments, and the association between longitudinal changes in MRE-based measures and adverse outcomes remains unexplored. Further longitudinal studies are needed to address this gap. MRE also has practical limitations: high cost and limited availability, which restrict its routine use in primary care and resource-limited settings.