To ease the navigation through the sections of the present systematic review, each prognostic biomarker is categorized into four main sections: serum biomarkers, imaging biomarkers, combination of serum and imaging biomarkers, and genomic biomarkers. Moreover, the findings are presented based on the two key outcomes predicted by these biomarkers: CVEs (Table 1) and LREs (Table 2). If not otherwise stated, CVEs are defined as fatal or nonfatal ischemic stroke, myocardial infarction, cardiac or peripheral revascularization, arterial fibrillation, cardiovascular death, or a combination thereof. The same principle applies to LREs, which are defined as ascites, portosystemic encephalopathy, hepatopulmonary syndrome, spontaneous bacterial peritonitis, hepatorenal syndrome, or a combination thereof. Finally, for each biomarker, an evaluation of its potential utility as treatment monitoring biomarkers is summarized in Table 3.

The NAFLD Fibrosis Score (NFS) is a non-invasive score that was initially developed and validated as a diagnostic tool by Angulo and colleagues to discriminate between the presence or absence of advanced fibrosis (F3-F4) in NAFLD patients [26]. Its formula includes age, body mass index (BMI), presence of impaired fasting glucose or diabetes mellitus, aspartate aminotransferase (AST)/alanine aminotransferase (ALT) ratio, platelet count, and serum albumin levels. In the current review, a total of nine studies identified NFS as prognostic biomarker of LREs and four of CVEs. Several threshold values are suggested for NFS to be prognostic of CVEs, LREs, and/or mortality. NFS > 0.676 is the most commonly used threshold and predicted CVEs occurring within 3.5–5.2 years follow-up (HR: 2.29–4.61, Table 1) [21, 27] as well as mortality due to CVEs over 14.5 years follow-up (HR: 3.46 (95% CI: 1.91–6.25), Table 1) [28]. In addition, NFS > 0.676 predicted LREs occurring within 5–8.7 years follow-up (HR: 5.12–34.20, Table 2) [29,30,31] as well as overall mortality occurring within 100–104.8 months follow-up (HR: 1.58–9.80, Table 2) [29, 31]. Furthermore, NFS > 0.676 predicted the increased occurrence of malignancies other than HCC (HR: 1.27 (95% CI: 1.05–1.42), Table 2), increased hospital admissions (HR: 1.74 (95% CI: 1.31–2.31), Table 2), and duration of hospitalization within 100 months (HR: 1.61 (95% CI: 1.23–2.10), Table 2) [29]. While the above-mentioned outcomes for NFS > 0.676 were predicted in cohorts of mixed ethnicities, lower NFS thresholds values were predictive of events in Asian patients. NFS >  − 1.455 predicted increased overall mortality in Japanese (HR: 12.87 (95% CI: 1.35–122.30), Table 2) [32] and Chinese (HR: 2.74 (95% CI: 1.67–4.50), Table 2) [33] participants followed-up for 4.6 and 6.6 years, respectively. In the Chinese study, a lower cut-off value (NFS >  − 1.836) is recommended to increase the prognostic accuracy of overall mortality in its population (sensitivity: 88.3%; specificity: 61.9% for the prediction of 6.6-year mortality, Table 2) [33]. An even lower threshold (NFS ≥  − 2.08) is set in a Korean study for optimal prognosis of overall and liver-specific mortality (HR: 1.43 (95% CI: 1.21–1.68), Table 2) [34]. Additionally, a non-specified threshold value for NFS predicted CVEs (Harrel’s c-index = 0.65, Table 1) [35], as well as LREs (HR: 1.78 (95% CI: 1.43–2.21), Table 2) [11], (Harrel’s c-index = 0.80 ± 0.02, Table 2) [35] within 30.9 and 81 months, respectively. Taken together, NFS can be considered a biomarker prognostic of decompensation events and mortality in patients with NASH cirrhosis. However, this score includes an invariable parameter such as age that will not change significantly during a typical trial duration of 6–12 months of treatment. In addition, patients with higher fibrosis tend to be significantly older than patients with milder fibrosis [29], and this could be reflected as a higher NFS value just as a function of age. Therefore, we consider the NFS a biomarker of rather low treatment monitoring utility (Table 3).

The Fibrosis-4 index (FIB-4) is a non-invasive score initially developed to stage liver disease in subjects with HIV-HCV co-infection [36] and with HCV infection alone [37]. Its formula includes age, AST, ALT, and platelet count. FIB-4 was later on applied as a marker of advanced fibrosis in NAFLD as well [38]. In the current review, a total of ten studies identified FIB-4 as prognostic biomarker of LREs and three of CVEs. Several threshold values are suggested for FIB-4 to be prognostic of CVEs, LREs, and/or mortality. FIB-4 > 2.67 is the most commonly used threshold and predicted CVEs (HR: 4.57 (95% CI: 1.60–12.98), Table 1) and mortality due to CVEs (HR: 2.68 (95% CI: 1.44–4.99), Table 1) occurring within 3.5–14.5 years [27, 28]. In addition, FIB-4 > 2.67 predicted LREs (HR: 13.1–14.6, Table 2) within 34.8–100 months follow-up [29, 31]. More in detail, FIB-4 > 2.67 predicted the development of HCC (HR: 3.66 (95% CI: 2.71–4.94), Table 2), liver transplantation (HR: 7.98 (95% CI: 4.62–13.79), Table 2) and end-stage liver disease (HR: 1.86 (95% CI: 1.68–2.05), Table 2) [39], overall mortality (HR: 2.49–10.52, Table 2) [29, 31, 39], increased hospital admissions (HR: 3.80 (95% CI: 2.79–5.19), Table 2), as well as duration of hospitalization (HR: 2.69 (95% CI: 1.92–3.78), Table 2) [29]. FIB-4 > 3.25 predicted LREs (HR: 6.33 (95% CI: 1.98–20.2), Table 2) [30] and the range 1.30–2.66 was used to predict overall mortality (HR: 1.13–1.46, Table 2) [28, 39] and end-stage liver disease (HR: 1.14 (95% CI: 1.07–1.22), Table 2) [39]. FIB-4 ≥ 1.30 predicted the development of HCC (HR: 8.46 (95% CI: 1.06–67.37), Table 2) over 2.5 years follow-up [40]. Similarly to what was the case for NFS, the lowest threshold (FIB-4 ≥ 1.22) was used in a study of Korean subjects where it predicted overall mortality (HR: 1.41 (95% CI: 1.18–1.68), Table 2) [34]. Finally, a non-specified threshold value for FIB-4 predicted CVEs (Harrel’s c-index = 0.60 ± 0.03, Table 1) [35] and LREs (Harrel’s c-index = 0.78 ± 0.03, Table 2) [35]. Taken together, FIB-4 can be considered a prognostic biomarker for decompensation events and mortality in patients with NASH cirrhosis. Although FIB-4 was also utilized as treatment monitoring biomarker [41], its formula contains age, which is an invariable parameter considering a typical trial duration. Hence, an observed treatment effect in FIB-4 is most likely attributable to changes in the values of the liver enzymes ALT and AST, as well as platelet count. We therefore consider FIB-4 of medium value as treatment monitoring biomarker (Table 3).

The AST to platelet ratio index (APRI) is a non-invasive score initially developed for the prediction of F3 and F4 in patients with chronic hepatitis C infection [42]. In the current review, one study investigates APRI as prognostic biomarker of CVEs, whereas four studies focus on LREs. Those studies identify baseline APRI > 1.5 as prognostic threshold for the outcomes considered [28,29,30,31]. In a large, multiethnic study including more than 11,000 patients, APRI > 1.5 predicted CVEs (HR: 2.53 (95% CI: 1.33–4.83), Table 1) [28]. In addition, APRI > 1.5 predicted LREs (HR: 5.02–20.90, Table 2) [29,30,31], increased overall mortality (HR: 3.1 (95% CI: 1.1–8.4), Table 2) [31], the occurrence of malignancies (HR: 4.94 (95% CI: 1.92–12.82), Table 2), and increased hospital admissions (HR: 2.49 (95% CI: 1.80–3.43), Table 2) as well as hospitalizations (HR: 2.90 (95% CI: 2.11–3.98), Table 2) [29]. Finally, a non-specified threshold value for APRI predicted LREs (HR: 1.88 (95% CI: 1.45–2.46), Table 2) [11]. Taken together, APRI can be considered a biomarker prognostic of decompensation events and mortality in patients with NASH cirrhosis. Given the formula of the APRI score, consisting of AST to platelet ratio, we consider this biomarker of high utility for treatment monitoring in patients with advanced liver disease (Table 3), and indeed evidence for the use of APRI as treatment monitoring biomarker exists [41]. It remains to be determined whether treatment-related changes in APRI associate with better outcomes.

The BARD (BMI, AST/ALT ratio, type 2 diabetes (T2D)) score was initially developed in a cohort of 823 NALFD patients of various ethnicities (Caucasian, Black, Hispanic, Asian Pacific Islander) considering BMI, AST/ALT ratio and T2D, where it showed a positive predictive value (PPV) = 43% and a negative predictive value (NPV) = 96% for the diagnosis of advanced fibrosis [43]. Because of its high NPV, this score seems to be more suited for ruling out the presence of fibrosis as to predict the occurrence of long-term outcomes, reflected by the low number of studies reporting on the prognostic ability of BARD. A non-specified threshold value for BARD predicted CVEs (Harrel’s c-index = 0.64 ± 0.04, Table 1), LREs (Harrel’s c-index = 0.73 ± 0.02, Table 2), HCC (Harrel’s c-index = 0.77 ± 0.03, Table 2), and extrahepatic cancer (Harrel’s c-index = 0.62 ± 0.04, Table 2) within 81 months [35]. BARD = 4 predicted the development of LREs (HR: 6.6 (95% CI: 1.4–31.1), Table 2) over a median of 104.8 months follow-up [31]. In the multicenter cohort study from Younes and colleagues, BARD was significantly outperformed by NFS and FIB-4 in the prognosis of long-term outcomes according to univariate cox proportional hazard models [35]. Given that the BARD formula includes T2D, i.e. invariable parameters considering a typical trial duration, we consider its treatment monitoring utility to be rather low (Table 3).

The enhanced liver fibrosis (ELF™) test is a non-invasive test developed and patented by Siemens Healthineers that combines three serum biomarkers of fibrosis: hyaluronic acid (HA), tissue inhibitor of metalloproteinase-1 (TIMP-1), and amino-terminal peptide of procollagen III (PIIINP). The algorithm for its calculation was initially identified by Rosenberg and colleagues [44] and a population of patients with liver fibrosis of diverse etiology was used to determine threshold levels for the diagnosis of moderate liver fibrosis (≥ 7.7– < 9.8; sensitivity = 85%) and cirrhosis (≥ 11.3; specificity = 95%) [45]. In the USA, the ELF™ Test has been granted FDA authorization as prognostic risk assessment tool for patients with chronic liver disease by the FDA [46]. It can be used as prognostic marker in conjunction with other laboratory findings and clinical assessment tools in patients with advanced fibrosis due to NASH to assess the likelihood of progression to cirrhosis and liver-related clinical events. In a study of 475 Caucasian and Hispanic cirrhotic patients, ELF™ ≥ 11.27 predicted LREs (HR: 2.11 (95% CI: 1.53–2.90), Table 2) within 30.9 months follow-up [11]. Conversely, lower baseline ELF™ was associated with fibrosis regression [11]. The ELF™ test is widely used as treatment monitoring biomarker in recent clinical trials investigating new NASH treatments [41, 47,48,49,50,51]. In a phase IIa study of patients with compensated NASH cirrhosis, 16-week treatment with efruxifermin was associated with significant reduction of ELF score (− 0.4 efruxifermin vs. + 0.4 placebo; p = 0.0036) [52]. Hence, we consider the ELF™ test of high treatment monitoring utility as this parameter might be well suited to study treatment responses (Table 3), given that its constituents (i.e., HA, TIMP-1, PIIINP) are direct markers of liver fibrosis that are sensitive to change from baseline following treatment [51, 53].

The Hepamet Fibrosis Score (HFS) is a recently developed formula including age, sex, AST, albumin, homeostatic model assessment (HOMA), diabetes mellitus and platelet count [54]. Values of HFS ≥ 0.47 were used to identify advanced fibrosis (sensitivity: 35.2%; specificity: 97.2; PPV: 76.3%; NPV: 85.2%) and in doing so HFS demonstrated greater diagnostic accuracy compared to NFS and FIB-4 [54]. In the multicenter cohort study from Younes and colleagues, HFS was predictive of increased overall mortality in Caucasian subjects over a median follow-up of 81 months [35]. Given that the HFS contains invariable parameters such as age, sex, and diabetic status, we consider its treatment monitoring utility to be rather low (Table 3).

Two studies were found where the liver enzymes measured were ALT and AST/√ALT. In a study of 42,282 American NAFLD patients of various ethnicities (Caucasian, Black, Hispanic), patients with liver steatosis + elevated ALT (> 40 IU/mL in men and > 31 IU/mL in women) were compared to patients with liver steatosis + normal ALT and those with no liver steatosis + normal ALT. Patients with liver steatosis + elevated ALT had a significantly increased incidence of HCC over a median follow-up of 8.4 years (HR: 4.35 (95% CI: 1.90–9.94), Table 2) [55]. In this group, 5-year and 8-year cumulative incidence rates of HCC were 1.0 and 1.4 per 1000 patients, respectively [55]. In another study including 7068 cirrhotic patients of various ethnicities the AST/√ALT was used as predictor of HCC development. Several ranges were tested and those with AST/√ALT > 6.45 (> 6.45–8.80, > 8.80–12.83, > 12.83) were predictive of HCC (HR > 1.99, Table 2) over a mean of 3.7 years follow-up [56]. Both liver enzymes are utilized as established treatment monitoring biomarkers. Nevertheless, it is also well accepted that ALT is not always elevated in patients with NASH [57]. Therefore, a reduction in ALT levels following treatment might not occur despite an effective therapy. For this reason, we consider the use of liver enzymes of rather low treatment monitoring utility in patients with NASH cirrhosis (Table 3).

Alpha-fetoprotein is considered a diagnostic and prognostic biomarker of HCC, and high serum levels are associated with increased risk of HCC development and poor prognosis [