Acute-on-chronic liver failure (ACLF) is a syndrome that occurs in the context of chronic liver disease and is triggered by a variety of internal and external liver insults (1). Characterized by acute decompensation, multi-organ failure and high short-term mortality, ACLF poses a significant challenge to healthcare systems worldwide, particularly in Asia where liver disease is prevalent (2, 3). Patients with ACLF often present with complications such as hepatorenal syndrome, ascites, hepatic encephalopathy and bacterial infections, which further complicate their clinical course.

The pathogenesis of ACLF is multifaceted, involving immune responses, hepatic inflammation and gut microbiome dysbiosis, among other factors. Early diagnosis of ACLF can be challenging due to overlapping symptoms and signs with other liver diseases (3). Liver transplantation remains the only curative option for ACLF; however, the shortage of donor organs limits its widespread use. Therefore, early diagnosis, prompt treatment and comprehensive management are essential to improve patient outcomes and survival.

Several prognostic models for ACLF have been developed, including the APASL-AARC score, EASL-CLIF C ACLF score, COSSH-ACLF II score, Model for End-Stage Liver Disease (MELD) and MELD-Na scores (1). These models provide clinicians with objective criteria to assess the severity of ACLF and facilitate informed treatment decisions and resource allocation. The Albumin-Bilirubin (ALBI) score, originally proposed by Johnson et al. as a model to assess liver function in patients with hepatocellular carcinoma (HCC), has shown limited predictive power for the prognosis of ACLF (4).

Our aim is to use existing case data to develop and validate a novel ALBI prognostic model and nomogram to predict the prognosis of ACLF patients. Nomograms offer a more intuitive and user-friendly approach compared to statistical models, helping clinicians to make more accurate medical decisions. By integrating ALBI with other clinically relevant parameters, we aim to improve the prognostic accuracy and clinical utility of this model, ultimately contributing to improved treatment strategies and patient outcomes in the management of ACLF.

ACLF represents a critical transition in chronic liver disease, triggered by an acute insult, leading to rapid deterioration of liver function and potentially culminating in multi-organ failure within a short period of time (17, 18). Given the high mortality associated with ACLF and the paucity of effective therapeutic interventions, early diagnosis and prognostic assessment are of paramount importance in clinical practice (19). The development and evaluation of prognostic models is essential to improve our understanding of disease progression, refine treatment strategies and provide patients with more accurate prognostic information.

Despite the widespread use of the CTP score and the MELD score in liver disease, their predictive power in China is not optimistic due to the differences in aetiological factors and environmental conditions between regions (20–22). To address this gap, we have developed a survival prediction model that integrates the ALBI with CR, gender and ascites, complemented by a nomogram to facilitate clinical application. Our model has undergone rigorous internal validation and has demonstrated commendable efficacy.

The ALBI score, introduced by Professor Philip J. Johnson and colleagues (4), emerged from a landmark study in 2015 that analyzed clinical data from 1,313 patients with hepatocellular carcinoma (HCC) of various stages in Japan. By incorporating serum albumin and total bilirubin, the ALBI score was developed to measure liver reserve function. The score has been shown to be an effective predictor of overall survival in patients with mid-stage HCC. However, its efficacy in the prognosis of ACLF was suboptimal (23–25). In our study, the ALBI score was −1.05 (−1.33, −0.86) in the improvement group and − 0.77 (−0.97, −0.53) in the mortality group, with an AUC of 0.729, Our findings are in agreement with those of Bo Chen’s study (26), which evaluated the relative efficacy of the ALBI score in predicting the 90-day mortality rate of patients with acute-on-chronic liver failure (ACLF) by establishing a ROC curve. The area under the ALBI score curve (AUC) was 0.784 ± 0.049 (p < 0.001), indicating that they also highlighted the significance of the ALBI score in predicting the exacerbation of chronic liver failure, similar to the AUC of 0.729 observed in our study. This model offers a novel tool that can assist clinicians in more accurately assessing patient risk and formulating personalized treatment plans. However, the study has certain limitations, such as the sample size and selection bias toward specific populations. Consistent with these findings, the ALBI score alone had suboptimal predictive power for ACLF in our study.

Our investigation revealed that CR levels differed significantly between the improvement group (median: 62.60 μmol·L⁻¹, interquartile range [IQR]: 50.60, 73.80) and the mortality group (median: 79.10 μmol·L⁻¹, IQR: 59.00, 115.00) in the training cohort, highlighting CR as an independent prognostic factor in ACLF patients, with higher levels predicting a worse prognosis. Elevated CR may indicate renal dysfunction, a hallmark of the multi-organ failure characteristic of ACLF, involving the kidneys, respiratory and circulatory systems (27–29). The increase in creatinine may not only indicate renal failure, but may also predict the deterioration of other organ functions, thereby influencing the overall clinical outcome (30–32). This finding is consistent with previous studies identifying CR as a marker of adverse prognosis in ACLF (33). In several prognostic models, CR is integrated with other clinical parameters to form complex predictive algorithms for ACLF (34–36). Notably, the ALBI score does not incorporate renal function metrics; therefore, we integrated CR with the ALBI score to potentially improve the predictive power of our model.

The role of Gender in ACLF is complex, involving enzyme activity, hormone levels, lifestyle factors and genetic predisposition (37). At the hormonal level, estrogen is known for its anti-fibrotic effects, which may contribute to the better prognosis observed in women with certain liver diseases (38–40). Lifestyle factors, such as alcohol consumption, are more prevalent in men, with excessive alcohol consumption directly damaging hepatocytes and indirectly accelerating the progression of liver failure through metabolic (41), inflammatory and nutritional effects. Genetically, recent evidence suggests that male-specific genes, such as the determinant region Y gene (42), act as potent profibrotic factors, which may explain the preponderance of males in our study (43). In conclusion, although gender plays a role in ACLF, it is typically not an independent prognostic factor and existing models often omit this variable. However, given the ease of assessing gender and the absence of this parameter in current models, we experimentally included it in our model. In the nomogram, women receive approximately 5 points, a relatively small contribution compared to other factors. We suggest that the complex interplay of multiple factors associated with gender may account for its suboptimal performance as an independent predictor of ACLF prognosis.

Ascites, a common complication of ACLF, typically results from portal hypertension (44), renal sodium and water retention, and decreased colloid osmotic pressure (45, 46). Although ascites is included in the Child-Pugh classification (47–49), our study simplifies its categorization as present or absent without assessing the volume of ascites and its impact on prognosis. Future research could include the quantification of ascites using ultrasound to refine prognostic assessments.

Our model, by assessing the 90-day prognosis of patients with ACLF, provides a robust decision-support tool for clinical practice. In clinical decision-making, by identifying those at high risk, physicians can formulate personalized treatment plans in advance, optimize resource allocation, and intervene promptly. In patient management, the application of this model can facilitate more precise resource allocation. By identifying patients who are most likely to benefit from specific treatments, the healthcare system can make more effective use of limited resources.

Due to the limitations of the study design and sample size, we included only a limited number of clinical variables. This may have constrained the model’s ability to comprehensively capture the complexity of patients’ conditions, thus affecting its robustness. From a model validation perspective, we stratified the 290 patients based on timeline to create a training set and a validation set, representing an internal validation process. This approach did not include data from patients in other hospitals for external validation. We hope that future research will involve collaborative multicentre studies to develop more accurate predictive models. The issue of gender in relation to ascites has been addressed previously and is therefore not discussed here. In terms of study design, our research is a retrospective cohort study. In order for the nomogram to be translated into clinical practice, we believe that further prospective clinical trials are essential to establish its clinical utility. Furthermore, the limited number of clinical variables included in our study may affect the robustness of the generated nomogram. Finally, our study lacks an aetiological classification for ACLF. The majority of our patients were infected with HBV, and although statistically categorizing the etiology and creating prognostic models for patients with different aetiologies should theoretically yield more accurate results, the current case numbers are insufficient for this purpose.

To improve the model’s predictive accuracy and clinical utility, we suggest that future research should consider incorporating a wider array of variables, including biomarkers, genetic markers, and lifestyle factors. Furthermore, we recommend conducting multicenter and multipopulation studies to verify the model’s universality and robustness. Concurrently, with the advancement of technology, we should explore the use of more advanced statistical and machine learning algorithms to optimize the model.

We suggest that in clinical application, the model could be integrated into electronic health record systems to provide clinicians with real-time prognostic assessments and treatment recommendations. In addition, the use of nomograms for patient education may improve patients’ understanding of their disease status and prognosis, thereby increasing their engagement in treatment. Finally, public health policy decisions based on the results of the model will provide data to support the prevention and control of liver disease.

In conclusion, the ALBI score combined with CR, gender and ascites to construct a survival prediction model not only improves the quality of clinical decision making, but also promotes the development of personalized and precise treatment and management of liver diseases, which has significant clinical importance and broad application prospects.