Sepsis is a life-threatening organ dysfunction caused by infection and is one of the common causes of death in patients admitted to the intensive care unit (ICU) (1). The production of endotoxins and the release of inflammatory factors in sepsis lead to an abnormal immune response that impairs the functioning of many organs, including the liver and kidneys (2). The liver plays a central role in metabolic and immune homeostasis, and liver injury is one of the common complications in patients with septic shock and one of the risk factors for death in septic patients (3, 4). Early liver damage is caused by inflammation and microcirculation disorders, which can be recovered later with treatment. The mechanism of late injury may be related to organ failure caused by sepsis (5).

Sepsis-associated liver injury (SALI) is one of the most common complications in patients with sepsis. The mechanism is that the onset of primary dysfunction is usually associated with inadequate hepatic perfusion, which leads to liver injury and further complications such as diffuse intravascular coagulation and multi-organ failure (6). Secondly, sepsis also causes dysfunction of the intestinal microcirculation, prompting leakage of intestinal toxins and bacteria into the liver through the portal vein, which can lead to hepatitis (7). There are several main types of SALI: ischemic-hypoxic liver injury, cholestatic liver injury, and hepatocellular injury (8). Ischemic-hypoxic liver injury results from hepatic hypoperfusion and hypoxemia, characterized by a significant elevation in serum aminotransferase levels, along with coagulation abnormalities and minimal jaundice (9). Cholestasis is primarily defined by hyperbilirubinemia or jaundice, which may be linked to impaired bile formation and reduced motility (10). The classification of hepatocellular injury types predominantly relies on drug-related liver injury criteria established in the literature (11). There is a lack of clarity about the mechanisms by which SALI occurs, which may be related to the following pathways, including microcirculation and endothelial damage (12), intestinal flora dysbiosis (13), inflammatory factor imbalance (7), oxidative stress (14), and mitochondrial metabolic disorders (15), among other points. Studies have shown that liver failure is an independent risk factor for multi-organ dysfunction and high mortality in patients with sepsis (16). However, there are no studies of in-ICU mortality prediction models for patients with SALI. Therefore, the purpose of this paper is to explore the independent risk factors for the development of SALI.

Sepsis is most prevalent among the elderly, infants, and individuals with certain underlying conditions that compromise the immune system. As society ages, the population of elderly patients is growing, leading to an increase in the incidence of sepsis (17). Reducing the mortality and disability rates of SALI in the elderly population is one of the current issues that needs to be addressed. Nomograms are visual statistical tools that integrate various data to develop continuous scoring systems reflecting individual risk probabilities accurately (18). Therefore, it is crucial to establish an accurate dynamic nomogram, which has important implications for both individuals and society through prognostic and clinical management decisions.

Sepsis continues to be a critical concern in modern medicine, demanding urgent attention. If left unchecked, it can escalate into MODS, significantly increasing patient morbidity and mortality (21). Liver failure is an important aspect of septic MODS and usually portends a poor prognosis (22, 23). Despite advances in treatment, mortality rates continue to rise progressively, a problem further exacerbated by the significant aging of today’s population (24). Therefore, it is essential to assess the mortality of elderly SALI patients admitted to the ICU.

This study primarily utilized the MIMIC database to evaluate the incidence and progression of mortality in SALI patients following their admission to the ICU. Lasso regression analysis was employed to identify independent risk factors associated with ICU mortality in these patients. Ultimately, six clinical variables of ALB, AG, PTT, ALT/ALP, APRI and mechanical ventilation were determined by the construction of the model.

Albumin has binding, translocation, and detoxification of endogenous and exogenous molecules, killing effects, antioxidant activity, and modulation of immune and inflammatory responses (25, 26). The mechanisms of SALI are unknown, and previous studies have shown that cytokine-induced inflammation and mitochondrial oxidative stress are key factors contributing to SALI (27). Previous studies have shown that ALB reduces the release of cytochrome c from mitochondria by inhibiting the leakage of the cysteine protease histone B from lysosomes while reducing the release of cytochrome c from mitochondria, concluding that ALB not only protects hepatocytes from the cytotoxic effects of the cytokine TNFα but also prevents inflammatory mediators from further damaging tissues (28). Evidence-based indications for albumin administration in clinical guidelines are primarily liver disease (29). In patients with sepsis or septic shock who remain hemodynamically unstable after resuscitation with 30 ml/kg crystalloid, albumin infusion may be considered clinically for further treatment. It can be added to a large crystalloid infusion rather than used alone for crystalloid therapy (30, 31). In conclusion, when the concentration of ALB in the patient’s body decreases, the mortality rate of the patient will increase, and our modeling further validates that ALB has a strong ability for the prognosis of the patients, and further by strengthening the management of the patient in terms of ALB.

Anion gap is a comprehensive outcome indicator determined by the measured anion and cation concentrations (32). However, AG is also significantly affected by ALB levels (33). In the setting of severe sepsis, the development of hyperlactatemia can trigger a high AG metabolic acidosis. Previous studies have shown that AG in early ICU admission is one of the risk factors for the outcome of sepsis patients (34). In this study, AG variables were included through lasso review to predict the mortality of patients admitted to the ICU, and the effect of ALB on AG was considered, so after screening ALB was also used as a part of the model, which complemented the previous study and thus predicted in several ways.

Alanine transaminase and ALP are sensitive indicators of liver damage, and liver injury can cause elevated serum levels of ALT and ALP (35). ALT is a specific marker of hepatocellular injury, whereas bilirubin and PT-INR represent liver function (36). ALP is an enzyme found primarily in the hepatobiliary tract and is elevated in the cholestatic form of liver injury (37). However, mild to moderate elevations of ALP are common in all types of hepatitis are common, even in hepatocellular injury (36). The ALT/ALP ratio, which combines both ALT and ALP, is more closely related to the pathologic features of liver injury and changes in disease (38). PLT is involved in sepsis combined with multiorgan dysfunction by regulating inflammation, tissue integrity, and defense against infection (39). Studies have shown that APRI is a good predictor of SALI in children that platelets accumulate in the liver when the body is attacked by inflammatory cytokines, leading to liver injury and a decrease in platelet counts in the peripheral blood, and that platelets can influence the regeneration of damaged liver tissue (40). Previous studies have shown that two novel indicators, ALT/ALP and APRI, can be good predictors of SALI, but in the modeling developed in this study, both accounted for a small proportion of the modeling, which may be related to the fact that we did not include enough sample sizes of the elderly. Secondly, ALT and ALP may increase simultaneously in liver injury, resulting in no significant difference in their ratios, and the P-value of APRI tends to be close to 0.05 when predicting SALI alone at the beginning of the modeling, which means that it does not have a significant meaning when predicting alone. Therefore, the weighting of the scores in the modeling was not high, which means that we need to include more cases in the future to explore the predictive value of APRI for SALI.

Common complications at the onset of severe sepsis include coagulation abnormalities and liver injury. Overall coagulation indices such as PTT and prothrombin time are generally used in the clinical setting, and a single index may reflect only part of the coagulation system (41). DCIs that occur when there is decreased production of coagulation factors associated with hepatocellular injury or sepsis-associated coagulation abnormalities can exacerbate the depletion. Studies have shown that a longer PTT may reflect the most severe sepsis due to coagulation factor depletion or high-dose heparin therapy and that a prolonged PTT is associated with higher mortality (42), and PTT prolongation is also an important marker of severe hyperfibrinolysis (43). This is consistent with our study. As the PTT value prolongs, the percentage of PTT increases, leading to a gradual rise in patient mortality rates. The occurrence of SALI further impacts the consumption of coagulation factors, resulting in sepsis-related coagulation abnormalities. In severe cases, DIC develops, accelerating patient mortality. These two conditions exacerbate each other, so it is crucial to clinically monitor changes in coagulation function when SALI occurs.

The last variable incorporated in our model is mechanical ventilation. When a patient is in a critical situation, such as respiratory failure, respiratory arrest, or oxygen desaturation, clinical work is mostly done with a ventilator. Mechanical ventilation is widely used in the management of septic patients but is prone to complications. Ventilator-associated pneumonia, a lung infection during mechanical ventilation, is the most common nosocomial lung infection in intensive care patients. Whereas sepsis itself, as a severe systemic inflammatory disease, its combination with ventilator-associated pneumonia may lead to deterioration of the patient’s condition and increase the risk of death (44, 45).

Simplified Acute Physiology Score II and SOFA scores are commonly used to evaluate the prognosis of patients in clinical practice (5, 46), so the two variables are not included in the model construction. In this study, SOFA and SAPS2 scores were used to compare their performance with that of predictive models. We found that nomogram performed the best. In addition, the DCA curve supports this finding. Through the above comparison, it is further shown that the efficiency of the constructed model is better.

Intensive care unit mortality in elderly SALI patients was the endpoint of this study. Independent risk factors were screened by lasso regression, and a risk prediction model was successfully constructed based on these factors. Lasso regression analysis can solve the problem of multicollinearity among variables and is superior to univariate analysis. The nomogram constructed based on the Lasso regression model has a certain reference value for medical workers to visually analyze individual prognosis. As a visual scoring tool suitable for clinical research, the nomogram can synthesize various influencing factors and present their results intuitively. Decision makers can formulate individualized treatment plans to prolong survival based on the predicted risk of death. Through the predictive model, doctors can find out in advance whether a patient is likely to have a poor prognosis and take appropriate measures to reduce the risk of mortality, thus improving treatment outcomes, lowering healthcare costs, and enhancing patients’ quality of life.

By establishing a robust predictive model, clinicians can proactively identify health issues at an early stage, thereby improving both treatment efficacy and the overall quality of life for patients. However, it is crucial to recognize the limitations inherent in this study. Firstly, it relies on a single-center retrospective analysis, and the findings require validation through prospective cohort studies. Secondly, despite the approach of randomly dividing the data into training and validation sets, and the highly reliable models produced during the development process, the generalizability of the models is still limited by the lack of external validation of the centralized dataset. This limitation emphasizes the need for further evaluation of its stability and applicability. Lastly, while this analysis focused on 934 elderly patients, future research would benefit from multi-center approaches and larger-scale validations to enhance both stability and performance.

We successfully constructed a model of in-ICU mortality of SALI elderly patients. The nomogram created based on the above factors assessed the risk of death in patients, which is of great informative value and clinical applicability.