Sepsis in liver failure patients: Diagnostic challenges and recent advancements

Abstract

Acute liver failure (ALF) and acute-on-chronic LF (ACLF) are prevalent hepatic emergencies characterized by an increased susceptibility to bacterial infections (BI), despite significant systemic inflammation. Literature indicates that 30%–80% of ALF patients and 55%–81% of ACLF patients develop BI, attributed to immunological dysregulation. Bacterial sepsis in these patients is associated with adverse clinical outcomes, including prolonged hospitalization and increased mortality. Early detection of bacterial sepsis is critical; however, distinguishing between sterile systemic inflammation and sepsis poses a significant challenge due to the overlapping clinical presentations of LF and sepsis. Conventional sepsis biomarkers, such as procalcitonin and C-reactive protein, have shown limited utility in LF patients due to inconsistent results. In contrast, novel biomarkers like presepsin and sTREM-1 have demonstrated promising discriminatory performance in this population, pending further validation. Moreover, emerging research highlights the potential of machine learning-based approaches to enhance sepsis detection and characterization. Although preliminary findings are encouraging, further studies are necessary to validate these results across diverse patient cohorts, including those with LF. This article provides a comprehensive review of the magnitude, impact, and diagnostic challenges associated with BI in LF patients, focusing on novel advancements in early sepsis detection and characterization.

Key Words: Liver failure; Sepsis; Bacterial infection; Acute liver failure; Acute-on-chronic liver failure

Core Tip: Patients with liver failure (LF) are prone to bacterial sepsis due to immune dysregulation. Up to 80% of such patients develop bacterial infection, which is associated with various complications and poor outcomes. Therefore, it is imperative to diagnose bacterial sepsis at the earliest. However, differentiating between patients with and without sepsis can be challenging, as LF itself can mimic sepsis by inducing systemic inflammation and organ failure. Several novel biomarkers for sepsis and machine learning techniques are being investigated, as conventional biomarkers have shown inconsistent results in LF patients. This article addresses the magnitude, impact, challenges, and recent developments in understanding bacterial sepsis in LF patients.

INTRODUCTION

Sepsis is a global health problem, and despite advancements in therapeutics, patients carry a substantial risk of in-hospital mortality[1]. A Global Burden of Disease Study reported an incidence of sepsis at 677.5 cases per 100000 age-standardized population, contributing to 19.8% (11 million) of global deaths[2].In 2019, sepsis resulted in 13.7 million deaths worldwide, with significant regional variations in mortality rates, ranging from 52.2 deaths per 100000 people in high-income regions to 230 deaths per 100000 people in sub-Saharan Africa[3]. The mortality rate from sepsis is significantly higher among intensive care unit (ICU) patients; a meta-analysis by Fleischmann et al[4] reported a pooled mortality rate of 41.9% in ICU-treated sepsis, vis-a-vis 26.7% in hospital-treated sepsis. Consequently, the World Health Organization declared sepsis a global health problem in 2017. The hallmark of sepsis is a dysregulated host response to infection, resulting in life-threatening organ failure[5]. Although BI are the most common cause of sepsis, infections from viruses, parasites, or fungi can also lead to the condition. The liver serves as a vital first line of defense against various pathogens[6]. Thus, patients with liver failure (LF) are at an increased risk of microbial infections and face a high risk of death from sepsis[7-9]. Early and accurate recognition of sepsis is essential for improving outcomes through more targeted medical treatment.

Acute LF (ALF) and acute-on-chronic LF (ACLF) are two common hepatic crises that can occur in patients with normal liver function and those with chronic liver diseases, respectively. These conditions are distinct but share several common features, including acute onset, jaundice, coagulopathy, hepatic encephalopathy, and a high short-term mortality rate[10,11]. Additionally, these conditions clinically resemble sepsis in diverse ways, including the development of systemic inflammatory response syndrome (SIRS) and organ failure[12-15]. Therefore, distinguishing between LF patients with and without sepsis can be challenging. ALF and ACLF patients paradoxically exhibit heightened susceptibility to BI despite significant systemic inflammation. This vulnerability appears to be related to immunological dysregulation, resulting in either a suppressed or exhausted adaptive immune system[9,16]. Once SIRS is triggered by an acute insult, the body develops a compensatory anti-inflammatory response, which may lead to immune paralysis and bacterial infections (BI) (Figure 1).

Figure 1 Schematic diagram showing immune dysregulation in liver failure patients. Acute insult results not only in systemic inflammation, but also compensatory anti-inflammatory response which leads to immune paralysis and heightened susceptibility to bacterial infection. ALF: Acute liver failure; ACLF: Acute-on-chronic liver failure; CLD: Chronic liver disease; CARS: Compensatory anti-inflammatory response syndrome; SIRS: Systemic inflammatory response syndrome; IL: Interleukin; TNF: Tumor necrosis factor; TGF: Transforming growth factor.

BI is one of the most common causes of ACLF and a frequent complication thereof too. Nearly one-third of ACLF patients present with BI, while half develop them during follow-up[6]. Notably, ACLF caused by BI exhibits greater mortality rates compared to other precipitating events[17-19]. In one study, the 30-day survival rate was only 33.8% in patients with BI as a trigger, compared to 71.6% among those without BI[18]. It is challenging but vital to distinguish between sepsis and sterile inflammation in LF patients. Without appropriate antibiotic therapy, mortality rates in sepsis patients generally rise by 3.3% to 7.6% every hour[20,21]. Thus, early identification of BI and timely antibiotic therapy are crucial for LF patients. This article addresses the magnitude, impact, challenges, and recent developments in understanding sepsis in LF patients, primarily due to BI.

IMMUNE DYSFUNCTION IN LF PATIENTS

The liver plays a critical role in the proper functioning of the immune system, as it neutralizes pathogens and enhances immunological tolerance. When innate immune cells are activated by an acute hepatic insult, a cascade of cytokines and chemokine is triggered, resulting in a severe SIRS. Uncontrolled systemic inflammation may lead to a vicious cycle of immunological dysfunction at multiple cell levels[22]. A compensatory anti-inflammatory response syndrome (CARS) develops throughout, leading to functional monocyte deactivation, a critical event in the development of systemic immunological dysfunction[1]. Depending on the stage of LF, macrophages may exhibit both tissue-destructive and tissue-repair effects. Additionally, the proliferation of suppressor cells produced from monocytic myeloids, through the programmed cell death protein 1/programmed death-ligand 1 axis, and a decrease in the bacterial clearance of Kupffer cells further compromise antimicrobial responses in LF patients[23]. Overall, an inappropriate CARS, along with the dysfunction and exhaustion of both innate and adaptive immune systems in LF patients leads to functional immune paralysis[24].Therefore, a dysregulated immune response in LF not only contributes to the progression of liver disease but also increases the risk of overwhelming sepsis, organ failure, and mortality[25,26].

Further, liver transplantation (LT) remains the definitive treatment for ALF and ACLF patients, offering a chance of long-term survival. However, post-transplant sepsis poses a significant challenge, particularly within the first month post-transplant[27]. The risk of infection correlates with the immunosuppressive regimen and graft function and is further exacerbated by the SIRS triggered by reperfusion injury and surgical stress. Moreover, immunosuppressant therapy after LT can lead to delayed presentation of infectious clinical symptoms, underscoring the importance of prompt diagnosis and rapid treatment for favorable outcomes.

CHARACTERISTICS OF BI IN ALF AND ACLF

There is a high prevalence of BI in both ALF and ACLF patients (Table 1). In ALF patients, BI occur in 30%–80%[8,28-30]. In a prospective observational study from King's College Hospital, London, Rolando et al[29] found culture-positive BI in 80% of 50 ALF patients. The most common site of infection was the respiratory tract (47.1%), with gram-positive bacteria accounting for 69.8% of all infections, and Staphylococcus aureus being the most common isolate (35.8%). In a retrospective study by Karvellas et al[30] from Canada, BI were found in 35% of ALF patients (72/206). Gram-negative organisms were observed in 52% of isolates, while gram-positive organisms were found in 44%. In a large cohort of ALF patients (n = 540) from India, BI were present at admission or developed within 48 hours in 22.2% of patients, while 34.3% developed them after 48 hours. Among infected patients, 54.9% were culture-positive, with 90% of isolates being gram-negative bacteria, the most common being Acinetobacter (32.7%). The respiratory tract was the most common site of isolation of organisms (45.8%), followed by bacteraemia in 28.2%[31].In a retrospective cohort study from the United States, Zider et al[28] found BI in 41% (62/150) of ALF patients, with the respiratory tract being the most common site (64.5%), followed by the urinary tract (54.8%).Notably, 40% of patients had more than one site of infection. In a recent study from India, multi-drug resistant (MDR) organisms accounted for 70% of BI, and 7% of infections were caused by pan-drug resistant organisms, underscoring the seriousness of this situation. The most prevalent presentations were pneumonia (50%), followed by urinary tract infections (UTIs) in 22%[8]. BI leading to sepsis can manifest at any point during ALF, with the median time of infection reported to vary from 3 to 10 days[29,30]. Immune dysfunction primarily causes early infections, while invasive procedures contribute to late episodes[30]. Early infections are mainly caused by bacteria from the endogenous normal flora, whereas later infections involve exogenous microorganisms.

Table 1 Prevalence of bacterial infections amongacute liver failure or acute-on-chronic liver failure patients in different countries.

Ref.	Country	Patients	n	Overall infection	Remarks
Cai et al[32], 2017	China	ACLF	389	81.2%	Pneumonia 49.4%, SBP 37.3%, UTI 13.3%. Gram positive sepsis 59.8%, Gram negative 55.3%
Fernández et al[18], 2018	Europe	ACLF	407	66.1%	BI at diagnosis 37%, BI during follow-up 46%. BI was associated with higher inflammation and worse outcomes
Karvellas et al[30], 2009	Canada	ALF	206	35%	Gram positive organisms in 44% of isolates, gram negatives in 52%, and 4% was fungal
Kaur et al[8], 2024	India	ALF	143	64%	Overall infection was 77% including fungal infection in 17.3%, MDR 70%
Liu et al[33], 2021	China	ACLF	140	69.2%	SBP 36.1%, pneumonia 23.7%, and multi-site infection 22.7%. Gram-negative bacteria 68.4%, and Gram-positive 31.6%
Moreau et al[35], 2013	Europe	ACLF	303	32.6%	SBP 10.6%, Pneumonia 61%, UTI 6.1%
Mücke et al[19], 2018	Europe	ACLF	173	41%	SBP 32.4%, pneumonia 25.4%
Rolando et al[29], 1990	England	ALF	50	80%	Respiratory tract infection 47%,and gram-positive bacteria 69.8%
Shalimar et al[31], 2017	India	ALF	540	49%	BI at diagnosis 22.2%, BI during follow-up 34%
Shalimar et al[34], 2018	India	ACLF	572	66.7%	Gram negative sepsis 91.6%. Pneumonia 45%, SBP 21.1%, UTI (15.2%)
Zhai et al[43], 2020	China	ACLF	289	64%	Gram negative sepsis 58.3%, Pneumonia 55.7%, SBP 47.6%
Zhang et al[37], 2022	China	ACLF	539	58.8%	SBP 31.54%, UTI 26.53%, Pneumonia 12.9%. Gram-positive sepsis 23.76%, Gram-negative sepsis 62.87%
Zider et al[28], 2016	United States	ALF	150	41%	One site infection 60%, multi-site infection 40%. Pneumonia 64% and UTI 55%

ALF: Acute liver failure; ACLF: Acute-on-chronic liver failure; BI: Bacterial infection; MDR: Multi-drug resistance; SBP: Spontaneous bacterial peritonitis; UTI: Urinary tract infection.

ACLF patients have an extraordinarily high prevalence of BI, ranging from 55% to 81%[9]. In a large European cohort study consisting of 407 ACLF patients, 37% presented BI at the time of ACLF diagnosis, and 46% of the remaining patients developed BI during follow-up (4 weeks). The most common site of infection at presentation was spontaneous bacterial peritonitis (SBP) (26.9%), while UTIs were the most common during follow-up (23%)[18]. In a retrospective cohort study from China, 81.2% of 389 patients with ACLF had BI, respiratory tract infections (49.4%) and SBP (37.3%) being the most prevalent forms. Gram-positive organisms (59.8%) were more frequently observed than gram-negative organisms (55.3%)[32].In another retrospective cohort study from China, BI occurred in 69.2% of 140 hepatitis B virus (HBV)-related ACLF patients, with SBP (36.1%) being the most common form of infection, followed by the lung infection (23.7%). Gram-negative BI were more common than gram-positive ones (68.4% vs 31.6%)[33]. In a large retrospective cohort study from India, 66.7% of 572 ACLF patients had BI, with gram-negative bacteria accounting for 91.6% of infections and pneumonia being the most common infection (45%)[34].

The risk of BI rises with the increasing severity of ACLF. Local epidemiological factors appear crucial in imparting the risk of BI in ACLF patients. Asian nations have reported a greater incidence of BI in ACLF than in Europe[19,32-38]. According to a multicenter international study, the incidence of BI-triggered ACLF was 59% in Asia, 75% in the Indian subcontinent, and 39% in Europe[28]. Thus, the most common forms of BI among ACLF patients appear to be SBP and pneumonia. Gram-negative bacteria are the predominant pathogens worldwide, while gram-positive bacteria have been reported to be more prevalent in Northern Europe (39%)[17]. Of particular concern is the increasing prevalence of MDR and extensively drug-resistant (XDR) organisms in ACLF patients. An international study of hospitalized patients with cirrhosis, including those with ACLF (n = 1302), found that 34% of MDR bacteria were present in culture-positive infections, with frequency varying geographically, from < 20% in the United States to > 70% in India. XDR organisms were found in 33% of cases in India[39], while in Europe (CANONIC study), BI were MDR in 15.8% at diagnosis and 18.8% on follow-up[18].

IMPACT OF BI ON OUTCOMES IN LF

A normal liver plays an essential regulatory role in sepsis and homeostasis. When the liver is already dysfunctional, sepsis can exacerbate hepatic injury, amplify systemic inflammation, and lead to multiple organ dysfunction. Bacterial sepsis is the leading cause of mortality in ALF patients, accounting for 10% to 52% of deaths. Infections caused by MDR organisms are associated with a higher incidence of advanced HE, multiorgan failure, longer hospitalization, and increasing mortality rates[8]. BI have been identified as independent predictors of persistent hyperammonemia in ALF, which in turn is associated with worsening HE and poor outcomes[40,41].

In ACLF, BI cause clinical deterioration, prolonged hospitalization, and increased mortality[9,42,43]. The mortality rates of ACLF patients at 28 days (35.5%–45.5%) and 90 days (50.7%–56.9%) are worse in patients with BI than those without[18,33,43-45]. BI are independent predictors of survival in patients with ACLF grade 1 and grade 2[9]. ACLF patients with multiple site infections have been found to have a higher incidence of septic shock, grade-3 ACLF, and 28-day mortality rates[9]. The failure of empirical antibiotic treatment in BI by MDR organisms further increases short-term mortality[18,19,46,47]. The site of infection also appears to affect mortality rates in ACLF; for instance, SBP is linked to increased risks of 90-day mortality compared to cutaneous, bone, or soft tissue infections[7].

CHALLENGES IN DIAGNOSING SEPSIS IN LF PATIENTS

Both ALF and ACLF trigger marked systemic inflammation driven by tissue damage, which mimics a dysregulated host response to pathogens. Thus, differentiating sterile systemic inflammation from sepsis becomes challenging in such patients. Fever can be absent in up to 30% of patients with ALF and 56% of patients with ACLF[31,48]. Total leucocyte counts (TLC) and neutrophil-lymphocyte ratios can both be elevated in inflammatory states without infection, making them unreliable for discrimination. Blood culture is the gold standard for bloodstream infections; however, its long processing time (72 hours or more) and low sensitivity make it unreliable. Conventional culture methods are often not sensitive enough to identify many unusual infections. Moreover, distinguishing between colonization and infection can be challenging when cultures are positive.

C-reactive protein (CRP) is an acute-phase protein synthesized in the liver following inflammatory stimuli; hence, CRP lacks specificity for BIs. In a meta-analysis, Tan et al[49] found CRP having a pooled sensitivity of 80% for overall sepsis, but a pooled specificity of only 61%. Further, there is a lag time of 12 to 24 hours before CRP concentrations begin to rise, limiting its usefulness as an early biomarker. Due to reduced hepatic parenchyma, CRP levels are often low in ALF patients. Silvestre et al[50], in a case series of ALF patients, discovered CRP to be markedly decreased and even undetectable in some ALF patients, despite the presence of sepsis. Procalcitonin (PCT) upregulation has been suggested as a means to distinguish between BI and other inflammatory conditions[51]. In healthy subjects, PCT is produced in the medullary C-cells of the thyroid gland. However, BI cause a profound increase in the expression of the CALC-I gene, leading to the release of PCT from various cell types. Following infection, PCT levels increase faster than CRP, usually within 2 to 3 hours, peaking at 24 hours. PCT levels are highest in patients with gram-negative sepsis, while they are only slightly elevated in patients with fungal infection[52]. In a meta-analysis, PCT demonstrated an area under the receiver operating characteristics (AUROC) curve of 85%, making it a viable biomarker to distinguish sepsis from other non-inflammatory conditions[53]. However, non-infectious inflammation can also cause an increase in PCT. While PCT appears to be a good assay for BI detection in the general population, it has limited discriminatory value in critically ill patients[54]. PCT levels can be elevated in ALF patients due to massive hepatocyte necrosis and systemic inflammation. In a study on ALF patients, PCT was unable to distinguish between those with or without BI[55]. In ACLF subjects, PCT was higher in the sepsis group, but its discriminating value was only modest, with the AUROC curve of 69%[48]. Additionally, PCT shows poor diagnostic sensitivity and AUROC curve when predicting BI in patients with impaired renal function, as well as in immunocompromised and autoimmune patients[56]. Lin et al[57] proposed an infection score comprising serum PCT, CRP, and neutrophils%, to predict BI in ACLF patients; however, the discriminating potential of this score remained modest, with an AUROC curve of 74%. A meta-analysis found that interleukin (IL)-6 had good diagnostic value for differentiating BI in patients with cirrhosis, with pooled sensitivity and specificity of 85% and 91%, respectively[58]. Serum IL-6 has been identified as an independent predictor of mortality in patients with HBV-ACLF[59]. Nevertheless, IL-6 is a pro-inflammatory cytokine, and its predictive role for sepsis in LF—a condition marked by intense systemic inflammation—requires further study.

NEWER BIOMARKERS AND FUTURE PERSPECTIVE

There is an urgent need for novel biomarkers to improve the diagnosis of BI in patients with LF. Before effective biomarkers can be developed and used in clinical settings, significant work is needed regarding methodology, standardization, and validation. Although numerous novel biomarkers for sepsis exist, based on proteomic, metabolomic, and genomic variables, only a few have been studied concerning liver disease (Table 2). Moreover, many of these biomarkers share the same drawbacks as traditional ones, and it is challenging to determine which is superior due to the lack of comparability caused by different methodologies and heterogeneous study populations.

Table 2 Studies on biomarkers of sepsispatients within acute liver failure or acute-on-chronic liver failure.

Ref.	Country	Patients	Biomarker(s)	Results and remarks
Silvestre et al[50], 2010	Europe	ALF	CRP	CRP levels are markedly decreased in ALF, even in presence of sepsis
Igna et al[62], 2022	Europe	ACLF	Presepsin, PCT and CRP	Presepsin, CRP, and PCT levels were higher in sepsis patients. Presepsin ≥ 2300 pg/mL had excellent (AUROC 0.95) for sepsis
Rule et al[55], 2015	USA	ALF	PCT	PCT levels > 2.0 ng/mL could not differentiate ALF patients with or without BIs
Huang et al[68], 2017	China	ACLF	Prostaglandin E2	Serum Prostaglandin E2 Level 141pg/mL predicted infection with AUROC curve of 0.83
Chen et al[48], 2021	China	ACLF	sTREM-1, Presepsin, and PCT	sTREM-1 and presepsin were significantly higher in sepsis patients, with higher accuracy compared to CRP and PCT
Cavazza et al[63], 2024	United States	ALF	sCD206	sCD206, a soluble markers of macrophage activation, independently predictedinfection
Yadav et al[69], 2022	India	ACLF	IL-1Ra, IL-18, TREM1, PD-L1, and TIM3	Higher baseline and rising levels of IL-1Ra, IL-18, TREM1 soluble factors, and suppressive monocytes (PDL1+ve, TIM3+ve)predictedrisk of sepsis within 72 hours
Lin et al[57], 2020	China	ACLF	PCT, neutrophils% and CRP	The AUROC of the infection score,comprisingPCT, neutrophils% and CRP, for discriminatingBI was 0.740
Yuet al[80], 2024	China	ACLF	BTLA	BTLA levels, a member of the CD28Igsuperfamily, significantly increased in the CD4+ T cells andwere positively correlated with infection complications

ALF: Acute liver failure; ACLF: Acute-on-chronic liver failure; AUROC: Area under receiver operating characteristic curve; BI: Bacterial infection; CRP: C-reactive protein; PCT: Procalcitonin; IL-1Ra: Interleukin-1 receptor antagonist; IL-18: Interleukin-18; TREM1: Triggering receptor expressed on myeloid cells 1; PD-L1: Programmed death ligand 1; TIM3: T-cell immunoglobulin and mucin domain-containing protein 3; BTLA: B- and T-lymphocyte attenuator.

Newer biomarkers such as soluble triggering receptor expressed on myeloid cell-1 (sTREM-1) and presepsin, a soluble CD14 subtype, have shown good sensitivity (83–85%) and moderate specificity (78%–79%) in differentiating sepsis from SIRS[60,61]. In a recent study, sTREM-1 and presepsin levels were significantly higher in ACLF patients with sepsis vis-a-vis those without. Further, sTREM-1 and presepsin outperformed TLC, PCT, and CRP in predicting sepsis. Additionally, diagnostic efficiency improved when sTREM-1 or presepsin was combined with the CLIF-SOFA score, achieving AUROC curves of 87% and 91%, respectively[48]. In another study, a presepsin level ≥ 2300 pg/mL was significantly associated with the early diagnosis of BI in ACLF patients[62]. Extra-hepatic BI can trigger hepatic macrophage activation. Recently, macrophage activation markers have been significantly associated with infection in ALF patients. Specifically, sCD206 was elevated in serum and upregulated on CD14+ monocytes[63]. Moreover, ACLF patients with BI and non-survivors exhibited higher levels of CD206[64]. Thus, sCD206 should be investigated as a potential biomarker of sepsis and mortality in LF patients. High-density lipoprotein cholesterol (HDL-C), which can bind to bacterial lipopolysaccharides, has an inverse correlation with BI[65]. Low HDL-C levels have been associated with poor outcomes in patients with cirrhosis and ACLF[66,67]. Therefore, HDL-C could also serve as a potential biomarker of BI in LF patients. In a study from China, serum prostaglandin E2 (> 140 pg/mL) predicted infection with modest accuracy (AUROC curve of 0.83) in ACLF patients[68]. In another recent study from India, higher baseline and rising levels of soluble factors such as IL-1 receptor antagonist, IL-18, sTREM-1, and suppressive monocytes predicted the risk of early sepsis in ALF patients[69]. Bacterial DNA testing, which primarily detects 16S ribosomal ribonucleic acid genes, enables rapid and wide-ranging detection of bacteria. However, limitations include inconsistent results, low comparability due to the wide range of PCR techniques, and compromised interpretation because the tests detect DNA rather than live pathogens[70,71]. Some other novel markers of sepsis include mid-regional pro-adrenomedullin and microfluidic assays for the spontaneous motility of neutrophils[72,73]. However, their performance in patients with LF remains to be tested. A summary of the performance, advantages, and limitations of various traditional and novel biomarkers of sepsis is provided in Table 3.

Table 3 Summary of the performance, advantages, and limitations of various traditional and novel biomarkers of sepsis.

Biomarkers	Overall performance for sepsis	Advantages	Limitations
CRP	Pooled sensitivity 80% and pooled specificity 61% for BI in general[49]	Wide availability. Low cost	Low specificity for BI. Falsely low levels in liver failure. Lag time: 12-24 hours. False positive in inflammation
PCT	Pooled sensitivity 77% and pooled specificity of 79% in general[53]	Easily available. Validated across multiple studies. Rapid elevation, 3–4 hours after BI	Modest to poor discriminatory role in liver failure patients. Varying cut-off levels. False positive in inflammation. Poor performance in renal failure and immunocompromised patients
IL-6	Pooled sensitivity 85% and specificity 91% for BI in cirrhosis[58]	Estimation is accurate, fast, and simple	It is a non-specific pro-inflammatory cytokine. Needs further studies in liver failure patients
HDL-C	HDL-C has an inverse correlation with BI[65,66]	Simple test. Low-cost. Widely available	Inverse correlation also exists between HDL-C and liver disease per se. Needs further studies as a biomarker for sepsis
Presepsin	Overall diagnostic sensitivity 83% and specificity 78%[59]	Better performance in liver failure patients than CRP and PCT[48]. Specific association with gram negative sepsis-Detectable within 2 hours of BI	Limited availability. Expensive test. More effective as an adjunct biomarker than when used alone. Requires further validation studies
sTREM-1	Pooled sensitivity 85% andspecificity 79% for differentiating sepsis from SIRS[61]	Early detection, < 2 hours after BI. Short half-life, making it useful for treatment response	Not routinely available. Varying cut-off levels. Requires further validation studies. Only modest performance when used alone
Bacterial DNA testing	Next-Generation Sequencing methodenables the identification of all bacteria in the blood and body fluid[70,71]	Quick and wide-ranging detection of bacteria	Not routinely available. Primer cross-reactivity with human DNA. Limited specificityand inconsistent results. DNA without a live pathogen compromises interpretation
sCD206	Significant association with infection (AUROC 71%) and mortality in ALF (AUROC 81%)[63]	It is among few novel biomarker evaluated in ALF patients	Not routinely available for use. Levels also increases in fungal and viral infection.Requires further validation studies

CRP: C-reactive protein; BI: Bacterial infection; PCT: Procalcitonin; IL: Interleukin; HDL: High density lipoprotein; sTREM-1: Soluble triggering receptor expressed on myeloid cell-1; SIRS: Systemic inflammatory response syndrome; AUROC: Area under receiver operating characteristics; ALF: Acute liver failure.

Recent years have seen advancements in the early detection and characterization of sepsis through machine learning techniques. These techniques involve the assessment of multiple sepsis biomarkers, such as cytokines, metabolites, damage-associated molecular patterns, microRNAs, and soluble or membrane receptors. Wang et al[74] developed a random forest model using 20 Laboratory parameters, achieving an AUROC of 91% for predicting sepsis in ICU patients. Zhao et al[75] employed XGBoost and LightGBM algorithms, with the LightGBM model yielding an AUROC of 97.1% for early sepsis prediction. A comparative study evaluating six machine learning algorithms found that the random forest model exhibited superior performance, with an F-measure of 99.9% and an AUROC curve of 91.8%[76]. Further, a recent comprehensive network meta-analysis of 73 articles, involving 457932 septic patients and 256 models, demonstrated the superior predictive performance of machine learning models, with a pooled AUROC of 82.5%. The analysis highlighted the substantial influence of inherent characteristics and algorithms of different models on their effectiveness in predicting sepsis. Neural network and decision tree models demonstrated the highest AUROC metrics[77]. To distinguish between sterile inflammation and early sepsis, Cahill et al[78] proposed a machine-learning classifier based on circulating levels of 31 cytokines. This investigation yielded several new findings, including the identification of macrophage-derived chemokine (MDC/CCL22) as a potential marker of sepsis. Despite the potential of machine learning models, such as neural networks, decision trees, and random forests, in clinical settings, their effectiveness is hindered by study heterogeneity, variability in sepsis definitions, and the lack of standardized validation procedures. To address this gap, there is a pressing need for standardized reporting and validation frameworks to ensure the reliability and generalizability of machine learning tools in diverse clinical contexts, including those with ALF and ACLF[79]. It was recently shown that CD4+ T cells in ACLF patients have significantly higher levels of B- and T-lymphocyte attenuator (BTLA), a member of the CD28 Ig superfamily, which is positively associated with BI. Further, administering a neutralizing anti-BTLA antibody reduced BI and mortality in a mouse ACLF model. These findings may provide new targets for therapeutic interventions[80]. Another proof-of-concept study identified Synaptotagmin 13 and IL-1 family member 10 as potential biomarkers of sepsis using advanced technologies, such as matrix-assisted laser desorption/ionisation and multiplex antibody arrays[81]. It is necessary to assess the usefulness of these potential biomarkers of sepsis in patients with ALF and ACLF.

CONCLUSION

Immune dysregulation not only contributes to the pathogenesis of ALF and ACLF but also increases the risk of infection. BI represent a frequent complication in patients with LF that negatively impact survival. Therefore, early detection and effective treatment of BI are fundamental to improving the survival of such patients. Many efforts are underway to distinguish sepsis from SIRS in patients with acute hepatic injury. Unfortunately, conventional biomarkers have produced inconsistent and disappointing results. Extensive research is being conducted to aid in the identification of sepsis biomarkers. While many novel circulating biomarkers have been discovered recently, only a few have been studied in the context of LF. Thus, the validation of current sepsis biomarkers in liver disease patients, and the search for new, accurate, and cost-effective ones, must remain a priority in human research. Emerging research has highlighted the potential of machine learning-based approaches for enhancing the early detection and characterization of sepsis. Although preliminary results are promising, further prospective research is warranted to validate these findings across diverse patient populations, including individuals with LF, and to establish the clinical utility and generalizability of these approaches.