Qi R, Wang X, Kuang ZD, Shang XY, Lin F, Chang D, Mu JS. Dynamic nomogram predicts sepsis risk in patients with acute liver failure: Analysis of intensive care database with external validation. World J Gastroenterol 2025; 31(31): 105229 [DOI: 10.3748/wjg.v31.i31.105229]
Corresponding Author of This Article
Jin-Song Mu, MD, PhD, Associate Professor, Chief Physician, Peking University 302 Clinical Medical School, No. 100 Xisihuan Middle Road, Fengtai District, Beijing 100039, China. jinsongmu302@126.com
Research Domain of This Article
Gastroenterology & Hepatology
Article-Type of This Article
Retrospective Study
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Rui Qi, Jin-Song Mu, Peking University 302 Clinical Medical School, Beijing 100039, China
Rui Qi, Xin Wang, Zhi-Dan Kuang, Xue-Yi Shang, Fang Lin, Dan Chang, Jin-Song Mu, Department of Critical Care Medicine, The Fifth Medical Center of Chinese PLA General Hospital, Beijing 100039, China
Author contributions: Qi R contributed to writing the original draft, data collection, and statistical analysis; Wang X contributed to statistical analysis and interpretation; Kuang ZD and Chang D contributed to data collection and data curation; Shang XY and Lin F contributed to reviewing and editing; Mu JS contributed to study conception and design, reviewing, and editing; All authors critically reviewed and approved the final manuscript; All authors were responsible for the decision to submit the manuscript for publication.
Supported by National Key Research and Development Program, No. 2022YFA1103501.
Institutional review board statement: Medical Information Mart for Intensive Care IV is a public database, and all patient information is anonymized to protect privacy. Therefore, the requirement for approval by the local ethics committee was waived. The Fifth Medical Center of Chinese PLA General Hospital cohort strictly complied with the ethical standards outlined in the Declaration of Helsinki and received approval from the Ethics Committee of the Fifth Medical Center of Chinese PLA General Hospital (No. KY-2024-10-163-1).
Informed consent statement: The Medical Information Mart for Intensive Care IV database obtained ethical approval from the Institutional Review Boards at both Beth Israel Deaconess Medical Center and Massachusetts Institute of Technology. Informed consent from patients in The Fifth Medical Center of Chinese PLA General Hospital cohort was waived due to the retrospective nature of the study.
Conflict-of-interest statement: All authors report no relevant conflicts of interest for this article.
Data sharing statement: The datasets used or analyzed during the current study are available from the corresponding author upon reasonable request at jinsongmu302@126.com.
Open Access: This article is an open-access article that was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution NonCommercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: https://creativecommons.org/Licenses/by-nc/4.0/
Corresponding author: Jin-Song Mu, MD, PhD, Associate Professor, Chief Physician, Peking University 302 Clinical Medical School, No. 100 Xisihuan Middle Road, Fengtai District, Beijing 100039, China. jinsongmu302@126.com
Received: January 16, 2025 Revised: April 8, 2025 Accepted: July 16, 2025 Published online: August 21, 2025 Processing time: 214 Days and 15.5 Hours
Abstract
BACKGROUND
Acute liver failure (ALF) with sepsis is associated with rapid disease progression and high mortality. Therefore, early detection of high-risk sepsis subgroups in patients with ALF is crucial.
AIM
To develop and validate an accurate nomogram model for predicting the risk of sepsis in patients with ALF.
METHODS
We retrieved data from the Medical Information Mart for Intensive Care (MIMIC) IV database and the Fifth Medical Center of Chinese PLA General Hospital (FMCPH). Univariate and multivariate logistic regression analysis were used to identify risk factors for sepsis in ALF and were subsequently incorporated to construct a nomogram model [sepsis in ALF (SIALF)]. The discrimination ability, calibration, and clinical applicability of the SIALF model were evaluated by the area under receiver operating characteristic curve, calibration curves, and decision curve analysis, respectively. The Kaplan-Meier curves were used for robustness check. The SIALF model was internally validated using the bootstrapping method with the MIMIC validation cohort and externally validated by the FMCPH cohort.
RESULTS
A total of 738 patients with ALF patients were included in this study, with 510 from the MIMIC IV database and 228 from the FMCPH cohort. In the MIMIC IV cohort, 387 (75.89%) patients developed sepsis. Multivariate logistic regression analysis revealed that age [odds ratio (OR) = 1.016, 95% confidence interval (CI): 1.003-1.028, P = 0.017], total bilirubin (OR = 1.047, 95%CI: 1.008-1.088, P = 0.017), lactate dehydrogenase (OR = 1.001, 95%CI: 1.000-1.001, P < 0.001), albumin (OR = 0.436, 95%CI: 0.274-0.692, P = 0.003), and mechanical ventilation (OR = 1.985, 95%CI: 1.269-3.105, P = 0.003) were independent risk factors associated with sepsis in patients with ALF. The SIALF model demonstrated satisfactory accuracy and clinical utility with area under receiver operating characteristic curve values of 0.849, 0.847, and 0.835 for the internal derivation, internal validation, and external validation cohort, respectively, which outperformed the Sequential Organ Failure Assessment scores of 0.733, 0.746, and 0.721 and systemic inflammatory response syndrome scores of 0.578, 0.653, and 0.615, respectively. The decision curve analysis and calibration curves indicated superior clinical utility and efficiency than other score systems. Based on the risk stratification score derived from the SIALF model, the Kaplan-Meier curves effectively discriminated the real high-risk subpopulation. To enhance the clinical utility, we constructed an online dynamic version, enabling physicians to evaluate patients’ condition and track disease progression in real-time.
CONCLUSION
Based on easily identifiable clinical data, we developed the SIALF model to predict the risk of sepsis in patients with ALF. The model demonstrated robust predictive efficiency, outperformed Sequential Organ Failure Assessment and systemic inflammatory response syndrome scores, and was validated in an external cohort. The model-based risk stratification and online calculator might further facilitate the early detection and appropriate treatment for this subpopulation.
Core Tip: The study developed and validated a dynamic nomogram to predict sepsis risk in patients with acute liver failure using the Medical Information Mart for Intensive Care database and an external cohort. Key predictors included age, total bilirubin, lactate dehydrogenase, albumin, and mechanical ventilation. The model demonstrated satisfactory accuracy, outperformed Sequential Organ Failure Assessment and systemic inflammatory response syndrome scores. Decision curve analysis and calibration curves confirmed superior clinical utility and efficiency. The model-based risk stratification and online calculator might further facilitate risk evaluation and guide rational management.
Citation: Qi R, Wang X, Kuang ZD, Shang XY, Lin F, Chang D, Mu JS. Dynamic nomogram predicts sepsis risk in patients with acute liver failure: Analysis of intensive care database with external validation. World J Gastroenterol 2025; 31(31): 105229
Acute liver failure (ALF) is a rare condition characterized by a rapid decline of liver function, coagulopathy, and encephalopathy in the absence of preexisting liver disease[1,2]. Owing to the hepatic dysfunction, patients with ALF experience immunodeficiency, making them more susceptible to severe bacterial infections[2]. Infection is one of the most common complications of ALF, and approximately 50%-70% of patients may progress to sepsis[3]. When combined with sepsis, the aberrant immune response to infection can induce fatal multiple organ dysfunction syndrome (MODS), which is a leading cause of mortality in intensive care units (ICUs)[3,4]. The pathophysiological mechanism of sepsis in ALF remains unclear, and it may be associated with nutritional deficiencies, reduced intestinal barrier function, and weakened immune function.
The liver plays a key role in the immune response and metabolic regulation during sepsis[5]. In patients with ALF the capacity of the liver to clear endotoxins is compromised, leading to an elevated blood endotoxin level. Cytokine storms and endotoxins triggered by sepsis can further damage hepatocytes, leading to hepatic inflammation, necrosis, and disorders in regeneration[6]. Damaged hepatocytes release progenitor-associated molecular patterns and damage-progenitor-associated molecular patterns, which may exacerbate systemic inflammatory responses, leading to MODS or death[7]. Clinically, ALF presents as a systemic inflammatory response due to massive hepatocyte necrosis and cytokine release. The symptoms are similar to those of sepsis, such as high excretion and low resistance, renal failure, metabolic abnormalities, and multiorgan damage. Delayed antibiotic therapy reduces survival in patients with sepsis by 7.6% per hour[8], while premature antibiotic use may increase the risk of hospital-acquired infections and infections with multiresistant organisms[9].
The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) suggest that an increase in the Sequential Organ Failure Assessment (SOFA) score of ≥ 2 from baseline may indicate the presence of sepsis in patients[10]. However, the definition based on the SOFA score lacks necessary sepsis-specific indicators and may still lack the precision required for accurate diagnosis. Moreover, ALF has a distinct clinical course and features, including sepsis-like manifestations, making the identification of sepsis in patients with ALF more challenging[11-13]. In recent years nomograms have been widely used as visualization tools for clinical prognostic studies in survival analysis of patients who were critically ill[14].
In this study we aimed to develop an online nomogram model for predicting the risk of sepsis in patients with ALF from the ICU. We used data from the Marketplace for Information in Critical Care Medicine (MIMIC) database and conducted external validation in a cohort of ICU wards at the Fifth Medical Center of Chinese PLA General Hospital (FMCPH).
MATERIALS AND METHODS
Patients
Patients included in the retrospective study were from the MIMIC-IV database and the FMCPH cohort. The MIMIC-IV database is a freely available database that contains anonymous medical records of patients admitted to ICUs of Beth Israel Deaconess Hospital (Bowers, MA, United States) from 2008 to 2019[15]. The FMCPH cohort consists of patients with ALF admitted to the Department of Critical Care Medicine between January 2009 and March 2023. The MIMIC-IV data were used for developing and internally validating the sepsis in ALF (SIALF) nomogram model, while the FMCPH cohort served for external validation. We included the patients who were admitted to the ICU with a diagnosis of ALF and without preexisting liver diseases.
Exclusions criteria: (1) Suspected sepsis prior to ICU admission; (2) Age under 18 years; (3) Use of immunosuppressive drugs; (4) HIV infection; (5) Severe chronic extrahepatic organ disease, such as chronic obstructive pulmonary disease with respiratory failure, coronary heart disease with New York Heart Association III, or chronic kidney disease with renal failure; (6) Malignant tumors; (7) Liver transplantation history; (8) Pregnancy; and (9) ICU stay shorter than 24 h as these patients may not have sufficient clinical data for analysis.
Data collection
This study included two cohorts of patients with ALF admitted to the ICU. Data for the FMCPH cohort, including demographic, clinical, biochemical, and outcome information, were retrieved from the electronic medical record system. We extracted data on baseline patient characteristics (age, sex), laboratory parameters on the first day of admission [lactate dehydrogenase (LDH), international normalized ratio, albumin (ALB), serum alanine aminotransferase, total bilirubin (TBil), platelets, creatinine, etc.], disease severity scores [model for end-stage liver disease, Simplified Acute Physiology Score II, systemic inflammatory response syndrome (SIRS)], treatments [mechanical ventilation (MV), continuous renal replacement therapy, vasopressor therapy], and survival outcomes from the MIMIC-IV database. The baseline SOFA scores in MIMIC-IV were calculated using the worst values of key parameters (respiratory, hepatic, renal, cardiovascular, neurological) recorded within the first 24 h of ICU admission, strictly adhering to Sepsis-3 criteria. These data were extracted through Postgres Structured Query Language programming using Navicat Premium software.
Definition of ALF and sepsis
ALF: ALF was defined by the following criteria[1,2,16]: (1) Coagulation abnormality (international normalized ratio ≥ 1.5); (2) Hepatic encephalopathy of any grade (West Haven Criteria); (3) Acute illness onset within 26 weeks after hepatic injury; and (4) Absence of preexisting cirrhosis.
Sepsis: Sepsis was diagnosed using the Sepsis-3 criteria: (1) Patients with pathogenically positive infections and new organ dysfunction; and (2) With a SOFA score increase ≥ 2 points from baseline.
Statistical analysis
Continuous variables were summarized as the median and analyzed with the Mann-Whitney U test. Categorical data were presented as counts (percentages) and assessed using the χ2 test. Univariate and multivariate logistic regression analysis identified independent risk factors for sepsis in patients with ALF, estimating odds ratios (OR) and 95% confidence intervals (CI). These factors were then incorporated into the construction of nomograms with each multivariate regression coefficient being proportionally scaled to a 0-100-point range. The summation of points across independent variables yielded total points, which were subsequently converted to predicted probabilities for sepsis in ALF at multiple time points. The predictive accuracy of the nomogram was evaluated using receiver operating characteristic analysis with pROC package in R. The discriminatory ability was assessed using the area under the receiver operating characteristic curve (AUROC). Bootstrap resampling was employed to construct the calibration curve to minimize overfitting bias. Calibration curves were generated by comparing predicted probabilities with observed outcomes where alignment with the diagonal line indicated optimal calibration. Overall performance was evaluated using the Brier score and R2 value where a lower Brier score or higher R2 value indicated better performance. DCA quantified the net clinical benefit of using our nomogram across threshold probabilities, demonstrating its superiority over treat-all or treat-none strategies. Kaplan-Meier (KM) curves were employed to plot cumulative survival, and overall survival proportions were compared by log-rank test. Variance inflation factor (VIF) analysis was used to evaluate multicollinearity among variables. Model validation was confirmed in an external cohort using the above methods. A two-tailed P < 0.05 was considered statistically significant. Data analysis was conducted using R software version 4.2.1.
RESULTS
Clinical characteristics
As shown in Figure 1, the MIMIC cohort initially enrolled 1225 patients. According to the specified inclusion and exclusion criteria, 715 patients were excluded: 63 patients with HIV infection or undergoing immunosuppressive therapy; 159 patients with malignant tumors; 68 patients with severe chronic extrahepatic organ diseases; 46 patients with liver transplants; 221 patients suspected of having sepsis before ICU admission; 137 patients with incomplete data; 11 pregnant patients; and 10 patients under 18 years old. Finally, 510 patients with ALF were enrolled in the study. Patients were divided into sepsis and non-sepsis groups according to whether they developed sepsis after ICU admission.
Figure 1 Flow chart of patient enrollment and study design.
MIMIC: Medical Information Mart for Intensive Care; FMCPH: The Fifth Medical Center of Chinese PLA General Hospital; ICU: Intensive care unit; ALF: Acute liver failure; SIALF: Sepsis in acute liver failure; ROC: Receiver operating characteristic; DCA: Decision curve analysis.
The external FMCPH cohort eventually included 228 patients applying the same exclusion criteria. The baseline demographic and laboratory characteristics of the MIMIC cohort and the FMCPH cohort are summarized in Table 1. We compared the clinical and biochemical characteristics of enrolled patients in the MIMIC cohort stratified by sepsis (n = 387) and non-sepsis (n = 123) (Table 1). The sepsis group had a median age of 68 (52, 75) years compared with 59 (48, 71) years in the non-sepsis group. The sepsis group generally exhibited more severe abnormalities in vital signs and laboratory parameters, including lower levels of hemoglobin, platelets, hematocrit, red blood cells, and calcium, as well as worse liver function indicators (aspartate aminotransferase, TBil, LDH, lactate, and ALB) (all P < 0.05). They also had higher infection-related markers (WBC, neutrophils) and worse disease severity scores (model for end-stage liver disease, SOFA, Simplified Acute Physiology Score II, and SIRS) than the non-sepsis group. In addition we also observed a significant increase in the treatment of continuous renal replacement therapy, MV, and vasopressors on the first day of ICU admission in the sepsis group (all P < 0.05). Lastly, in the study endpoints the 28-day and 90-day mortality rates were significantly higher in the sepsis group compared with the non-sepsis group (all P < 0.05).
Table 1 Baseline characteristics of patients with acute liver failure patients in the Medical Information Mart for Intensive Care database and external cohort.
Characteristics
MIMIC cohort
FMCPH cohort (n = 228)
Total (n = 510)
Sepsis (n = 387)
Non-sepsis (n = 123)
P value
Patient demographics
Age (year)
61.0 (48.2-73.0)
68.0 (52.5-75.0)
59.0 (48.0-71.0)
0.003
58.0 (47.5-70.5)
Male
325 (63.72)
248 (64.00)
77 (62.60)
0.849
173 (75.87)
Vital signs
MAP
78.0 (72.5-84.4)
77.0 (72.2-83.7)
79.3 (73.2-85.7)
0.071
77.4 (70.5-83.2)
Heart rate
88.1 (76.3-101.7)
89.3 (77.9-103.9)
84.5 (71.7-93.2)
0.005
85.9 (74.4-98.6)
Respiratory rate
20.5 (17.2-24.3)
20.9 (17.3-24.9)
19.2 (16.9-22.5)
< 0.001
20.2 (17.0-23.8)
Electrolytes and metabolic indicators
Na
138.6 (135.8-141.2)
139.0 (136.3-141.8)
137.6 (134.6-140.0)
< 0.001
137.0 (133.0-141.0)
Ca
7.9 (7.3-8.4)
7.8 (7.3-8.4)
8.2 (7.7-8.6)
< 0.001
7.6 (7.4-8.5)
K
4.8 (4.2-5.4)
4.9 (4.4-5.8)
4.7 (4.2-5.3)
0.014
4.8 (4.5-5.1)
Cl
101.6 (98.2-105.6)
101.7 (98.3-105.8)
101.2 (97.9-104.4)
0.098
100.9 (98.1-104.8)
Anion gap
19.5 (16.0-25.0)
20.0 (16.0-26.0)
17.0 (14.5-22.0)
< 0.001
18.3 (15.7-25.6)
Glucose
135.3 (112.4-167.0)
140.6 (115.8-170.8)
121.6 (104.2-151.2)
< 0.001
138.19 (116.2-169.2)
HbA1c
5.7 (5.3-6.4)
5.8 (5.3-6.8)
5.6 (5.3-5.8)
0.103
5.7 (5.2-6.3)
Fibrinogen
231.1 (159.4-340.4)
226.9 (159.5-343.5)
232.2 (157.3-322.1)
0.766
227.4 (152.7-337.2)
Granulocytes
0.9 (0.6-1.4)
0.9 (0.6-1.5)
0.7 (0.5-1.4)
0.019
0.9 (0.6-1.3)
CK
478.5 (142.5-1739.8)
561.5 (169.0-1922.4)
276.0 (114.5-1056.8)
0.026
486.4 (134.0-1817.5)
CRP
74.8 (25.8-133.8)
77.9 (36.5-145.4)
56.2 (18.2-113.7)
0.378
80.7 (24.1-125.0)
Blood routine examination
WBC
15.5 (11.0-21.5)
16.2 (12.2-22.7)
13.1 (8.3-18.2)
< 0.001
16.3 (11.9-23.9)
Neu
10.2 (7.2-15.3)
10.5 (7.6-15.9)
8.6 (5.8-12.5)
0.002
10.8 (7.0-14.8)
RBC
3.2 (2.8-3.8)
3.1 (2.8-3.7)
3.3 (2.9-4.0)
0.033
3.8 (2.6-4.1)
RDW
15.9 (14.3-18.4)
16.2 (14.4-18.4)
15.4 (14.0-17.6)
0.023
14.8 (13.6-17.3)
HCT
28.9 (26.3-34.4)
28.6 (26.1-33.8)
30.9 (27.3-36.6)
0.003
27.8 (24.9-35.0)
HGB
9.4 (8.5-11.3)
9.3 (8.4-11.0)
10.0 (8.8-12.3)
0.004
9.1 (8.1-12.2)
PLT
151.9 (93.3-210.1)
144.4 (88.2-207.4)
170.0 (107.8-224.2)
0.042
155.0 (101.0-269.0)
HCO3-
17.0 (13.0-21.0)
16.0 (12.0-20.0)
19.0 (15.5-22.0)
< 0.001
18.0 (14.5-22.5)
Blood gas analysis
PaO2
95.8 (69.9-124.2)
96.9 (74.2-123.4)
92.0 (53.0-132.0)
0.161
91.0 (74.0-117.2)
PaCO2
39.5 (35.6-43.7)
39.8 (35.7-43.7)
39.0 (34.5-43.2)
0.229
32.5 (27.8-36.0)
Renal function indicators
BUN
30.2 (18.9-42.3)
31.9 (20.2-46.7)
28.6 (16.0-40.9)
0.090
30.7 (17.7-38.7)
Cr
1.6 (1.0-2.5)
1.7 (1.1-2.6)
1.3 (0.8-1.9)
< 0.001
1.6 (1.0-2.2)
Liver function indicators
AST
333.5 (120.3-1079.6)
347.3 (127.8-1144.5)
276.8 (82.5-805.5)
0.027
315.0 (97.5-1010.5)
TBil
1.5 (0.7-4.8)
1.5 (0.7-6.0)
1.3 (0.7-2.4)
0.036
1.4 (0.8-3.8)
LDH
593.1 (356.4-1234.3)
647.4 (378.6-1385.0)
466.6 (307.5-877.9)
< 0.001
623.0 (387.4-1197.4)
Lactate
2.7 (1.9-4.4)
2.8 (2.0-4.7)
2.4 (1.8-3.5)
0.012
2.4 (1.6-3.3)
ALT
271.0 (80.9-752.7)
285.1 (90.0-765.0)
266.4 (67.8-741.2)
0.394
327.5 (79.5-896.8)
ALB
3.0 (2.6-3.4)
2.9 (2.6-3.3)
3.2 (2.8-3.5)
< 0.001
3.0 (2.5-3.3)
Coagulation function indicators
PT
18.4 (14.8-24.2)
18.4 (14.9-24.6)
18.1 (14.6-23.2)
0.526
19.2 (14.3-28.2)
INR
1.7 (1.4-2.2)
1.8 (1.6-2.2)
1.7 (1.5-2.1)
0.456
2.6 (1.7-3.9)
First day of ICU admission
CRRT
63 (12.4)
56 (14.5)
7 (5.7)
0.016
24 (10.5)
Vasopressin
85 (16.7)
79 (20.4)
8 (6.5)
< 0.001
28 (12.3)
Vasoactive agents
41 (8.0)
29 (7.5)
12 (9.8)
0.539
17 (7.5)
Mechanical ventilation
355 (69.6)
284 (73.4)
71 (57.7)
0.002
138 (60.5)
Severity Score
MELD
23.0 (16.0-31.0)
23.0 (17.0-33.0)
20.0 (14.0-27.5)
< 0.001
23.4 (21.0-33.9)
SAPS II
44.0 (34.0-57.0)
46.0 (36.0-58.0)
37.0 (29.0-50.0)
< 0.001
40.5 (31.0-54.5)
SIRS
3.0 (2.0-3.0)
3.0 (2.0-4.0)
3.0 (2.0-3.0)
< 0.001
3.0 (2.0-3.0)
SOFA
9.0 (6.0-11.0)
9.0 (7.0-12.0)
5.0 (4.0-8.0)
< 0.001
9.0 (6.0-10.0)
Outcome
28-day mortality
220 (43.1)
185 (47.8)
35 (28.5)
< 0.001
144 (63.1)
90-day mortality
257 (50.4)
207 (53.5)
50 (40.7)
0.017
163 (71.5)
Novel nomogram development
Through univariate and multivariate logistic regression analysis, we selected statistically significant variables (Table 2) including age (OR = 1.016, 95%CI: 1.003-1.028, P = 0.017), TBil (OR = 1.047, 95%CI: 1.008-1.088, P = 0.017), LDH (OR = 1.001, 95%CI: 1.000-1.001, P < 0.001), ALB (OR = 0.436, 95%CI: 0.274-0.692, P = 0.003), and MV (OR = 1.985, 95%CI: 1.269-3.105, P = 0.003) as candidates to construct the novel prediction model. These five indicators were plotted on a single plane with scaled lines to depict their interrelationships in the predictive model. We performed VIF analysis to assess multicollinearity. All predictor variables showed acceptable VIF values ranging from 1.2 to 3.8, well below the threshold of 5, confirming minimal multicollinearity in our final model. The predictive values for outcome events were calculated by transforming the total score into a probability function for the occurrence of the outcome event. The generated nomogram, named SIALF, was developed to predict the risk of sepsis in patients with ALF (Figure 2).
Figure 2 Sepsis in acute liver failure nomogram for predicting the risk of sepsis in acute liver failure.
The nomogram was constructed with five admission variables. Each patient admitted would receive an individualized score for each variable. Summing all scores generated a potential risk of sepsis in patients with acute liver failure (red dot). MV: Mechanical ventilation; ALB: Albumin; LDH: Lactate dehydrogenase; TBil: Total bilirubin.
Table 2 Risk factors for sepsis in patients with acute liver failure from the Medical Information Mart for Intensive Care derivation cohort.
Parameters
Univariate
Multivariate
OR (95%CI)
P value
OR (95%CI)
P value
Age (year)
0.984 (0.972-0.996)
0.007
1.016 (1.003-1.028)
0.017
Male
1.066 (0.713-1.622)
0.766
Heart rate
1.016 (1.004-1.027)
0.006
Respiratory rate
1.081 (1.033-1.131)
< 0.001
Na
1.098 (1.048-1.153)
< 0.001
Ca
0.718 (0.575-0.896)
0.003
K
1.317 (1.062-1.633)
0.012
Anion gap
1.085 (1.048-1.123)
< 0.001
Glucose
1.007 (1.002-1.012)
0.006
Granulocytes
1.652 (1.139-2.391)
0.008
WBC
1.076 (1.044-1.112)
< 0.001
Neu
1.081 (1.036-1.128)
< 0.001
RBC
0.765 (0.579-1.012)
0.058
PLT
0.998 (0.996-1.015)
0.085
RDW
1.065 (0.994-1.141)
0.074
HGB
0.877 (0.796-0.967)
0.008
TBil
1.054 (1.015-1.096)
0.007
1.047 (1.008-1.088)
0.017
LDH
1.000 (1.000-1.001)
0.001
1.001 (1.000-1.001)
< 0.001
Cr
1.343 (1.106-1.632)
0.003
Lactate
1.129 (1.032-1.237)
0.009
ALB
0.385 (0.248-0.599)
< 0.001
0.436 (0.274-0.692)
0.003
PH
0.004 (0.003-0.103)
< 0.001
HCO3-
0.910 (0.873-0.948)
< 0.001
Mechanical ventilation
2.019 (1.323-3.082)
< 0.001
1.985 (1.269-3.105)
0.003
Vasopressin
3.687 (1.728-7.869)
< 0.001
Vasoactive
0.749 (0.374-1.517)
0.423
Prediction accuracy of the SIALF model
Patients in the MIMIC cohort were randomly assigned in a 7:3 ratio to a derivation cohort (n = 357) and an internal validation cohort (n = 153). Additionally, 228 patients from the FMCPH cohort served as the external validation cohort. The SIALF model was compared with SOFA and SIRS to evaluate its prediction efficiency. In the derivation cohort the SIALF model achieved the highest AUROC of 0.849, significantly outperforming the SOFA AUROC of 0.733 and the SIRS AUROC of 0.578 (P < 0.05; Figure 3A). Across the internal validation and external validation cohorts, the AUROC values of the SIALF model were 0.847 and 0.835, respectively, which were higher than other scores (all P < 0.05), suggesting a superior predictive accuracy for sepsis risk in patients with ALF (Figure 3B and C). The SIALF model demonstrated consistent predictive performance for sepsis risk in patients with ALF regardless of etiology, showing superior discrimination in both infectious (internal validation AUROC = 0.854 vs SOFA 0.751 and SIRS 0.643; external validation AUROC = 0.842) and non-infectious etiology groups (internal validation AUROC = 0.846 vs SOFA 0.741 and SIRS 0.658; external validation AUROC = 0.832; Supplementary Figure 1) with all comparisons being statistically significant (P < 0.05), indicating its robust clinical applicability across different ALF etiologies.
Figure 3 Receiver operating characteristic and calibration curves to assess the accuracy and calibration of sepsis in the acute liver failure model.
A: Receiver operating characteristic (ROC) comparison for sepsis risk in the internal derivation cohort; B: ROC comparison in the internal validation cohort; C: ROC comparison in the external validation cohort; D: The calibration curve of the internal derivation cohort; E: The calibration curve of the internal validation cohort; F: The calibration curve of the external validation cohort. SOFA: Sequential Organ Failure Assessment; SIRS: Systemic inflammatory response syndrome; SIALF: Sepsis in acute liver failure.
To assess the calibration we conducted the Hosmer-Lemeshow test and generated calibration curves. The actual and predicted probabilities of sepsis risk were found to be highly consistent across the deciles of the SIALF score (Hosmer-Lemeshow test P = 0.989; Brier score = 0.128; R2 = 0.411; Table 3). The SIALF model demonstrated excellent calibration performance with its calibration curves showing a close match between the predictions of the model and the actual occurrence of sepsis in patients with ALF (Figure 3D-F).
Table 3 Predictive value of various scoring models for sepsis risk in patients with acute liver failure in the internal derivation, validation cohort, and external validation cohort.
Predictive model
Youden index
Cutoff
SEN
SPE
PPV
NPV
Brier score
R2 value
P value in H-L test
Derivation set in MIMIC cohort
SIALF model
0.573
0.726
0.770
0.802
0.924
0.527
0.128
0.411
0.989
SOFA
0.386
6.500
0.781
0.605
0.861
0.468
0.157
0.183
0.620
SIRS
0.183
2.500
0.729
0.453
0.807
0.348
0.180
0.022
0.022
Validation set in MIMIC cohort (internal validation)
SIALF model
0.600
0.717
0.815
0.786
0.918
0.589
0.087
0.665
0.943
SOFA
0.384
7.500
0.694
0.690
0.869
0.433
0.160
0.202
0.752
SIRS
0.265
2.500
0.718
0.548
0.824
0.397
0.178
0.082
0.088
FMCPH cohort (external validation)
SIALF model
0.553
0.719
0.784
0.769
0.918
0.518
0.078
0.703
0.801
SOFA
0.183
7.500
0.778
0.405
0.805
0.366
0.158
0.182
0.421
SIRS
0.352
2.500
0.650
0.703
0.874
0.388
0.177
0.045
0.067
Clinical utility of the SIALF model
To evaluate the clinical utility of the SIALF model, we performed the DCA. In the internal derivation cohort, interventions guided by the SIALF model demonstrated greater net benefits than the SOFA and SIRS scores across the entire range of threshold probability (Figure 4A). Similarly, in both the internal and external validation cohort, the SIALF model provided more net benefit than others when the threshold probability was within the ranges of 0.35-0.95 and 0.55-0.96, respectively (Figure 4B and C), indicating that SIALF had superior clinical utility compared with the other two traditional predictive scores.
Figure 4 Comparison of the clinical utility of the sepsis in acute liver failure model for predicting sepsis risk in acute liver failure with other scoring systems using decision curve analysis.
A: The decision curve analysis (DCA) curve of the internal derivation cohort; B: The DCA curve of the internal validation cohort; C: The DCA curve of the external validation cohort. SOFA: Sequential Organ Failure Assessment; SIRS: Systemic inflammatory response syndrome; SIALF: Sepsis in acute liver failure.
Risk stratification based on the SIALF model
The results indicated that when patients with ALF were comorbid with sepsis, the 28-day/90-day mortality rates were significantly elevated compared with the non-sepsis group with 185 (47.8%)/207 (53.5%) and 35 (28.5%)/73 (40.7%) patients, respectively (Figure 5). Patients with ALF were stratified according to the optimal cutoff values (0.72), and survival analysis was conducted on the enrolled patients. Based on the SIALF model risk stratification score, the KM curves effectively discriminated the real high-risk subpopulation in both the 28-day mortality (Figure 6A) and 90-day mortality (Figure 6B).
Figure 5 Survival analysis of the sepsis and non-sepsis groups in acute liver failure.
A: Survival analysis for 28-day mortality; B: Survival analysis for 90-day mortality.
Figure 6 Survival analysis for risk stratification based on sepsis in acute liver failure score.
A: Survival analysis for 28-day mortality stratified by the sepsis in acute liver failure score; B: Survival analysis for 90-day mortality stratified by the sepsis in acute liver failure score. The cutoff point was 0.72. SIALF: Sepsis in acute liver failure.
Online dynamic nomogram
To facilitate the clinical application of SIALF model, we constructed an online dynamic nomogram model (a screenshot is shown in Figure 7). To use the SIALF nomogram, users only need to select “Yes” or “No” for the applicable options, enter the required laboratory test results, and click “Predict”. Subsequently, the risk of sepsis in patients with ALF can be easily obtained.
Figure 7 The dynamic online sepsis in acute liver failure nomogram for predicting sepsis risk in acute liver failure.
LDH: Lactate dehydrogenase; ALB: Albumin; MV: Mechanical ventilation; TBil: Total bilirubin; SIALF: Sepsis in acute liver failure.
DISCUSSION
The complication of sepsis in patients with ALF significantly escalates the risk of multiple organ failure and results in an exceedingly high short-term mortality rate. Therefore, it is crucial to identify high-risk sepsis subgroups among patients with ALF at an early stage. In this study we constructed a new predictive nomogram named the SIALF model. The robustness analysis demonstrated that the SIALF model exhibited superior discrimination, calibration, and clinical utility, outperforming traditional SOFA and SIRS scores in both internal and external validation cohorts. After risk stratification based on the SIALF model, the KM curves showed satisfactory discrimination not only in the internal cohort but also within the external validation cohort, effectively identifying the true high-risk subgroup. Notably, our model demonstrated consistent performance across various etiologies of ALF, indicating its broad clinical applicability. Moreover, to facilitate clinical practice we constructed an interactive online version, enabling physicians to assess patient conditions and track disease progression in real-time.
ALF is a rare and specific clinical syndrome marked by extensive hepatocyte necrosis and a rapid decline in liver function, resulting in a high short-term mortality rate. Immune dysfunction during liver failure increases the risk of secondary bacterial infections with up to 70% of patients at risk of developing sepsis[3]. When ALF complicates sepsis, damage-progenitor-associated molecular patterns are activated, further intensifying the proinflammatory response and increasing the risk of multiorgan failure[17]. The Sepsis-3 redefined sepsis by incorporating two decades of research in pathobiology, epidemiology, and management[10]. However, the new definition may still lack the precision required for accurate diagnosis, and the reliance on SOFA and quick SOFA scores may not fully reflect the complexity of sepsis. The validation of SOFA and quick SOFA in clinical guidelines was based on expected mortality rates and evaluated disease severity rather than the diagnostic accuracy for sepsis[18]. In addition ALF has distinct clinical features and may present with sepsis-like symptoms[19]. Therefore, we established an accurate predictive model to identify high-risk sepsis populations among patients with ALF at an early stage, enabling timely intervention and organ support, potentially improving patient outcomes. Our model demonstrated satisfactory accuracy with an AUROC value of 0.849 and outperformed the SOFA (AUROC: 0.733) and SIRS (AUROC: 0.578) scores. The DCA analysis and calibration curves also indicated superior clinical utility and efficiency over traditional scoring systems in both the internal and external validation cohorts.
Through univariate and multivariate regression analysis, we finally identified five independent risk factors for sepsis in patients with ALF, including age, TBil, LDH, ALB, and MV. These factors were subsequently included in the development of the SIALF model. Our study confirmed that the TBil level was an independent risk factor for sepsis in patients with ALF and is consistent with previous reports. In ALF endotoxins, inflammatory mediators, and inadequate tissue perfusion can impact bilirubin metabolism and excretion, causing hyperbilirubinemia and cholestasis[20]. Field et al[21] noted that patients in the ICU with TBil > 3 mg/dL had a three-fold increased risk of infection. Brienza et al[22] reported that patients in the ICU with TBil > 2 mg/dL were at higher risk of infections and sepsis. The TBil level offered a convenient and rapid method for early sepsis identification in patients with ALF as early and effective management was crucial. The TBil level was more widely used in ALF and other areas of liver injury, but this was the first time it had been used in patients with ALF to predict the risk of sepsis in our study.
In our analysis high LDH levels were strongly linked to an increased sepsis risk in patients with ALF. Yu et al[23] found a positive correlation between serum LDH levels and in-hospital mortality in patients in the ICU with sepsis, indicating that LDH is a significant risk assessment indicator. While the exact mechanism linking elevated LDH to sepsis is unclear, several factors may contribute[24]. LDH is extensively present in hepatocytes and is closely associated with energy metabolism, which is often used to reflect the extent of hepatic necrosis. When ALF is complicated by sepsis, energy metabolism disorders occur and are characterized by an increased rate of glycolysis and a significant decrease in the pyruvate-malate dependent oxygen consumption rate. In addition a strong inflammatory response occurs in the acute phase of sepsis[25]. During the process, cell integrity is compromised, leading to the release of substantial cellular contents into the bloodstream, including LDH. Therefore, persistently elevated LDH levels can serve as a predictive biomarker for the occurrence of sepsis in patients with ALF.
Compared with recent studies on patients with ALF, we validated several clinical indicators in the subpopulation of patients with sepsis. We identified age as a powerful predictor of sepsis occurrence in patients with ALF. As age increases, the immune system undergoes a process known as immunosenescence, resulting in a decline in immune function. This makes the elderly more susceptible to infections and complicates their post-infection recovery[26]. Furthermore, the elderly usually have a diminished organ reserve capacity, which makes them more susceptible to developing sepsis during ALF[27].
Similarly, hypoproteinemia was also correlated with multiple adverse complications and increased risk of adverse outcomes in ALF[28]. Cao et al[29] indicated that when the ALB level was ≤ 2.6 g/dL, each 1 g/dL rise in ALB corresponded to a 59% and 62% decrease in the risk of mortality at 28 days and 1 year, respectively. A prospective cohort study demonstrated that the ALB level was a crucial predictor of 28-day mortality in sepsis, with 29.2 g/L identified as the optimal threshold, showing both high sensitivity and specificity[30]. However, the connection between low ALB levels and the risk of sepsis in patients with ALF remains unexplored. Some researchers queried whether the standard cutoff value was also adapted for the prediction in patients with sepsis[31]. Our study uncovered a significant association between reduced ALB levels and an increased risk of sepsis in patients with ALF. Several potential mechanisms could account for this association[32]. Firstly, lower plasma ALB levels signal severe liver dysfunction, characterized by a decline in metabolic and immune functions, which impairs the ability of the body to resist infections and eliminate pathogens, thus raising the risk of sepsis[33]. Secondly, ALF induces SIRS, which impairs vascular endothelium and increases capillary vessels. This leads to ALB leakage from the blood vessels, causing a decrease in plasma ALB levels and a range of pathological changes, such as tissue and organ edema, hypovolemia, and potentially shock and MODS[34,35].
Consistent with the study by Freundlich et al[36], we observed that MV was positively associated with infection in patients with sepsis. MV, while providing airway protection and respiratory support, also disrupts the normal defense mechanisms of the upper respiratory tract, thereby increasing the risk of ventilator-associated pneumonia in patients with ALF. Moreover, a recent study also highlighted that the requirement for MV was associated with increased mortality and morbidity in patients with sepsis and septic shock[37].
Based on identified prognostic factors, we developed a new individualized nomogram, the SIALF model to assess the condition of patients with ALF with sepsis at admission. The optimal C-index and AUROC values demonstrated that the model has superior discrimination and accuracy in identifying patients at high-risk and outperformed the SOFA and SIRS scores both in the internal and external validation cohorts. The most fundamental and routinely assessed clinical data was integrated and enhanced the generalizability of the model in the ICU. Importantly, patients with ALF were stratified according to the SIALF score, and the KM curves demonstrated effective discrimination in identifying the real high-risk subpopulation. Moreover, to facilitate clinical use we developed an online SIALF nomogram. This online nomogram model, combined with the risk grades, will be very useful in assisting clinicians in the ICU to evaluate the risk of sepsis in patients with ALF. After risk stratification patients in low and moderate risk groups can continue intensive therapy to prevent ALF progression[38-40], while those in the high-risk sepsis group require early and rational antibiotic administration, prompt recognition, fluid resuscitation, immunomodulation, and organ support therapy[41,42]. Furthermore, the SIALF model was further validated in an external validation cohort that included 228 patients with ALF from a critical care database in China. Even though the training cohort data originated from Western nations, we illustrated the applicability of the model across diverse populations.
Nevertheless, our study had some limitations: Firstly, although this is a large-scale cohort study with over 1453 patients involved, the retrospective and observational design may inherently introduce selection bias. Secondly, our study only included parameters from the first day of ICU admission. It might be better to analyze the indicators during the ICU stay in with dynamic and continuous observation. Thirdly, according to the definition of sepsis, incorporating the specific pathogen culture results might further enhance the predictive strength of the model. Despite these limitations our models still achieved satisfactory predictive performance and demonstrated robustness and generalizability in identifying high-risk sepsis subgroups in ALF within the external validation cohort.
CONCLUSION
We developed and validated a new nomogram model (SIALF model) for predicting the risk of sepsis in patients with ALF. The robustness analysis demonstrated the superior accuracy, discrimination, calibration, and clinical utility of the SIALF model, outperforming traditional SOFA and SIRS scores in both internal and external validation cohorts. The risk stratification based on the SIALF model effectively discriminated the real high-risk subpopulation. Furthermore, we constructed an online dynamic version of this model, enabling physicians to track disease progression in real-time, that might further facilitate the early detection and rational management for this subpopulation.
ACKNOWLEDGEMENTS
We express our gratitude to the contributions of the Medical Information Mart for Intensive Care (MIMIC) Program registries for creating and updating the databases. Additionally, we extend our thanks to the colleagues within the FMCPH cohort for their substantial support throughout the research.
Footnotes
Provenance and peer review: Unsolicited article; Externally peer reviewed.
Peer-review model: Single blind
Corresponding Author’s Membership in Professional Societies: Beijing Society of Critical Care Medicine, Chinese Medical Association.
Specialty type: Gastroenterology and hepatology
Country of origin: China
Peer-review report’s classification
Scientific Quality: Grade B, Grade C
Novelty: Grade B, Grade B
Creativity or Innovation: Grade B, Grade B
Scientific Significance: Grade B, Grade C
P-Reviewer: Ning ZX; Zhang L S-Editor: Li L L-Editor: Filipodia P-Editor: Zhao S
Yang NB, Ni SL, Li SS, Zhang SN, Hu DP, Lu MQ. Endotoxin tolerance alleviates experimental acute liver failure via inhibition of high mobility group box 1.Int J Clin Exp Pathol. 2015;8:9062-9071.
[PubMed] [DOI]
Wendon J, Cordoba J, Dhawan A, Larsen FS, Manns M, Samuel D, Simpson KJ, Yaron I, Bernardi M; European Association for the Study of the Liver; Clinical practice guidelines panel; EASL Governing Board representative. EASL Clinical Practical Guidelines on the management of acute (fulminant) liver failure.J Hepatol. 2017;66:1047-1081.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 417][Cited by in RCA: 621][Article Influence: 77.6][Reference Citation Analysis (1)]
Kumar A, Roberts D, Wood KE, Light B, Parrillo JE, Sharma S, Suppes R, Feinstein D, Zanotti S, Taiberg L, Gurka D, Kumar A, Cheang M. Duration of hypotension before initiation of effective antimicrobial therapy is the critical determinant of survival in human septic shock.Crit Care Med. 2006;34:1589-1596.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 3798][Cited by in RCA: 3968][Article Influence: 208.8][Reference Citation Analysis (0)]
Tamma PD, Heil EL, Justo JA, Mathers AJ, Satlin MJ, Bonomo RA. Infectious Diseases Society of America 2024 Guidance on the Treatment of Antimicrobial-Resistant Gram-Negative Infections.Clin Infect Dis. 2024;ciae403.
[PubMed] [DOI] [Full Text]
Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, Bellomo R, Bernard GR, Chiche JD, Coopersmith CM, Hotchkiss RS, Levy MM, Marshall JC, Martin GS, Opal SM, Rubenfeld GD, van der Poll T, Vincent JL, Angus DC. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3).JAMA. 2016;315:801-810.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 15803][Cited by in RCA: 17179][Article Influence: 1908.8][Reference Citation Analysis (2)]
Pérez-Torres D, Merino-García PA, Canas-Pérez I, Cuenca-Rubio C, Posadas-Pita GJ, Colmenero-Calleja C, Ticona-Espinoza TG, Díaz-Rodríguez C. Real-world inter-observer variability of the Sequential Organ Failure Assessment (SOFA) score in intensive care medicine: the time has come for an update.Crit Care. 2023;27:160.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 3][Reference Citation Analysis (0)]
Yu H, Yang Q, Qian Y, Luo S, Kong T, Yang G, Chen W, Xie F, Chen J, Xiong X, Wen D, Zhang Z. A positive correlation between serum lactate dehydrogenase level and in-hospital mortality in ICU sepsis patients: evidence from two large databases.Eur J Med Res. 2024;29:525.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 3][Reference Citation Analysis (0)]
Kawazoe Y, Miyamoto K, Morimoto T, Yamamoto T, Fuke A, Hashimoto A, Koami H, Beppu S, Katayama Y, Itoh M, Ohta Y, Yamamura H; Dexmedetomidine for Sepsis in Intensive Care Unit Randomized Evaluation (DESIRE) Trial Investigators. Effect of Dexmedetomidine on Mortality and Ventilator-Free Days in Patients Requiring Mechanical Ventilation With Sepsis: A Randomized Clinical Trial.JAMA. 2017;317:1321-1328.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 134][Cited by in RCA: 167][Article Influence: 20.9][Reference Citation Analysis (0)]
Evidence review for managing and treating suspected sepsis in acute hospital settings; antibiotic treatment in people with suspected sepsis: Suspected sepsis: recognition, diagnosis and early management: Evidence review B. London: National Institute for Health and Care Excellence (NICE); 2024 Jan-.
[PubMed] [DOI]
Sarin SK, Kedarisetty CK, Abbas Z, Amarapurkar D, Bihari C, Chan AC, Chawla YK, Dokmeci AK, Garg H, Ghazinyan H, Hamid S, Kim DJ, Komolmit P, Lata S, Lee GH, Lesmana LA, Mahtab M, Maiwall R, Moreau R, Ning Q, Pamecha V, Payawal DA, Rastogi A, Rahman S, Rela M, Saraya A, Samuel D, Saraswat V, Shah S, Shiha G, Sharma BC, Sharma MK, Sharma K, Butt AS, Tan SS, Vashishtha C, Wani ZA, Yuen MF, Yokosuka O; APASL ACLF Working Party. Acute-on-chronic liver failure: consensus recommendations of the Asian Pacific Association for the Study of the Liver (APASL) 2014.Hepatol Int. 2014;8:453-471.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 541][Cited by in RCA: 499][Article Influence: 45.4][Reference Citation Analysis (0)]
