CLIF-AKI score predicts KDIGO stage 3 acute kidney injury in intensive care unit patients with acute-on-chronic liver failure

Abstract

BACKGROUND

Reliable tools for the early detection of stage 3 acute kidney injury (AKI) in patients with acute-on-chronic liver failure (ACLF) are limited. We developed a scoring system to predict stage 3 AKI within 14 days of intensive care unit (ICU) admission.

AIM

To externally validate the predictive score for stage 3 AKI in patients with ACLF within 14 days of ICU admission.

METHODS

Patients with cirrhosis admitted to the ICU who met the diagnostic criteria for ACLF were included. Two independent cohorts were established for the model derivation (1314 patients from the Medical Information Mart for Intensive Care IV database) and the external validation (223 patients from Zhenjiang). The primary outcome was stage 3 AKI occurring within 14 days of ICU admission. Predictors were selected using least absolute shrinkage and selection operator regression and were subsequently entered into a multivariable logistic regression model.

RESULTS

Stage 3 AKI occurred in 34% of the derivation cohort and 30% of the validation cohort. The chronic liver failure organ failure (CLIF)-AKI score was developed using eight variables: Mechanical ventilation, vasopressor use, serum albumin, mean arterial pressure, white blood cell count, blood urea nitrogen, model for end-stage liver disease (MELD) score, and body mass index. The score demonstrated good discrimination [area under the receiver operating characteristic curve: 0.79 (95% confidence interval: 0.76-0.82) in the derivation cohort and 0.86 (95% confidence interval: 0.81-0.91) in the validation cohort], and showed acceptable calibration in both. The score stratified patients into four risk categories. In the validation cohort, CLIF-AKI score outperformed PIRO, chronic liver failure consortium ACLF, Child-Turcotte-Pugh, MELD, and MELD-sodium in predicting stage 3 AKI (all P < 0.05).

CONCLUSION

The CLIF-AKI score enables the early identification of ICU patients with ACLF who are at high risk of developing Kidney Disease: Improving Global Outcomes stage 3 AKI within 14 days of admission.

Key Words: Acute kidney injury; Acute-on-chronic liver failure; Predictive score; Risk stratification; Intensive care unit

Core Tip: Patients with acute-on-chronic liver failure admitted to the intensive care unit are at high risk of developing stage 3 acute kidney injury (AKI); however, reliable tools for early prediction remain limited. We developed and externally validated the chronic liver failure organ failure-AKI score to predict stage 3 AKI in patients with acute-on-chronic liver failure. The score demonstrated good discrimination and clinical utility, and outperformed the PIRO model and existing prognostic scores in the external validation cohort. We also developed a point-based scoring system and an online calculator to facilitate its implementation in clinical practice.

INTRODUCTION

Acute-on-chronic liver failure (ACLF) is a clinical syndrome in patients with cirrhosis that is characterized by rapid deterioration of liver function. It is frequently accompanied by multiorgan failure and is associated with high short-term mortality (approximately 33%-50%)[1,2], often requiring intensive care unit (ICU) admission. Among the extrahepatic organs affected in ACLF, the kidneys are the most vulnerable, with approximately 22.8%-34% of patients developing acute kidney injury (AKI)[3]. AKI is not only a common complication but also a key driver of disease progression and poor outcomes, substantially increasing short-term mortality[4]. Moreover, renal failure appears to have a more detrimental impact on prognosis than other forms of single-organ failure in ACLF[5]. For example, the 28-day mortality rate is approximately 6.3% in patients without renal failure but increases to 18.6% in those with isolated renal failure[6]. Collectively, these findings indicate that renal impairment plays a central role in the progression and prognosis of ACLF.

Notably, stage 3 AKI is common in patients with ACLF. In a large European multicenter study of 510 patients with ACLF, 98 (19.2%) developed AKI, of whom 34.7% had stage 1 AKI, 4.1% had stage 2 AKI, and 61.2% had stage 3 AKI[7]. Consistent with these findings, an Indian study of 1022 patients with ACLF demonstrated that among the 649 patients without AKI at baseline, 247 (38%) developed new-onset AKI during 1 month of follow-up, the majority of whom progressed to stage 3 AKI (206/247, 83.4%).

Previous studies suggest that mortality risk in ACLF increases progressively with advancing AKI stage, with stage 3 AKI conferring the poorest prognosis[8]. To further evaluate this association, we analyzed data from 1623 patients with ACLF in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and assessed 7-, 14-, and 28-day mortality. Mortality increased across AKI stages at all time points, with stage 3 AKI consistently associated with the highest mortality, consistent with previous reports (Supplementary Figure 1). Despite its clinical significance, reliable models for predicting progression to stage 3 AKI in patients with ACLF remain limited. The PIRO model, recommended by the Asian Pacific Association for the Study of the Liver ACLF guidelines, has demonstrated good performance in predicting AKI in ACLF[9]; however, this recommendation is weak (level of evidence 4), largely owing to limited external validation. Moreover, the ability of the PIRO model to predict stage 3 AKI remains unclear[3]. Therefore, developing an early risk prediction model for stage 3 AKI is clinically important, as it may facilitate risk stratification, guide treatment decisions, and potentially improve patient outcomes. In this study, we developed and externally validated a novel risk score to predict progression to stage 3 AKI within 14 days in patients with ACLF. We also assessed the performance of the PIRO model for identifying patients at risk of progression to stage 3 AKI and predicting its occurrence.

MATERIALS AND METHODS

Study design and participants

The derivation cohort was obtained from the MIMIC-IV database (v3.0), which contains detailed clinical data from patients admitted to the ICU at Beth Israel Deaconess Medical Center (Boston, MA, United States) between 2008 and 2022[10]. Eligible patients were retrospectively identified and analyzed in July 2025. Use of the MIMIC-IV database was approved by the Institutional Review Board of Beth Israel Deaconess Medical Center, which waived the requirement for informed consent in accordance with applicable regulations. Authorized access to the database was granted to one author of this study (ID No. 66161870). The use of the MIMIC-IV public database for this study was reviewed by the Ethics Committee of Zhenjiang Third People’s Hospital and was exempted from additional ethical approval.

The external validation cohort was derived from Zhenjiang Third People’s Hospital and included patients diagnosed with ACLF between January 2018 and May 2025. The inclusion and exclusion criteria were identical to those applied in the derivation cohort. Eligible patients were retrospectively identified and analyzed in July 2025. The study protocol was approved by the Ethics Committee of Zhenjiang Third People’s Hospital and was registered in the Chinese Clinical Trial Registry (No. ChiCTR2500103594).

Patients with cirrhosis were identified using ICD-9 and ICD-10 diagnosis codes recorded in the hospital diagnosis records of the MIMIC-IV database. Eligible participants were adults with ACLF admitted to the ICU for acute decompensation, which was defined by the presence of ascites, bacterial infection, acute variceal hemorrhage, or hepatic encephalopathy. Patients were excluded if they were younger than 18 years, pregnant, had malignancy or HIV infection, died within 24 hours of ICU admission, had an ICU stay of less than 24 hours, were not first-time ICU admissions, had pre-existing chronic kidney disease, or had Kidney Disease: Improving Global Outcomes (KDIGO) stage 3 AKI at ICU admission (Supplementary Figure 2). This study was conducted and reported in accordance with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis statement.

Study variables and definitions

Demographic and clinical variables included age, sex, body mass index (BMI), mean arterial pressure (MAP), and total urine output during the first 24 hours after ICU admission. Laboratory data included liver and kidney function tests, coagulation parameters, arterial blood gases, complete blood counts, electrolytes, bicarbonate, anion gap, glucose, and other relevant measurements. Treatment-related variables included the use of antibiotics, human albumin, furosemide, and spironolactone, while organ support measures included mechanical ventilation and vasoactive agents (norepinephrine, epinephrine, and dopamine). Comorbidities included fatty liver disease, hypertension, diabetes mellitus, and congestive heart failure. Disease severity was assessed using the sequential organ failure assessment (SOFA), model for end-stage liver disease (MELD), and MELD-sodium (MELD-Na) scores. For this study, ICU admission was defined as the index time point for prediction. Baseline laboratory variables were defined as the first available measurements obtained within 24 hours of ICU admission.

ACLF was diagnosed according to the European Association for the Study of the Liver-Chronic Liver Failure Consortium criteria and graded using the chronic liver failure organ failure (CLIF) score. Organ failures were defined as follows: Liver failure, total bilirubin ≥ 12 mg/dL; coagulation failure, international normalized ratio ≥ 2.5; renal failure, serum creatinine (SCr) level > 2 mg/dL or need for renal replacement therapy (RRT); circulatory failure, need for vasopressor therapy to maintain blood pressure; cerebral failure, grade III-IV hepatic encephalopathy according to the West Haven criteria; and respiratory failure, a PaO₂/FiO₂ ratio ≤ 200, an SpO₂/FiO₂ ratio ≤ 214, or the need for mechanical ventilation for reasons other than airway protection in patients with cerebral failure.

ACLF grade 1 was defined by any of the following: Isolated kidney failure; isolated liver, coagulation, circulatory, or respiratory failure accompanied by a SCr level of 1.5-1.9 mg/dL and/or mild-to-moderate hepatic encephalopathy; or isolated cerebral failure accompanied by a SCr level of 1.5-1.9 mg/dL. ACLF grade 2 was defined as the presence of two organ failures, whereas ACLF grade 3 was defined as the presence of three or more organ failures.

Stage 3 AKI was defined according to the KDIGO criteria. A diagnosis of stage 3 AKI was made if any of the following criteria were met: (1) SCr ≥ 3 times the baseline value; (2) SCr ≥ 4.0 mg/dL (353.6 μmol/L); (3) Initiation of RRT; (4) Urine output < 0.3 mL/kg/hour for ≥ 24 hours; or (5) Anuria for ≥ 12 hours. Baseline SCr was defined as the lowest value recorded within 7 days before admission; if unavailable, the first SCr measurement obtained on admission was used. Stage 3 AKI at ICU admission was defined by any of the following: (1) SCr ≥ 3 times the baseline value; (2) SCr ≥ 4.0 mg/dL (353.6 μmol/L); or (3) Receipt of RRT before ICU admission. Patients meeting any of these criteria at ICU admission were classified as having stage 3 AKI and excluded from the study. As complete 24-hour urine output data before ICU admission were not reliably available, patients who may have already met the urine output criterion for KDIGO stage 3 AKI at ICU admission could not be identified and were therefore not excluded. The MELD-Na score was calculated as MELD + 1.59 × (135 - Na), with serum sodium constrained to 120-135 mEq/L. The neutrophil-to-lymphocyte ratio was calculated as the ratio of neutrophil percentage to lymphocyte percentage.

Clinical outcomes

The primary outcome was the development of KDIGO stage 3 AKI within 14 days of the first ICU admission. For sensitivity analyses, a composite outcome was defined as the occurrence of either KDIGO stage 3 AKI or in-hospital all-cause mortality within 14 days of the first ICU admission.

Model development and statistical analysis

Continuous variables are presented as mean ± SD or median (interquartile range), as appropriate, and categorical variables are presented as n (%). Between-group comparisons were performed using Student’s t test or the Mann-Whitney U test for continuous variables and the χ² test or Fisher’s exact test for categorical variables, as appropriate. Missing data were handled using multiple imputation by chained equations implemented in the mice package in R, with random forests specified as the imputation method. Five imputed datasets were generated, each with five iterations, and the random seed of 1234 was used. The pattern of missing data is shown in Supplementary Figure 3.

To identify predictive variables and develop the risk model, we applied least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation[11]. Predictors were selected using the lambda.1se criterion and entered into a parsimonious multivariable logistic regression model. The lambda.1se criterion was selected primarily to obtain a more concise and stable set of predictors, reduce the potential risk of overfitting, and improve the clinical interpretability and practicality of the derived scoring system. The CLIF-AKI scoring system was developed based on the predictors retained in the final model. To generate clinically applicable point scores, predictors requiring stratification were categorized, after which a logistic regression model was refitted using the categorized variables. The raw point value for each predictor category was calculated by multiplying the corresponding regression coefficient by a predefined scaling factor. To facilitate clinical application, raw point values were rounded to the nearest 5-point increment. The reference category was assigned 0 points, negative point values were set to 0, and monotonicity was maintained for ordered categorical predictors. The total CLIF-AKI score was calculated by summing the points assigned to each predictor. Based on the distribution of total scores and predicted risks, patients were classified into four risk groups: Low, moderate, high, and very high risk. The cutoff values were set at 80, 110, and 140 points, corresponding approximately to the 25^th percentile, median, and 75^th percentile of the total score distribution in the derivation cohort, respectively. These thresholds were associated with predicted risks of approximately 15%, 30%, and 50%. The final cutoff values were selected after considering the sample size distribution across risk groups and their ease of clinical application.

In the derivation and validation cohorts, predicted probabilities were compared with observed incidence across risk groups. Model discrimination was assessed using receiver operating characteristic curves and the area under the curve (AUC). Calibration was evaluated using calibration plots, and overall model fit was assessed using the Hosmer-Lemeshow test. Clinical utility across threshold probabilities was examined using decision curve analysis (DCA), with net benefit quantified at each threshold probability[12]. To assess the internal stability and the potential risk of overfitting of the CLIF-AKI model, bootstrap resampling was performed for internal validation. Specifically, 1000 bootstrap samples were drawn from the derivation cohort. For each bootstrap sample, the model was refitted, and performance metrics, including the AUC, calibration intercept, calibration slope, and Brier score, were calculated. Statistical analyses were performed using R software (version 4.4.1; R Foundation for Statistical Computing, Vienna, Austria). LASSO-based feature selection was performed using the glmnet package. All statistical tests were two-sided, with P < 0.05 considered statistically significant.

RESULTS

Clinical characteristics and univariate analysis

A derivation cohort comprising 1314 patients with ACLF was identified from the MIMIC-IV database, and an external validation cohort of 223 patients with ACLF was assembled from Zhenjiang (Supplementary Figure 2). Baseline characteristics of both cohorts are summarized in Table 1 and include demographic characteristics, clinical variables, disease severity scores, laboratory variables, treatments, and comorbidities. The two cohorts showed distinct baseline profiles, allowing evaluation of model performance across clinically heterogeneous populations. Most baseline variables differed significantly between the cohorts; only age, urine output, albumin, chloride, sodium, MELD-Na, and coagulation failure were comparable (P ≥ 0.05), whereas the remaining variables showed significant between-cohort differences (all P < 0.05). Compared with the external validation cohort, patients in the derivation cohort had a higher BMI [28 (24-33) kg/m² vs 23 (21-27) kg/m²] and a lower MAP [60 (53-67) mmHg vs 87 (80-93) mmHg]. The proportions of patients receiving antibiotics, mechanical ventilation, and vasopressors were also higher in the derivation cohort (78.9% vs 46.2%, 26.7% vs 6.7%, and 34.7% vs 5.8%, respectively). In addition, patients in the MIMIC-IV cohort had higher white blood cell (WBC) counts, blood urea nitrogen (BUN) and creatinine levels, and SOFA scores than those in the Zhenjiang cohort [12 (8-17) × 10⁹/L vs 6 (4-9) × 10⁹/L; 26 (16-39) mg/dL vs 15 (10-24) mg/dL; 1.10 (0.80-1.70) mg/dL vs 0.78 (0.64-1.02) mg/dL; and 8 (6-11) vs 6 (5-8), respectively]. Conversely, prothrombin time, total bilirubin, and MELD scores were higher in the Zhenjiang cohort [22 (19-26) seconds vs 20 (16-26) seconds; 12 (7-20) mg/dL vs 4 (2-9) mg/dL; and 25 (22-32) vs 21 (14-28), respectively].

Table 1 Baseline characteristics and study outcomes of the derivation and validation cohorts, n (%)/median (interquartile range).

Variable	Derivation cohort, n = 1314	Validation cohort, n = 223	P value
Sex			0.153
Female	526 (40.0)	78 (35.0)
Male	788 (60.0)	145 (65.0)
Age, years	57 (50-65)	56 (49-69)	0.983
BMI, kg/m2	28 (24-33)	23 (21-27)	< 0.001a
MAP, mmHg	60 (53-67)	87 (80-93)	< 0.001a
Urine output, mL	1225 (750-1900)	1250 (950-1515)	0.421
Comorbidities
Congestive heart failure	177 (13.5)	11 (4.9)	< 0.001a
Diabetes	307 (23.4)	34 (15.2)	0.007a
Hypertension	235 (17.9)	59 (26.5)	0.004a
Fatty liver disease	89 (6.8)	34 (15.2)	< 0.001a
Organ support
Mechanical ventilation	351 (26.7)	15 (6.7)	< 0.001a
Vasopressor use	456 (34.7)	13 (5.8)	< 0.001a
Medications
Antibiotic	1037 (78.9)	103 (46.2)	< 0.001a
Furosemide	166 (12.6)	107 (48.0)	< 0.001a
Spironolactone	122 (9.3)	107 (48.0)	< 0.001a
Human albumin	309 (23.5)	74 (33.2)	0.003a
Organ failure
Liver	242 (18.4)	108 (48.4)	< 0.001a
Kidney	401 (30.5)	24 (10.8)	< 0.001a
Coagulation	312 (23.7)	53 (23.8)	0.994
Brain	326 (24.8)	39 (17.5)	0.018a
Circulation	456 (34.7)	13 (5.8)	< 0.001a
Lung	591 (45.0)	25 (11.2)	< 0.001a
ACLF grade			< 0.001a
Grade 1	601 (45.7)	192 (86.1)
Grade 2	425 (32.3)	19 (8.5)
Grade 3	288 (21.9)	12 (5.4)
Outcomes
In-hospital mortality	305 (23.2)	36 (16.1)
KDIGO stage 3 AKI within 14 days	452 (34.4)	68 (30.5)
Laboratory tests
Hemoglobin, g/dL	8.50 (7.20-10.10)	10.70 (8.40-12.80)	< 0.001a
Platelet, × 109/L	82 (52-133)	71 (50-105)	0.001a
WBC, × 109/L	12 (8-17)	6 (4-9)	< 0.001a
Albumin, g/dL	2.80 (2.40-3.20)	2.82 (2.42-3.16)	0.545
Anion gap, mmol/L	16.0 (13.0-19.0)	11.3 (9.6-13.8)	< 0.001a
BUN, mg/dL	26 (16-39)	15 (10-24)	< 0.001a
Calcium, mmol/L	1.98 (1.83-2.10)	2.07 (1.96-2.16)	< 0.001a
Chloride, mmol/L	101 (96-106)	102 (97-105)	0.874
Creatinine, mg/dL	1.10 (0.80-1.70)	0.78 (0.64-1.02)	< 0.001a
Glucose, mmol/L	8.0 (6.5-10.7)	5.5 (4.8-7.0)	< 0.001a
Sodium, mmol/L	136 (131-139)	136 (133-139)	0.552
Potassium, mmol/L	4.40 (4.00-4.90)	3.80 (3.38-4.21)	< 0.001a
INR	1.90 (1.50-2.40)	1.97 (1.61-2.44)	0.014a
PT, seconds	20 (16-26)	22 (19-26)	< 0.001a
aPTT, seconds	40 (33-54)	44 (39-52)	< 0.001a
ALT, U/L	38 (23-71)	63 (29-361)	< 0.001a
ALP, U/L	112 (76-156)	143 (108-195)	< 0.001a
AST, U/L	80 (46-165)	108 (57-322)	< 0.001a
TBil, mg/dL	4 (2-9)	12 (7-20)	< 0.001a
Lymphocytes, %	8 (5-14)	14 (8-22)	< 0.001a
Monocytes, %	5.0 (3.1-7.3)	8.3 (6.4-10.1)	< 0.001a
Neutrophils, %	82 (75-88)	75 (67-83)	< 0.001a
NLR	4 (3-7)	5 (3-10)	0.016a
RBC, × 1012/L	2.55 (2.08-3.08)	3.15 (2.43-3.88)	< 0.001a
RDW, %	18.1 (15.9-20.8)	15.8 (14.1-18.9)	< 0.001a
Severity score
MELD	21 (14-28)	25 (22-32)	< 0.001a
SOFA	8 (6-11)	6 (5-8)	< 0.001a
MELD-Na	27 (17-40)	27 (23-33)	0.573

^aP < 0.05.

BMI: Body mass index; MAP: Mean arterial pressure; ACLF: Acute-on-chronic liver failure; AKI: Acute kidney injury; WBC: White blood cell; BUN: Blood urea nitrogen; INR: International normalized ratio; PT: Prothrombin time; aPTT: Activated partial thromboplastin time; ALT: Alanine aminotransferase; ALP: Alkaline phosphatase; AST: Aspartate aminotransferase; TBil: Total bilirubin; NLR: Neutrophil-to-lymphocyte ratio; RBC: Red blood cell; MELD: Model for end-stage liver disease; SOFA: Sequential organ failure assessment; MELD-Na: Model for end-stage liver disease-sodium; RDW: Red cell distribution width.

In the derivation cohort, stage 3 AKI occurred in 34% of patients (n = 452), and the in-hospital mortality rate was 23% (n = 305). In the external validation cohort, the incidence of stage 3 AKI was 30% (n = 68), and the in-hospital mortality rate was 16% (n = 36) (Table 1). Using clinical and laboratory variables collected at enrollment, we performed univariable analyses comparing patients who developed stage 3 AKI within 14 days with those who did not (Supplementary Table 1). Albumin, urine output, WBC count, BUN, creatinine, SOFA score, and MELD score differed significantly between the groups (all P < 0.05).

Feature determination and risk score construction

A total of 31 candidate variables were entered into a LASSO regression model to identify predictors of the study endpoint (Figure 1). Eight independent predictors were selected and incorporated into the final risk model: Mechanical ventilation, vasopressor use, albumin, MAP, WBC count, BUN, MELD score, and BMI. The coefficients of the multivariable logistic regression model are provided in Table 2, and the corresponding standardized effect sizes are shown in Supplementary Figure 4. No significant multicollinearity was detected among the variables included in the final model, with all variance inflation factors below 5 in both the derivation cohort (range: 1.01-1.34) and the validation cohort (range: 1.02-2.45).

Figure 1 Predictor selection using the least absolute shrinkage and selection operator regression method. A: Coefficient profile plot showing the trajectories of predictor coefficients across the log (λ) sequence. The dotted vertical lines indicate the optimal λ values selected according to the minimum criterion and the one-standard-error (1-SE) criterion. Predictors included in the final model were selected based on the 1-SE criterion; B: Ten-fold cross-validation plot used to determine the optimal tuning parameter (λ) in the least absolute shrinkage and selection operator model. The dotted vertical lines represent the λ values corresponding to the minimum criterion and the 1-SE criterion.

Table 2 Final multivariable logistic regression model used to develop the chronic liver failure organ failure-acute kidney injury score.

Intercept and variable	β coefficient	OR (95%CI)	P value
WBC, × 109/L	0.018	1.02 (1.00-1.04)	0.044
MAP, mmHg	-0.020	0.98 (0.97-0.99)	< 0.001
BUN, mg/dL	0.011	1.01 (1.00-1.02)	0.005
MELD score	0.057	1.06 (1.04-1.08)	< 0.001
BMI, kg/m2	0.054	1.05 (1.04-1.07)	< 0.001
Albumin, g/dL	-0.426	0.65 (0.52-0.81)	< 0.001
Mechanical ventilation	0.977	2.66 (1.99-3.54)	< 0.001
Vasopressor use	0.797	2.22 (1.68-2.93)	< 0.001
Intercept	-2.409

The predicted probability of acute kidney injury stage 3 was calculated as: P = 1/{1 + exp[-(-2.409 + 0.797 × “vasopressor use = 1” + 0.977 × “mechanical ventilation = 1” - 0.020 × mean arterial pressure + 0.018 × white blood cell + 0.011 × blood urea nitrogen - 0.426 × albumin + 0.054 × body mass index + 0.057 × model for end-stage liver disease)]}. Vasopressor use and mechanical ventilation were coded as binary variables (yes = 1, no = 0), whereas the remaining six predictors were treated as continuous variables. OR: Odds ratio; CI: Confidence interval; WBC: White blood cell; MAP: Mean arterial pressure; BUN: Blood urea nitrogen; MELD: Model for end-stage liver disease; BMI: Body mass index.

The model was translated into a clinical risk scoring system (CLIF-AKI score; Table 3), with item scores derived from the multivariable logistic regression coefficients. The incidence of stage 3 AKI increased progressively with higher CLIF-AKI scores. In the derivation cohort, patients were stratified into four risk categories based on their total scores: Low (≤ 80 points; n = 377), moderate (81-110 points; n = 326), high (111-140 points; n = 323), and very high (> 140 points; n = 288; Figure 2).

Figure 2 Distribution of chronic liver failure organ failure-acute kidney injury risk scores and predicted risk of the primary outcome in the derivation cohort. Bar heights represent the number of patients at each total risk-score value (left y-axis), whereas the solid line denotes the predicted probability of the primary outcome (right y-axis).

Table 3 Chronic liver failure organ failure-acute kidney injury risk score and risk stratification.

Variables	Points	Variables	Points
WBC, × 109/L		BUN, mg/dL
≤ 5	0	≤ 15	0
5-10	15	16-25	5
11-15	25	26-40	10
> 15	30	> 40	15
Albumin, g/dL		BMI, kg/m2
≥ 3.5	0	≤ 23.9	0
3.0-3.4	10	24.0-27.9	10
2.5-2.9	15	28.0-34.9	20
< 2.5	20	≥ 35.0	30
MAP in mmHg		Mechanical ventilation
> 65	0	Yes	35
56-65	10	No	0
≤ 55	20
MELD score		Vasopressor use
≤ 15	0	Yes	30
16-25	15	No	0
26-35	35
> 35	60
Risk stratification, CLIF-AKI score	Risk category
> 140	Very high risk
111-140	High risk
81-110	Moderate risk
≤ 80	Low risk

WBC: White blood cell; MAP: Mean arterial pressure; MELD: Model for end-stage liver disease; CLIF-AKI: Chronic liver failure organ failure-acute kidney injury; BMI: Body mass index; BUN: Blood urea nitrogen.

To evaluate the impact of variable categorization and score simplification on model performance, we compared the discriminative performance of three models: A model based on the original continuous variables, a model based on categorized variables, and the final rounded point-score model. The AUCs of the three models were 0.789 [95% confidence interval (CI): 0.763-0.815], 0.787 (95%CI: 0.761-0.813), and 0.784 (95%CI: 0.758-0.810), respectively. These results indicate that variable categorization and point-score rounding had minimal impact on model discrimination.

In clinical practice, prespecified CLIF-AKI score thresholds may be used for risk stratification to identify patients at high risk of progression to stage 3 AKI (Table 4). For example, a CLIF-AKI score > 140 classified 21.9% of patients with ACLF as being at very high risk for progression to stage 3 AKI within 14 days. At this cutoff, the positive predictive value was 71% and the negative predictive value was 76%. To facilitate clinical implementation, we constructed a nomogram (Figure 3) and developed an online risk calculator (https://zpfwnzyt.shinyapps.io/CLIFAKI/). This tool enables clinicians to generate individualized predictions of the 14-day risk of stage 3 AKI by entering the relevant patient variables.

Figure 3 Nomogram of the chronic liver failure organ failure-acute kidney injury score for predicting the 14-day risk of KDIGO stage 3 acute kidney injury in patients with acute-on-chronic liver failure. Points assigned to each predictor are summed to generate a total score, which corresponds to the predicted probability of stage 3 acute kidney injury within 14 days of intensive care unit admission. MAP: Mean arterial pressure; WBC: White blood cell; BMI: Body mass index; BUN: Blood urea nitrogen; MELD: Model for end-stage liver disease; AKI: Acute kidney injury.

Table 4 Proportion of patients identified at each chronic liver failure organ failure-acute kidney injury score threshold and diagnostic performance for predicting KDIGO stage 3 acute kidney injury within 14 days, 95% confidence interval.

Variable	Chronic liver failure organ failure-acute kidney injury score
	> 80	> 110	> 140
No. of patients, n (%)	937 (71.3)	611 (46.5)	288 (21.9)
Sensitivity	0.91 (0.88-0.94)	0.73 (0.69-0.77)	0.45 (0.41-0.50)
Specificity	0.39 (0.36-0.42)	0.68 (0.64-0.71)	0.90 (0.88-0.92)
Positive predictive value	0.44 (0.41-0.47)	0.54 (0.50-0.58)	0.71 (0.66-0.76)
Negative predictive value	0.89 (0.86-0.92)	0.83 (0.80-0.86)	0.76 (0.73-0.79)

The CLIF-AKI score stratified patients into four risk categories: Low, moderate, high, and very high. In the derivation cohort, the observed incidence of stage 3 AKI across these categories was 10.6%, 24.5%, 39.3%, and 71.2%, respectively, and the predicted risks were generally consistent with the observed risks (Figure 4A). Similar risk gradients were observed in the validation cohort, with corresponding incidences of 20.7%, 48.0%, 58.3%, and 71.4% (Figure 4B). In both cohorts, the incidence of stage 3 AKI increased progressively across higher risk categories, supporting the ability of the CLIF-AKI score to effectively stratify risk. The model showed good discrimination, with an AUC of 0.79 (95%CI: 0.76-0.82) in the derivation cohort and 0.86 (95%CI: 0.80-0.91) in the validation cohort (Figure 5). Calibration was also acceptable in both cohorts, as indicated by the Hosmer-Lemeshow goodness-of-fit test (derivation: χ² = 3.38, P = 0.91; validation: χ² = 12.64, P = 0.24). We further evaluated model calibration and overall prediction error using the calibration intercept, calibration slope, and Brier score. In the derivation cohort, the CLIF-AKI model exhibited a calibration intercept of 0.000 (95%CI: -0.148 to 0.148), a calibration slope of 1.000 (95%CI: 0.873-1.127), and a Brier score of 0.170. In the validation cohort, the calibration intercept was -0.038 (95%CI: -0.466 to 0.390), the calibration slope was 0.953 (95%CI: 0.583-1.323), and the Brier score was 0.181. These findings indicate that the CLIF-AKI model maintained good calibration and a low overall prediction error in the external validation cohort.

Figure 4 Comparison of observed and predicted risks of KDIGO stage 3 acute kidney injury across chronic liver failure organ failure-acute kidney injury risk categories in the derivation and validation cohorts. A: Derivation cohort; B: Validation cohort. Bars represent observed and predicted event rates within 14 days of intensive care unit admission. The number of patients and events in each risk category is shown below the x-axis. AKI: Acute kidney injury.

Figure 5 Receiver operating characteristic curves of the chronic liver failure organ failure-acute kidney injury score in the derivation and validation cohorts. A: Derivation cohort; B: Validation cohort. The score demonstrated good discriminative performance for predicting KDIGO stage 3 acute kidney injury within 14 days, with areas under the curve of 0.79 (95% confidence interval: 0.76-0.82) and 0.86 (95%CI: 0.80-0.91), respectively. AUC: Area under the curve; CI: Confidence interval.

Calibration plots demonstrated good agreement between predicted and observed probabilities (Figure 6). Bootstrap internal validation further confirmed the stability of the CLIF-AKI model. The apparent AUC in the derivation cohort was 0.79, and the optimism-corrected AUC was 0.78 (95%CI: 0.76-0.81). Regarding calibration, the bootstrap-corrected calibration intercept was -0.016, the calibration slope was 0.971, and the Brier score was 0.173. Clinical utility was assessed using DCA, which showed that the CLIF-AKI model provided greater net benefit compared with alternative strategies for predicting the primary outcome across threshold probabilities ranging from 10% to 70% (Figure 7).

Figure 6 Calibration plots for the chronic liver failure organ failure-acute kidney injury score predicting KDIGO stage 3 acute kidney injury within 14 days in the derivation and validation cohorts. A: Derivation cohort; B: Validation cohort. The apparent and bias-corrected calibration curves are shown in blue and orange, respectively. The dashed diagonal line represents ideal calibration, where predicted probabilities correspond exactly to observed event probabilities.

Figure 7 Decision curve analysis of the chronic liver failure organ failure-acute kidney injury score in the derivation and validation cohorts. A: Derivation cohort; B: Validation cohort. The orange curve represents the net benefit of the chronic liver failure organ failure-acute kidney injury model for predicting KDIGO stage 3 acute kidney injury within 14 days. The gray line indicates the strategy of treating all patients, whereas the black line indicates the strategy of treating no patients.

In the external validation cohort, we compared the discriminative performance of the CLIF-AKI score with that of the PIRO model[9], chronic liver failure consortium (CLIF-C) ACLF[13], Child-Turcotte-Pugh (CTP)[14], MELD[15], and MELD-Na scores[16] for predicting stage 3 AKI. The CLIF-AKI score achieved an AUC of 0.86 (95%CI: 0.80-0.91), outperforming the CTP score (AUC 0.67; 95%CI: 0.61-0.74; P < 0.001), CLIF-C ACLF score (AUC 0.73; 95%CI: 0.66-0.80; P = 0.0004), MELD score (AUC 0.63; 95%CI: 0.56-0.71; P < 0.001), and MELD-Na score (AUC 0.66; 95%CI: 0.58-0.73; P < 0.001) (Figure 8; Supplementary Table 2). The CLIF-AKI score also showed better discrimination than the PIRO score (AUC 0.79; 95%CI: 0.73-0.86), and the paired DeLong test confirmed that this difference was statistically significant (P = 0.04).

Figure 8 Receiver operating characteristic curves comparing the performance of the chronic liver failure organ failure-acute kidney injury score with that of the PIRO, Child-Turcotte-Pugh, chronic liver failure consortium acute-on-chronic liver failure, model for end-stage liver disease, and model for end-stage liver disease-sodium scores in the external validation cohort. The curves illustrate the ability of each model to predict KDIGO stage 3 acute kidney injury within 14 days of intensive care unit admission. The area under the curve for each model is indicated in the figure. CLIF-AKI: Chronic liver failure organ failure-acute kidney injury; CTP: Child-Turcotte-Pugh; CLIF-C-ACLF: Chronic liver failure consortium acute-on-chronic liver failure; MELD: Model for end-stage liver disease; MELD-Na: Model for end-stage liver disease-sodium; AUC: Area under the curve.

As key variables required to calculate the comparator scores were missing in the MIMIC-IV database, their discriminative performance could not be reliably evaluated in the derivation cohort. To assess the incremental value of the CLIF-AKI model for risk classification, we calculated the continuous net reclassification improvement and integrated discrimination improvement. As shown in Table 5, the CLIF-AKI model significantly improved both reclassification and discrimination compared with the PIRO, CTP, CLIF-C ACLF, MELD, and MELD-Na scores.

Table 5 Continuous net reclassification improvement and integrated discrimination improvement of the chronic liver failure organ failure-acute kidney injury model compared with PIRO, Child-Turcotte-Pugh, chronic liver failure consortium acute-on-chronic liver failure, model for end-stage liver disease, and model for end-stage liver disease-sodium scores for predicting stage 3 acute kidney injury in the external validation cohort.

Comparator model	NRI (95%CI)	P value	IDI (95%CI)	P value
PIRO	0.408 (0.126-0.696)	0.008	0.093 (0.037-0.149)	0.004
CTP	1.162 (0.916-1.391)	< 0.001	0.271 (0.209-0.334)	< 0.001
CLIF-C-ACLF	1.000 (0.758-1.242)	< 0.001	0.212 (0.151-0.275)	< 0.001
MELD	1.117 (0.858-1.353)	< 0.001	0.307 (0.237-0.377)	< 0.001
MELD-Na	1.061 (0.802-1.305)	< 0.001	0.294 (0.224-0.362)	< 0.001

CI: Confidence interval; IDI: Integrated discrimination improvement; NRI: Net reclassification improvement; CTP: Child-Turcotte-Pugh; CLIF-C-ACLF: Chronic liver failure consortium acute-on-chronic liver failure; MELD: Model for end-stage liver disease; MELD-Na: Model for end-stage liver disease-sodium.

Subgroup analyses

In the derivation cohort, subgroup analyses were performed according to ACLF grade, mechanical ventilation status, and vasopressor use. The results indicated that the CLIF-AKI model demonstrated consistent and acceptable discriminative performance across different subgroups, with AUCs ranging from 0.74 to 0.82 (Supplementary Table 3).

Sensitivity analyses

To reduce potential overlap between predictor measurement and early outcome occurrence, we conducted a 24-hour landmark sensitivity analysis. Patients with ACLF who developed stage 3 AKI within the first 24 hours after ICU admission were excluded. The CLIF-AKI model achieved an AUC of 0.77 (95%CI: 0.74-0.80) in the derivation cohort and 0.85 (95%CI: 0.78-0.91) in the external validation cohort (Supplementary Figure 5). These findings suggest that the CLIF-AKI model maintained acceptable discriminative performance after excluding patients with very early-onset stage 3 AKI.

We further assessed the predictive performance of the CLIF-AKI model for early stage 3 AKI, defined as the first occurrence of stage 3 AKI within 7 days of ICU admission. In the derivation cohort, 271 patients developed stage 3 AKI within 7 days, and the model achieved an AUC of 0.80 (95%CI: 0.77-0.83). In the external validation cohort, 40 patients developed stage 3 AKI within 7 days, and the model achieved an AUC of 0.81 (95%CI: 0.74-0.88). These findings suggest that the CLIF-AKI model maintained good discriminative performance for predicting early-onset stage 3 AKI.

To assess the influence of renal function-related predictors on model performance, two additional sensitivity analyses were performed. After excluding the MELD score, the modified model achieved an AUC of 0.77 (95%CI: 0.75-0.80) in the derivation cohort and 0.85 (95%CI: 0.80-0.91) in the external validation cohort. After excluding BUN, the corresponding AUCs were 0.79 (95%CI: 0.76-0.81) and 0.84 (95%CI: 0.78-0.90), respectively. These findings indicate that the CLIF-AKI model maintained acceptable discriminative performance after the removal of individual renal function-related predictors.

Sensitivity analyses were performed using the composite outcome. Without model refitting, the CLIF-AKI model developed for the primary outcome was applied to generate predicted probabilities and evaluate performance for the composite outcome in both cohorts. The AUC for the composite outcome was 0.77 (95%CI: 0.75-0.80) in the derivation cohort and 0.81 (95%CI: 0.74-0.87) in the validation cohort. The CLIF-AKI score maintained good discrimination and calibration for the composite outcome in both cohorts (Supplementary Figure 6). DCA showed a greater net benefit across threshold probabilities ranging from 20% to 75% (Supplementary Figure 7). These findings suggest that the CLIF-AKI score predicts not only progression to stage 3 AKI but also short-term serious adverse outcomes, including early mortality, in patients with ACLF.

DISCUSSION

In this study, we developed and externally validated the CLIF-AKI score to predict 14-day progression to KDIGO stage 3 AKI in ICU patients with ACLF. To our knowledge, this is one of the first studies to develop and externally validate a scoring system specifically designed to estimate the short-term risk of progression to stage 3 AKI in critically ill patients with ACLF. The CLIF-AKI score comprises eight routinely collected clinical and laboratory variables and showed good discrimination and calibration in both the derivation and external validation cohorts. In sensitivity analyses, the score also identified patients at increased risk of early mortality. The score is parsimonious, can be readily calculated at the bedside, and has the potential to facilitate early risk stratification and clinical decision-making in this high-risk population.

The CLIF-AKI score integrates routinely available variables that reflect systemic inflammation and infection, hemodynamic status, organ support requirements, liver disease severity, and baseline renal function. WBC count was included as a pragmatic marker of systemic inflammation, as leukocytosis is an independent predictor of mortality in patients with decompensated cirrhosis and is associated with ACLF progression[17]. In cirrhosis, inflammation and infection are tightly intertwined, and ACLF commonly develops in this setting[18]. Clinically, leukocytosis often indicates concomitant bacterial infection, a common precipitant of AKI in decompensated cirrhosis that may both trigger and exacerbate renal injury[19].

Mechanistically, accumulating evidence suggests that AKI in cirrhosis is largely driven by bacterial translocation and overt infection[20]. Bacterial translocation is central to the pathogenesis of spontaneous bacterial peritonitis and other infections[21], with bacterial components activating monocytes via pathogen-associated molecular patterns and inducing pro-inflammatory cytokines such as tumor necrosis factor-α, interleukin-6, and interleukin-1β[22]. These inflammatory mediators have been closely linked to renal dysfunction and disease progression in patients with cirrhosis and ACLF[23].

The CLIF-AKI score also incorporates markers of hemodynamic status and organ support, including MAP, mechanical ventilation, and vasoactive agent use. In cirrhosis and ACLF, renal complications often arise from a common pathophysiological cascade in which splanchnic vasodilation reduces effective arterial blood volume. The resulting arterial underfilling activates systemic vasoconstrictor pathways, leading to sodium retention, impaired free-water excretion, and renal vasoconstriction, thereby reducing renal blood flow and promoting renal dysfunction[24]. Accordingly, MAP is a key hemodynamic target in cirrhosis and ACLF, as reduced renal perfusion pressure is a major driver of AKI development and progression.

The use of vasoactive agents typically reflects circulatory failure refractory to fluid resuscitation[25]. Consistent with this, lower MAP has been associated with stage 2 AKI in cirrhosis, whereas each 5-mmHg increase in MAP is associated with a 1.17-fold greater likelihood of AKI reversal[26,27]. Mechanical ventilation, although often necessary, may impair renal function by increasing intrathoracic and central venous pressures[28], thereby reducing effective renal perfusion pressure and activating neurohormonal and pulmonary inflammatory pathways. These effects may exacerbate renal venous congestion and tubular injury, contributing to AKI development and progression to more severe stages[29].

The CLIF-AKI also incorporates markers of baseline hepatic functional reserve and liver failure severity, including the MELD score and serum albumin level. In a prospective study of 1022 patients with ACLF, higher MELD scores were associated with an increased risk of stage 3 AKI. Among patients who developed stage 3 AKI, MELD independently predicted 1-month mortality and the need for RRT, with risk increasing markedly when the MELD score was ≥ 35. The same study also reported lower baseline serum albumin levels among patients with stage 3 AKI[30], suggesting that impaired hepatic functional reserve may predispose patients to severe AKI. In addition, baseline BUN, a widely used marker of renal function, was included in the CLIF-AKI model[31]. Elevated BUN reflects reduced glomerular filtration and impaired renal function has previously been incorporated into AKI prediction models in patients with ACLF. Higher BUN levels are associated with more advanced AKI stages and poorer clinical outcomes[32].

In this study, BMI remained an independent predictor of progression to stage 3 AKI in multivariable analysis, and higher BMI was associated with an increased risk of this outcome. Previous studies have identified overweight status as an independent risk factor for 30-day mortality in patients with ACLF[33]. However, BMI has not been systematically evaluated as a predictor of AKI risk in this population. Our findings suggest that BMI may be a previously underrecognized predictor of progression to stage 3 AKI in patients with ACLF. One possible explanation is that, in critically ill patients with ACLF, a higher BMI may reflect not only adiposity but also fluid retention and venous congestion. For example, severe ascites can increase intra-abdominal pressure and contribute to abdominal compartment syndrome, thereby elevating central venous pressure, impairing cardiac filling and output, and ultimately precipitating or exacerbating AKI[34]. Therefore, BMI should be interpreted with caution in patients with ACLF. In this population, a higher BMI may partly reflect volume overload and fluid retention rather than true body composition. Notably, BMI was significantly lower in the validation cohort than in the derivation cohort, whereas the incidence of stage 3 AKI was comparable between the two cohorts. This finding suggests that the predictive role of BMI may be influenced by population characteristics and volume status. Consequently, the stability and generalizability of BMI as a predictor of stage 3 AKI require further validation in additional cohorts.

Several AKI prediction models have been developed for patients with ACLF or cirrhosis. Huang et al[35] developed the KP-AKI model, whereas Zhang et al[31] developed and compared multiple machine-learning models for predicting in-hospital AKI in patients with ACLF. In populations outside ACLF, Lin et al[36] and Zheng et al[37] developed AKI prediction models for hospitalized patients or ICU patients with cirrhosis and validated these models using data from the MIMIC database. Among the currently available AKI prediction models, the PIRO model appears to be the most closely aligned with an ACLF-specific framework and is recommended by the Asian Pacific Association for the Study of the Liver ACLF guidelines[9]. Nevertheless, external validation data remain limited, and its performance in predicting progression to stage 3 AKI has not been well established. In contrast, the CLIF-AKI score was specifically developed and externally validated to predict progression to stage 3 AKI in critically ill patients with ACLF.

In addition, previous biomarker studies have suggested that cystatin C and urinary neutrophil gelatinase-associated lipocalin may detect kidney injury earlier than conventional clinical variables do. However, these biomarkers were not routinely available in either the MIMIC-IV database or the external validation cohort used in the present study. As a result, they could not be incorporated into model development or validation.

In this study, the CLIF-AKI score showed significantly better accuracy for predicting stage 3 AKI than traditional prognostic scores, including CLIF-C ACLF, CTP, MELD, and MELD-Na. Notably, these scores were originally developed to assess disease severity and prognosis rather than to predict AKI in patients with ACLF. Therefore, we further compared the CLIF-AKI score with the PIRO model, which was specifically developed for AKI prediction in ACLF.

For predicting progression to stage 3 AKI within 14 days, the CLIF-AKI score demonstrated superior discriminatory performance compared with the PIRO model. This difference may partly be explained by differences in the prediction targets. The PIRO model was developed to predict AKI of any stage within 7 days, whereas the CLIF-AKI score was specifically designed to predict progression to stage 3 AKI within 14 days. Thus, the two models address related but distinct clinical questions, with the CLIF-AKI score focusing on a more severe endpoint over a longer observation period.

Compared with patients in the external validation cohort, those in the MIMIC-IV cohort appeared to have greater overall disease severity. The two cohorts differed in several important clinical characteristics, including hemodynamic status, organ support, ACLF grade, liver function, and infection burden. Therefore, part of the predictive performance of the CLIF-AKI model may reflect organ support and illness severity at ICU admission. It is important to note that this model was primarily intended for early risk stratification at ICU admission rather than for establishing causality or identifying the predictors as modifiable treatment targets. However, subgroup analyses showed that the model retained acceptable discriminative performance across different ACLF grades, mechanical ventilation statuses, and vasopressor use categories. In addition, the model maintained acceptable discrimination after excluding either the MELD score or BUN. Collectively, these findings suggest that the predictive performance of the CLIF-AKI model was not driven solely by differences in disease severity or by any single renal function-related predictor.

This study has several limitations. First, the retrospective design and the use of the MIMIC database for model derivation may have introduced bias, as several clinically important factors associated with stage 3 AKI were not fully captured, including detailed information on nephrotoxic drug exposure, contrast agent dose and timing, and fluid resuscitation strategies. In addition, furosemide use differed significantly between the derivation and external validation cohorts, suggesting potential differences in volume management and diuretic treatment strategies. These differences in treatment strategies may be associated with the risk of progression to stage 3 AKI. Second, the external validation cohort was derived from a single center, which may limit generalizability of the findings to other regions and healthcare settings. Furthermore, in the exploratory assessment of baseline renal status, no patients with possible stage 1 AKI at ICU admission were identified in the external validation cohort. Consequently, the performance of the CLIF-AKI score could not be externally evaluated in this subgroup, thereby limiting the evidence supporting its generalizability to this population. Therefore, further multicenter studies are needed to validate the CLIF-AKI score.

Third, this study identified stage 3 AKI using the full KDIGO criteria, including SCr, urine output, and initiation of RRT. However, the decision to initiate RRT may be influenced by clinical workflow, physician judgment, and availability of treatment resources. Therefore, variation in RRT practices across institutions may introduce heterogeneity in the ascertainment of stage 3 AKI and affect the external generalizability of the model across healthcare settings. Fourth, the primary outcome was defined as the first occurrence of, or progression to, stage 3 AKI within 14 days of ICU admission, because this study focused on risk stratification for severe kidney injury during the early phase of critical illness. This outcome is clinically relevant and closely related to the early identification of high-risk patients and the development of clinical management strategies. However, intermediate- and long-term renal outcomes were not further evaluated, including changes in renal function at 28 and 90 days, duration of RRT, long-term RRT dependence, renal recovery, and the risk of progression to chronic kidney disease.

In addition, some patients with ACLF may have died before developing stage 3 AKI, making death a potential competing event. Although a sensitivity analysis using the composite endpoint of stage 3 AKI or death was performed, time-to-event and competing-risk analyses were not conducted. Therefore, the potential impact of death before stage 3 AKI on risk estimation could not be fully excluded. Further studies should evaluate the predictive value of the CLIF-AKI model for intermediate- and long-term renal outcomes in multicenter prospective cohorts with comprehensive follow-up and validate its performance using time-to-event and competing-risk methods.

CONCLUSION

We developed and externally validated the CLIF-AKI score using baseline data available at ICU admission to identify patients with ACLF at high risk of progression to stage 3 AKI within 14 days. The score demonstrated good discrimination and calibration and may serve as a practical tool for early risk stratification in the ICU. Further prospective multicenter studies are warranted to validate its clinical utility and generalizability.

ACKNOWLEDGEMENTS

We sincerely acknowledge the research team at the Massachusetts Institute of Technology and the clinical staff of Beth Israel Deaconess Medical Center for their contributions to the development and maintenance of the Medical Information Mart for Intensive Care-IV database. We also express sincere gratitude to the data collection team at Zhenjiang Third People’s Hospital for their assistance with data acquisition and management.