Risk factors for 90-day mortality in patients with acute-on-chronic liver failure and establishment of a predictive model
- VernacularTitle:慢加急性肝衰竭患者90天死亡的危险因素分析及预测模型构建
- Author:
Jing SUN
1
;
Tingji WANG
1
;
Zhijiao DUAN
2
;
Li ZHANG
3
;
Yanmei LI
2
Author Information
- Publication Type:Journal Article
- Keywords: Acute-on-Chronic Liver Failure; Prognosis; Risk Factors; Nomogram
- From: Journal of Clinical Hepatology 2026;42(1):151-159
- CountryChina
- Language:Chinese
- Abstract: ObjectiveTo investigate the independent predictive factors for 90-day mortality in patients with acute-on-chronic liver failure （ACLF）， to establish a risk predictive model， and to assess its predictive efficacy in comparison with MELD， MELD-Na， MELD 3.0， and COSSH-ACLF Ⅱ. MethodsA retrospective analysis was performed for the clinical data of 394 patients with ACLF who were admitted to The Affiliated Hospital of Inner Mongolia Medical University and Hohhot Second Hospital from July 2018 to July 2024， and general information and laboratory markers on admission were collected from all patients. The independent-samples t test or the Mann-Whitney U test was used for comparison of quantitative data between two groups， and the chi-square test or the adjusted chi-square test was used for comparison of qualitative data between two groups. The LASSO regression analysis was used to identify related variables， and the multivariate logistic regression analysis was used to establish a predictive model and generate a nomogram. The receiver operating characteristic （ROC） curve， the area under the ROC curve （AUC）， calibration curve， and clinical decision curve were used to assess the performance of the model. ResultsA total of 394 patients with ACLF were included in this study， with 136 patients in the training set， 58 in the internal validation set， and 200 in the external validation set. The cohort had a mean age of 52.9±11.7 years， among whom male patients accounted for 72.84% （287/394）， the patients with HBV infection accounted for 22.33% （88/394）， the patients with alcohol-related causes accounted for 45.94% （181/394）， and the patients with other causes （including drug-induced and autoimmune diseases） accounted for 31.73% （125/394）. The overall 90-day mortality rate was 27.41% （108/394）. The multivariate logistic regression analysis showed that diabetes （odds ratio ［OR］= 5.831， 95% confidence interval ［CI］： 1.587 — 21.424， P=0.008）， cystatin C （Cys-C） （OR=2.984， 95%CI： 1.501 — 5.933， P=0.002）， and spontaneous peritonitis （SBP） （OR=5.692， 95%CI： 2.150 — 15.071， P<0.001） were independent risk factors， and a nomogram was generated based on these factors. This model had an AUC of 0.836 in the training set， 0.881 in the internal validation set， and 0.878 in the external validation set， showing a good discriminatory ability. The calibration curve showed a good degree of fitting， with a relatively high net clinical benefit. The subgroup analysis based on etiology showed that the model had an AUC of 0.850 in the patients with HBV infection， 0.858 in the patients with alcohol-induced ACLF， and 0.908 in the patients with other etiologies， indicating that the model had a good discriminatory ability across the populations with different etiologies. Compared with traditional scores， the model （AUC=0.836） had a significantly better predictive value than MELD （AUC=0.619， Z=3.197， P=0.001）， MELD-Na （AUC=0.651， Z=2.998， P=0.003）， MELD 3.0 （AUC=0.601， Z=3.682， P<0.001）， and COSSH-ACLF Ⅱ （AUC=0.719， Z=2.396， P=0.017） alone. ConclusionDiabetes， SBP， and Cys-C are independent risk factors for 90-day mortality in patients with ACLF. Compared with MELD， MELD-Na， MELD 3.0， and COSSH-ACLF Ⅱ scores， this model has a higher predictive value for 90-day prognosis in patients with ACLF and is suitable for patients with ACLF caused by various etiologies.
