Dataset Information

Development and Validation of a Proteomics and Machine Learning-Based Predictive Model for Secondary Infection in Patients with HBV-Related Liver Failure: A Prospective Multicenter Study

ABSTRACT: Abstract Background and Aims: Patients with HBV-related cirrhosis are highly susceptible to secondary infections (SI) during hospitalization, which substantially increases mortality risk. Reliable tools for early prediction remain limited. This study aimed to develop and validate a multi-parameter predictive model based on plasma proteomic profiling to assess the risk of SI. Methods: In a prospective, multicenter cohort study, 114 patients were enrolled in the discovery cohort and 60 in the validation cohort. Differentially expressed proteins were identified using untargeted plasma proteomics. Feature selection was performed using the minimum redundancy maximum relevance (mRMR) algorithm, and logistic regression was applied to construct the predictive model. Model performance was subsequently validated in an independent cohort using targeted proteomics. Results: Proteomic analysis indicated that dysregulation of inflammatory and coagulation pathways contributes to SI development. Ten infection-associated proteins were selected, and a final predictive model incorporating LYZ, CALM1, SERPIND1, DPT, total bilirubin (Tbil), and aspartate aminotransferase (AST) was established. The model achieved an area under the receiver operating characteristic curve (AUROC) of 0.980 in the discovery cohort and 0.873 in the validation cohort, significantly outperforming conventional inflammatory markers such as C-reactive protein (CRP; 0.674 and 0.702), white blood cell count (WBC; 0.645 and 0.694), and neutrophil percentage (NE%; 0.642 and 0.793). The model also demonstrated strong prognostic performance for predicting 28-day mortality, with an AUROC of 0.957, exceeding that of the Chronic Liver Failure Consortium Acute-on-Chronic Liver Failure (CLIF-C ACLF) score (0.545) and the Model for End-stage Liver Disease (MELD) score (0.824). Conclusions: The proteomics-based predictive model accurately identifies patients with HBV-related liver failure at high risk of SI, offering promising clinical applicability and generalizability.

ORGANISM(S): Homo Sapiens

SUBMITTER: Yixin Hou

PROVIDER: PXD067174 | iProX | Fri Aug 08 00:00:00 BST 2025

REPOSITORIES: iProX

ACCESS DATA

Similar Datasets

Project description:Importance: Severe alcoholic hepatitis (AH) is a highly lethal disease with a 3-month mortality up to 50%. Corticosteroids are the current standard of care although nearly 40% of the patients do not respond and accurate pre-treatment predictors of survival are lacking. Objective: To develop a prognostic score based on molecular and clinical variables before initiation of corticosteroids. Design, Setting, and Participants: Gene expression profiling of fixed liver biopsy obtained for diagnosis of severe AH assessed in 3 independent cohorts (1 prospective cohort for prognostic gene signature and integrative score derivation, and 2 retrospective cohorts for validation). Samples were obtained from 4 European and 1 US medical centers between July 2006 and February 2015. Exposures: Biopsy proven severe AH patients treated with corticosteroids. Main Outcome Measures: Identification of a gene signature combined with a validated clinical index to develop an integrative prognostic score of survival without death or liver transplantation in a derivation cohort (n=71). Prognostic performance of the score was validated in an independent multi-center validation cohort 1 (n=48). Finally, the score was implemented in an FDA-approved clinical diagnostic platform (NanoString) and tested in another independent validation cohort 2 (n=20). Results: A prognostic score, integrating a 123-gene signature and the model for end-stage liver disease (gs-MELD score), was defined in the derivation cohort, in which poor- and good-prognosis patients were discriminated with a cut-off value of 2.66 at 3 months (event-free survival rates of 32% vs. 76%, respectively, p<.001) and 6 months (26% vs. 65%, respectively, p<.001). In the validation cohort 1, the score similarly discriminated poor- and good-prognosis patients at 3 months (43%, [95% CI, 26%-70%] vs. 96% [95% CI, 89%-100%], respectively, p<.001, c-index of 0.83 [95% CI, 0.66-0.99]) and 6 months (34% [95% CI, 18%-61%] vs 84% [95% CI, 72%-100%], respectively, p<.001, c-index of 0.80 [95% CI, 0.66-0.94]), and outperformed other existing clinical indices. The NanoString-based gs-MELD score remained predictive of 3- (p=.03) and 6-month (p=.009) even-free survival in the validation cohort 2. Conclusion and Relevance: The gs-MELD score, incorporating clinically applicable gene signature test and the MELD score, enables pre-treatment survival prediction in severe AH patients.

Dataset Information

Development and Validation of a Proteomics and Machine Learning-Based Predictive Model for Secondary Infection in Patients with HBV-Related Liver Failure: A Prospective Multicenter Study

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets