Integrated immuno-proteomic, cell death, and clinical trajectories with machine learning to predict outcomes in Acute-on-Chronic Liver Failure patients (IMP-ACLF)

Acute-on-chronic liver failure (ACLF) is acatastrophic syndrome with a significant burden in cirrhosis patients,characterized by multi-organ failures and high 90-day mortality (15-100%). Itscourse is dynamic, and improvement/worsening over the first seven days determinesthe patient’s outcomes. Mortality prediction is essential to prognosticatepatients, allocate resources, and guide definite therapies e.g. livertransplantation. Existing predictive models can misclassify up to 20-30% ofpatients, indicating a need for precision in prediction. Integrating the immuno-biologyof disease with clinical variables can improve such predictions.

Dysregulated systemic inflammation and celldeath are central to the pathobiology of ACLF. Both pro- and anti-inflammatoryresponses exist and the balance between the two determines the outcomes in ACLFpatients. The existing hypothesis of inflammatory cell damage and unbridled immunoparesisleading to death is primarily derived from the cross-sectional assessment of the immune system in ACLF. The knowledge of dynamicity of immune-balance is urgentlyrequired, which can predict the outcomes more precisely and guide the novel immune-modulatorytherapies.

Advanced flow cytometry and ELISA techniques can precisely estimate immune cells, their functions, cell death, and variouscytokines. "Omics" platform allows an unbiased understanding of thecomplex biology of a condition and proteomics exhibits a better opportunity to discoverscalable biomarkers and develop novel therapies. Machine learning can beintegrated to analyze large-scale, multi-dimensional, complex data.

Therefore, we first aim to analyze thetrajectories in the immuno-proteomic, cell death, and clinical profile of ACLFpatients to understand the dynamic pathobiology of disease and associate themwith mortality. Second, we aim to establish a machine learning modelintegrating clinical, cell death, and immuno-proteomic variables to predictmortality in ACLF.

We will conduct a cohort study in threephases. Firstly; in the discovery phase, eligible ACLF patients will be enrolledand followed up till 90-days or transplantation or death, whichever early.Clinical details will be serially noted for 90-days. The immunological andcell death profile will be recorded at baseline and day-7 through measuringmonocyte/neutrophil/T-/B-/NK-cells-numbers, subsets and functions, cell deathmarkers, and pro-/anti-inflammatory cytokines. Plasma proteome (protein levelsand pathways) will be assessed at baseline and day-7 of presentation.

Second; in the computational phase, thecomparisons of dynamic changes in immunological, proteomic, cell death, andclinical profile over time will be made between/within survivors andnon-survivors. We will employ machine learning on these variables to derive amortality prediction model. The model’s performance will be compared withexisting predictive models.

Third; in the validationphase, the model will be validated in a separate cohort of ACLF patients.