Prediction of Personalized Prognosis in Patients with Acute-on-chronic Liver Failure: A Joint Model for Dynamic Risk Profiling

Abstract

Abstract Background: Short-term mortality is high in patients with acute-on-chronic liver failure (ACLF), defined by the rapid deterioration of underlying chronic liver diseases. Current prediction models cannot estimate dynamic prognosis adequately. This study used both longitudinal and survival data to develop and validate a dynamic prediction model for ACLF. Methods: Adult patients with ACLF from a retrospective cohort, including 943 patients from 2014 to 2019 at the Third Affiliated Hospital of Sun Yat-sen University, were included. The progression of temporal indices was described using a mixed-effects model, and subject-specific prediction risk models with time-to-event data were constructed using a joint model (JM). The model was validated by testing the data using the area under the curve (AUC) and Brier score. Results: The AUCs for JM ranged from 0.808 to 0.840 when predicting 28-day mortality and from0.747 to 0.811 when predicting 90-day mortality in patients with ACLF. The linearity of the calibration curves was good, with the Brier scores ranging from 0.083 to 0.205. The performance of the ACLF-JM for 90-day predictions was superior (P < 0.001) to that of the Model for End-Stage Liver Disease score. Conclusions: It is possible to construct individualized dynamic event prediction models for patients with ACLF by jointly modeling longitudinal data with time-to-event outcomes. This JM provides a potentially valuable evidence-based tool for liver transplantation in clinical practice.