A0939
Title: Evaluating prognostic biomarkers for censored survival data with covariate adjustment
Authors: Ainesh Sewak - University of Zurich (Switzerland) [presenting]
Vanda Inacio - University of Edinburgh (United Kingdom)
Torsten Hothorn - Ludwig-Maximilians-Universitaet Muenchen (Germany)
Abstract: Identifying reliable biomarkers for predicting clinical events in longitudinal studies is important for enabling accurate disease prognosis and supporting the development of new therapies. Traditional receiver operating characteristic (ROC) curve analysis has been adapted for time-dependent and censored outcomes. However, accounting for clinical heterogeneity in patient characteristics remains a challenge in assessing the prognostic accuracy of biomarkers. Prior methods have relied on the proportional hazards assumption or model only the summary statistics of the ROC curve. A flexible conditional bivariate model is proposed to quantify biomarkers' prognostic accuracy for censored survival data while accounting for covariates. The model uses separate marginal regression models for the time-to-event and biomarker outcomes, accounting for their dependence structure on a latent normal transformed scale. By parameterizing the marginal models, the maximum likelihood for estimation and inference is used. The application of the method in a study is demonstrated by examining serum neurofilament biomarkers for predicting survival in amyotrophic lateral sclerosis (ALS) patients.
