Statistical advancements improve biomarker and recurrent event modeling

Recent statistical research has introduced new methodologies for improving the assessment of prognostic biomarkers and the modeling of recurrent clinical events.

One study addresses the challenges of nonignorable missing data and heterogeneity in biomarker assessment. By leveraging instrumental variables and integrating inverse probability weighting into a pseudo partial likelihood, researchers developed a method to estimate the impact of covariates on time-dependent area-under-curve (AUC) performance. This approach was applied to data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.

Another study proposes a joint modeling framework designed for recurrent event data, such as disease relapses, that incorporates longitudinal internal covariates. This framework utilizes a Bayesian approach to integrate both proportional hazards and accelerated failure time models, providing a flexible alternative for analyzing how patient-specific factors, like tumor size, influence the recurrence of events.