Modelling Dynamic PSA Trajectories and Survival in Prostate Cancer Using Joint Longitudinal Survival Models

Abstract

Biologically and clinically, prostate cancer is significantly heterogeneous; still, much of the existing empirical research on prognosis proceeds within the two-stage paradigm, which conceptualises prostate-specific antigen (PSA) as a static baseline variable and neglects its dynamic relationship to survival. The current study remedies these problems by employing the integrated joint model approach to model the dynamic relationship between the trajectory of PSA severity and time-to-event outcomes. For this study, a retrospective design included a sample of 300 prostate cancer patients; the study used models for overall PSA severity, the Cox model for complete survival, and the joint model approach with shared random effects to handle informative censoring. The descriptive analysis showed equal proportions across cancer stage categories and total PSA severity. For the general model, a modest change in population-averaged PSA was observed, yet significant individual heterogeneity remained. For general survival outcomes, the Gleason score was a significant predictor (HR=0.847; 95% CI:0.732-0.980). Kaplan-Meier curves also showed significant visual separation in total survival by overall PSA categories. Nevertheless, in the total joint model analysis, the trajectories for overall PSA showed significant predictive relationships beyond complete static clinicopathological factors. The originality of this research study lies in its approach to effectively integrate dynamic biomarker models and complete survival models to strategically eliminate bias, as found in static and Two-Stage models. Generally, this study provides strong support for specific clinician recommendations on the use of dynamic biomarker models to predict prostate cancer outcomes within the context of evidence-based oncology practice.