1644P - PSA trajectories and prediction of time to no-longer clinically benefitting in metastatic castration-resistant prostate cancer: Insights from the ProBio trial

Background

The utility of prostate-specific antigen (PSA) in predicting when a patient no longer clinically benefits (NLCB) in metastatic castration-resistant prostate cancer (mCRPC) remains uncertain. We aimed to estimate longitudinal PSA trajectories along with their association with NCLB risk, and their predictive value using data from the ProBio trial (NCT03903835).

Methods

ProBio is a platform trial designed to personalise treatments for mCRPC patients based on predefined genomic biomarker signatures. Randomised patients were monitored until NLCB. We analysed longitudinal PSA measurements for patients receiving AR Pathway Inhibitors (ARPI), Taxanes, or Carboplatin. Bayesian joint models were employed to estimate PSA trajectories. We quantified the impact of PSA trajectories on NLCB using hazard ratio (HR) for current PSA levels and PSA velocity, i.e. the rate of change between the previous and current PSA level. Based on the selected models, we computed dynamic predictions of NLCB risk from 4 months post-randomization up until 10 months, comparing their performance with landmark approaches. Repeated 20x5 cross-validated time-varying AUC and Brier score were used as predictive accuracy measures.

Results

We included 180 patients (60 ARPI, 88 Taxanes, 32 Carboplatin) with 1012 PSA measurements; 159 patients reached NLCB with a median follow-up time of 6.8 months. PSA trajectories were associated with NLCB risk and varied by therapy received. Each doubling of PSA was associated with an HR of 1.23 (90% Credible Intervals (CrI) 1.16-1.30), while each 1-unit increase in PSA velocity conferred an HR of 3.81 (90% CrI 2.10-6.98). Predictive performances up until 10 months were similar between joint and landmark models, with an AUC of 0.73 (90% CrI 0.68-0.76) and Brier score of 0.22 (90% CrI 0.21-0.23) for the best-fitting models.

Conclusions

Elevated PSA levels and velocity were associated with increased NLCB risk. Dynamic predictions from joint and landmark models effectively predicted NLCB up to 10 months, showing how PSA may be used as an early indicator of NLCB thus allowing for personalised treatment strategies and monitoring.

Clinical trial identification

NCT03903835, EudraCT 2018-002350-78.

Editorial acknowledgement

Legal entity responsible for the study

Karolinska Institutet.

Funding

Karolinska Institutet.

Disclosure

All authors have declared no conflicts of interest.