PSA trajectories and prediction of NLCB in mCRPC: insights from the ProBio trial

International Investigator’s Meeting, October 7-8, 2024
ProBio sponsors

Alessio Crippa

Thanks to Lana Broer

Outline

 

  • Background and aims
     

  • Patients population and methods
     

  • Results
     

  • Conclusions
     

  • Q&A

Background and aims

Background

 

  • PSA measurements are the cornerstone of follow-up for localized and mHSPC patients.

  • PSA is also monitored for mCRPC patients but less used for treatment discontinuation.

  • Time to no-longer clinically benefitting (NLCB) as recommended by PCWG3:

    • PSA increase;
    • Radiological progression;
    • Clinical progression.

Aim

 

  • Aims of this study:

    • PSA trajectories for different therapies in mCRPC patients.
    • Can PSA trajectories predict NLCB.

Patients population and methods

Patients population

  • First randomization data for mCRPC randomized until November 2022 to ARPI, Taxanes, Carboplatin, and receiving ARPI or Taxanes in Control group.

Methods (1)

 

Part 1) Longitudinal PSA trajectories

Bayesian joint models for longitudinal and survival data:

  1. Longitudinal sub-model: mixed-models to describe how PSA changes over time (different time parametrization), adjusted by baseline PSA and received therapy.

  2. Survival sub-model: Cox PH model for the effect of PSA trajectories on NLCB (different functional forms) adjusted by received therapy, and other prognostic factors (treatment line, ECOG status, timing of metastatic disease, ISUP at diagnosis, type of progressive disease, disease volume, and ctDNA fraction).

Methods (2)

 

Part 2) Dynamic predictions

Predict NLCB up to 10 months using PSA measurements until 4 months:

  1. Bayesian joint models.

  2. Landmark Bayesian Cox regression models.

Results

PSA descriptives

  • Different shapes of PSA trajectories.
  • Patients kept on treatment after PSA progression.

Outcome data

  • 1012 longitudinal PSA measurements, 151 (84%) NLCB events (6.8 month median follow-up).
Driving progression n
PSA and radiological 57
All types 48
Only PSA 19
PSA and clinical 14
Only radiological 9
Only clinical 3
Radiological and clinical 3
None 1

Modelled PSA trajectories

Effect of PSA dynamics on NLCB risk

  • Adjusted HR (90% Credible Intervals)

 

name

current value

slope

current value and slope

Functional forms

Current value of log2(PSA)

1.23 (1.16, 1.3)

1.21 (1.15, 1.28)

Velocity of log2(PSA)

5.61 (3.05, 12.54)

3.81 (2.1, 6.98)

Therapy class

Taxane

1.99 (1.18, 3.38)

3.60 (2.19, 5.89)

2.29 (1.38, 3.86)

Platinum

3.80 (1.97, 7.00)

4.38 (2.32, 8.30)

3.10 (1.73, 5.67)

Example cases

Predictive performances

Dynamic predictions (evolution)

Dynamic predictions (at 4 months)

Conclusions

Conclusions

  • Differential PSA trajectories, with greater initial decline for ARPI.

  • Both the final PSA value and changes in PSA influence NLCB.
     

  • Landmark and JM demonstrated good predictive performance.

  • Dynamic predictions offer valuable insights for anticipating favourable responses or early progression.
     

  • There remains residual heterogeneity — could incorporating ctDNA dynamics further enhance predictive accuracy?

PSA dynamics