International Investigator’s meeting

Outline

• ProBio rationale and design
• Biomarker subgroup vs signatures
• Adaptive randomization
• Evaluation of therapies
• Operating characteristics
• Simulation example
• Summary

Study definition

ProBio - An

• Outcome-adaptive randomized

• Multi-arm

• Biomarker-driven

• Platform-trial

in patients with advanced prostate cancer

Changes in the theapeutic landscape

Traditional vs platform trials

Technical solutions

• Open ended: new hypotheses can be tested perpetually

• Master protocol, ideally a central IRB and approval process

• Leverage design, statistical engine, infrastructure, logistics, IT, …

• Universal consent process

• New hypotheses described in amendments

• Enables a common control arm (increased efficiency)

• Standing infrastructure for research

Platform-trial design

ProBio Life cycle

Biomarker subgroup (combination)

A genetic biomarker: specific driver mutations or multiple gene alterations.

The biomarker subgroup of a subject is given by the combination of the binary genetic biomarkers.

Biomarker signatures

A biomarker signature: specific and complex combination of biomarkers, grouping of previous biomarker subgroups.

Observations:
- Each patient belongs to one and only one biomarker subgroup.
- Patients may belong to more than one biomarker signature.

Note:
- Prevalence of some biomarker subgroups is too low for meaningful analysis.

Randomization & Therapies evaluation

Key points

• Patients are stratified based on their biomarker subgroup combination, and only then randomized to either the control (standard of care) or one of the active arms.

• Outcome-adaptive randomization is implemented to assign more patients to more promising (effective) therapies within the biomarker subgroup combinations.

• Therapies are constantly (monthly) evaluated within the biomarker signatures.

Adaptive randomization (subgroup)

• Depend upon the patient’s biomarker subgroup combination and treatment history.

• Men with too little tumor burden will be excluded.

• Adaptive:

  • Fixed before enrolling 50 patients in the active arms.
    
  • Thereafter, randomization probabilities are updated monthly based on the accumulated data, i.e. based how long subjects respond within the subgroup.
    
  • How to measure the effectiveness of a therapy vs the control in a biomarker subgroup?

How to compare therapies?

Bayesian survival model

\[S_{lis}(t) = \exp \left(-t^\gamma \lambda_{lis} \right)\]

Scale parameter \(\lambda_{lis} = \exp \left( -\gamma \eta_{lis} \right)\); shape parameter \(\gamma > 0\)

Linear predictor: \(\eta_lis = {\bf{X}}_{lis}\beta_{is}\)

Probability of superiority: \(\pi_{is} = P \left( \beta_{is} > 0 \right)\)

Chosen prior:

• \(\beta_{is} \sim N \left(0, .5 \right)\)

• \(\gamma \sim \textrm{Exp} \left( 1 \right)\)

• \(\beta_0 \sim N \left(3, 1\right)\)

Prior quantities

Probability of superiority

Rule of adaptation

Randomization adaptation

Evaluation of therapies (signature)

• Therapies are compared within the biomarker signatures of interest using a common control group as comparator.

• The main outcome is a survival time (PFS). We will use Bayesian parametric model to contrast the distributions of the mean PFS.

• Monthly, we will decide if continuing enrollment of new patients to a therapy-signature combination, or to early stop (graduation, futility, max patients).

• Expert in the Data Safety and Monitoring Board will advice on the action to take.

Comparative analyses within ProBio

Decision rules for therapy \(i\) in the biomarker signature \(s\)

• graduation:
  • \(n_{is} \ge 20\)
    
  • \(\pi_{is} \ge .85\)
    
  • \(\pi_{ij} \ge .65\), with \(j \in s\)
• stop futility:
  • \(n_{is} \ge 20\)
    
  • \(\pi_{is} \le .15\)
    
  • \(\pi_{ij} \le .25\), with \(j \in s\)
• stop max patients:
  • \(n_{is} \ge 150\)

Evaluation pf therapies over time

Evaluation Therapies

https://stweb.meb.ki.se/

Automatic Report

Operating characteristics

• Key operating characteristics include error and power.

• Complicated design and assessment of operating characteristics.

• Errors:

  • strict: graduating a therapy only in a signature where it was not superior to standard-of-care;
    
  • liberal: graduating a therapy only in subgroups where it was not superior to standard-of-care.

• Power:

  • strict: graduating a therapy for the signature where it was superior to standard-of-care;
    
  • liberal: graduating a therapy for any of the subgroups where it was superior to standard-of-care.

Results

• Overall error rate ~35%-40%, <15% for individual therapies (reduced in the liberal definition).

• Overall power was ~95% and ranged from 45% to 80% for individual therapies (increased in the liberal definition).

• Discovery trial, relatively liberal graduation criteria.

• Graduated combinations will be validated in a side trial nested within the ProBio platform.

Statistical simulations

Simulations ProBio

Assumptions

• 4 active treatments, 4 biomarker signatures (16 subgroups).

• Enrollment of 30 patients/month for an evaluation period of 3 years.

• Standard-of-care mixed of therapies (ARSi: 40%; Taxane: 35%; Other: Other = 25%).

• Different possible treatment effects.

• Additional model assumptions.

Assumed treatment effect

Errors and Power

Results over time