Androgen receptor pathway inhibitors and taxanes in metastatic prostate cancer: an outcome-adaptive randomized platform trial

Abstract

ProBio is the first outcome-adaptive platform trial in prostate cancer utilizing a Bayesian framework to evaluate efficacy within predefined biomarker signatures across systemic treatments. Prospective circulating tumor DNA and germline DNA analysis was performed in patients with metastatic castration-resistant prostate cancer before randomization to androgen receptor pathway inhibitors (ARPIs), taxanes or a physician's choice control arm. The primary endpoint was the time to no longer clinically benefitting (NLCB). Secondary endpoints included overall survival and (serious) adverse events. Upon reaching the time to NLCB, patients could be re-randomized. The primary endpoint was met after 218 randomizations. ARPIs demonstrated ~50% longer time to NLCB compared to taxanes (median, 11.1 versus 6.9 months) and the physician's choice arm (median, 11.1 versus 7.4 months) in the biomarker-unselected or 'all' patient population. ARPIs demonstrated longer overall survival (median, 38.7 versus 21.7 and 21.8 months for taxanes and physician's choice, respectively). Biomarker signature findings suggest that the largest increase in time to NLCB was observed in AR (single-nucleotide variant/genomic structural rearrangement)-negative and TP53 wild-type patients and TMPRSS2-ERG fusion-positive patients, whereas no difference between ARPIs and taxanes was observed in TP53-altered patients. In summary, ARPIs outperform taxanes and physician's choice treatment in patients with metastatic castration-resistant prostate cancer with detectable circulating tumor DNA. ClinicalTrials.gov registration: NCT03903835 .

Fig. 1. Trial design.

a, The ProBio platform design, including control and investigational arms, with adaptive randomization after screening and ctDNA analysis. Upon reaching progression, patients were re-randomized. Follow-up and treatment response evaluation followed the PCWG3 recommendations. Investigational arms in transparent gray denote other investigational arms to which patients could have been randomized, which do not fall within the scope of the current paper. b, The CONSORT diagram detailing patient screening, randomization and follow-up. Inclusion and randomization updated until 25 November 2022, with results for ARPIs and taxanes. Data on other investigational arms pending. As of this date, 7 patients awaited liquid biopsy results and had not been randomized. An extra 4 months of follow-up were allowed until March 2023 for 193 randomized patients. Of 129 patients in the investigational arms, 49 reached the progression endpoint, remained in the trial, and were re-randomized. In the physician’s choice group, 28 of 64 patients continued with new standard-of-care treatments. Follow-up included all-cause mortality from electronic health records. MSI+, microsatellite instability positive.

Fig. 2. Time to NLCB and overall survival.

a,b, Posterior survival curves (smooth) and Kaplan–Meier (stepped) estimates for the time to NLCB (a) and overall survival (b) in all patients (that is, biomarker-unselected or any biomarker subgroup combination), grouped by randomization. Tick marks on Kaplan–Meier curves denote censored patients. Weibull accelerated failure time models were used for survival estimates. Colored lines depict median posterior distribution, and shaded areas show 90% CrI. Insets: Patients at risk, cumulative events and estimated median survival times (in months) with 90% CrI.

Fig. 3. Posterior survival curves and Kaplan–Meier estimates by therapy arm and biomarker signature status (positive versus negative).

Posterior survival curves (smooth) and Kaplan–Meier estimates (step function) for the time to NLCB with ARPIs and taxanes. Left: Patients negative for a specific biomarker signature. Right: Patients positive for the respective biomarker signature. Survival curves were estimated using Weibull accelerated failure time survival models with an interaction term, enabling the assessment of differential treatment effects. The smoothed colored lines represent medians of the posterior distribution, while shaded areas depict corresponding 90% CrI. Top: AR (SNV/GSR) positive or TP53 altered versus AR (SNV/GSR) negative and TP53 wild type. Middle: TMPRSS2–ERG fusion negative versus TMPRSS2–ERG positive. Bottom: TP53 wild type versus TP53 altered. Homologous recombination deficiency was left out due to very few randomizations to ARPIs; preliminary results are available in Supplementary Fig. 5 and Table 3.