Development and external validation of a computational approach for accelerated read out of randomized clinical trials in metastatic prostate cancer.

Abstract

1502 Background: While overall survival (OS) is the gold standard endpoint for clinical trials in metastatic prostate cancer (mPC), it requires years of follow-up. We sought to develop and externally validate a computational model for accelerated OS readout based on short term (4 months) PSA kinetic data. Methods: Longitudinal PSA data were obtained from 8 completed phase 3 mPC trials. Using data from the first 4 months on trial, 18 different PSA kinetic variables were developed, including 50% and 90% decline in PSA, PSA=0.1, and PSA=0.2 at 1, 2, 3, and 4 months, and slope and base of the exponential fit of PSA. TITAN, COU-AA-301, and ACIS comprised the training cohort. Data from those trials was used to model OS using a relaxed LASSO approach. The resulting model was applied to 5 validation trials (see Table 1) to simulate 1,000 potential outcomes for each trial (after Harden and Kropko). For the CHAARTED trial, a model using only PSA-kinetic data was used as most baseline variables were missing. The distribution of 1,000 simulated Hazard Ratios (HR) for each trial was compared against the reported HR. Results: Overall, > 100,000 PSA values were used with a total > 7,500 eligible study pts, with ~ 70,000 PSA values and ~5,000 patients comprised the validation cohort. The model identified baseline PSA at treatment, PSA50 at 4 months, and the PSA slope as the most informative PSA predictors of OS. Table 1 below compares the median predicted HR and 95%-interpercentile range from the 1,000 simulations with the reported HR for each validation trial. The model correctly predicted the OS point estimate and 95%-confidence interval using the first 4 months of PSA data in every validation trial. Conclusions: A computational simulation approach to predict the OS outcome of phase 3 mPC trials based on PSA kinetics from the first 4 months on trial was developed. This model correctly predicted OS outcomes in 5 completed validation phase 3 trials. This included phase 3 trials with both positive and negative outcomes for OS as well as in both hormone sensitive and resistant mPC, and involving AR signaling inhibitors, chemotherapy and a PARP inhibitor. This model is being further validated with additional completed phase 3 trials, and once prospectively validated, has the potential to significantly shorten the follow up required for OS readout from phase 3 trials in mPC. Predicted vs. reported HR of each trial. Validation Trial Predicted HR (median [95 th interpercentile range]) Actual HR [95%-CI] LATITUDE 0.64 [0.54 – 0.74] 0.66 [0.56 – 0.78] COU-AA-302 0.69 [0.60 – 0.80] 0.81 [0.70 – 0.93] MAGNITUDE 1.06 [0.88 – 1.26] 0.97 [0.79 –1.19]* CALGB 90401 0.89 [0.78 – 1.01] 0.91 [0.78 – 1.05] CHAARTED (Unselected) 0.79 [0.61 – 0.98] 0.72 [0.59 – 0.89] CHAARTED (High Volume) 0.72 [0.53 – 0.92] 0.63 [0.50 – 0.79] CHAARTED (Low Volume) 0.95 [0.61 – 1.52] 1.04 [0.70 – 1.55] *Observed HR used as overall OS HR not reported in manuscript.