Skip to main content

Table 5 Virtual re-execution of shortlisted designs using bootstrapped ANZ 9311 data

From: Improving clinical trials using Bayesian adaptive designs: a breast cancer example

Design Duration (months), mean (SD) Sample size, mean (SD) Probability of success (%) Probability of inconclusive (%) Hazard ratio, median (IQR)
Fixed 76.7 (3.93) 233 (0.0) 0.13 1.16 (1.06, 1.28)
Small sample size
  Posterior 55.2 (32.3) 164 (51.3) 0.10 15.7 1.27 (1.15, 1.43)
  PPS 36.9 (13.1) 134 (51.8) 0.13 0.05 1.32 (1.10, 1.62)
  CPS 42.3 (9.7) 154 (28.9) 0.07 0.09 1.15 (1.03, 1.36)
  Goldilocks 36.9 (13.1) 134 (51.8) 0.14 0.05 1.32 (1.10, 1.62)
  PPBS 41.0 (5.6) 152 (26.6) 0.12 0.00 1.15 (1.02, 1.36)
High power
  Posterior 88.1 (35.4) 214 (32.5) 0.10 48.5 1.22 (1.07, 1.43)
  PPS 53.6 (9.2) 218 (14.6) 0.06 0.10 1.17 (1.07, 1.33)
  CPS 52.0 (8.1) 213 (15.1) 0.05 0.09 1.17 (1.03, 1.33)
  Goldilocks 53.5 (9.2) 218 (14.6) 0.06 0.09 1.17 (1.07, 1.33)
  PPBS 50.9 (4.5) 214 (15.1) 0.18 0.00 1.17 (1.03, 1.33)
  1. PPS Predictive probability of success, CPS Conditional probability of success, PPBS Predictive probability of Bayesian success. The median and interquartile range (IQR) of the distribution of 100,000 bootstrapped hazard ratios are shown, unadjusted for any adaptations, for illustrative purposes only