Figure 4
Statistical power as a function of residual variance and sample size. Straight growth trajectories were simulated with sp(pow) covariance structure at small (0.1), medium (0.2) and large (0.3) residual variance values representing approximate Q25, Q50 and Q75 percentiles of the NCT02312245 clinical trial PDX model ln(At) residual variance distribution, sp(pow) structure with parameter 0.98, and two-sided type I error rate of α = 0.05. Hypothetical growth trajectories are shown in panel (a) for a diluent arm with tumor growing to 2cm2 day 28 area and active arms shrinking to 0.8cm2, 0.64cm2, 0.48cm2, and 0.3cm2 day 28 area. Tumor area is plotted on the natural log scale and labeled on the raw cm2 scale. We calculated statistical power for the specified two degree of freedom coincident curve hypothesis tests in panels (b–g) as the proportion of times out of 500 simulated datasets that the p-value was less than 0.05. Panels (b–d) represent hypothetical primary hypothesis tests versus arm 1, and panels (e–f) represent hypothetical secondary hypothesis tests versus diluent. Specifically, (b) arm 1 vs. arm 2 (day 28 area represents a 20% decrease from arm 1), (c) arm 1 vs. arm 3 (day 28 area represents a 40% decrease from arm 1), (d) arm 1 vs. arm 4 (day 28 area represents a 63% decrease from arm 1), (e) diluent vs. arm 1, (f) diluent vs. arm 2, (g) diluent vs. arm 3. Diluent vs. arm 4 has the same power as diluent vs. arm 3 so is not shown.
