We can/could do so much more with publicly shared data Tito Fojo, - PowerPoint PPT Presentation

We can/could do so much more with publicly shared data Tito Fojo, MD, PhD Columbia University

Theory for regression and growth f(t) = e (g · t) + e (-d · t) -1 Where f = tumor measurement in t days d = regression rate constant; g = growth rate constant

Exponential regression and growth: 4 equations gd [growth AND f(t) = e (g · t) + e (-d · t) -1 regression] gx [only growth] f(t) = e (g · t) dx [only regression] f(t) = e (-d · t) gd Ø [growth, regression + sensitive fraction] f(t) = (1 - Ø) • exp (g • t) + Ø • exp (-d • t) - 1

14 Models

Analysis of independently conducted colorectal cancer trials in PDS confirms results from randomized trials and provides robust data sets against which to benchmark smaller subsets Bevacizumab FOLFIRI FOLFOX FOLFIRI Gregorio 1 st line 1 st line 2 nd line + FOLFOX Marañon 1 st line Log g N = 451 N = 243 N = 172 N = 390 N = 58 g = -6.71 g = -6.52 g = -6.51 g = -6.03 g = -6.62 (-7.1, -6.29) (-7.12, -6.06) (-7.11, -6.17) (-6.65, -5.51) (-7.23, -6.17) Red = Statistically inferior; all others in black including data from Gregorio Marañon Hospital in Madrid were statistically indistinguishable

Project Data Sphere: Colorectal Cancer Predicting overall survival with g values estimated using scans 1.0 Slowest g values] Intermediate g values] Fastest g values] Survival Probability 0.75 0.5 0.25 P < 0.0001 0 6 0 12 24 30 36 18 Survival in Months Number at risk Slowest 414 412 337 147 44 6 0 Intermediate 419 413 283 103 32 7 1 Fastest 423 346 169 53 13 4 0

Colorectal Cancer FOLFIRI + Afilbercept OS: 13.5 vs 12.06 months ∆ = 1.44 months p = 0.0032 Unidimensional Bidimensional Volumetric g g p = 0.0000000743 p = 0.000000000282 p = 0.00000000000591 [Wilcoxon] [Wilcoxon] [Wilcoxon]

Colorectal Cancer Unidimensional Data FOLFIRI + Afilbercept OS: 13.5 vs 12.06 months ∆ = 1.44 months N = 40 Power Simulations 80% P Reference = Afilbercept + FOLFIRI Experimental = FOLFIRI 1000 resamples with replacement of Exp Bidimensional Data Volumetric Data N = 27 N = 27 80% P 80% P

Towards a Virtual Control – Data in PDS Docetaxel Sample Docetaxel Sample N g value Size for 80% Power Size for 90% Power n = 1132 0.0032 n = 21 n = 26 n = 100 0.0031 n = 26 n = 36 n = 200 0.0030 n = 18 n = 25 n = 300 0.0030 n = 18 n = 24

Great for real world data ➜ f(t) = e (g · t) + e (-d · t) -1

Power Simulations: B = Reference; A = Experimental 1000 resamples with replacement of A Volumetric Data