Motivating Trial Vidaza ® in AML patients undergoing allotx  Activates genes for apoptosis  Modifies phenotype of leukemic cells to potentiate a GVL effect Dose-toxicity profile of Vidaza ® is unknown The cumulative risk of toxicity from repeated administrations is unknown Goal: Jointly optimize (Dose, Schedule) Optimizing Dose and Schedule (Braun, Thall, Nguyen, deLima, 2007)

“Schedule” is a predetermined number of courses and days of administration within each course “Dose” = dose per administration - Example, starting at time 0 : s (1) = (0, 1, 2, 7, 8, 9) s (2) = ( s (1), s (1) + 14 days ) = (0, 1, 2, 7, 8, 9, 14, 15, 16, 21, 22, 23), etc. -The agent may be administered as frequently as desired to each patient, provided the (dose,schedule) pair is sufficiently SAFE - A patient’s actual administration times in the trial may deviate from the scheduled times

One cycle of an agent is administered at a fixed sequence of successive times s 1 < s 2 <... < s k, the patient rests, & this is repeated one or more times. How many cycles can be given “safely?” _______________________________________________ 0 s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 rest

Hazard of toxicity from one administration : A simple 3-parameter piecewise linear model Each dose has its own 3-parameter triangle for the one-administration hazard t = time in days h(t)

Cumulative hazard of toxicity from multiple administrations at a given dose t = time in days H(t)

Cumulative hazard of toxicity at day 10 t = time in days H(t)

Cumulative Hazard of Toxicity by t* e = time of study entry s j = time of j th administration after e  (t* | dose, schedule)} = ∫ [0, t*]  j h{ u - ( e + s j ) | dose} d u Pr(Toxicity by time t* | dose, schedule) = F(t* | d,s,  ) = 1 – e -  (t* | dose, schedule) replaces the usual Pr(toxicity | dose) used for binary outcomes, ignoring schedule

Dose per Administration (mg/m2)

Treat 1 st patient at the lowest (dose, schedule) pair Based on current Time-to-Toxicity data, treat each patient at the best (dose, schedule) pair Do not “skip” untried (dose, schedule) pairs If no (dose, schedule) pair is acceptable  Stop the trial Trial Conduct

Dose per Administration (mg/m2)

What Actually Happened in The Vidaza ® Trial

Since only 1 toxicity occurred in the first 27 patients, Dr. de Lima decided to add 4 higher dose levels of Vidaza: 32, 40, 48, 56 mg/m 2 After receiving IRB approval, the trial was re- started with 4x7 = 28 (dose,schedule) combinations

16 new (dose, schedule) pairs

Final Optimal Combination (40 mg/m 2 per administration, 3 cycles)

A case where all (dose,schedule) combinations are unacceptably toxic

After pat. #2 began treatment at (8 mg/m 2, 2 cycles), pat. # 1 treated at (8 mg/m 2, 1 cycle) experienced toxicity

All (dose,schedule) pairs are too toxic  The Trial is Stopped Early T Dose-schedule toxicity cop

Hazard of toxicity from one administration b a = area c

Final Data Analysis Posterior mean and standard deviation (SD) of per-administration hazard parameters in the Bayesian model for Pr(toxicity| PAD, number of cycles). Area a Days to Peak of Hazard b Days from Peak of Hazard to End c Duration (Days) b + c PADMean (SD) (0.0034) 14.5 (24.1)8.7 (12.8)23.2 (27.8) (0.0041) 14.9 (22.9)14.4 (21.6)29.4 (31.5) (0.0049) 11.7 (25.7)20.3 (38.9)32.0 (47.3) (0.0054) 15.9 (12.4)31.3 (29.4)47.3 (26.6) (0.0160) 14.0 (11.9)32.0 (29.0)46.0 (26.5)

Final Data Analysis Posterior mean ptox = probability of toxicity by day 116 Per Administration Dose of Vidaza (mg/m 2 ) Number of Cycles

For ptox = Pr(toxicity within 116 days), each entry is the posterior value of Prob( ptox > 0.30 ). For each combination of (Number of Cycles, Per-Administration Dose), A = acceptable toxicity, T = unacceptable toxicity. Final Data Analysis Per Administration Dose of Vidaza (mg/m 2 ) Number of Cycles A A A A A T T A A A A A A T A A A A A A A A A A

Take-Away Messages 1) The optimal combination (3 cycles, 32 mg/m 2 per day) would not have been found using ANY other phase I methods. 2) Current work is to incorporate progression- free survival time, to be used along with time- to-toxicity, into the method.

Conclusions Dose-Schedule Algorithm The Dose-Schedule Algorithm reliably 1) Finds (Dose,Schedule) pairs having specified Pr(Toxicity by day t) 2) Stops if no (Dose,Schedule) is acceptable Implementation is Hard Work, but a free computer program “Dose Schedule Finder” is available from