Beyond TraditionalDesigns in Early DrugDevelopment

MaRS Centre Toronto – Feb 2006

Miklos Schulz, PhD.

St. Clare Chung, MMath.

Early Drug Development

! Phase Ib – maximum tolerated dose / schedule

! Phase I/II efficacy-toxicity trade-off studies

Later-phase Studies

! Phase II or Proof-of-Concept trials

! A trials to “explore” clinical efficacy with a smallnumber of targeted subjects : provide earlierevidence of the potential to show clinical efficacy

! Seamless Phase II / III Designs

Design Approaches

! Frequentist (traditional)

! Bayesian

Frequentist vs. Bayesian

Frequentist vs. Bayesian in ClinicalTrials

Objective of Phase I trials

Phase I: Traditional Design! The only traditional adaptive dose/treatment allocation design

! “1-in-3” Design (3+3 design)

! Treat 3 patients at the starting dose level Di

! If 0 patients experience dose-limiting toxicity (DLT),escalate to dose Di+1

! If 1 or more patients experiences DLT, treat 3 morepatients at dose level Di

! If 1 of 6 experiences DLT, escalate to dose Di+1

! If 2 or more experiences DLT, MTD = Di-1

! Dose escalation stops when ! 1/3 patients have DLT at agiven dose level; MTD is next lower dose level

Phase I: Traditional Adaptive Design

Phase I: Traditional Design

! Limitations of “1-in-3” Design

! Inflexible; what to do if:

! Number of subjects treated at a dose differ fromalgorithm of (3 or 6)

! Outcome (DLT) re-assessed after dose-escalationdecision made

! Sample size is variable

! Confidence in MTD is usually poor

Continual Reassessment Method! O’Quigley et al., 1990

! Reconcile practical constraints and ethical demands of Phase I studies

! Treat patients at the dose which all currently available evidence indicates tobe the best estimate of the MTD

! Two features of CRM:

! Estimate the MTD after every patient has been dosed and has completedthe follow-up segment

! Allocate next patient to the dose-level suggested to be the MTD

! Currently available evidence:

! Prior knowledge of MTD

! Beliefs in the initial data

! Bayesian procedure: one parameter model

! Binary response: toxicity vs. no toxicity

Continual Reassessment Method

! Method accounts for different number of patients per dose

! Targets a pre-selected DLT rate

! Variants of design:

! Two-parameter CRM (Schulz & Chung, 1995)

! Modified CRM (Goodman et al. 1995)

! Extended CRM [2 stage] (Moller, 1995)

! Restricted CRM (Moller, 1995)

! Tri-CRM (Zhang et al. 2005)

Continual Reassessment Method

CRM – Case Study

! Original CRM (one parameter model) adequatewhen dose response curve is typical ‘s-shaped’

! Not efficient when toxicity increases at a slowerrate over the dose-range tested

! Deficiency compensated by 2-parameter model

CRM – One vs. Two parameter model

CRM – Case Study - Background

! Cancer patients treated at combination doses of 2 drugs

! Objective: determine the most efficacious treatment combination whichproduces at most, 33% toxicity

! 8 dose combination levels were tested

! Patients were on 4 cycles of treatment before outcome was determined

! Dose-limiting toxicity was any Grade III or IV toxicity in hematologicalparameters

! Patients were allocated to dose levels based on the traditional 1-in-3 approach

! Re-analysis was performed with the 2-parameter CRM model

CRM – Case Study

CRM – Case Study

Efficacy/Toxicity Trade-offs

! Thall PF, Cook J (2004)

! Problems with usual Phase I " Phase II paradigm

! Phase I designs ignore Response, but no patient hopesonly for “No Toxicity”

! For Biologic Agents Pr(Response) may be non-monotone in dose

! If Pr(Toxicity) is low for all doses but Pr(Response)increases with dose, then the superior higher dose willnot be found

Efficacy/Toxicity Trade-offs

! Thall PF, Cook J (2004)

! Phase I/II dose-finding strategy

! Patient outcome = {response, toxicity}

! Investigator defines:

! a lower limit P(Res)

! an upper limit P(Tox)

! three equally desirable("R, "T) targets - used toconstruct an Efficacy-Toxicity Trade-off Contour

! Dose x is acceptable if:

! Pr{"E (x,!) > "E* | data } > .10 or

! Pr{"T (x,!) < "T* | data } > .10

Other upper cutoff limits may be used

Efficacy/Toxicity Trade-offs

! Thall PF, Cook J (2004)

! Demo and Simulation results from Thall & Cook program

! Program may be downloaded from:

http://biostatistics.mdanderson.org/SoftwareDownload/

! The Trade-Off-Based Algorithm reliably:

! Finds Safe Doses having High Efficacy

! Stops if no dose is acceptable

Efficacy/Toxicity Trade-offs! Yin G, Li Y and Ji Y (2006)

! Phase I/II design

! Curve-free; not dependent on a specific response curve

! Incorporate bivariate outcomes, toxicity and efficacy

! Model the data to account for the correlation between toxicityand efficacy

! Dose for the next cohort of patients is determined fromresponses of previous cohorts and based on odds ratio criteriafrom posterior toxicity and efficacy probabilities

Summary

Clinical Trial Designs: Bayesian /Adaptive

! Learn faster " more efficient trials

! More efficient drug development

! More effective treatment of patients in the trial

! Drop or add doses

! Early stopping for futility

Traditional Approaches

! Robust, but inflexible: design parameters cannot bechanged without affecting robustness / interpretation

! Inefficient / time-wasting (e.g., treating patients inineffective studies arms)

! May focus only on single patient populations -therapeutic strategies

! Restricts statistical inferences to information in thecurrent trial