1. DOSE SPACING IN EARLY DOSE RESPONSE CLINICAL TRIAL DESIGNS
Naitee Ting, Ph. D.
Associate Director
Pfizer Global Research & Development

5. STUDY 1 - WHAT’S NEXT?

6. STUDY 2

8. Questions in Designing the First Dose Response Study
How many doses to be tested?
What are the high and low doses?
What are the spaces between the test doses (what is the dose spacing)?
How frequent should subjects be dosed?
How many subjects for the study?

9. Dose Response Study Design
Selection of control
Selection of endpoints
Fixed dose vs titration dose
Two-stage designs vs Two designs

10. Dose Response Study Design
In early Phase II, drug first tested in patients
Assume maximum tolerable dose (MTD) known, assume monotonicity
Efficacy response, toxicity response
Range and spacing of doses

11. Limited Number of Fixed Doses
Concerns in interpreting titration dose
Multiple center designs
Formulation considerations
Placebo and maximum tolerable dose (MTD)
Incorporate active control?

12. Considerations in Dose Allocation
Selecting a wide range of doses
Find doses to capture the steepest increasing portion of efficacy dose response curve
Use of some low doses to help identify the minimum effective dose (MinED)
Not very high to be too close to MTD

17. Binary Dose Spacing For 2 active doses, one above 1/2, one below
Continue with this fashion to the lower end
Any cut for 1/p, where p  2
Non-parametric, model independent
Applies to titration design, sequential design, active control, early or late Phase

18. Compare to Optimal Designs
For a given model, optimal design allocate doses according to D-optimality
The most frequently used model is logistic
Another model is normal cdf (with parameters m and s)

19. Criteria for Comparison
Provide a steeper slope from placebo
Identify a minimum effective dose (MinED)

20. Simulation Procedure Ten and fifty obs. generated from each dose
Each obs. is normal with mean from model and standard deviation of 0.1 & 0.025
Slopes from placebo to each dose
Assuming minimum effect is 0.2, MinED is obtained from lower 95% confidence limit

21. Table 1. Treatment Group Means, Slopes From Placebo To Each Dose, And 95% Confidence Limits From Simulated Data Under Different Sample Sizes And Standard Deviations
Simulation based on Model: with q1 =1.565 and q2 =4.174
BDS doses: Low = 0.125, Medium = 0.375, High = 0.750
Optimal doses: Low = 0.082, Medium = 0.375, High = 0.668   
n/group = n/group = 50
std. dev. = std. dev. = std. dev. = std. dev. =0.25
BDS Opt BDS Opt BDS Opt BDS Opt.
Mean (Pcbo)
Mean (Low)
Lower L
Mean (Med)
Lower L
Mean (High)
Lower L

22. Table 2. Simulation Results with 3 Active Doses Using a Logistic Model to Compare D-optimal Dose Spacing and Binary Dose Spacing of ½ distance
  Low Medium High Overall MinED
B O B B BD2
B O O O OP2
B B O O OP2
B B O O BD1
B O O O BD1
B O O O BD1
B O O O OP2
B O O O OP2
O O B O BD2
O B O O OP2
O B O O OP2
O B O O OP2
B B O O OP2
B O O O OP1
B O O O BD1
O O O O OP1
O O O O OP1
O O O O OP1
O B O O OP1
O B O O OP1
O B O O OP1
O B O O OP1

23. Table 3. Simulation Results with 4 Active Doses Using a Normal Model to Compare D-optimal Dose Spacing and Binary Dose Spacing of ½ distance
  Low Low Med. Hign Med. High Overall MinED .
B B O O O BD1
O B O O O OP2
B B O O O OP2
O O B O O OP1
O O B O O BD2
B B B O O BD2
O O O O O OP1
O O O O O OP2
B O O O O OP2
O O O B B OP2
O O O O B OP1
O O O O O OP1
O O O O O BD3
O O O B B BD3
B B O B B OP2
O O O B B OP1
O O O O B OP1
B O O B O BD3
B O O B O BD3

24. Table 4. Treatment Group Means, Slopes From Placebo To Each Dose, And 95% Confidence Limits From Simulated Data When The Underlying Model Is Mis-specified
True Model: q1 = .38 and q2 = .32.
Using optimal design, this model was mis-specified as q1 = .32 and q2 = .38
BDS doses: Low = 0.125, Medium = 0.375, High = 0.750
Optimal doses (mis-specified model): Low = 0.010, Medium = 0.320, High = 0.630
(the correct doses should be Low = 0.119, Medium = 0.380, High = 0.641)
Simulation performed with n=20 per group and std. dev. = 0.25.
BDS Dose Allocation Optimal Dose Allocation
Point Point
Lower L estimate Upper L Lower L estimate Upper L
Mean (Pcbo)
Mean (Low)
Mean (Med)
Mean (High)
Slope Low
Slope Med
Slope High

25. Conclusion Assume MTD known and monotonic relationship
Intuitive and with wide applications
Model independent approach vs parametric optimality - Not much of a comparison
A general recommendation, not one size fits all

26. Analysis of Dose Response Studies
Multiple comparison adjustment
Placebo control, active control, or both
Dunnett’s method, Step down method
Linear, quadratic dose response
Minimum effective dose (MinED)

27. Analysis of Dose Response Studies
Drug safety in dose response studies
Estimation vs hypothesis testing
Exploratory vs confirmatory
Analysis of the entire database