Microsoft powerpoint - ivanova_iwsm_2013.ppt [compatibility mode]
Treatment Selection with the Sequential Parallel Comparison Design Anastasia Ivanova University of North Carolina Chapel Hill aivanova@bios.unc.edu
• “The fact that an increasing number of medications are unable to beat
sugar pills has thrown the industry into crisis… Some products that have been on the market for decades, like Prozac, are faltering in more recent follow-up tests. It's not that the old meds are getting weaker… It's as if the placebo effect is somehow getting stronger” (Silberman, 2009). Yellow pills make the most effective antidepressants, like little doses of pharmaceutical sunshine. Red pills can give you a more stimulating kick More is better Placebos taken four times a day deliver greater relief than those taken twice daily. Data from 86 major depressive disorder trials submitted to FDA during 1986 to 2008 (Khin et al., 2011) + US trials - Non US trials + US trials - Non US trials
The Sequential Parallel Comparison Design (SPCD)
Response SPCD (Fava, et al. 2003) • After parallel first stage, re-study the patients who do not respond to placebo Efficacy analysis includes data from both stages
The Sequential Parallel Comparison Design (SPCD)
q1 q Response p1 p 0 p1 q1 p2 q2 0 p1 q1 p2 q2 1 p1 q1 p2 q2 Response HIGH DOSE HIGH DOSE 4 week trt 4 week trt
Advantages • We are able to estimate drug and placebo response in both overall population and placebo non-responders to aid the planning of future studies
• Increase in power for any given sample size
– From potentially larger effect size in placebo non-
Disadvantages of the SPCD for dose-finding
Disadvantages • Longer trial duration for individual subjects Two-stage, adaptive design
posterior probability of that dose being
• Performs better than a single stage and
Combining SPCD with 2-stage dose-finding
q q2 Response δ p δ p2 HIGH DOSE HIGH DOSE p1 p2
Testing for SPCD treatment effect of selected
0 p1 q1 p2 q2 H : δ ∩ δ p1 q1 p2 q2 ∩ δ δ 0 p1 q1 p2 q2 p1 q1 p2 q2
Testing for SPCD treatment effect of selected
• Multiplicity of treatments and the selection processes in stage 1?• Multiple stages?
Hochberg and weighted Z statistics
o Hochberg’s method (1988) to obtain adjusted p-value for
o Similarly Hochberg, then compute ܼ2 from adjusted p-value
• After both stages use weighted Z statistics (Liu et al., 2012)
• No closed form for sample size for Hochberg
o Stage 1: High dose + Low dose vs. placebo
o Stage 2: High dose + Low dose vs. placebo + 0.5 δ ) = ∩ (π p1 p2 q2 p1 - π) δp2 q2 where π is the proportion assigned to the high dose in stage 2 + 0.5 δ ) = ∩ (π p1 p2 q2 p1 - π) δp2 q2
can use sample size formula from Liu et al. (2012)
or score test formula (Ivanova et al., 2012) athttp://www.rctlogic.com/
Approximates required sample size with Hochberg well when
otherwise Hochberg needs less sample size
p = 0.4, q1 p2 q2 = 0.1, δ = 0.75
retention rate s = 0.9allocation 3:1:1 (that is, allocation prop to placebo = 0.6)w = 0.6one-sided 0.05 testpower of 80%
π = ?, depends on sample size and true parameters
Step 1. With the initial value π = 0.75, the score test formula yields the total sample size of n = 189.
Step 2. The total sample size of n = 189 yields 37 patients allocated to each of the two doses in stage 1.
37 patients per dose and response rates 0.4 and 0.75×0.4 = 0.3, yield π = 0.81
Step 1 (repeated). π = 0.81 yields the total sample size of
Example: Alkermes SPCD Dose-Finding Study
• ALKS 5461, a novel opioid modulator, in patients with major
depressive disorder and inadequate response to standard therapies
- Hamilton Depression Rating Scale (HAM-D17)
- Montgomery-Åsberg Depression Rating Scale (MADRS)- Clinical Global Impression (CGI-S)
• ALKS 5461 was significantly better than placebo based on the
n = 142 Response HIGH DOSE HIGH DOSE 4 week trt 4 week trt
Fava M, Evins A, Dorer D, Schoenfeld D (2003). The problem of the placebo response in
clinical trials for psychiatric disorders: culprits, possible remedies and a novel study
design approach. Psychotherapy and Psychosomatics 72, 115-127.
Fava M (2013). Results from Phase 2 clinical study of ALKS 5461 in major depressive
disorder. 53 rd Annual NCDEU Meeting, Hollywood, FL.
Ivanova, A, Qaqish, B, Schoenfeld D (2011). Sample size and power calculations for the
sequential parallel design. Statistics in Medicine 30, 2793–2803.
Ivanova A, Xiao C, Tymofyeyev, Y (2012). Two-stage designs for Phase 2 dose-finding
trials. Statistics in Medicine 31, 2872–2881.
Liu Q, Lim P, Singh J, Lewin D, Schwab B, Kent J. Doubly randomized delayed-start design
for enrichment studies with responders or nonresponders. Journal of Biopharmaceutical
Silberman, S Placebos are getting more effective. Drugmakers are desperate to know
why. (2009). Wired Magazine http://www.wired.com/wired/issue/17-09
Wang, J.J., Ivanova, A. Dose finding with the sequential parallel comparison design.
Quelle: http://www.naturepower.ch/blog/ Medikamente schädigen Ihren Vitalstoffhaushalt Wussten Sie, dass schon ein ganz normales Kopfschmerzmittel dafür sorgen kann, dass der Vitalstoff-haushalt in Ihrem Körper aus den Fugen gerät? Wie etliche andere Medikamente nämlich können auch Kopfschmerzmittel zu einer verstärkten Vitalstoff-Ausscheidung und einer verringerten Vitalstoffauf-na
Shahzad_Naseem@yahoo.com director.cssp@pu.edu.pk Personal Information Present Employment Centre of Excellence in Solid State Physics Lahore-54590, Pakistan Tel: +92 42 9923 1136 References Microelectronics Research Centre, Cavendish Laboratories Device Development Group, Applied Sciences Branch 1617 Cole Boulevard, Golden, Colorado 80401, USA Director, Spintronics & Nano
Treatment Selection with 
• “The fact that an increasing number of medications are unable to beat
sugar pills has thrown the industry into crisis… Some products that have 
Data from 86 major depressive disorder trials submitted to 
+ US trials