SlideShare ist ein Scribd-Unternehmen logo
1 von 34
Efficient Design and Analysis of
 Combination Experiments to Improve
    Early Stage Clinical Development

            Hong-Bin Fang, Ph.D.

                 Division of Biostatistics
   University of Maryland Greenebaum Cancer Center
and Department of Epidemiology and Preventive Medicine
               Baltimore, MD 21201, USA
           Email: hfang@som.umaryland.edu


                             •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
•First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Outline
• Introduction
• Existing Methods for Combination Study Design
• Maximal Power Experimental Design
• Existing Analysis Methods for Synergy
• Statistical Analysis of Interaction Index
• Conclusion and Further Research



                           •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Introduction: Combination Therapy
  • Reduce the frequency of acquired resistance
  • Achieve greater efficacy with lower doses and reduced toxicity
  • Achieve enhanced potency (or sensitization) exploring synergistic ac-
    tivities
  • Provide a firmer basis for potential clinical trials


Joint action is divided into three types:
 1. Independent joint action
 2. Simple similar (additive) action
 3. Synergistic/antagonistic action

Refs.: Berenbaum MC. (1989). Pharmacological Reviews 41: 93-141.
       Greco WR, et al. (1995). Pharmacological Reviews 47, 331-385.
       Fitzgerald et al.(2006). Nature Chemical Biology 2, 458 - 466

                                                   •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Interaction Index
To assess the interaction of two drugs at the combination of (xA, xB ),
Berenbaum (1977) defined an interaction index (τ ),
                                          xA   xB
                                    τ=       +
                                          XA XB
XA and XB are the doses of drugs A and B that when administered alone
yield the same effect as does the combination (xA, xB ).
  • τ = 1, A and B are additive at (xA, xB );
    Loewe Independence (Additivity)
  • τ < 1, A and B are synergistic at (xA, xB );
  • τ > 1, A and B are antagonistic at (xA, xB ).


Refs.: Berenbaum MC. (1977). Clin. Exp. Immunol. 28: 1-18.
       Berenbaum MC. (1989). Pharmacological Reviews 41: 93-141.


                                                  •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Bliss Independence (Additivity)
• The joint effect of inhibitors A and B is the product of the effect of
  each
                             EAB = EA ∗ EB
• Assumption: Inhibitors can bind simultaneously and mutually nonex-
  clusively through distinct mechanism




                                     •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Fitzgerald et al. Nature 2006
                 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Statistical Approaches for Design
• Statistical (instead of mechanistic) evaluation is very valuable in
  clinical trials because it is impractical for a measure of success (such
  as synergism between two drugs) to change with every biochemical
  advance.
• Variation: the administration of precisely the same dose to aliquots
  (or virtually genetically identical animals) may result in different levels
  of dose effect.
• Sample Size: how many samples are needed, namely, how to identify
  doses in the combinations and how many replicates at each combina-
  tion and how to analyze the data produced




                                        •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Current Methods of Design
 • Equal regression lines (Finney, 1971)
 • An optimal design by fixing the total dose for
   specific models (Abdelbasit and Plackett, 1982)
 • Fixed ratio/ray design (Tallarida et al., 1992)
 • Checkerboard (Lattice) design (Martinez-Irujo et
   al., 1996)
 • A D-optimal design for in vitro combination
   studies in linear models (Greco et al., 1995) but
   n too large
Refs.: Finney DJ (1971). Probit Analysis. Cambridge University Press.
       Abdelbasit KM, Plackett RL (1982). Biometrics 38, 171-179.
       Greco WR, et al. (1995). Pharmcological Reviews 47, 331-385.
       Martinez-Irujo, et al.(1997). Biochemical Pharmacology 51, 635-644.
       Tallarida RJ, et al.(1997). Life Science 61, 417-425.
                                                    •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Fixed Ratio/Ray Design
• Assume that two drugs are synergistic, additive or antagonistic for all
  doses of a fixed ratio;
• Suboptimal dose allocations result in false synergistic combinations or
  miss an apparent interaction at a particular combination;
• Statistical power to detect additivity is undermined substantially.




                                      •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Maximal Power Design
                 Statistical Model
Single dose effect: y = fA(XA), y = fB (XB )
                                      −1
Potency of B relative to A: ρ(XB ) = fA fB (XB )/XB
The joint model at the combination x is assumed
     y = fA(xA + ρ(XB )xB ) + g(xA, xB ) + ε

 • g(xA, xB ) = 0, A and B are additive at x;
 • g(xA, xB ) > 0, A and B are synergistic at x;
 • g(xA, xB ) < 0, A and B are antagonistic at x.


Experimental design is based on testing the additive
action of drugs A and B H0 : g = 0 versus H1 :
g = 0.
                                  •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Statistical Inference
After transformation, the additive model
 y = fA(xA + ρ(XB )xB ) = (or ≈)g(z1) + g(z2)
              (i)    (i)
Let z(i) = (z1 , z2 )T , i = 1, 2, . . . , m,
y = (y11, · · · , y1k , · · · , ymk )T ,
                                          (i) (i)
Z: m × 2 matrix, its ith row: (g1(z1 ), g2(z2 ))
Then, under H0,
      yT (J − V )y/(m − 2)
  F = T                    ∼ Fm−2,m(k−1)
     y (I − J)y/(mk − m)
where U = I   1 k , V = U Z(Z T U T U Z)−1Z T U T , J = U (U T U )−1U T

                                     •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Statistical Inference
If z(1), . . . , z(m) are uniformly scattered in S, under
H1, the F statistic has a noncentral F -distribution
with degrees of freedom m − 2 and m(k − 1) and
the noncentrality parameter,
                          mk
                     δ= 2        g 2(z)dz,
                          σ S
is maximized. Thus, the power for detecting de-
partures from additivity of two drugs is maximized
when m mixtures z(1), . . . , z(m) are uniformly scat-
tered in S.
Refs.: Wiens (1991). Statistics & Probability Letters, 12: 217-221.
       Tan et al.(2003). Statistics in Medicine, 22: 2091-221.

                                                      •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Sample Size Determination
Given the type I error rate: α, power: 1 − β
the smallest meaningful difference: η
the measurement variation: σ 2
the sample sizes can be obtained from the noncen-
tral F -distribution function,
                ∞
                      e−δ/2(δ/2)k                              (m − 2)x
  P (F ≤ x) =                     P   Fm−2+2k,n−m ≤                                  ,
                           k!                                 m − 2 + 2k
                k=0


δ = nη 2/σ 2.

                                       •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Sample Size Calculation
                        α = 0.05, 1 − β = 0.8,
d = η 2/σ 2                         # of Replications n0
              1           2          3                 4                5               6
 d = 0.1      −           −          139(556)          87(435)          61(366)         46(322)
 d = 0.2      −           78(234)    42(168)           27(135)          19(114)         14(98)
 d = 0.3      107(214)    40(120)    21(84)            14(70)           10(60)          7(49)
 d = 0.4      68(136)     25(75)     14(56)            9(45)            3(18)           3 (21)
 d = 0.5      48(96)      18(54)     10(40)            6(30)            3(18)           3(21)
 d = 0.8      24(48)      9(27)      4(16)             3(15)            3(18)           3(21)
  d=1         18(36)      6(18)      3(12)             3(15)            3(18)           3(21)


                                         •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Experimental Design I
The sample sizes are based on the F-test to detect
departures from additive action with 80% power at
a significance level of 0.05
• Choose the dose range of significance: e.g., ED20-
  ED80 (based on pharmacology)
• Choose the meaningful difference in the dose-
  effect outcome to be detected: η
• Choose the number of replicates at each combi-
  nation
• The variance is estimated based on the pooled
  variations from the two single drug experiments.
                           •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Experimental Design II
• The additive model depends on the individual
  dose-response;
• Different individual dose-responses result in dif-
  ferent experimental designs;

We have considered three classes of drugs classified
by the shape of individual dose-response curves:
• both log-linear;
• both linear;
• linear + log-linear.

                           •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Log-linear + Log-linear
Let the dose-response of drugs A and B be
y(xA) = αA+βA log xA,       y(xB ) = αB +βB log xB
The potency ρ of B relative to A is
            βB /βA−1
ρ(xB ) = ρ0xB        , ρ0 = exp[(αB − αA)/βA].
The additive model at combination (xA, xB ):
      y(xA, xB ) = αA + βA log(xA + ρxB )
                                 (βB −βA)/βA
       ρ = ρ0   ρ−1x   A + xB                                 .

                           •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Log-linear + Log-linear
Ray Design: Using lattice points undermines the power to detect the
additivity;
Maximum Power Design: Using uniformly scattered points achieves
maximum power to detect departures from additivity.




                                    •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Results of Simulation Studies
         Number  of combinations                                         9                  16
     Maximum      Discrepancy (CL2)                                      0.003583           0.001190
     Power        Type I error                                           0.0501             0.0493
     Design       Power                                                  0.7949             0.8345
                  Discrepancy (CL2)                                      0.022719           0.017986
     Ray/Lattice Type I error                                            0.0532             0.0501
     Design       Power                                                  0.6540             0.4062
                  Discrepancy (CL2) average:                             0.035943           0.020812
                                    SD:                                  0.0173668          0.0147279
     Monte        Type I error      average:                             0.0503             0.0502
     Carlo                          SD:                                  0.00311            0.00227
     method       Power             average:                             0.5330             0.7439
                                    SD:                                  0.31456            0.26184
y = 20 log(z1 ) + 70 log(z2 ) + g(z1 , z2 ) + ε, g(z1 , z2 ) = 50(z1 − 2) sin(z1 ) cos(z2 ), ε ∼ N (0, σ 2 ) and
σ 2 = 250.
η 2 = 100, with 10,000 replications

                                                            •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: SAHA + Ara-C against K562




  y = 43.69 − 10.93 log(S), y is the viability




           y = 35.98 − 8.25 log(C)
  Shiozawa et al. (2009). CCR. 15:1698-1707
                           •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: SAHA + Ara-C against K562
      Mixtures SAHA and Ara-C for combination experiment

            Exper. SAHA Ara-C Exper. SAHA Ara-C
               #     (µM)    (µM)      #         (µM)         (µM)
               1     0.706   0.101     7         1.427        3.193
               2     3.035   0.099     8         3.402        2.681
               3     2.160   2.030     9         1.008        0.924
               4     0.635   5.122    10         3.139        0.781
               5     0.393   0.727    11         4.524        0.724
               6     0.118   2.684

dose range: ED20-ED80; η = 15%(viability); 5 replicates; σ 2 = 804.564
                                                         ˆ

                                      •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Analysis of Synergy
Estimation of the Interaction Index Surface
                                    (i)   (i)
 yij : the jth response at (xA , xB ),
                                                                    (i) (i)
 With the single dose-response curves, the interaction indexes at (xA , xB )
 are
                                   (i)               (i)
                                  xA                xB
                τij =                   +
                      exp{(yij − αA)/βA} exp{(yij − αB )/βB }
 j = 1, . . . , k, i = 1, . . . , m.

 The method of two-dimensional B-splines (thin plate splines) is employed
 to estimate the interaction index surface τ = h(xA, xB ),

 Ref.: Fang, et al. Stat. Medicine 2008




                                                •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
SAHA: Estimation of relative potency
                                                                  0.842
The potency of Vorinostat relative to Etoposide is ρ(XB ) = 0.368XB ,
which is non-constant and depends on dose.
The predicted additive model is

          y(xA, xB ) = 41.52 − 13.02 log(xA + ψ(xA, xB )xB ),

where ψ(xA, xB ) is determined by ψ(xA, xB ) = 0.368(ψ −1(xA, xB )xA +
xB )0.842.




                                      •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: SAHA + Ara-C against K562
                      Dose-response surface




Observations: 66(11 combinations with 5 replicates at each combination)
Maximum viability: 82.81%; Minimum viability: 17.72%; Mean: 30.67%;
                      Standard deviation: 13.40.
                Output: F9,55 = 8.14, p-value< 0.0001
                                      •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: SAHA + Ara-C against K562
      Interaction Index Surface




                  •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Linear + Log-linear
Let
z1 = xA,    z2 = βAξ(xA, xB )/[βB ψ(xA, xB )xB ],
then, the additive model becomes
        y(xA, xB ) = αA + βAz1 + βB z2.
The m experimental points should be uniformly
scattered in the tetragon,
  z1
     : a < αA + βAz1 + βB z2 < b, z1 > 0, z2 > 0
  z2
for given a and b.
                          •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: LY-168 with Sorafnib against WM164




       y = 111.85 − 9.56xA, y is the viability




               y = 101.91 − 31.17xB
                             •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
dose range: ED20-ED80; η = 15%(viability); 6 replicates; σ 2 = 1352.724
                                                         ˆ

                                      •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: LY-168 + Sorafnib against WM164
                      Dose-response surface




      Observations: 133(19 combinations with 6 replicates at each
  combination) Maximum viability: 93.58%; Minimum viability: 3.53%;
              Mean: 30.45%; Standard deviation: 25.347.
             Output: F17,114 = 162.9696, p-value< 0.0001
                                     •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Example: LY-168 + Sorafnib against WM164
         Interaction Index Surface




                     •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Follow-up in Early Clinical Stage Trials
Three cases classified by the shape of individual
dose-response curves considered:
• both log-linear (SAHA + Ara-C; Clini-
  cal trial ongoing);
• both linear (potentially resurrect a
  promising drug combination) ;
• linear + log-linear (In vivo experiment
  ongoing).


                          •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Conclusion
• Maximal power design produces data efficiently.
  It is optimal in that statistical power to detect
  departures from additivity is maximized.
• The F test can be used to test departures from
  additivity and the thin plate splines to estimate
  the interaction index surface effectively with data
  generated by the MP design.
• SYNSTAT R package at
  www.umgcc.org/research/biostatistics.htm

                           •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Acknowledgments
Ming Tan, PhD, UMGCC
Guo-Liang Tian, PhD, University of Hong Kong
Dr. Douglas D. Ross’s Lab, U Maryland School of Medicine
Dr. Wei Li’s Lab, U Tennessee College of Pharmacy
Dr. Pei Feng’s Lab, U Maryland Dental School
Dr. Peter Houghton, St Jude Children’s Research Hospital




                                     •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit

Weitere ähnliche Inhalte

Was ist angesagt?

17 ch ken black solution
17 ch ken black solution17 ch ken black solution
17 ch ken black solution
Krunal Shah
 
05 ch ken black solution
05 ch ken black solution05 ch ken black solution
05 ch ken black solution
Krunal Shah
 

Was ist angesagt? (20)

PG STAT 531 Lecture 6 Test of Significance, z Test
PG STAT 531 Lecture 6 Test of Significance, z TestPG STAT 531 Lecture 6 Test of Significance, z Test
PG STAT 531 Lecture 6 Test of Significance, z Test
 
One-way ANOVA for Randomized Complete Block Design (RCBD)
One-way ANOVA for Randomized Complete Block Design (RCBD)One-way ANOVA for Randomized Complete Block Design (RCBD)
One-way ANOVA for Randomized Complete Block Design (RCBD)
 
One-way ANOVA for Completely Randomized Design (CRD)
One-way ANOVA for Completely Randomized Design (CRD)One-way ANOVA for Completely Randomized Design (CRD)
One-way ANOVA for Completely Randomized Design (CRD)
 
Chapter4
Chapter4Chapter4
Chapter4
 
Student’s t test
Student’s  t testStudent’s  t test
Student’s t test
 
Student t t est
Student t t estStudent t t est
Student t t est
 
Chapter12
Chapter12Chapter12
Chapter12
 
Exploratory Data Analysis - Checking For Normality
Exploratory Data Analysis - Checking For NormalityExploratory Data Analysis - Checking For Normality
Exploratory Data Analysis - Checking For Normality
 
Particle filter
Particle filterParticle filter
Particle filter
 
Chapter8
Chapter8Chapter8
Chapter8
 
P G STAT 531 Lecture 8 Chi square test
P G STAT 531 Lecture 8 Chi square testP G STAT 531 Lecture 8 Chi square test
P G STAT 531 Lecture 8 Chi square test
 
17 ch ken black solution
17 ch ken black solution17 ch ken black solution
17 ch ken black solution
 
AnalysisOfVariance
AnalysisOfVarianceAnalysisOfVariance
AnalysisOfVariance
 
Chapter11
Chapter11Chapter11
Chapter11
 
Introduction tocausalinference april02_2020
Introduction tocausalinference april02_2020Introduction tocausalinference april02_2020
Introduction tocausalinference april02_2020
 
Probability Distribution
Probability DistributionProbability Distribution
Probability Distribution
 
International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)
 
Design of experiments(
Design of experiments(Design of experiments(
Design of experiments(
 
Bbs11 ppt ch11
Bbs11 ppt ch11Bbs11 ppt ch11
Bbs11 ppt ch11
 
05 ch ken black solution
05 ch ken black solution05 ch ken black solution
05 ch ken black solution
 

Ähnlich wie Fang

Integration of biological annotations using hierarchical modeling
Integration of biological annotations using hierarchical modelingIntegration of biological annotations using hierarchical modeling
Integration of biological annotations using hierarchical modeling
USC
 

Ähnlich wie Fang (20)

Probit and logit model
Probit and logit modelProbit and logit model
Probit and logit model
 
Statistics
StatisticsStatistics
Statistics
 
Basic Concepts of Experimental Design & Standard Design ( Statistics )
Basic Concepts of Experimental Design & Standard Design ( Statistics )Basic Concepts of Experimental Design & Standard Design ( Statistics )
Basic Concepts of Experimental Design & Standard Design ( Statistics )
 
Basic Concepts of Standard Experimental Designs ( Statistics )
Basic Concepts of Standard Experimental Designs ( Statistics )Basic Concepts of Standard Experimental Designs ( Statistics )
Basic Concepts of Standard Experimental Designs ( Statistics )
 
Methods for High Dimensional Interactions
Methods for High Dimensional InteractionsMethods for High Dimensional Interactions
Methods for High Dimensional Interactions
 
Introduction and crd
Introduction and crdIntroduction and crd
Introduction and crd
 
Randomized complete block_design_rcbd_
Randomized complete block_design_rcbd_Randomized complete block_design_rcbd_
Randomized complete block_design_rcbd_
 
Tugasan kumpulan anova
Tugasan kumpulan anovaTugasan kumpulan anova
Tugasan kumpulan anova
 
Anova by Hazilah Mohd Amin
Anova by Hazilah Mohd AminAnova by Hazilah Mohd Amin
Anova by Hazilah Mohd Amin
 
Statistical analysis by iswar
Statistical analysis by iswarStatistical analysis by iswar
Statistical analysis by iswar
 
Dr.Dinesh-BIOSTAT-Tests-of-significance-1-min.pdf
Dr.Dinesh-BIOSTAT-Tests-of-significance-1-min.pdfDr.Dinesh-BIOSTAT-Tests-of-significance-1-min.pdf
Dr.Dinesh-BIOSTAT-Tests-of-significance-1-min.pdf
 
Integration of biological annotations using hierarchical modeling
Integration of biological annotations using hierarchical modelingIntegration of biological annotations using hierarchical modeling
Integration of biological annotations using hierarchical modeling
 
Variance component analysis by paravayya c pujeri
Variance component analysis by paravayya c pujeriVariance component analysis by paravayya c pujeri
Variance component analysis by paravayya c pujeri
 
Anova; analysis of variance
Anova; analysis of varianceAnova; analysis of variance
Anova; analysis of variance
 
Analysis of variance anova
Analysis of variance anovaAnalysis of variance anova
Analysis of variance anova
 
Determination of sample size in scientific research.pptx
Determination of sample size in scientific research.pptxDetermination of sample size in scientific research.pptx
Determination of sample size in scientific research.pptx
 
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
 
2008 JSM - Meta Study Data vs Patient Data
2008 JSM - Meta Study Data vs Patient Data2008 JSM - Meta Study Data vs Patient Data
2008 JSM - Meta Study Data vs Patient Data
 
Sct2013 boston,randomizationmetricsposter,d6.2
Sct2013 boston,randomizationmetricsposter,d6.2Sct2013 boston,randomizationmetricsposter,d6.2
Sct2013 boston,randomizationmetricsposter,d6.2
 
2.0.statistical methods and determination of sample size
2.0.statistical methods and determination of sample size2.0.statistical methods and determination of sample size
2.0.statistical methods and determination of sample size
 

Mehr von University of Maryland Baltimore

Mehr von University of Maryland Baltimore (20)

Shu
ShuShu
Shu
 
Twaddell
TwaddellTwaddell
Twaddell
 
Sausville
SausvilleSausville
Sausville
 
Wang
WangWang
Wang
 
Penn
PennPenn
Penn
 
Scheibner
ScheibnerScheibner
Scheibner
 
Ross - ET Overview
Ross - ET OverviewRoss - ET Overview
Ross - ET Overview
 
Lapidus
LapidusLapidus
Lapidus
 
Meiller
MeillerMeiller
Meiller
 
Hosame Gcc09 Retreat 15min 100209
Hosame   Gcc09 Retreat 15min 100209Hosame   Gcc09 Retreat 15min 100209
Hosame Gcc09 Retreat 15min 100209
 
Edelman - Clinical Trials
Edelman - Clinical TrialsEdelman - Clinical Trials
Edelman - Clinical Trials
 
Fenselau
FenselauFenselau
Fenselau
 
Edelman- NCI
Edelman- NCIEdelman- NCI
Edelman- NCI
 
Dorsey
DorseyDorsey
Dorsey
 
Daniel
DanielDaniel
Daniel
 
Aurelian
AurelianAurelian
Aurelian
 
Badros
BadrosBadros
Badros
 
Coop
CoopCoop
Coop
 
Ma
MaMa
Ma
 
Swaan
SwaanSwaan
Swaan
 

Kürzlich hochgeladen

The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
heathfieldcps1
 

Kürzlich hochgeladen (20)

Unit-IV; Professional Sales Representative (PSR).pptx
Unit-IV; Professional Sales Representative (PSR).pptxUnit-IV; Professional Sales Representative (PSR).pptx
Unit-IV; Professional Sales Representative (PSR).pptx
 
Python Notes for mca i year students osmania university.docx
Python Notes for mca i year students osmania university.docxPython Notes for mca i year students osmania university.docx
Python Notes for mca i year students osmania university.docx
 
INDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptx
INDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptxINDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptx
INDIA QUIZ 2024 RLAC DELHI UNIVERSITY.pptx
 
Class 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfClass 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdf
 
Unit-V; Pricing (Pharma Marketing Management).pptx
Unit-V; Pricing (Pharma Marketing Management).pptxUnit-V; Pricing (Pharma Marketing Management).pptx
Unit-V; Pricing (Pharma Marketing Management).pptx
 
Measures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeMeasures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and Mode
 
Web & Social Media Analytics Previous Year Question Paper.pdf
Web & Social Media Analytics Previous Year Question Paper.pdfWeb & Social Media Analytics Previous Year Question Paper.pdf
Web & Social Media Analytics Previous Year Question Paper.pdf
 
Key note speaker Neum_Admir Softic_ENG.pdf
Key note speaker Neum_Admir Softic_ENG.pdfKey note speaker Neum_Admir Softic_ENG.pdf
Key note speaker Neum_Admir Softic_ENG.pdf
 
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxBasic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
 
This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.
 
General Principles of Intellectual Property: Concepts of Intellectual Proper...
General Principles of Intellectual Property: Concepts of Intellectual  Proper...General Principles of Intellectual Property: Concepts of Intellectual  Proper...
General Principles of Intellectual Property: Concepts of Intellectual Proper...
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
 
On National Teacher Day, meet the 2024-25 Kenan Fellows
On National Teacher Day, meet the 2024-25 Kenan FellowsOn National Teacher Day, meet the 2024-25 Kenan Fellows
On National Teacher Day, meet the 2024-25 Kenan Fellows
 
psychiatric nursing HISTORY COLLECTION .docx
psychiatric  nursing HISTORY  COLLECTION  .docxpsychiatric  nursing HISTORY  COLLECTION  .docx
psychiatric nursing HISTORY COLLECTION .docx
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy Consulting
 
Advanced Views - Calendar View in Odoo 17
Advanced Views - Calendar View in Odoo 17Advanced Views - Calendar View in Odoo 17
Advanced Views - Calendar View in Odoo 17
 
ICT role in 21st century education and it's challenges.
ICT role in 21st century education and it's challenges.ICT role in 21st century education and it's challenges.
ICT role in 21st century education and it's challenges.
 
ComPTIA Overview | Comptia Security+ Book SY0-701
ComPTIA Overview | Comptia Security+ Book SY0-701ComPTIA Overview | Comptia Security+ Book SY0-701
ComPTIA Overview | Comptia Security+ Book SY0-701
 
microwave assisted reaction. General introduction
microwave assisted reaction. General introductionmicrowave assisted reaction. General introduction
microwave assisted reaction. General introduction
 
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17  How to Extend Models Using Mixin ClassesMixin Classes in Odoo 17  How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
 

Fang

  • 1. Efficient Design and Analysis of Combination Experiments to Improve Early Stage Clinical Development Hong-Bin Fang, Ph.D. Division of Biostatistics University of Maryland Greenebaum Cancer Center and Department of Epidemiology and Preventive Medicine Baltimore, MD 21201, USA Email: hfang@som.umaryland.edu •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 2. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 3. Outline • Introduction • Existing Methods for Combination Study Design • Maximal Power Experimental Design • Existing Analysis Methods for Synergy • Statistical Analysis of Interaction Index • Conclusion and Further Research •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 4. Introduction: Combination Therapy • Reduce the frequency of acquired resistance • Achieve greater efficacy with lower doses and reduced toxicity • Achieve enhanced potency (or sensitization) exploring synergistic ac- tivities • Provide a firmer basis for potential clinical trials Joint action is divided into three types: 1. Independent joint action 2. Simple similar (additive) action 3. Synergistic/antagonistic action Refs.: Berenbaum MC. (1989). Pharmacological Reviews 41: 93-141. Greco WR, et al. (1995). Pharmacological Reviews 47, 331-385. Fitzgerald et al.(2006). Nature Chemical Biology 2, 458 - 466 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 5. Interaction Index To assess the interaction of two drugs at the combination of (xA, xB ), Berenbaum (1977) defined an interaction index (τ ), xA xB τ= + XA XB XA and XB are the doses of drugs A and B that when administered alone yield the same effect as does the combination (xA, xB ). • τ = 1, A and B are additive at (xA, xB ); Loewe Independence (Additivity) • τ < 1, A and B are synergistic at (xA, xB ); • τ > 1, A and B are antagonistic at (xA, xB ). Refs.: Berenbaum MC. (1977). Clin. Exp. Immunol. 28: 1-18. Berenbaum MC. (1989). Pharmacological Reviews 41: 93-141. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 6. Bliss Independence (Additivity) • The joint effect of inhibitors A and B is the product of the effect of each EAB = EA ∗ EB • Assumption: Inhibitors can bind simultaneously and mutually nonex- clusively through distinct mechanism •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 7. Fitzgerald et al. Nature 2006 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 8. Statistical Approaches for Design • Statistical (instead of mechanistic) evaluation is very valuable in clinical trials because it is impractical for a measure of success (such as synergism between two drugs) to change with every biochemical advance. • Variation: the administration of precisely the same dose to aliquots (or virtually genetically identical animals) may result in different levels of dose effect. • Sample Size: how many samples are needed, namely, how to identify doses in the combinations and how many replicates at each combina- tion and how to analyze the data produced •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 9. Current Methods of Design • Equal regression lines (Finney, 1971) • An optimal design by fixing the total dose for specific models (Abdelbasit and Plackett, 1982) • Fixed ratio/ray design (Tallarida et al., 1992) • Checkerboard (Lattice) design (Martinez-Irujo et al., 1996) • A D-optimal design for in vitro combination studies in linear models (Greco et al., 1995) but n too large Refs.: Finney DJ (1971). Probit Analysis. Cambridge University Press. Abdelbasit KM, Plackett RL (1982). Biometrics 38, 171-179. Greco WR, et al. (1995). Pharmcological Reviews 47, 331-385. Martinez-Irujo, et al.(1997). Biochemical Pharmacology 51, 635-644. Tallarida RJ, et al.(1997). Life Science 61, 417-425. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 10. Fixed Ratio/Ray Design • Assume that two drugs are synergistic, additive or antagonistic for all doses of a fixed ratio; • Suboptimal dose allocations result in false synergistic combinations or miss an apparent interaction at a particular combination; • Statistical power to detect additivity is undermined substantially. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 11. Maximal Power Design Statistical Model Single dose effect: y = fA(XA), y = fB (XB ) −1 Potency of B relative to A: ρ(XB ) = fA fB (XB )/XB The joint model at the combination x is assumed y = fA(xA + ρ(XB )xB ) + g(xA, xB ) + ε • g(xA, xB ) = 0, A and B are additive at x; • g(xA, xB ) > 0, A and B are synergistic at x; • g(xA, xB ) < 0, A and B are antagonistic at x. Experimental design is based on testing the additive action of drugs A and B H0 : g = 0 versus H1 : g = 0. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 12. Statistical Inference After transformation, the additive model y = fA(xA + ρ(XB )xB ) = (or ≈)g(z1) + g(z2) (i) (i) Let z(i) = (z1 , z2 )T , i = 1, 2, . . . , m, y = (y11, · · · , y1k , · · · , ymk )T , (i) (i) Z: m × 2 matrix, its ith row: (g1(z1 ), g2(z2 )) Then, under H0, yT (J − V )y/(m − 2) F = T ∼ Fm−2,m(k−1) y (I − J)y/(mk − m) where U = I 1 k , V = U Z(Z T U T U Z)−1Z T U T , J = U (U T U )−1U T •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 13. Statistical Inference If z(1), . . . , z(m) are uniformly scattered in S, under H1, the F statistic has a noncentral F -distribution with degrees of freedom m − 2 and m(k − 1) and the noncentrality parameter, mk δ= 2 g 2(z)dz, σ S is maximized. Thus, the power for detecting de- partures from additivity of two drugs is maximized when m mixtures z(1), . . . , z(m) are uniformly scat- tered in S. Refs.: Wiens (1991). Statistics & Probability Letters, 12: 217-221. Tan et al.(2003). Statistics in Medicine, 22: 2091-221. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 14. Sample Size Determination Given the type I error rate: α, power: 1 − β the smallest meaningful difference: η the measurement variation: σ 2 the sample sizes can be obtained from the noncen- tral F -distribution function, ∞ e−δ/2(δ/2)k (m − 2)x P (F ≤ x) = P Fm−2+2k,n−m ≤ , k! m − 2 + 2k k=0 δ = nη 2/σ 2. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 15. Sample Size Calculation α = 0.05, 1 − β = 0.8, d = η 2/σ 2 # of Replications n0 1 2 3 4 5 6 d = 0.1 − − 139(556) 87(435) 61(366) 46(322) d = 0.2 − 78(234) 42(168) 27(135) 19(114) 14(98) d = 0.3 107(214) 40(120) 21(84) 14(70) 10(60) 7(49) d = 0.4 68(136) 25(75) 14(56) 9(45) 3(18) 3 (21) d = 0.5 48(96) 18(54) 10(40) 6(30) 3(18) 3(21) d = 0.8 24(48) 9(27) 4(16) 3(15) 3(18) 3(21) d=1 18(36) 6(18) 3(12) 3(15) 3(18) 3(21) •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 16. Experimental Design I The sample sizes are based on the F-test to detect departures from additive action with 80% power at a significance level of 0.05 • Choose the dose range of significance: e.g., ED20- ED80 (based on pharmacology) • Choose the meaningful difference in the dose- effect outcome to be detected: η • Choose the number of replicates at each combi- nation • The variance is estimated based on the pooled variations from the two single drug experiments. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 17. Experimental Design II • The additive model depends on the individual dose-response; • Different individual dose-responses result in dif- ferent experimental designs; We have considered three classes of drugs classified by the shape of individual dose-response curves: • both log-linear; • both linear; • linear + log-linear. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 18. Log-linear + Log-linear Let the dose-response of drugs A and B be y(xA) = αA+βA log xA, y(xB ) = αB +βB log xB The potency ρ of B relative to A is βB /βA−1 ρ(xB ) = ρ0xB , ρ0 = exp[(αB − αA)/βA]. The additive model at combination (xA, xB ): y(xA, xB ) = αA + βA log(xA + ρxB ) (βB −βA)/βA ρ = ρ0 ρ−1x A + xB . •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 19. Log-linear + Log-linear Ray Design: Using lattice points undermines the power to detect the additivity; Maximum Power Design: Using uniformly scattered points achieves maximum power to detect departures from additivity. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 20. Results of Simulation Studies Number of combinations 9 16 Maximum Discrepancy (CL2) 0.003583 0.001190 Power Type I error 0.0501 0.0493 Design Power 0.7949 0.8345 Discrepancy (CL2) 0.022719 0.017986 Ray/Lattice Type I error 0.0532 0.0501 Design Power 0.6540 0.4062 Discrepancy (CL2) average: 0.035943 0.020812 SD: 0.0173668 0.0147279 Monte Type I error average: 0.0503 0.0502 Carlo SD: 0.00311 0.00227 method Power average: 0.5330 0.7439 SD: 0.31456 0.26184 y = 20 log(z1 ) + 70 log(z2 ) + g(z1 , z2 ) + ε, g(z1 , z2 ) = 50(z1 − 2) sin(z1 ) cos(z2 ), ε ∼ N (0, σ 2 ) and σ 2 = 250. η 2 = 100, with 10,000 replications •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 21. Example: SAHA + Ara-C against K562 y = 43.69 − 10.93 log(S), y is the viability y = 35.98 − 8.25 log(C) Shiozawa et al. (2009). CCR. 15:1698-1707 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 22. Example: SAHA + Ara-C against K562 Mixtures SAHA and Ara-C for combination experiment Exper. SAHA Ara-C Exper. SAHA Ara-C # (µM) (µM) # (µM) (µM) 1 0.706 0.101 7 1.427 3.193 2 3.035 0.099 8 3.402 2.681 3 2.160 2.030 9 1.008 0.924 4 0.635 5.122 10 3.139 0.781 5 0.393 0.727 11 4.524 0.724 6 0.118 2.684 dose range: ED20-ED80; η = 15%(viability); 5 replicates; σ 2 = 804.564 ˆ •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 23. Analysis of Synergy Estimation of the Interaction Index Surface (i) (i) yij : the jth response at (xA , xB ), (i) (i) With the single dose-response curves, the interaction indexes at (xA , xB ) are (i) (i) xA xB τij = + exp{(yij − αA)/βA} exp{(yij − αB )/βB } j = 1, . . . , k, i = 1, . . . , m. The method of two-dimensional B-splines (thin plate splines) is employed to estimate the interaction index surface τ = h(xA, xB ), Ref.: Fang, et al. Stat. Medicine 2008 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 24. SAHA: Estimation of relative potency 0.842 The potency of Vorinostat relative to Etoposide is ρ(XB ) = 0.368XB , which is non-constant and depends on dose. The predicted additive model is y(xA, xB ) = 41.52 − 13.02 log(xA + ψ(xA, xB )xB ), where ψ(xA, xB ) is determined by ψ(xA, xB ) = 0.368(ψ −1(xA, xB )xA + xB )0.842. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 25. Example: SAHA + Ara-C against K562 Dose-response surface Observations: 66(11 combinations with 5 replicates at each combination) Maximum viability: 82.81%; Minimum viability: 17.72%; Mean: 30.67%; Standard deviation: 13.40. Output: F9,55 = 8.14, p-value< 0.0001 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 26. Example: SAHA + Ara-C against K562 Interaction Index Surface •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 27. Linear + Log-linear Let z1 = xA, z2 = βAξ(xA, xB )/[βB ψ(xA, xB )xB ], then, the additive model becomes y(xA, xB ) = αA + βAz1 + βB z2. The m experimental points should be uniformly scattered in the tetragon, z1 : a < αA + βAz1 + βB z2 < b, z1 > 0, z2 > 0 z2 for given a and b. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 28. Example: LY-168 with Sorafnib against WM164 y = 111.85 − 9.56xA, y is the viability y = 101.91 − 31.17xB •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 29. dose range: ED20-ED80; η = 15%(viability); 6 replicates; σ 2 = 1352.724 ˆ •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 30. Example: LY-168 + Sorafnib against WM164 Dose-response surface Observations: 133(19 combinations with 6 replicates at each combination) Maximum viability: 93.58%; Minimum viability: 3.53%; Mean: 30.45%; Standard deviation: 25.347. Output: F17,114 = 162.9696, p-value< 0.0001 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 31. Example: LY-168 + Sorafnib against WM164 Interaction Index Surface •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 32. Follow-up in Early Clinical Stage Trials Three cases classified by the shape of individual dose-response curves considered: • both log-linear (SAHA + Ara-C; Clini- cal trial ongoing); • both linear (potentially resurrect a promising drug combination) ; • linear + log-linear (In vivo experiment ongoing). •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 33. Conclusion • Maximal power design produces data efficiently. It is optimal in that statistical power to detect departures from additivity is maximized. • The F test can be used to test departures from additivity and the thin plate splines to estimate the interaction index surface effectively with data generated by the MP design. • SYNSTAT R package at www.umgcc.org/research/biostatistics.htm •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
  • 34. Acknowledgments Ming Tan, PhD, UMGCC Guo-Liang Tian, PhD, University of Hong Kong Dr. Douglas D. Ross’s Lab, U Maryland School of Medicine Dr. Wei Li’s Lab, U Tennessee College of Pharmacy Dr. Pei Feng’s Lab, U Maryland Dental School Dr. Peter Houghton, St Jude Children’s Research Hospital •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit