Modified maximum contrast method to detect pharmacokineticsrelated genes in pharmacogenomics studies

1
Modified maximum contrast method to detect
pharmacokinetics-related genes in
pharmacogenomics studies
Y Sato1, N Laird1, R Kato3, K Nagashima2,3, C
Hamada3, I Yoshimura3, H Sakamoto2, T Yoshida2

• 1 Department of Biostatistics, Harvard School of
  Public Health, Boston, United States
• 2 Genetics Division, National Cancer Center
  Research Institute, Tokyo, Japan
• 3 Faculty of Engineering, Tokyo University of
  Science, Tokyo, Japan

Email ysato_at_hsph.harvard.edu
28th Annual Conference of International Society
for Clinical Biostatistics
2
Contents

• Background
• Purpose
• Proposed method
• Simulation study and result
• Questionnaire survey and result
• Discussion and conclusion

3
Interindividual variation in drug response

• The interindividual variability could be
  influenced by genetic and environmental factors.
• These factors affect drug absorption,
  distribution, metabolism and excretion, side
  effects and efficacy.

BMI, AGE, Gender, SNPs
Blood drug concentration (µg/mL)
0.5 1.0 1.5 2.0
2.5 3.0 3.5 Time(hr)
4
What is PK and ADME

• Pharmacokinetics (PK)
• The quantitative description of the disposition
  of a drug in the body or a body compartment over
  time
• ADME

• Absorption The process of substance entering the
  body
• Distribution The dispersion of substance
  throughout the fluids and tissues of the body.
• Metabolism the transformation of the substance
  and its daughter metabolites
• ExcretionThe elimination of the substance from
  the body

5
What are SNP?

• Mutation of a single nucleotide (A, T, G, C)
• Some can be associated with various phenotypic
  differences
• Drug response
• Disease susceptibility

6
Our Purpose

• To propose a statistical method for identifying
  the SNPs which relate to the pathways of drug
  metabolism by using PK data in clinical trial
• We apply the maximum contrast method (Yoshimura
  et al., (1997)) to detect biologically possible
  response patterns
• We propose a modified maximum contrast method
  under unbalanced sample size

7
Standard statistical analysis

• Pharmacologists have been testing the null
  hypothesis that there is no difference of the
  population mean of target PK parameters (AUC,
  Cmax, t1/2) among genotypes (AA, Aa, aa).
• H0 µAA µAa µaa, H1 µAA? µAa ? µaa
• e.g. Kruskal-Wallis test

8
Standard methods problem

• Biologically possible

• Not biologically possible

Elimination rate constant(Kel)
Standard method identified both profiles!
AA Aa aa
AA Aa aa
9
Contrast coefficient vector for identifying the
PK-related SNPs

• Biologically possible response patterns
• Not biologically possible response patterns

Linear
Dominant
Recessive









AA
AA
Aa
aa
AA
Aa
aa
Aa
aa









AA
AA
Aa
aa
Aa
aa
AA
Aa
aa
10
Maximum contrast method(Yoshimura et al., 1997)

• For detecting a monotonic dose-response
  relationship, maximum contrast method has been
  used in toxicological experiments clinical
  trials
• Maximum contrast statistics
• Vector of group means
• Contrast coefficient
  vector,

11
Modified maximum contrast method

• The sample size of each genotype is considerably
  unbalanced, because minor allele frequency is
  less than 20 in common diseases
• Modified maximum contrast statistics
• Multiplicity adjusted P-value for the probability
  distribution of under H0 by using a
  resampling technique

12
Framework of simulation study

• To compare performance among modified method and
  the original method in unbalanced sample size
• Statistical decision rule for identifying
  PK-related SNPs
• Positive judgment when two-sided P-value is less
  than 0.05
• Probability of identifying PK-related SNPs
• Power NT/N
• NT of rejection by hypothesis test
• N of simulation replications

13
Simulation condition

• Generate random numbers of PK data by each
  genotype
• ?0.0, 0.25, 0.5, 0.75, 1.0
• (CAA, CAa, Caa) (1,0,1), (-2,1,1), (-1,-1,2)
• Unbalanced sample size

14
Simulation result (?0.50)
15
Influence of Minor AF

• Power (?0.75, n300 )

16
Questionnaire survey on judgment for identifying
PK-related genes

• Object To compare statistical judgment with
  judgment by experts
• Respondent to a questionnaire 6 experts
  (pharmacologists, molecular biologists,
  geneticists)
• Method Based on the summary statistics (mean
  SD) and box-and-whisker plots of 13 SNPs in real
  data, 6 experts identify response pattern
• Evaluation Kendall's rank correlation coefficient

17
Questionnaire survey result 1

• Q1
• Judgment

AA Aa aa
18
Questionnaire survey result 2

• Q2
• Judgment

AA Aa aa
19
Discussion Conclusion

• PowerWhen the degree of unbalance was large
  (MAFlt 0.25), the power of the modified method
  was higher than the original method.
• JudgmentThe judgment of modified method was
  closer to the judgment of expert than original
  method
• Kendall's rank correlation coefficient
• Between expert and modified method 0.731
• Between expert and original method 0.423

The modified maximum contrast method is useful
in identifying PK-related SNPs
20
Acknowledgments

• National Cancer Center
• Research Institute Dr. S.Onami, Dr. O.Kawaguchi,
  Dr. M. Andoh and Dr. H.Totsuka
• Hospital Dr. H.Ueno, Dr. T.Okusaka and Dr.
  N.Saijo
• National Institute of Health Sciences
• Dr. N. Kaniwa, Dr. Y. Saito and Dr. J. Sawada
• Gunma University
• Dr. H. Sakai
• Hamano Statistical Analysis Ltd.
• Dr. T.Hamano
• This study was partially supported by
• the Program for Promotion of Fundamental Studies
  in Health Sciences of the National Institute of
  Biomedical Innovation of Japan.
• the Japanese Society of Clinical Pharmacology and
  Therapeutics.

21
Reference
1 Abelson RP, Tukey JW. Efficient utilization
of non-numerical information in
quantitative analysis General theory and the
case of simple order. The Annals of
Mathematical Statistics 1963 341347-1369.
2 Chirac J, Bush J et al.. Joint proclamation
by the heads of government of six countries
regarding the completion of the human genome
sequence 2003.
3 Genz, A. and Bretz, F. Numerical computation
of multivariate t-probabilities with
application to power calculation of multiple
contrasts. Journal of Statistical
Computation and Simulation 1999 63, 361-378.
4 Licinio, J., Wong, M. Pharmacogenomics The
search for individualized therapies.
2002WILEY.
5 Nishiyama, H., Yanagihara, H. and Yoshimura,
I. SAS/IML program for computing
probabilities related to maximum contrast
methods. JJB 2003 24, 57-70.
22
Reference
6 Ruberg SJ. Contrasts identifying the minimum
effective dose. Journal of the American
Statistical Association 1989 82816-822.
7 Stewart, H. and Ruberg, S. J. Detecting dose
response with contrasts. Statistics in
Medicine 200019, 913-921.
8 Wakana A, Yoshimura I, Hamada C. A method for
therapeutic dose selection in a phase
?clinical trial using contrast statistics.
Statistics In Medicine 2007 26498-511.
9 Westfall P. H. and Young S. S.
Resampling-based multiple testing Examples
and methods for p-value adjustment.Wiley, New
York 1993.
10 Yoshimura I , Wakana A, Hamada C. A
performance comparison of maximum
contrast methods to detect dose dependency.
Drug Information Journal 1997 31423--432.
23
Thank you for your kind attention!
???????????????
24
Discussion1

• Contrast coefficient vector
• Linear pattern (-1, 0, 1)
• When MAF was less than 0.33, the modified method
  has higher power than the original method.
• Dominant pattern (-1, -1, 2)
• The modified method has higher power than the
  original method in every case.
• Recessive pattern (-2, 1, 1)
• The modified method has lower power than the
  original method in every case.

25
Discussion 2

• Kendall's rank correlation coefficient
• Between expert and modified method 0.731
• Between expert and original method 0.423
• The judgment by the modified method better
  represented the one by experts

26
Simulation result (?0.0)
27
Simulation result (?1.0)
28
Simulation result (Recessive)

• Power ( (CXX, CXY, CYY)(-2, 1, 1) )

29
Geometric interpretation of Tr
Constant
?
When cos? is large, Y is goodness fit of C