1. Pharmacogenetics of response to antiresorptive therapy: Vitamin D receptor gene Tuan V. Nguyen, Associate Professor John A. Eisman, Professor and Director Bone and Mineral Research Program Garvan Institute of Medical Research Sydney, Australia

2. Variability in BMD = 0.12 x Mean Variability in  BMD = 3 x Mean Nguyen et al, JBMR 1999 Average rate of BMD loss: -0.6  1.8 %/yr Adapted from Cummings et al, JAMA 1998 Variability in response to therapy: 1-3 x Mean  BMD Osteoporosis heterogenous pathophysiological mechanisms and response to therapy

3. Individual vs average Clinical –efficacy and tolerance –Duration –new pharmacologic targets Theoretical –genetics of BMD –genetics of BMD change –environmental factors Available data genetic polymorphisms and response to antiresorptive therapy

4. Genetics of BMD and body composition Nguyen, et al, Am J Epidemiol 1998

5. VDR genotype and BMD VDR genotype and osteocalcin levels (PNAS, 1992) VDR genotype and BMD (Nature, 1994) Contentious association Meta-analysis: 15 cross-sectional, cohort studies Bayesian modelling

6. VDR genotype and lumbar spine BMD

7. Pooled effects of VDR genotype on BMD: Bayesian analysis Overall difference: 14.7 (95% CI: 0.8 to 42.3) mg/cm 2 Overall difference: 5.8 (95% CI: -6.5 to 18.0) mg/cm 2 BsmI b allele associated with higher BMD

8. Model of drug response Drug response Adverse reaction Drug effect Activity of other biological systems Target responsiveness Drug concentration at target Drug concentration at other biological systems Responsiveness at other biological systems Other predisposition Adapted from Meisel, et al. J Mol Med 2003

9. Heritability of BMD change 21 MZ and 19 DZ twin pairs over 3 years Changes in lumbar spine BMD: rMZ = 0.93 vs rDZ = 0.51 (Kelly et al. JBMR 1993; 8:11-7) 25 MZ and 21 DZ male twin pairs over 14 years Changes in distal radius BMD: rMZ=0.61, rDZ=0.41 (NS) (Christian, et al. 1989)

10. VDR genotype and BMD change Significant associationNo significant association Rapuri, J Steroid Biochem & Mol Biol 2004 Gomez, Osteoporosis Int 1999 Guardiola, Ann Int Med 1999 Gough, J Rheumatol 1998 Deng, Hum Genet 1998 Zmuda, JBMR 1997 Ferrari, Lancet 1995 Krall, JBMR 1995 Gunnes, JCEM 1997 Garnero, JBMR 1996 Hansen, Bone 1998 Publication bias? In “positive” studies, BsmI b allele associated with lesser loss or greater increase in BMD

11. Inter-subject variability in response to antiresorptive therapy Adapted from Cummings et al, JAMA 1998 Placebo: n=2218, mean change in LSBMD: 1.5 ± 8.1 % Alendronate: n=2214, mean change in LSBMD: 8.3 ± 7.8 %

12. Pharmacogenetics of response to antiresorptive treatments Few studies Candidate gene approach

13. VDR genotypes and response to Raloxifene Rx n=66 osteoporotic women; duration of Rx: 1 yr Palomba et al. Human Reprod 2003; 18:192-8 BMDBone turnover markers

14. VDR genotypes and response to Alendronate Rx n=68 osteoporotic women; duration of Rx: 1 yr Palomba et al. Clin Endocrinol 2003; 58:365-71 BMD Bone turnover markers

15. VDR genotype and BMD response to treatment ALNRLXALN+RLX Adapted from Palomba et al. Clin Endocrinol 2003; Hum Reprod 2003; and Palomba et al, OI 2005 (Epub). HRTALN+HRT

16. Response to antiresorptive therapy is multifactorial (VDR genotypes explained 5-10% of the variability) Genetic factors and response to antiresorptive therapy SNPs profile could allow individualization of treatment Issues of study design and interpretation Bayesian decision approach

17. SNP association studies: Bayesian approach to decision Alternatives 1. Abandon study 2. Continue data collection 3. Evidence strong enough for molecular exploration Rationale for decision True positive assoc. / False positive assoc. = 20/1 (NOT the same as p-value)

18. A hypothetical scenario 20 SNPs (out of 1000 SNPs) are actually associated with BMD response to Rx Study power = 80% (i.e., type II error = 20%) Type I error = 5% Finding: Significant association for 1 SNP (P = 0.05) What is the probability that there is indeed an association?

19. 20 SNPs involved; Power = 80%; False +ve = 5% 1000 SNPs Association (n=20) No association (n=980) Significant N=16 Non- significant Significant (n=49) Non- significant True positive / False positive = 16/49 P(True association | Significant result) = 16/(16+49) = 25%  =5% power=80%

20. The need for lower P-value About 25% of all findings with “p<0.05” should, if viewed in a scientifically agnostic light, properly be regarded as nothing more than chance findings (1). Proportion of significant associations depends on : –p-value, –overall proportion of hypotheses being tested are true –statistical power For a ratio (true +ve) / (false +ve) association = 20:1, p-value should be lowered by 400 times For a ratio (true +ve) / (false +ve) association = 50:1, p-value should be lowered by 1000 times (1) J Berger (1987); R Matthews (2001)

21. Bayesian resolution of conflicting finding Change in LSBMD in response to ALN Rx: bb vs BB genotypes Palomba 2003, 2005 P(bb-BB>3%) = 0.91 Marc OI 1999 P(bb-BB>3%) = 0.01 P(bb-BB>3%) = 0.73

22. Genetic markers could allow identification of those more or less likely to –fracture –respond to a specific treatment –suffer side effects from a specific treatment With cost-benefits in relation to intervention, Bayesian method offers a powerful approach to individualise inference

23. Acknowledgments Nguyen D. Nguyen Garvan Institute of Medical Research Regia Congressi Organizing C’tee

24. Reserved slides

25. Misunderstanding of P-value Bisphosphonate treatment was associated with a 5% increase in BMD compared to placebo (p<0.05) 1.It has been proved that bisphosphonate is better than placebo? 2.If the treatment has no effect, there is less than a 5% chance of obtaining such result 3.The observed effect is so large that there is less than 5% chance that the treatment is no better than placebo 4.I don’t know

27. P value is NOT the likelihood that findings are due to chance the probability that the null hypothesis is true given the data P-value is 0.05, so there is 95% chance that a real difference exists With low p-value (p < 0.001) the finding must be true The lower p-value, the stronger the evidence for an effect

28. P-value Grew out of quality control during WWII Question: the true frequency of bad bullets is 1%, what is the chance of finding 4 or more bad bullets if we test 100 bullets? Answer: With some maths (binomial theorem), p=2% So, p-value is the probability of getting a result as extreme (or more extreme) than the observed value given an hypothesis

29. Process of Reasoning The current process of hypothesis testing is a “proof by contradiction” If the null hypothesis is true, then the observations are unlikely. The observations occurred ______________________________________ Therefore, the null hypothesis is unlikely If Tuan has hypertension, then he is unlikely to have pheochromocytoma. Tuan has pheochromocytoma ______________________________________ Therefore, Tuan is unlikely to have hypertension

30. What do we want to know? Clinical P(+ve | Diseased): probability of a +ve test given that the patient has the disease P(Diseased | +ve): probability of that the patient has the disease given that he has a +ve test Research P(Significant test | No association): probability that the test is significant given that there is no association P(Association | Significant test): probability that there is an association given that the test statistic is significant

31. Diagnostic and statistical reasoning DiagnosisResearch Absence of diseaseThere is no real difference Presence of diseaseThere is a difference Positive test resultStatistical significance Negative test resultStatistical non-significance Sensitivity (true positive rate) Power (1-  ) False positive rateP-value Prior probability of disease (prevalence) Prior probability of research hypothesis Positive predictive valueBayesian probability

32. For a given sample size, posterior probability increases with p-value

33. Distribution of sample sizes Ioannidis et al, Trends Mol Med 2003

34. Distribution of effect sizes Ioannidis et al, Trends Mol Med 2003

35. Correlation between the odds ratio in the first studies and in subsequent studies Ioannidis et al, Nat Genet 2001

36. Evolution of the strength of an association as more information is accumulated Ioannidis et al, Nat Genet 2001

37. Predictors of statistically significant discrepancies between the first and subsequent studies of the same genetic association PredictorOdds ratio – univariate analysis Odds ratio – multivariate analysis Total no. of studies (per association) 1.17 (1.03, 1.33)1.18 (1.02, 1.37) Sample size of the first study 0.42 (0.17, 0.98)0.44 (0.19, 0.99) Single first study with clear genetic effect 9.33 (1.01, 86.3)NS Ioannidis et al, Nat Genet 2001