Nguyen D. Nguyen, John A. Eisman and Tuan V. Nguyen Garvan Institute of Medical Research, Sydney, Australia Indirect comparison of anti-vertebral fracture efficacy among available drugs: A Bayesian meta-analysis

Treatment of Osteoporosis Several drugs currently available for the treatment of osteoporosis. Which treatment appropriate? –efficacy –safety –cost considerations.

Which drug is better? Decision makers need more head-to-head comparison trials. Reluctance of industry –Uncertainty of result –Sample size and costs Issues of head-to-head comparison trial –Active control: which one? –Margin of non-inferiority –Parameter of comparison

Required sample size for a noninferiority comparative trial FractureRatio of efficacy for new drug in standardcompared to standard drug drug (%) Sample size was based on 5% significant level with power of 80% (de Boo and Zielhuis, Statist. Med., 2004)

Current status No head-to-head comparative trial in the field of osteoporosis. Indirect comparison based on meta-analysis, a useful approach to make simultaneous inference on the relative efficacy of various drugs.

Indirect comparison Placebo Drug A Direct comparison Indirect comparison Direct comparison Placebo Drug B PlaceboDrug C RR A RR B RR C RR A RR B RR A RR C RR B RR C RR: relative risk

Traditional and Bayesian approaches TreatmentPlacebo Collect data (D) P(D|given hyphothesis) TraditionalBayesian Treatment Placebo Existing knowledge: effect/no effect/ no idea TreatmentPplacebo Collect data (D) P(Hyphothesis|given D) Prob. of observing data D given no effect Prob. of an effect given observed data D TreatmentPlacebo Hypothesis: effect TreatmentPlacebo Hypothesis: no effect

Traditional and Bayesian approaches TraditionalBayesian 1 Point estimate Rely on p-value significant difference not significant difference Relative risk Favours treatmentFavours placebo 0.6 Favours treatmentFavours placebo Posterior distribution Not rely on p-value, comprehensive information Relative risk confidence interval credible interval

Ln(OR A /OR B ) Favours A Favours B Favours A Favours B Favours A Favours B Analysis of “noninferiority” OR values between ( ) : A = B OR values to the left of 1.1 : A = acceptable OR values to the left of 1.0 : A > B OR values to the right of 1.0 : A < B Tolerence limit  10% (LnOR between -0,1 & 0.1 or OR between 0.9 & 1.1) (Diamond GA and Kaul S, 2007)

To compare the anti-vertebral fracture efficacy among therapies by using Bayesian approach.

Search strategy and study inclusion A systematic search, electronic resource: PubMed, Ovid, and Cochrane Controlled Trials. Inclusion criteria – Published randomized placebo-control trials (RCT) in English. – Postmenopausal women receiving osteoporotic therapy. – Vertebral fracture outcome.

Characteristics of studies IDDrugStudy (n)Sample (n)Duration (y) 1Alendronate Etidronate Risedronate Ibandronate HRT Strontium ranelate Raloxifene Calcitonin Fluoride PTH All drugs

Efficacy of individual drugs on vertebral fracture reduction Odds-ratio Favours treatment Favours placebo DrugOR(95% CrI) Alendronate0.50(0.35,0.74) Etidronate0.52(0.29,0.85) Risedronate0.57(0.40,0.85) Ibandronate0.50(0.25,0.96) HRT0.48(0.27,0.84) Strontium ranelate0.54(0.30,1.00) Raloxifene0.55(0.34,0.80) Calcitonin0.66(0.36,1.14) Fluoride0.59(0.35,0.96) PTH0.30(0.15,0.58) OR, odds-ratio CrI, credible interval Posterior distribution Point estimation

Probability that a drug reduces fracture risk at least 30% (Coefficient of Efficacy)

Indirect comparison the antivertebral fracture efficacy between Raloxifene and Etidronate Ln(OR Z /OR A ) Favours Ralox. Favour Etid. Tolerance limit was defined as ± 10%

Drug Probability (%) of the relative efficacy between two drugs Alendronate (1)Etidronate (2)Risedronate (3) EquivalentEtidronate (2)23 Risedronate (3)2421 Ibandronate (4) AcceptableEtidronate (2)53 Risedronate (3)4043 Ibandronate (4)53 66 BetterEtidronate (2)41 Risedronate (3)2731 Ibandronate (4) Comparison efficacy of vertebral fracture reduction among drug treatments No evidence that a drug was either much better or worse than the others.

Drug Probability (%) of the relative efficacy between two drugs (1)(2)(3)(4)(5)(6)(7)(8)(9) EquivalentStrontium r. (6) Raloxifene (7) Calcitonin (8) Fluoride (9) PTH AcceptableStrontium r. (6) Raloxifene (7) Calcitonin (8) Fluoride (9) PTH BetterStrontium r. (6) Raloxifene (7) Calcitonin (8) Fluoride (9) PTH (1), alendronate; (2), etidronate; (3), risedronate; (4), ibandronate; (5), HRT

Anti-vertebral fracture Most active therapies significantly reduced the risk of vertebral fx (vs. placebo), with variable magnitudes. Evidence of efficacy for calcitonin uncertain. No evidence for a drug either much better or worse than the others. Superiority for PTH vs. other drugs.

Bayesian approach Updating the existing knowledge or information. Comprehensive information. Clinically relevant inference of results.

Acknowledgements The Ho Chi Minh City Medical Association. Bridge Healthcare Co. Ltd., Australia for the untied sponsorship.

Thank you!

Anti-vertebral fracture Most active therapies significantly reduced the risk of vertebral fx (vs. placebo), with variable magnitudes: bisphosphonates, raloxifene, HRT, fluoride, strontium ranelate and PTH. Evidence of effect for calcitonin uncertain. No evidence for a drug neither much better nor worse than the others. Superiority for PTH vs. other drugs. Probabilities that PTH better than alendronate: 0.88, etidronate: 0.86; risedronate: 0.93, HRT: 0.91, and raloxifene: 0.94.

Indirect comparison –Not necessary, because “equivalence” depends on the tolerance limit. Drug ADrug B Drug C = Drug B = Drug A Drug C = ? Drug ADrug B Drug C > Drug B > Drug ADrug C > ? – Not necessary, because of different populations.

Anti-vertebral fracture Current therapies were efficacious in reducing vertebral fracture risk and their effect sizes were comparable. However, PTH appears to have higher anti-vertebral efficacy that all other treatments.

Traditional and Bayesian approaches A/placeboB/placebo Hypothesis: no effect A/placeboB/placebo Collect data P(D|given hyphothesis) TraditionalBayesian A/placeboB/placebo Hypothesis: effect A/placeboB/placebo Collect data P(Hyphothesis|given D) Prob. of observing data D given no effect Prob. of an effect given observed data D

Bayesian approach: Updating information P(H|D) Prior information Current likelihood (Observed data) = x Prior information:  Probability of effect based on existing knowledge.  Vague prior, or no difference/ no effect between drug and placebo or between two drugs. Posterior distribution

Posterior distribution of efficacy of drug therapy on vertebral fracture reduction AgentOR(95% CrI)Probability that OR ≤ Alendronate0.50(0.35,0.74) Etidronate0.52(0.29,0.85) Risedronate0.57(0.40,0.85) Ibandronate0.50(0.25,0.96) Zoledronic acid0.28(0.15,0.50) HRT0.48(0.27,0.84) Strontium ranelate0.54(0.30,1.00) Raloxifene0.55(0.34,0.80) Calcitonin0.66(0.36,1.14) Fluoride0.59(0.35,0.96) PTH0.30(0.15,0.58)

Large sample size is required! Consider a head-to-head comparison trial: Margin of inferiority = 20% (e.g. 20% worse than the control is ok) Background rate = 10% Drug would be equivalent if incidence <12% Power = 90%, significance level = 5% 5134 subjects per group!

Requested sample size for a comparative trial Fracture rateDifferences in efficacy from control drug (%) in treatedpower=80%power=90% patients (%) Sample size was base on 5% significant level (Kanis JA et al, 2002)

Probability (%) of the relative efficacy between two drugs AlendronateEtidronateRisedronateIbandronate Etidronate vs.equivalent23 better41 acceptable53 Risedronate vs.equivalent2421 better2731 acceptable4043 Ibandronate vs.equivalent better acceptable53 66 Zoledronic acidequivalent88410 vs.better acceptable Comparison efficacy of vertebral fracture reduction among drug treatments

Probability (%) of the relative efficacy between two drugs (1)(2)(3)(4)(5)(6)(7)(8)(9)(10) Stront. ranelate (7) vs.p p p Raloxifene (8) vs.p p p Calcitonin (9) vs.p p p Fluoride (10) vs.p p p PTH vs.p p p