Bayes-Verfahren in klinischen Studien
Dr. rer. nat. Joachim Gerß, Dipl.-Stat.
[email protected] of Biostatistics and Clinical Research
J. Gerß: Bayesian Methods in Clinical Trials 2
Popular Bayesian Methods in Clinical Trials
• Combination of knowledge from previous data or prior ‘beliefs’ with data from a current study
• Dose finding: Continual reassessment method
• Response-adaptive randomization
• Bayesian data monitoring / sequential stoppingin interim analyses
• Prediction of the study result using predictiveprobabilities
• Borrowing of information across relatedpopulations
Thomas Bayes(1702-1761)
J. Gerß: Bayesian Methods in Clinical Trials 3
Contents
1. Combination of prior beliefs with data from a study
2. Response-adaptive randomization
3. Borrowing of information across related populations
4. Summary and Conclusion
J. Gerß: Bayesian Methods in Clinical Trials 4
Contents
1. Combination of prior beliefs with data from a study
2. Response-adaptive randomization
3. Borrowing of information across related populations
4. Summary and Conclusion
J. Gerß: Bayesian Methods in Clinical Trials 5
1. Combination of prior ‘beliefs’ with data from a studyExample: Two groups, survival data
HR=2.227 95% CI 0.947-5.238p=0.0990
Classical „frequentist“ statistical analysis
Bayesian analysis
Survival after (years)1614121086420
Sur
viva
l rat
e
1,0
0,8
0,6
0,4
0,2
0,0
Group 2
Group 1
Normal-Normal Model
Let θ:=ln(Hazard Ratio)
Data Model: | ~ ,
with , total no. observed eventsPrior distribution: ~ μ ,
Posterior distribution:
| = ,
∝ ,
= | ∙
=> | ~∙ ∙
,
J. Gerß: Bayesian Methods in Clinical Trials 6
1. Combination of prior ‘beliefs’ with data from a study
HR=2.227 95% CI 0.947-5.238p=0.0990
Survival after (years)1614121086420
Sur
viva
l rat
e
1,0
0,8
0,6
0,4
0,2
0,0
Group 2
Group 1
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Data
1 2 3 4 5 6 87 9Hazardratio
Prior
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Prior+ Data= Posterior
95% Credible Interval: (1.074,4.285)
Example 1Bayesian analysisClassical „frequentist“
statistical analysis
J. Gerß: Bayesian Methods in Clinical Trials 7
1. Combination of prior ‘beliefs’ with data from a study
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Prior+ Data= Posterior
95% Credible Interval: (1.074,4.285)
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Prior+ Data= Posterior
95% Credible Interval: (1.822,4.264)
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Prior+ Data= Posterior
95% Credible Interval: (0.947,5.238)
Example 1 Example 2 Example 3„Noninformative“ prior
J. Gerß: Bayesian Methods in Clinical Trials 8
1. Combination of prior ‘beliefs’ with data from a studyFrequentist and Bayesian analysis
HR=2.227 95% CI 0.947-5.238p=0.0990
Classical „frequentist“ statistical analysis
Bayesian analysis
Survival after (years)1614121086420
Sur
viva
l rat
e
1,0
0,8
0,6
0,4
0,2
0,0
Group 2
Group 1
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Data
1 2 3 4 5 6 87 9Hazardratio
Prior
If p≤0.05 (<=> 1 Confidence Interval) => „significant“
Prob(HR>1|Data) ≥ 97.5% => „significant“
1 2 3 4 5 6 87 9
95% Confidence interval: (0.947,5.238)Hazardratio
Prior+ Data= Posterior
95% Credible Interval: (1.074,4.285)
J. Gerß: Bayesian Methods in Clinical Trials 9
1. Combination of prior ‘beliefs’ with data from a studyType I error and power
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
0.0
0.2
0.4
0.6
0.8
1.0
prior
hazard ratio >> favours experimental therapyfavours standard << |
n=100 events Bayesian power
Classical power
J. Gerß: Bayesian Methods in Clinical Trials 10
Contents
1. Combination of prior beliefs with data from a study
2. Response-adaptive randomization
3. Borrowing of information across related populations
4. Summary and Conclusion
J. Gerß: Bayesian Methods in Clinical Trials 11
2. Response-adaptive randomizationExample: Randomized 3-arm trial• Untreated patients aged ≥50 years with Adverse Karyotype Acute Myeloid Leukemia
• Random trt allocation: - Standard arm: Idarubicin + Ara-C (IA), with prob. π0
- 1st investigational arm: Troxacitabine + Ara-C (TA),with prob. π1
- 2nd investigational arm: Troxacitabine + Idarubicin (TI), with prob. π2
• Response: Time to Complete Remission (CR) within 49 days of starting treatment
• Denote mk := Current posterior median time to CR in arm k{0,1,2}qk := Current posterior Prob(mk<m0|data), k{1,2}r := Current posterior Pr(m1<m2|data)
• Algorithm
1. Initially, balanced randomization, with a probabilities π0 = π1 = π2 =1/3
2. Standard arm: Fixed probability π0 = 1/3, as long as all three arms remain in the trial.
3. If q1≥0.85 or q2≥0.85, drop standard arm, set randomization probabilities π1=r2, π2=1-r2
4. If q1<0.15 or r<0.15, drop arm 1 (TA), set randomization probabilities π2=q22, π0=1-q2
2
5. If q2<0.15 or r>0.85, drop arm 2 (TI), set randomization probabilities π1=q12, π0=1-q1
2
6. Otherwise assign investigational treatments with probabilities π1 q12 , π2 q2
2any
time
durin
gth
etri
al
J. Gerß: Bayesian Methods in Clinical Trials 12
2. Response-adaptive randomizationExample: Randomized 3-arm trial• Untreated patients aged ≥50 years with Adverse Karyotype Acute Myeloid Leukemia
• Random trt allocation: - Standard arm: Idarubicin + Ara-C (IA), with prob. π0
- 1st investigational arm: Troxacitabine + Ara-C (TA),with prob. π1
- 2nd investigational arm: Troxacitabine + Idarubicin (TI), with prob. π2
1 5 10 15 20 25 30 34
0.0
0.2
0.4
0.6
0.8
1.0
Pat.-No.
IA
TA
TI
Pro
b (T
reat
men
t ass
ignm
ent)
CR No CR Total
IA 10 (56%) 8 18
TA 3 (27%) 8 11
TI 0 (0%) 5 5
Fisher‘s exact test: p = 0.057
J. Gerß: Bayesian Methods in Clinical Trials 13
Contents
1. Combination of prior beliefs with data from a study
2. Response-adaptive randomization
3. Borrowing of information across related populations
4. Summary and Conclusion
J. Gerß: Bayesian Methods in Clinical Trials 14
3. Borrowing of information across related populationsBiomarkers in JIA
Gerß et al.Ann Rheum Dis 2012;71:1991–1997.
No. Flares / Patients (%)OR Fisher‘s Exact
TestMRP8/14 ≥690 ng/ml
MRP8/14 <690 ng/ml
All patients (n=188) 22 / 75 (29%) 13/ 113 (12%) 3.2 p=0.0036
Subgroup Oligoarthritis (n=86) 9 / 34 (26%) 8 / 52 (15%) 2.0 p=0.2700
Subgroup Polyarthritis (n=74) 11 / 25 (44%) 5 / 49 (10%) 6.9 p=0.0019
Subgroup Other (n=28) 2 / 16 (13%) 0 / 12 (0%) 4.3 p=0.4921
J. Gerß: Bayesian Methods in Clinical Trials 15
3. Borrowing of information across related populationsHierarchical model
Let := Observed ln(Odds Ratio) in subgroup i
Observed lnOR‘s: | ~ , , 1,2,3 with assumed known
Parameter model: ~ ,
Prior: f ∝ 1 (noninformative)
f ln ∝ 1 (noninformative)
MCMC Sampling (Gibbs sampler, Metropolis algorithm)
• Burn-in: n=5000
• No. samples: n=100000
• | , , , 1,2,3
• | , ,
J. Gerß: Bayesian Methods in Clinical Trials 16
3. Borrowing of information across related populationsBiomarkers in JIA: Results
Observed Odds Ratio
Fully Bayesian Estimator
0.25 0.5 1 2 5 10
SubgroupOligoarthritis(n=86)
SubgroupPolyarthritis(n=74)
SubgroupOther(n=28)
Pooled OR
J. Gerß: Bayesian Methods in Clinical Trials 17
3. Borrowing of information across related populationsBiomarkers in JIA: Results
Observed Odds Ratio
Empirical Bayes Estimator
Fully Bayesian Estimator
0.25 0.5 1 2 5 10
SubgroupOligoarthritis(n=86)
SubgroupPolyarthritis(n=74)
SubgroupOther(n=28)
Pooled OR
J. Gerß: Bayesian Methods in Clinical Trials 18
Contents
1. Combination of prior beliefs with data from a study
2. Response-adaptive randomization
3. Borrowing of information across related populations
4. Summary and Conclusion
J. Gerß: Bayesian Methods in Clinical Trials 19
4. Summary and ConclusionBayesian methods: Operating characteristics
1. Combination of prior beliefs with data from a study
Fully Bayesian final analysis usingposterior distribution• increased power• … but also increased type I error• Bayesian methods in a strict corset of
frequentist quality criteria are usually not much more powerful than classical frequentist methods.
J. Gerß: Bayesian Methods in Clinical Trials 20
4. Summary and ConclusionBayesian supplements
1. Combination of prior beliefs with data from a study
2. Response-adaptive randomization
# Bayesian interim analysis
Fully Bayesian final analysis usingposterior distribution
Fully Bayesian final analysis usingposterior distribution
Fully Bayesian final analysis usingposterior distribution
Bayesian „supplement“Final frequentist statistical analysis
UsesupplementaryBayesianinterimanalysisto determine the time to stop recruitmentFinal frequentist statistical analysis
J. Gerß: Bayesian Methods in Clinical Trials 21
Bayesian Methods in Clinical Trials
• Early phase clinical trials („in-house studies“ w/o strict regulatory control)
• Trials in small populations• Medical device trials• Exploratory studies
• Large scale confirmatory trials with strict type I error control
Use fully Bayesian approach, paying attention to• choose the appropriate model
carefully,• choose the inputted (prior)
information carefully and• check (classical) operating
characteristics (type I error, power)
Use of Bayesian supplements
J. Gerß: Bayesian Methods in Clinical Trials 22
Literature• Berry SM, Carlin BP, Lee JJ , Müller P (2010): Bayesian Adaptive Methods for Clinical Trials.
Chapman & Hall/CRC Biostatistics.
• Spiegelhalter DJ, Abrams KR, Myles JP (2004): Bayesian Approaches to Clinical Trials and Health-Care Evaluation. Wiley Series in Statistics in Practice.
• Tan SB, Dear KBG, Bruzzi P, Machin D (2003): Strategy for randomised clinical trials in rare cancers. British Medical Journal 327;47-49.
• Giles FJ et al. (2003): Adaptive randomized study of Idarubicin and Cytarabine versus Troxacitabine and Cytarabine versus Troxacitabine and Idarubicin in untreated patients 50 yearsor older with adverse karyotype Acute Myeloid Leukemia. Journal of Clinical Oncology 21(9);1722-1727
• Gerss J et al. (2012): Phagocyte-specific S100 proteins and high-sensitivity C reactive protein as biomarkers for a risk-adapted treatment to maintain remission in juvenile idiopathic arthritis: a comparative study. Annals of the rheumatic diseases 71(12);1991-1997.