a bayesian approach for event predictions in clinical trials...
TRANSCRIPT
![Page 1: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/1.jpg)
A Bayesian approach for event predictions in clinical trials with time-to-event outcomes
Marine Antigny
![Page 2: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/2.jpg)
Capitalist Slides
1
Context
1
Modelisation
3
Case Study
4
Simulation Study
5
Algorithm
2
Summary
![Page 3: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/3.jpg)
1 - Context
![Page 4: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/4.jpg)
Capitalist Slides
2
In event driven trials, statistical analyses are planned at a pre specified number of events ( )X
Context
![Page 5: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/5.jpg)
Capitalist Slides
2
In event driven trials, statistical analyses are planned at a pre specified number of events ( )X
Context
Main objective: predict the date of the event with an acceptable prediction intervalXth
![Page 6: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/6.jpg)
Capitalist Slides
2
In event driven trials, statistical analyses are planned at a pre specified number of events ( )X
Context
Main objective: predict the date of the event with an acceptable prediction intervalXth
3 identified methods:• Emilia Bagiella and Daniel F Heitjan. Predicting analysis times in randomized clinical trials.
Statistics in medicine, 20(14):2055-2063, 2001 • Gui-shuang Ying and Daniel F Heitjan. Weibull prediction of event times in clinical trials.
Pharmaceutical Statistics: The Journal of Applied Statistics in the Pharmaceutical Industry, 7(2):107-120, 2008
• Bayesian approach proposed by Servier
![Page 7: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/7.jpg)
2 - Algorithm
![Page 8: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/8.jpg)
Capitalist Slides
3Algorithm
Population at the cut-off
Patient who have experienced the event
Drop-out / Lost to follow-up patient
Patient still in follow-up
Patient waiting for recruitment
![Page 9: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/9.jpg)
Capitalist Slides
3Algorithm
Model 1: Time-to-enrollment
Population at the cut-off
Patient who have experienced the event
Drop-out / Lost to follow-up patient
Patient still in follow-up
Patient waiting for recruitment
![Page 10: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/10.jpg)
Capitalist Slides
3Algorithm
Model 1: Time-to-enrollment
Model 2: Time-to-event
Model 3: Time-to-censorship
Population at the cut-off
Patient who have experienced the event
Drop-out / Lost to follow-up patient
Patient still in follow-up
Patient waiting for recruitment
![Page 11: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/11.jpg)
Capitalist Slides
3Algorithm
Model 1: Time-to-enrollment
Model 2: Time-to-event
Model 3: Time-to-censorship
Population at the cut-off
X 1,000Patient who have experienced the event
Drop-out / Lost to follow-up patient
Patient still in follow-up
Patient waiting for recruitment
![Page 12: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/12.jpg)
3 - Modelisation
![Page 13: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/13.jpg)
Capitalist Slides
4Modelisation
MDL 1
MDL 3
MDL 2
Time-to-enrollment
Time-to-event
Time-to-censorship
![Page 14: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/14.jpg)
Capitalist Slides
5ModelisationTime-to-enrollment
Notations: : inter-arrival times = time elapsed between the recruitment of two consecutive patients
: time of the cut-off (enrollment period observed): number of patients enrolled by the time of the cut-off
iattcN (tc)
Statistical model: iat |μ ∼ Exp (μ)
Prior: μ ∼ Γ (aμ, bμ)Posterior: μ | iat ∼ Γ (N (tc) + aμ, tc + bμ)
Time-to-enrollment ∼ HPP (μ)
![Page 15: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/15.jpg)
Capitalist Slides
6ModelisationTime-to-event & Time-to-censorship
Notations: : time-to-event (resp. time-to-censorship) in treatment arm Xj j
Calculation of the posterior distribution
Bagiella & Heitjan
Likelihood
Method
Prior distribution
Analytical
λj ∼ Γ (Aj, Bj)
Xj |λj ∼ Exp (λj)
![Page 16: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/16.jpg)
Capitalist Slides
6ModelisationTime-to-event & Time-to-censorship
Notations: : time-to-event (resp. time-to-censorship) in treatment arm Xj j
ρj ∼ Γ (aρj, bρj)
λj ∼ Γ (aλj, bλj)
Calculation of the posterior distribution
Bagiella & Heitjan
Likelihood
Method
Prior distribution
Analytical MCMC
Ying & Heitjan
λj ∼ Γ (Aj, Bj)
Xj |λj ∼ Exp (λj) Xj |ρj, λj ∼ Weibull (ρj, λj)
![Page 17: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/17.jpg)
Capitalist Slides
6ModelisationTime-to-event & Time-to-censorship
Notations: : time-to-event (resp. time-to-censorship) in treatment arm : covariate: parameter associated to the covariate
Xj jXi ith
βi ith
ρ ∼ Γ (aρ, bρ)λ0 ∼ Γ (aλ, bλ)
ρj ∼ Γ (aρj, bρj)
λj ∼ Γ (aλj, bλj)
λ = λ0e ∑ βiXi
Calculation of the posterior distribution
Bagiella & Heitjan
Likelihood
Method
Prior distribution
Analytical MCMC MCMC
Ying & Heitjan Servier
X |ρ, λ ∼ Weibull (ρ, λ)
λj ∼ Γ (Aj, Bj)
Xj |λj ∼ Exp (λj) Xj |ρj, λj ∼ Weibull (ρj, λj)
![Page 18: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/18.jpg)
Capitalist Slides
7ModelisationLag
Lag: authorized time for the investigator to report the information in the database
Patient 1
Patient 2
cut-off
authorized lag of report(e.g. 30 days)
Randomisation Visit’s timesCensorship without lag Censorship with lag
![Page 19: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/19.jpg)
Capitalist Slides
8Methods’ comparison
Exponential
Weibull
Weibull
TreatmentDistribution ImplementationLag Covariates
Analytical
MCMC1
MCMC1
Source
Bagiella & Heitjan
Ying & Heitjan
Servier
1 Two implementation approaches (Sampling Importance Resampling and Hamiltonian Monte Carlo) have presented similar results (all other things being equal) on a set of simulated datasets.
![Page 20: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/20.jpg)
Capitalist Slides
9Prediction
• Sample in parameter posterior distribution • Generate inter arrival times to derive randomisation date for the patients to be
recruited
𝜇𝑝
Enrollment
![Page 21: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/21.jpg)
Capitalist Slides
10PredictionTime-to-event & Time-to-censorship
Bagiella & Heitjan
Ying & Heitjan
Servier
SampleMethod Predictive
λpj
ρpj , λp
j
ρp, λp0 , βp
i
Xj ∼ Exp (λpj )
Xj ∼ Weibull (ρpj , λp
j )X ∼ Weibull (ρp, λp
0 e ∑ βpi Xi)
![Page 22: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/22.jpg)
Capitalist Slides
10PredictionTime-to-event & Time-to-censorship
Bagiella & Heitjan
Ying & Heitjan
Servier
SampleMethod Predictive
λpj
ρpj , λp
j
ρp, λp0 , βp
i
Xj ∼ Exp (λpj )
Xj ∼ Weibull (ρpj , λp
j )X ∼ Weibull (ρp, λp
0 e ∑ βpi Xi)
Classical simulation
![Page 23: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/23.jpg)
Capitalist Slides
10PredictionTime-to-event & Time-to-censorship
Bagiella & Heitjan
Ying & Heitjan
Servier
SampleMethod Predictive
λpj
ρpj , λp
j
ρp, λp0 , βp
i
Xj ∼ Exp (λpj )
Xj ∼ Weibull (ρpj , λp
j )X ∼ Weibull (ρp, λp
0 e ∑ βpi Xi)
Inverse Transform Sampling
Classical simulation
![Page 24: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/24.jpg)
Capitalist Slides
PredictionInverse Transform Sampling
• Sample
• Calculates
q ∼ U [0,S (tfu)]S−1 (q) > tfu
S (t) = e−λtρ
S (t)
tfu S−1 (q)
𝑞
S (tfu)
0
1
observed follow-up
time
prediction period
11
![Page 25: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/25.jpg)
4 – Case Study
![Page 26: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/26.jpg)
Capitalist Slides
Case StudyPhase II
Phase II in oncology Recruitment: 153 patients (FVFP: 29/04/2016 – 12 months enrollment) Target: 100 events (completion target event : 15/01/2018) Primary outcome: Progression Free Survival Design hypotheses: Drop-out rate: 1 % per year
medianctrl = 9.1 months, HR = 0.77
12
![Page 27: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/27.jpg)
Capitalist Slides
Case StudyPhase II
et
f (x) =xα−1
βαΓ (α)e− x
β
𝔼 (X) = αβ 𝕍 (X) = αβ2
Gamma distribution (scale notation)
Design hypothesis: exponential distribution supposed for events so h (t) = λρtρ−1 = λ ρ ∼ Γ (1,1) 𝔼 (ρ) = 1 𝕍 (ρ) = 1Proposition:
13
![Page 28: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/28.jpg)
Capitalist Slides
Case StudyPhase II
λ ∼ Γ (1,log (2)median ) 𝔼 (λ) =
log (2)median
𝕍 (λ) = ( log (2)median )
2
14
Hypothesis: represents the speed of events’ occurenceλ ∼ Γ (C, D)
median = ( log (2)λ )
1ρ
=log (2)
λ⇒ 𝔼 (λ) = CD =
log (2)median
ρ = 1
![Page 29: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/29.jpg)
Capitalist Slides
Case StudyPhase II Cut-off: 29/08/2017
Number of events: 57
15
![Page 30: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/30.jpg)
Capitalist Slides
Case StudyPhase II
Bagiella & Heitjan Approach
Over prediction
Under prediction
16
![Page 31: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/31.jpg)
Capitalist Slides
Case StudyPhase II
Ying & Heitjan Approach
Over prediction
Under prediction
17
![Page 32: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/32.jpg)
Capitalist Slides
Case StudyPhase II
Servier Approach
Over prediction
Under prediction
18
![Page 33: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/33.jpg)
5 – Simulation Study
![Page 34: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/34.jpg)
Capitalist Slides
Results & ComparisonDistribution: exponential / Weibull
Exponential Weibull
Blinded
Unblinded
Lag
No Lag
Lag
No Lag
• Cut-off: 75% of the time of the simulated study
• Probability threshold: 0.5
19
![Page 35: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/35.jpg)
Capitalist Slides
Results & ComparisonUnblinded / Blinded
No interest in separating data by treatment arm in 2 different models
More parameters to estimate in Weibull models → pooling treatment arms may improve the estimations
20
![Page 36: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/36.jpg)
Capitalist Slides
Results & ComparisonLag
With or without lag, good predictions with Weibull models
Counter intuitive results: more investigations are needed
mixed
more precise without lag
Early predictions (cut-off 50%)
Mid-predictions (cut-off 75%)
Late predictions (cut-off 90%)
Exponential models
Weibull models
no difference no difference
no differencemixed
21
![Page 37: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/37.jpg)
6 – Conclusion
![Page 38: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/38.jpg)
Capitalist Slides
22ConclusionRecommendations
Pooling data increases the number of events in the model
BlindedImportant on case studies but inefficient in simulations
Lag?
Best results when there are enough events in database
Weibull
![Page 39: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/39.jpg)
Thanks for your attention
![Page 40: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/40.jpg)
Capitalist Slides
Weibull distribution
Density: Survival: Hazard rate:
Median:
f (t) = λρtρ−1e−λtρ = h (t) S (t)S (t) = e−λtρ = 1 − F (t)h (t) = λρtρ−1
median = ( log (2)λ )
1ρ
![Page 41: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/41.jpg)
Capitalist Slides
Simulation StudyScenarios
Phase II: • 200 patients • 12 months accrual • 150 events targeted
Exponential
Weibull
Exponential
Exponential
Exponential
Exponential
PhaseDistribution Treatment effect
Weibull
Phase II
Phase II
Phase III PFS
Phase III PFS
Weibull
Weibull
Weibull
Phase III OS
Phase III OS
Weibull
Exponential
Phase II
Phase II
Phase III PFS
Phase III PFS
Phase III OS
Phase III OS
HR = 0.75
HR = 0.9
HR = 0.65
HR = 0.9
HR = 0.65
HR = 0.9
HR = 0.75
HR = 0.9
HR = 0.65
HR = 0.9
HR = 0.65
HR = 0.9
Phase III: • 1,000 patients • 24 months accrual • 600 events targeted
18
![Page 42: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/42.jpg)
Capitalist Slides
Simulation StudyModels
Hamiltonian Monte Carlo
Lag
Exponential
Exponential
Exponential
TreatmentDistribution Implementation
Exponential
Analytical
Weibull
Weibull
Weibull
Weibull
Analytical
Analytical
Analytical
Hamiltonian Monte Carlo
Hamiltonian Monte Carlo
Hamiltonian Monte Carlo
Unblinded
Unblinded
Unblinded
Unblinded
Blinded
Blinded
Blinded
Blinded
Bagiella & Heitjan
Ying & Heitjan
Servier Approach
18
![Page 43: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/43.jpg)
Capitalist Slides
Simulation Study
• Repeat each simulation 1,000 times • Evaluation criterion: difference in month between the target date & the predicted date of the Xth
event
Results
• Heatmap produced with different cut-off & different probability thresholds
18
![Page 44: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/44.jpg)
Capitalist Slides
Case StudiesPhase II
![Page 45: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/45.jpg)
Capitalist Slides
Case Study
Example: median =mediantrt + medianctrl
2=
medianctrl
HR + medianctrl
2= 315 days
𝔼 (λ) =log (2)median
= 0.0022
Parameters calcul
Scale notation:
Rate notation:
λ ∼ Γ (1,0.0022)
λ ∼ Γ (1,1
0.0022 ) = Γ (1,455)
![Page 46: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/46.jpg)
Capitalist Slides
Case StudyPhase III
Phase III in oncology Recruitment: 320 patients (FVFP: 20/04/2011) Target: 216 events (completion target event : 09/05/2018) Primary outcome: Progression Free Survival Design hypotheses: Drop-out rate:
𝑚𝑒𝑑𝑖𝑎𝑛𝑐𝑡𝑟𝑙 = 2.8 months, 𝐻𝑅 = 0.65
![Page 47: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/47.jpg)
Capitalist Slides
Case StudyPhase III Cut-off: 09/09/2016
Number of events: 128
![Page 48: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/48.jpg)
Capitalist Slides
Case StudyPhase III – Ying & Heitjan Algorithm
![Page 49: A Bayesian approach for event predictions in clinical trials ...gdr-stat-sante.math.cnrs.fr/spip/IMG/pdf/marine_antigny.pdf3 identified methods: • Emilia Bagiella and Daniel F Heitjan](https://reader036.vdocuments.us/reader036/viewer/2022081623/614907f79241b00fbd674bba/html5/thumbnails/49.jpg)
Capitalist Slides
Case StudyPhase III – Servier Approach