Download presentation

Presentation is loading. Please wait.

Published byCharles Sanchez Modified over 3 years ago

1
BAYESIAN ADAPTIVE DESIGN & INTERIM ANALYSIS Donald A. Berry dberry@mdanderson.org Donald A. Berry dberry@mdanderson.org

2
2 2 Some references l Berry DA (2003). Statistical Innovations in Cancer Research. In Cancer Medicine e6. Ch 33. BC Decker. (Ed: Holland J, Frei T et al.) l Berry DA (2004). Bayesian statistics and the efficiency and ethics of clinical trials. Statistical Science. l Berry DA (2003). Statistical Innovations in Cancer Research. In Cancer Medicine e6. Ch 33. BC Decker. (Ed: Holland J, Frei T et al.) l Berry DA (2004). Bayesian statistics and the efficiency and ethics of clinical trials. Statistical Science.

3
3 3 Benefits l Adapting; examples n Stop early (or late!) n Change doses n Add arms n Drop arms l Final analysis n Greater precision (even full follow-up) n Earlier conclusions l Adapting; examples n Stop early (or late!) n Change doses n Add arms n Drop arms l Final analysis n Greater precision (even full follow-up) n Earlier conclusions

4
4 4 Goals l Learn faster: More efficient trials l More efficient drug/device development l Better treatment of patients in clinical trials l Learn faster: More efficient trials l More efficient drug/device development l Better treatment of patients in clinical trials

5
5 5 OUTLINE: EXAMPLES l Extraim analysis l Modeling early endpoints l Seamless Phase II/III trial l Adaptive randomization n Phase II trial in AML n Phase II drug screening process n Phase III trial l Extraim analysis l Modeling early endpoints l Seamless Phase II/III trial l Adaptive randomization n Phase II trial in AML n Phase II drug screening process n Phase III trial

6
6 6 EXTRAIM ANALYSES* l Endpoint: CR (detect 0.42 vs 0.32) l 80% power: N = 800 l Two extraim analyses, one at 800 l Another after up to 300 added pts l Maximum n = 1400 (only rarely) l Accrual: 70/month l Delay in assessing response l Endpoint: CR (detect 0.42 vs 0.32) l 80% power: N = 800 l Two extraim analyses, one at 800 l Another after up to 300 added pts l Maximum n = 1400 (only rarely) l Accrual: 70/month l Delay in assessing response *Modeling due to Scott Berry *Modeling due to Scott Berry

7
7 7 l After 800 pts accrued, have response info on 450 pts l Find pred prob of stat sig when full info on 800 pts available l Also when full info on 1400 l Continue if... l Stop if... l If continue, n via pred prob l Repeat at 2 nd extraim analysis l After 800 pts accrued, have response info on 450 pts l Find pred prob of stat sig when full info on 800 pts available l Also when full info on 1400 l Continue if... l Stop if... l If continue, n via pred prob l Repeat at 2 nd extraim analysis

8
vs 0.80

9
9 9 MODELING EARLY ENDPOINTS: LONGITUDINAL MARKERS l Example CA125 in ovarian cancer l Use available data from trial (& outside of trial) to model relationship over time with survival, depending on Rx l Predictive distributions l Use covariates l Seamless phases II & III l Example CA125 in ovarian cancer l Use available data from trial (& outside of trial) to model relationship over time with survival, depending on Rx l Predictive distributions l Use covariates l Seamless phases II & III

10
10 CA125 data & predictive distributions of survival for two of many patients* > *Modeling due to Scott Berry *Modeling due to Scott Berry

11
Days Patient #1 Treatment

12
Patient #1

13
Days Patient #2

15
15 Methods l Analytical l Multiple imputation l Analytical l Multiple imputation

16
16 SEAMLESS PHASES II/III* l Early endpoint (tumor response, biomarker) may predict survival? l May depend on treatment l Should model the possibilities l Primary endpoint: survival l But observe relationships l Early endpoint (tumor response, biomarker) may predict survival? l May depend on treatment l Should model the possibilities l Primary endpoint: survival l But observe relationships *Inoue, et al (2002 Biometrics)

17
17 Good resp Good resp No resp Survival advantage No survival advantage Phase 2 Phase 3 Conventional drug development 6 mos 9-12 mos > 2 yrs Stop Seamless phase 2/3 < 2 yrs (usually) Not Market

18
18 Seamless phases l Phase 2: 1 or 2 centers; 10 pts/mo, randomize E vs C l If pred probs look good, expand to Phase 3: Many centers; 50 pts/mo (Initial centers continue accrual) l Max n = 900 [Single trial: survival data combined in final analysis] l Phase 2: 1 or 2 centers; 10 pts/mo, randomize E vs C l If pred probs look good, expand to Phase 3: Many centers; 50 pts/mo (Initial centers continue accrual) l Max n = 900 [Single trial: survival data combined in final analysis]

19
19 Early stopping l Use pred probs of stat sig l Frequent analyses (total of 18) using pred probs to: n Switch to Phase 3 n Stop accrual for s Futility s Efficacy n Submit NDA l Use pred probs of stat sig l Frequent analyses (total of 18) using pred probs to: n Switch to Phase 3 n Stop accrual for s Futility s Efficacy n Submit NDA

20
20 Conventional Phase 3 designs: Conv4 & Conv18, max N = 900 (same power as adaptive design) Conventional Phase 3 designs: Conv4 & Conv18, max N = 900 (same power as adaptive design) Comparisons

21
21 Expected N under H 0

22
22 Expected N under H 1

23
23 Benefits l Duration of drug development is greatly shortened under adaptive design: n Fewer patients in trial n No hiatus for setting up phase 3 n All patients used for s Phase 3 endpoint s Relation between response & survival l Duration of drug development is greatly shortened under adaptive design: n Fewer patients in trial n No hiatus for setting up phase 3 n All patients used for s Phase 3 endpoint s Relation between response & survival

24
24 Possibility of large N l N seldom near 900 l When it is, its necessary! l This possibility gives Bayesian design its edge [Other reason for edge is modeling response/survival] l N seldom near 900 l When it is, its necessary! l This possibility gives Bayesian design its edge [Other reason for edge is modeling response/survival]

25
25 l Troxacitabine (T) in acute myeloid leukemia (AML) combined with cytarabine (A) or idarubicin (I) l Adaptive randomization to: IA vs TA vs TI l Max n = 75 l End point: Time to CR (< 50 days) l Troxacitabine (T) in acute myeloid leukemia (AML) combined with cytarabine (A) or idarubicin (I) l Adaptive randomization to: IA vs TA vs TI l Max n = 75 l End point: Time to CR (< 50 days) ADAPTIVE RANDOMIZATION Giles, et al JCO (2003)

26
26 Adaptive Randomization l Assign 1/3 to IA (standard) throughout (until only 2 arms) l Adaptive to TA and TI based on current results l Results l Assign 1/3 to IA (standard) throughout (until only 2 arms) l Adaptive to TA and TI based on current results l Results

27
27

28
28 Compare n = 75 Drop TI

29
29 Summary of results CR < 50 days: n IA:10/18 = 56% n TA: 3/11 = 27% n TI: 0/5 = 0% Criticisms... CR < 50 days: n IA:10/18 = 56% n TA: 3/11 = 27% n TI: 0/5 = 0% Criticisms...

30
30 SCREENING PHASE II DRUGS l Many drugs l Tumor response l Goals: n Treat effectively n Learn quickly l Many drugs l Tumor response l Goals: n Treat effectively n Learn quickly

31
31 Standard designs l One drug (or dose) at a time; no drug/dose comparisons l Typical comparison by null hypothesis: RR = 20% l Progress hopelessly slow! l One drug (or dose) at a time; no drug/dose comparisons l Typical comparison by null hypothesis: RR = 20% l Progress hopelessly slow!

32
32 Standard 2-stage design First stage 20 patients: l Stop if 4 or 9 responses l Else second set of 20 First stage 20 patients: l Stop if 4 or 9 responses l Else second set of 20

33
33 An adaptive allocation l When assigning next patient, find r = P(rate 20% | data) for each drug l Assign drugs in proportion to r l Add drugs as become available l Drop drugs that have small r l Drugs with large r phase 3 l When assigning next patient, find r = P(rate 20% | data) for each drug l Assign drugs in proportion to r l Add drugs as become available l Drop drugs that have small r l Drugs with large r phase 3

34
34 Suppose 10 drugs, 200 patients 9 drugs have mix of RRs 20% & 40%, 1 has 60% (nugget) 9 drugs have mix of RRs 20% & 40%, 1 has 60% (nugget) <70% >99% Identify nugget … With probability: In average n: Identify nugget … With probability: In average n: 110 50 Adaptive also better at finding 40%, & sooner Standard Adaptive Standard

35
35 Suppose 100 drugs, 2000 patients 99 drugs have mix of RRs 20% & 40%, 1 has 60% (nugget) 99 drugs have mix of RRs 20% & 40%, 1 has 60% (nugget) Adaptive also better at finding 40%, & sooner <70% >99% Identify nugget … With probability: In average n: Identify nugget … With probability: In average n: 1100 500 Standard Adaptive Standard

36
36 Consequences l Treat pts in trial effectively l Learn quickly l Attractive to patients, in and out of the trial l Better drugs identified sooner; move through faster l Treat pts in trial effectively l Learn quickly l Attractive to patients, in and out of the trial l Better drugs identified sooner; move through faster

37
37 PHASE III TRIAL l Dichotomous endpoint l Q = P(p E > p S |data) l Min n = 150; Max n = 600 l After n = 50, assign to arm E with probability Q n Except that 0.2 P(assign E) 0.8 l (Not optimal, but …) l Dichotomous endpoint l Q = P(p E > p S |data) l Min n = 150; Max n = 600 l After n = 50, assign to arm E with probability Q n Except that 0.2 P(assign E) 0.8 l (Not optimal, but …)

38
38 Recommendation to DSMB to l Stop for superiority if Q 0.99 l Stop accrual for futility if P(p E – p S PF n PF depends on current n... l Stop for superiority if Q 0.99 l Stop accrual for futility if P(p E – p S PF n PF depends on current n...

39
39 PF

40
40 Common prior density for p E & p S l Independent l Reasonably non-informative l Mean = 0.30 l SD = 0.20 l Independent l Reasonably non-informative l Mean = 0.30 l SD = 0.20

41
41

42
42 Updating After 20 patients on each arm n 8/20 responses on arm 1 n 12/20 responses on arm 2 After 20 patients on each arm n 8/20 responses on arm 1 n 12/20 responses on arm 2

43
43

44
44 Assumptions l Accrual: 10/month l 50-day delay to assess response l Accrual: 10/month l 50-day delay to assess response

45
45 Need to stratify. But how? Suppose probability assign to experimental arm is 30%, with these data...

46
46

47
47 One simulation; p S = 0.30, p E = 0.45 Final Std12/38 19/60 20/65 Exp38/83 82/16787/178 Superiority boundary

48
48 9 mos. End Final Std 8/39 15/57 18/68 Exp 11/42 32/81 22/87 One simulation; p E = p S = 0.30 Futility boundary

49
49 Operating characteristics

50
50 FDA: Why do this? Whats the advantage? l Enthusiasm of PIs l Comparison with standard design... l Enthusiasm of PIs l Comparison with standard design...

51
51 Adaptive vs tailored balanced design w/same false-positive rate & power (Mean number patients by arm) ORR Arm p S = 0.20 p E = 0.35 p S = 0.30 p E = 0.45 p S = 0.40 p E = 0.55 StdExpStdExpStdExp Adaptive681687917874180 Balanced171 203 216 Savings10331242514236

52
52 Consequences of Bayesian Adaptive Approach l Fundamental change in way we do medical research l More rapid progress l Well get the dose right! l Better treatment of patients l... at less cost l Fundamental change in way we do medical research l More rapid progress l Well get the dose right! l Better treatment of patients l... at less cost

53
53 OUTLINE: EXAMPLES l Extraim analysis l Modeling early endpoints l Seamless Phase II/III trial l Adaptive randomization n Phase II trial in AML n Phase II drug screening process n Phase III trial l Extraim analysis l Modeling early endpoints l Seamless Phase II/III trial l Adaptive randomization n Phase II trial in AML n Phase II drug screening process n Phase III trial

Similar presentations

Presentation is loading. Please wait....

OK

Time for a BREAK! You have 45 Minutes.

Time for a BREAK! You have 45 Minutes.

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google

Ppt on natural resources for class 4 Ppt on complex numbers class 11th Ppt on pricing policy of a company Ppt on hindu religion history Ppt on object-oriented programming concepts youtube lecture Ppt on conceptual art movement Download ppt on sectors of economy Ppt on gas power plant in india Ppt on self development skills Ppt on construction industry in india