SlideShare a Scribd company logo
at multiple time points in clinical trials 
Lu Mao, PhD Student 
University of North Carolina at Chapel Hill 
Supervisor: Paul Gallo, PhD 
Strategies for Futility Analyses
OUTLINE 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
2 
Background for Futility Analysis 
•Motivation 
•B-value theory 
•Conditional Power 
•Predictive Power 
Methods and Results 
•Two futility looks 
•Three futility looks 
Software (R) Demonstration 
Conclusion 
Stop for Futility
BACKGROUND - MOTIVATION 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
3 
Traditional Design: fix the sample size and perform analyses after all subjects have been recruited. 
Group Sequential Design: several interim analyses are conducted in the course of patient enrollment 
•Safety 
•Early determination of (in)efficacy 
•Ethics 
•Cost reduction 
Futility Analysis: a group sequential design to allow for early termination of the trial when the likelihood of establishing efficacy at final stage is decide to be low.
BACKGROUND - MOTIVATION 
An ideal futility analysis design: 
1.Considerably curtails the length of the trial when there is no/negative effect 
2.Do not substantially affect the operating characteristics (Type I and Type II errors) of the original fixed-sample design. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
4 
H0 
HA 
Futility rules 
Final Analysis 
STOP
BACKGROUND - STATISTICAL TOOLS 
5 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
 Group sequential trials – multiple testing on several 
statistics based on the first m observations, m=n1,...,nk 
 Let the test statistics, e.g. z scores, be . 
The rejection region R={|Zn|>z1-α/2} 
 If Z>0 means a positive effect for the treatment, we wish 
to stop the trial for futility when the interim Z scores are 
very low, e.g. set (zn1 
,...,znk 
), and at the i’th interim, 
 Needs to find the joint distribution of 
( ,..., , ) 
n1 n n Z Z Z 
k 
( ,..., , ) 
n1 n n Z Z Z 
k 
ni ni Z  z 
Stop to 
Accept H0
BACKGROUND - STATISTICAL TOOLS 
 Typically, the Z statistic calculated from the first m obs can 
be expressed in terms of a partial sum statistic 
in the form 
where is an estimate of , the variance of . 
 Example: Two sample normal test 
6 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
  
 
m 
m i i S 
1 
 
mˆ 
S 
Z m 
m  
0 
1 
0 2 ˆ 
( ) 
2 ˆ 
( ) 
 m  
X Y 
m X Y 
Z 
m 
i 
i i 
m 
 
 
 
 
 
2 ˆ i  
2 
BACKGROUND - B-VALUE THEORY 
 Easier to find the joint distribution of by 
the well known Brownian motion approximation to the 
partial sums. 
 Indeed, if is the effect size (true state of nature), 
where is related to the true state of nature and the overall 
sample size, and is the fraction of sample size at 
the j’th interim, call it information time. Obviously 
7 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
( ,..., , ) 
n1 n n S S S 
k 
  
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
1 
, 
1 
1 1 
1 
1 
1 1 1 1 
1 
1 
1 
1 
1 
k 
k k k 
d 
n 
i i 
n 
i i 
n 
i i 
n 
n 
n 
t t 
t t t 
t t t 
t 
t 
N 
n 
S 
S 
S 
n k 
k 
 
 
    
 
   
 
 
 
 
  
i   E 
  n  
t n n j k j j  , 1,, 
0  1. j t
BACKGROUND - B-VALUE THEORY 
 Call the left side of the previous formula the B values, 
denoted . 
 We have seen that the B scores follow the marginal 
distribution of a standard Brownian motion with drift term θ 
 The relationship between the Z scores and the B scores is 
simple: 
 Or: 
8 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
( ( ), , ( ), (1)) 1 B t B t B k  
( ) 
1 
ˆ ˆ j 
j 
n 
j j 
n 
n B t 
n t 
S 
n 
n 
n 
S 
Z j j 
j 
   
  
j j n j B(t )  t Z
BACKGROUND - CONDITIONAL POWER 
 Conditional Power: quantifies the notion of likelihood of 
success given the current data 
 Futility rule: stop at the j’th interim if 
 Since , we have 
 Therefore 
9 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
CP b P B z B t b j k j j j ( ; ) ( (1) | ( ) ; ), 1, , 1 / 2          
( ( ); ) . j j CP B t   
  
 
 
  
 
 
  
 
 
  
 
 
  
 
 
  
 
 
   
 
 
  
 
 
1 
, 
(1) 1 
( ) 
j 
j j j j 
t 
t t t 
N 
B 
B t 
 
  j j j j B(1) B(t ), N B(t )  (1t ) , 1t 
 
 
 
 
 
 
 
 
 
   
   
j 
j j 
j 
t 
z t b 
CP b 
1 
(1 ) 
( ; ) 1 1 / 2  
 
BACKGROUND - CONDITIONAL POWER 
Conditional power based on the hypothesized 휃 does not allow the state of nature to adapt to the observed data. 
Conditional power based on the current estimate 휃 푗 
퐶푃푏푗;휃 푗≡푃퐵1>푧1− 훼 2 퐵푡푗=푏푗;휃 푗 
Recall that 퐵푡푗∼푁푡푗휃,푡푗, we obtain 휃 푗=퐵(푡푗)/푡푗 
퐶푃푏푗;휃 푗=1−Φ 푧1−훼2 −휃 푗1−푡푗−푏푗 1−푡푗 =1−Φ 푧1−훼2 −푏푗푡푗 1−푡푗 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
10
BACKGROUND - PREDICTIVE POWER 
Another way of allowing the state of nature to adapt to the observed data is (Bayesian) predictive power. 
Predictive power – conditional power averaged over the posterior distribution of the state of nature: 
푃푃푏푗;휋⋅=∫퐶푃푏푗;휃휋휃푏푗푑휃 
Typically, we take the uniform prior: 
휋휃∝1 
After some derivation, we obtain 
푃푃푏푗;휋⋅≡1=1−Φ 푡푗푧1−훼2 −푏푗푡푗 1−푡푗 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
11
BACKGROUND - POWER FUNCTION 
We have seen three ways of choosing the futility boundary (푏1,⋯,푏푘) 
1.Conditional power: 퐶푃 
2.Conditional power based on current estimate: 퐶푃(휃 ) 
3.Predictive power: 푃푃 
Given the boundary, the power function is given by 
Ψ휃≡푃Reject 퐻0휃 
=P휃Btj≥푏푗,푗=1,⋯,푘,B1>z1−훼2 
<푃휃퐵(1)>z1−훼2 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
12
BACKGROUND - ERROR PROBABILITIES 
Thus the power function is reduced when either 휃=0 or equals hypothesized effect size 
In statistical language, both the type I error rate and power are decreased. 
Power loss: a fraction of successful trials are terminated by the futility rule. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
13
BACKGROUND - PREVIOUS STUDIES 
Chang et al. 2004 
•Assesses one futility look at midway of trial based on conditional power; 
•Shows that most power loss can be “reclaimed” by lowering the (final) critical value to achieve type I error rate exactly 훼; 
(recall that for critical value 푧1−훼2 the overall type I error < 훼) 
•Provides a graphical method of doing this 
Lachin et al. 2005 
•One futility look based on conditional power at midway 
•Suggests an iterative algorithm of determining the critical value that achieves 훼 to regain power. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
14
BACKGROUND - PREVIOUS STUDIES 
Snapinn et al. 2006 
•Reviews conditional power approach to futility rule; 
•Note the problem of reclaiming 훼: rules become binding, while it crosses the boundary it has to stop. 
Emerson et al. 2005 
•Considers CP, CP(휃 ), PP and various other scales in determining futility rules 
•Argues that the scale used is less important than the resulted operating characteristics. 
V. Shih & P. Gallo 2010 
•Investigate power loss vs sample size reduction for one futility rule at midway based on CP, CP(휃 ), PP. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
15
METHODS – GENERAL 
Rationale 
•Nonbinding futility rule: no type I error reclaiming; 
•ASN (average sample number) under 퐻0 is the optimization target, provided that other factors are controlled; 
•To regain power, we may enlarge the sample size – sample size inflation (SI) is the control target; 
•No power regaining – power loss is the control target. 
Setting 
•Multiple futility looks at arbitrary time points – For simplicity and practicality we only consider two and three looks at evenly spaced interims; 
•Different scales (CP, CP(휃 ), PP, etc ) of setting futility rules. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
16
METHODS – GENERAL 
Aim 
•Setting up a theoretical framework for nonbinding futility rules; 
•Comparing numerical results using different scales CP, CP(휃 ) and PP; 
•Develop easy-to-implement program, based on efficient numerical algorithms, to allow the user to choose different scales, values of the scale, and setup of interim time points. 
Facilities 
•Most numerical analyses are done in R; a few in MatLab; 
•We make use of the R package mvtnorm for multivariate normal distribution function evaluations. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
17
METHODS – SAMPLE SIZE INFLATION 
Sample size inflation (SI) approach 
•The trade-off is between SI factor and ASN (average sample size under 퐻0). 
•Recall that 휃=푛훿/휎; if we find the 휃푆퐼 that achieves power 1−훽, then since the hypothesized effect size is known, we obtain 푛; 
•Since in the fixed sample test 휃0=푧1−훼2 +푧1−훽, we have the sample size inflation factor: 
푅= 휃푆퐼 푧1−훼2 +푧1−훽 2 
•By definition 휃푆퐼=Ψ−1(1−훽); by monotonicity, use bisection method to obtain 휃푆퐼; 
•Note that the inflation factor 푅 is NOT dependent on the hypothesized effect size. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
18
METHODS – POWER LOSS 
Power loss approach: 
•Without SI to regain power; 
•We look at equal 
-CP 
-CP(휃 ) 
-PP -Equal power loss (Power loss at tj: 푙푗휃=푃휃(퐵푡푖≥푏푖,푖=1⋯,푗− 1,퐵푡푗<푏푗,퐵1>푧1−훼2 )) 
at two and three evenly spaced interims. 
•The trade-off is between power loss and ASN 
•We also assess the optimal rules (the rules that result in minimum ASN given certain power loss) 
-Optimization done by grid search 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
19
RESULTS – SAMPLE SIZE INFLATION 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
20 
SI factor 
•To achieve 1−훽=0.9 based on equal conditional power 훾 at 푘=1,⋯,5 evenly spaced interims 
•when 훾≤0.5, inflation is in fact negligible (≤1.1), indicating great practicality in applying these non-binding futility rules through SI. 
SI 
Conditional power 훾 
k 
0.2 
0.3 
0.4 
0.5 
0.6 
0.7 
0.8 
1 
1.002 
1.007 
1.018 
1.038 
1.073 
1.129 
1.221 
2 
1.006 
1.016 
1.033 
1.063 
1.106 
1.175 
1.282 
3 
1.009 
1.022 
1.045 
1.078 
1.130 
1.201 
1.314 
4 
1.011 
1.029 
1.054 
1.092 
1.145 
1.222 
1.337 
5 
1.015 
1.033 
1.061 
1.102 
1.159 
1.237 
1.355
RESULTS – SAMPLE SIZE INFLATION 
ASN 
•Same setting as the previous table 
•There is substantial reduction of sample size for k=2, 3, 훾≤0.5 , a practical range of futility looks with tolerable SI. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
21 
ASN 
Conditional power 훾 
k 
0.2 
0.3 
0.4 
0.5 
0.6 
0.7 
0.8 
1 
0.823 
0.766 
0.722 
0.691 
0.675 
0.674 
0.696 
2 
0.756 
0.715 
0.678 
0.647 
0.622 
0.609 
0.609 
3 
0.716 
0.677 
0.648 
0.622 
0.604 
0.588 
0.582 
4 
0.693 
0.657 
0.628 
0.606 
0.589 
0.578 
0.571 
5 
0.680 
0.643 
0.615 
0.595 
0.580 
0.569 
0.563
RESULTS – SAMPLE SIZE INFLATION 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
22 
Additional issue with SI 
•Power=1−훽 at hypothesized effect size, by construction; 
•Need to assess the global power behavior, especially those near the hypothesized 훿 
•The right figure shows that the power curves for the “SI”ed futility design are almost indistinguishable from those of the fixed sample (reference) design when 훿≥0.5∗(designed 훿)
RESULTS – POWER LOSS 
Two futility looks 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
23 
Power loss 
Equal CP 
Equal CP(휃 ) 
Equal PP 
Equal power loss 
Optimal 
0.002 
0.751 
0.769 
0.724 
0.722 
0.717 
0.003 
0.726 
0.727 
0.696 
0.692 
0.689 
0.005 
0.700 
0.692 
0.663 
0.665 
0.661 
0.007 
0.674 
0.658 
0.636 
0.640 
0.635 
0.010 
0.648 
0.627 
0.611 
0.615 
0.610 
0.015 
0.620 
0.596 
0.584 
0.591 
0.584 
0.021 
0.591 
0.565 
0.557 
0.567 
0.557 
0.029 
0.562 
0.537 
0.532 
0.544 
0.532 
0.041 
0.532 
0.508 
0.506 
0.520 
0.505
RESULTS – POWER LOSS 
Equal PP is the closest to the optimal bound over all; 
Equal power loss approximates the optimal bound only when power loss is very small. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
24
RESULTS – POWER LOSS 
Three futility looks 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
25 
Power loss 
Equal CP 
Equal CP(휃 ) 
Equal PP 
Equal power loss 
Optimal 
0.003 
0.708 
0.738 
0.680 
0.673 
0.661 
0.005 
0.683 
0.708 
0.644 
0.641 
0.631 
0.007 
0.659 
0.678 
0.618 
0.620 
0.611 
0.010 
0.635 
0.637 
0.593 
0.594 
0.584 
0.014 
0.610 
0.600 
0.563 
0.569 
0.559 
0.020 
0.584 
0.564 
0.536 
0.548 
0.534 
0.028 
0.556 
0.528 
0.508 
0.522 
0.507 
0.037 
0.527 
0.495 
0.482 
0.500 
0.481 
0.051 
0.495 
0.462 
0.454 
0.476 
0.453
RESULTS – POWER LOSS 
Overall, equal PP performs best of all, again; less satisfactory for small power loss; 
Similarly, equal power loss is doing well for small power losses but not so for greater ones. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
26
RESULTS – OPTIMAL BOUNDS 
Optimal bounds for two futility looks (precision 0.001) 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
27 
Power loss 
푧1(훾1) 
푧2(훾2) 
푧3 
ASN 
0.002 
-0.852 (0.361) 
0.371 (0.159) 
1.960 
0.717 
0.003 
-0.674 (0.409) 
0.449 (0.187) 
1.960 
0.689 
0.005 
-0.518 (0.452) 
0.527 (0.218) 
1.960 
0.661 
0.007 
-0.383 (0.490) 
0.615 (0.257) 
1.960 
0.635 
0.010 
-0.259 (0.525) 
0.693 (0.294) 
1.960 
0.610 
0.015 
-0.103 (0.569) 
0.732 (0.313) 
1.960 
0.584 
0.021 
0.032 (0.606) 
0.815 (0.355) 
1.960 
0.557 
0.029 
0.170 (0.643) 
0.884 (0.392) 
1.960 
0.532 
0.041 
0.314 (0.680) 
0.966 (0.437) 
1.960 
0.505
RESULTS – OPTIMAL BOUNDS 
Optimal bounds for three futility looks (precision 0.01) 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
28 
Power loss 
푧1(훾1) 
푧2(훾2) 
푧3(훾3) 
푧4 
ASN 
0.003 
-1.13(0.457) 
-0.09(0.284) 
0.63(0.112) 
1.96 
0.661 
0.005 
-0.98(0.491) 
0.00(0.316) 
0.79(0.174) 
1.96 
0.631 
0.007 
-0.86(0.519) 
0.08(0.345) 
0.84(0.201) 
1.96 
0.611 
0.010 
-0.66(0.565) 
0.16(0.375) 
0.86(0.209) 
1.96 
0.584 
0.014 
-0.49(0.603) 
0.21(0.394) 
0.90(0.229) 
1.96 
0.559 
0.020 
-0.38(0.627) 
0.34(0.444) 
0.94(0.250) 
1.96 
0.534 
0.028 
-0.22(0.662) 
0.43(0.480) 
0.94(0.251) 
1.96 
0.507 
0.037 
-0.08(0.691) 
0.48(0.500) 
1.05(0.312) 
1.96 
0.481 
0.051 
0.02(0.711) 
0.67(0.575) 
1.10(0.346) 
1.96 
0.453
RESULTS – DISCUSSION 
Equal conditional power is not a good idea for futility rule at multiple time points; 
Intuitively, the conditional power does not adapt to the observed data as it moves along; 
The same conditional at a later point means drastically different things comparing to an earlier point, if the early data are already contradicting the hypothesized 휃. 
Allowing the state of nature to adapt is probably the reason for the success of equal PP. 
Note that our findings SHOULD NOT be taken to mean that the idea of conditional power is bad. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
29
RESULTS – DISCUSSION 
Compare the bounds: 
•This is the futility bounds for power loss 0.01; 
•Equal PP and optimal bounds coincide very well; 
•Comparing to the optimal, equal CP is conservative in the beginning and aggressive in the end; equal CP(휃 ) is aggressive in the beginning and conservative in the end. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
30
DEMONSTRATION OF fut() 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
31 
In addition to 훼 and 훽, allows the user to choose interim time points, the corresponding conditional (predictive) power, the scale (CP, CP(휃 ), PP), and whether you want SI. 
Example: 
•훼=0.05,훽=0.1 
•Two futility looks at one third and one half of the sample size 
•Use predictive power 0.2 and 0.3 respectively 
•No sample size inflation to regain power
DEMONSTRATION OF fut() 
•Use the summary() function to print out the details: 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
32
DEMONSTRATION OF fut() 
•Plot the boundary: 
>plot(D) 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
33 
•Plot the power function 
>powerplot(D)
CONCLUSIONS 
We have established an SI framework to non-binding futility rule with uncompromised power; 
We have shown that in realistic situations (푘=2,3), equal PP across the time points results in approximately optimal (in terms of ASN) bounds. 
We have developed an easy-to-use R program fut() for design of nonbinding futility rules. 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
34
THANK YOU FOR YOUR ATTENTION! 
QUESTIONS 
| Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 
35

More Related Content

What's hot

CDISC-CDASH
CDISC-CDASHCDISC-CDASH
CDISC-CDASH
Gowthami6789
 
Trial Design Domains
Trial Design DomainsTrial Design Domains
Trial Design Domains
Ankur Sharma
 
CDISC journey using RECISIT 1.1
CDISC journey using RECISIT 1.1CDISC journey using RECISIT 1.1
CDISC journey using RECISIT 1.1
Kevin Lee
 
Protocol Understanding_ Clinical Data Management_KatalystHLS
Protocol Understanding_ Clinical Data Management_KatalystHLSProtocol Understanding_ Clinical Data Management_KatalystHLS
Protocol Understanding_ Clinical Data Management_KatalystHLS
Katalyst HLS
 
Feasibility Solutions to Clinical Trial Nightmares
Feasibility Solutions to Clinical Trial NightmaresFeasibility Solutions to Clinical Trial Nightmares
Feasibility Solutions to Clinical Trial Nightmares
jbarag
 
Planning And Development Of The Iss Ise Webinar Final
Planning And Development Of The Iss Ise Webinar   FinalPlanning And Development Of The Iss Ise Webinar   Final
Planning And Development Of The Iss Ise Webinar Final
Jay1818mar
 
Efficacy endpoints in Oncology
Efficacy endpoints in OncologyEfficacy endpoints in Oncology
Efficacy endpoints in Oncology
Angelo Tinazzi
 
Protocol and CRF in clinical trials.pptx
Protocol and CRF in clinical trials.pptxProtocol and CRF in clinical trials.pptx
Protocol and CRF in clinical trials.pptx
445AmitPal
 
Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...
Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...
Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...
Kevin Clauson
 
CDISC SDTM and ADaM for survival data
CDISC SDTM and ADaM for survival dataCDISC SDTM and ADaM for survival data
CDISC SDTM and ADaM for survival data
Angelo Tinazzi
 
Conducting clinical trials in japan chris merriam-leith
Conducting clinical trials in japan   chris merriam-leithConducting clinical trials in japan   chris merriam-leith
Conducting clinical trials in japan chris merriam-leith
Chris Merriam-Leith, MS, PMP, RAC
 
Interim analysis in clinical trials (1)
Interim analysis in clinical trials (1)Interim analysis in clinical trials (1)
Interim analysis in clinical trials (1)
ADITYA CHAKRABORTY
 
regulatory.pptx
regulatory.pptxregulatory.pptx
regulatory.pptx
ssuser660bb1
 
Personalized medicine in prostate cancer care
Personalized medicine in prostate cancer carePersonalized medicine in prostate cancer care
Personalized medicine in prostate cancer care
Europa Uomo EPAD
 
Adverse Events and Serious Adverse Events - Katalyst HLS
Adverse Events and Serious Adverse Events - Katalyst HLSAdverse Events and Serious Adverse Events - Katalyst HLS
Adverse Events and Serious Adverse Events - Katalyst HLS
Katalyst HLS
 
Clinical trial design
Clinical trial designClinical trial design
Clinical trial design
BhushanSurana2
 
ADaM - Where Do I Start?
ADaM - Where Do I Start?ADaM - Where Do I Start?
ADaM - Where Do I Start?
Dr.Sangram Parbhane
 
A Systematic Review of ADaM IG Interpretation
A Systematic Review of ADaM IG InterpretationA Systematic Review of ADaM IG Interpretation
A Systematic Review of ADaM IG Interpretation
Angelo Tinazzi
 
SDTM modelling: from study protocol to SDTM-compliant datasets
SDTM modelling: from study protocol to SDTM-compliant datasets SDTM modelling: from study protocol to SDTM-compliant datasets
SDTM modelling: from study protocol to SDTM-compliant datasets
Angelo Tinazzi
 
Handling Third Party Vendor Data_Katalyst HLS
Handling Third Party Vendor Data_Katalyst HLSHandling Third Party Vendor Data_Katalyst HLS
Handling Third Party Vendor Data_Katalyst HLS
Katalyst HLS
 

What's hot (20)

CDISC-CDASH
CDISC-CDASHCDISC-CDASH
CDISC-CDASH
 
Trial Design Domains
Trial Design DomainsTrial Design Domains
Trial Design Domains
 
CDISC journey using RECISIT 1.1
CDISC journey using RECISIT 1.1CDISC journey using RECISIT 1.1
CDISC journey using RECISIT 1.1
 
Protocol Understanding_ Clinical Data Management_KatalystHLS
Protocol Understanding_ Clinical Data Management_KatalystHLSProtocol Understanding_ Clinical Data Management_KatalystHLS
Protocol Understanding_ Clinical Data Management_KatalystHLS
 
Feasibility Solutions to Clinical Trial Nightmares
Feasibility Solutions to Clinical Trial NightmaresFeasibility Solutions to Clinical Trial Nightmares
Feasibility Solutions to Clinical Trial Nightmares
 
Planning And Development Of The Iss Ise Webinar Final
Planning And Development Of The Iss Ise Webinar   FinalPlanning And Development Of The Iss Ise Webinar   Final
Planning And Development Of The Iss Ise Webinar Final
 
Efficacy endpoints in Oncology
Efficacy endpoints in OncologyEfficacy endpoints in Oncology
Efficacy endpoints in Oncology
 
Protocol and CRF in clinical trials.pptx
Protocol and CRF in clinical trials.pptxProtocol and CRF in clinical trials.pptx
Protocol and CRF in clinical trials.pptx
 
Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...
Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...
Superiority Trials Versus Non-Inferiority Trials to Demonstrate Effectiveness...
 
CDISC SDTM and ADaM for survival data
CDISC SDTM and ADaM for survival dataCDISC SDTM and ADaM for survival data
CDISC SDTM and ADaM for survival data
 
Conducting clinical trials in japan chris merriam-leith
Conducting clinical trials in japan   chris merriam-leithConducting clinical trials in japan   chris merriam-leith
Conducting clinical trials in japan chris merriam-leith
 
Interim analysis in clinical trials (1)
Interim analysis in clinical trials (1)Interim analysis in clinical trials (1)
Interim analysis in clinical trials (1)
 
regulatory.pptx
regulatory.pptxregulatory.pptx
regulatory.pptx
 
Personalized medicine in prostate cancer care
Personalized medicine in prostate cancer carePersonalized medicine in prostate cancer care
Personalized medicine in prostate cancer care
 
Adverse Events and Serious Adverse Events - Katalyst HLS
Adverse Events and Serious Adverse Events - Katalyst HLSAdverse Events and Serious Adverse Events - Katalyst HLS
Adverse Events and Serious Adverse Events - Katalyst HLS
 
Clinical trial design
Clinical trial designClinical trial design
Clinical trial design
 
ADaM - Where Do I Start?
ADaM - Where Do I Start?ADaM - Where Do I Start?
ADaM - Where Do I Start?
 
A Systematic Review of ADaM IG Interpretation
A Systematic Review of ADaM IG InterpretationA Systematic Review of ADaM IG Interpretation
A Systematic Review of ADaM IG Interpretation
 
SDTM modelling: from study protocol to SDTM-compliant datasets
SDTM modelling: from study protocol to SDTM-compliant datasets SDTM modelling: from study protocol to SDTM-compliant datasets
SDTM modelling: from study protocol to SDTM-compliant datasets
 
Handling Third Party Vendor Data_Katalyst HLS
Handling Third Party Vendor Data_Katalyst HLSHandling Third Party Vendor Data_Katalyst HLS
Handling Third Party Vendor Data_Katalyst HLS
 

Similar to Strategies for setting futility analyses at multiple time points in clinical trials

Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
The Statistical and Applied Mathematical Sciences Institute
 
2008 JSM - Meta Study Data vs Patient Data
2008 JSM - Meta Study Data vs Patient Data2008 JSM - Meta Study Data vs Patient Data
2008 JSM - Meta Study Data vs Patient Data
Terry Liao
 
DMAIC
DMAICDMAIC
DMAIC
Shane Yeh
 
Sct2013 boston,randomizationmetricsposter,d6.2
Sct2013 boston,randomizationmetricsposter,d6.2Sct2013 boston,randomizationmetricsposter,d6.2
Sct2013 boston,randomizationmetricsposter,d6.2
Dennis Sweitzer
 
EUGM 2011 | DARCHY | Deployment & use of east within sanofi r & d
EUGM 2011 | DARCHY | Deployment & use of east within sanofi r & dEUGM 2011 | DARCHY | Deployment & use of east within sanofi r & d
EUGM 2011 | DARCHY | Deployment & use of east within sanofi r & d
Cytel USA
 
HST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdf
HST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdfHST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdf
HST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdf
DR BHADRAPPA HARALAYYA
 
ppt0320defenseday
ppt0320defensedayppt0320defenseday
ppt0320defenseday
Xi (Shay) Zhang, PhD
 
Whats New in SigmaXL Version 8
Whats New in SigmaXL Version 8Whats New in SigmaXL Version 8
Whats New in SigmaXL Version 8
John Noguera
 
panel regression.pptx
panel regression.pptxpanel regression.pptx
panel regression.pptx
Victoria Bozhenko
 
A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...
A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...
A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...
inventionjournals
 
Survadapt-Webinar_2014_SLIDES
Survadapt-Webinar_2014_SLIDESSurvadapt-Webinar_2014_SLIDES
Survadapt-Webinar_2014_SLIDES
therealreverendbayes
 
Design of Experiments
Design of Experiments Design of Experiments
Design of Experiments
Furk Kruf
 
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
The Statistical and Applied Mathematical Sciences Institute
 
DOE L-02 C-01 Introduction C02 Basic Statistical Methods.pptx
DOE   L-02   C-01 Introduction C02 Basic Statistical Methods.pptxDOE   L-02   C-01 Introduction C02 Basic Statistical Methods.pptx
DOE L-02 C-01 Introduction C02 Basic Statistical Methods.pptx
Jibin Babu
 
Temporal Graph Pattern Mining
Temporal Graph Pattern MiningTemporal Graph Pattern Mining
Temporal Graph Pattern Mining
Eugene Yang
 
Computational Pool-Testing with Retesting Strategy
Computational Pool-Testing with Retesting StrategyComputational Pool-Testing with Retesting Strategy
Computational Pool-Testing with Retesting Strategy
Waqas Tariq
 
A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...
A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...
A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...
Savas Papadopoulos, Ph.D
 
EUGM 2011 | JEHL | group sequential designs with 2 time to event endpoints
EUGM 2011 | JEHL | group sequential designs with 2 time to event endpointsEUGM 2011 | JEHL | group sequential designs with 2 time to event endpoints
EUGM 2011 | JEHL | group sequential designs with 2 time to event endpoints
Cytel USA
 
การสุ่มตัวอย่างในงานวิจัยสาธารณสุข
การสุ่มตัวอย่างในงานวิจัยสาธารณสุขการสุ่มตัวอย่างในงานวิจัยสาธารณสุข
การสุ่มตัวอย่างในงานวิจัยสาธารณสุข
Ultraman Taro
 
Lecture 2 doe
Lecture 2 doeLecture 2 doe
Lecture 2 doe
Quan Nguyen
 

Similar to Strategies for setting futility analyses at multiple time points in clinical trials (20)

Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
 
2008 JSM - Meta Study Data vs Patient Data
2008 JSM - Meta Study Data vs Patient Data2008 JSM - Meta Study Data vs Patient Data
2008 JSM - Meta Study Data vs Patient Data
 
DMAIC
DMAICDMAIC
DMAIC
 
Sct2013 boston,randomizationmetricsposter,d6.2
Sct2013 boston,randomizationmetricsposter,d6.2Sct2013 boston,randomizationmetricsposter,d6.2
Sct2013 boston,randomizationmetricsposter,d6.2
 
EUGM 2011 | DARCHY | Deployment & use of east within sanofi r & d
EUGM 2011 | DARCHY | Deployment & use of east within sanofi r & dEUGM 2011 | DARCHY | Deployment & use of east within sanofi r & d
EUGM 2011 | DARCHY | Deployment & use of east within sanofi r & d
 
HST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdf
HST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdfHST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdf
HST- 0621-151 PAPER ANALYSIS OF BANK PRODUCTIVITY USING PANEL CAUSALITY TEST.pdf
 
ppt0320defenseday
ppt0320defensedayppt0320defenseday
ppt0320defenseday
 
Whats New in SigmaXL Version 8
Whats New in SigmaXL Version 8Whats New in SigmaXL Version 8
Whats New in SigmaXL Version 8
 
panel regression.pptx
panel regression.pptxpanel regression.pptx
panel regression.pptx
 
A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...
A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...
A Moment Inequality for Overall Decreasing Life Class of Life Distributions w...
 
Survadapt-Webinar_2014_SLIDES
Survadapt-Webinar_2014_SLIDESSurvadapt-Webinar_2014_SLIDES
Survadapt-Webinar_2014_SLIDES
 
Design of Experiments
Design of Experiments Design of Experiments
Design of Experiments
 
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
 
DOE L-02 C-01 Introduction C02 Basic Statistical Methods.pptx
DOE   L-02   C-01 Introduction C02 Basic Statistical Methods.pptxDOE   L-02   C-01 Introduction C02 Basic Statistical Methods.pptx
DOE L-02 C-01 Introduction C02 Basic Statistical Methods.pptx
 
Temporal Graph Pattern Mining
Temporal Graph Pattern MiningTemporal Graph Pattern Mining
Temporal Graph Pattern Mining
 
Computational Pool-Testing with Retesting Strategy
Computational Pool-Testing with Retesting StrategyComputational Pool-Testing with Retesting Strategy
Computational Pool-Testing with Retesting Strategy
 
A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...
A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...
A PRACTICAL POWERFUL ROBUST AND INTERPRETABLE FAMILY OF CORRELATION COEFFICIE...
 
EUGM 2011 | JEHL | group sequential designs with 2 time to event endpoints
EUGM 2011 | JEHL | group sequential designs with 2 time to event endpointsEUGM 2011 | JEHL | group sequential designs with 2 time to event endpoints
EUGM 2011 | JEHL | group sequential designs with 2 time to event endpoints
 
การสุ่มตัวอย่างในงานวิจัยสาธารณสุข
การสุ่มตัวอย่างในงานวิจัยสาธารณสุขการสุ่มตัวอย่างในงานวิจัยสาธารณสุข
การสุ่มตัวอย่างในงานวิจัยสาธารณสุข
 
Lecture 2 doe
Lecture 2 doeLecture 2 doe
Lecture 2 doe
 

Recently uploaded

Gadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdfGadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdf
PirithiRaju
 
Modelo de slide quimica para powerpoint
Modelo  de slide quimica para powerpointModelo  de slide quimica para powerpoint
Modelo de slide quimica para powerpoint
Karen593256
 
(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...
(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...
(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...
Scintica Instrumentation
 
Compexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titrationCompexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titration
Vandana Devesh Sharma
 
The cost of acquiring information by natural selection
The cost of acquiring information by natural selectionThe cost of acquiring information by natural selection
The cost of acquiring information by natural selection
Carl Bergstrom
 
HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1
Shashank Shekhar Pandey
 
Eukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptxEukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptx
RitabrataSarkar3
 
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills MN
 
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...
Advanced-Concepts-Team
 
Describing and Interpreting an Immersive Learning Case with the Immersion Cub...
Describing and Interpreting an Immersive Learning Case with the Immersion Cub...Describing and Interpreting an Immersive Learning Case with the Immersion Cub...
Describing and Interpreting an Immersive Learning Case with the Immersion Cub...
Leonel Morgado
 
Randomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNERandomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNE
University of Maribor
 
快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样
快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样
快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样
hozt8xgk
 
Immersive Learning That Works: Research Grounding and Paths Forward
Immersive Learning That Works: Research Grounding and Paths ForwardImmersive Learning That Works: Research Grounding and Paths Forward
Immersive Learning That Works: Research Grounding and Paths Forward
Leonel Morgado
 
23PH301 - Optics - Optical Lenses.pptx
23PH301 - Optics  -  Optical Lenses.pptx23PH301 - Optics  -  Optical Lenses.pptx
23PH301 - Optics - Optical Lenses.pptx
RDhivya6
 
ESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptxESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptx
PRIYANKA PATEL
 
The binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defectsThe binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defects
Sérgio Sacani
 
Farming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptxFarming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptx
Frédéric Baudron
 
Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...
Leonel Morgado
 
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of ProteinsGBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
Areesha Ahmad
 
11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf
PirithiRaju
 

Recently uploaded (20)

Gadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdfGadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdf
 
Modelo de slide quimica para powerpoint
Modelo  de slide quimica para powerpointModelo  de slide quimica para powerpoint
Modelo de slide quimica para powerpoint
 
(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...
(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...
(June 12, 2024) Webinar: Development of PET theranostics targeting the molecu...
 
Compexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titrationCompexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titration
 
The cost of acquiring information by natural selection
The cost of acquiring information by natural selectionThe cost of acquiring information by natural selection
The cost of acquiring information by natural selection
 
HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1
 
Eukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptxEukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptx
 
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
 
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...
 
Describing and Interpreting an Immersive Learning Case with the Immersion Cub...
Describing and Interpreting an Immersive Learning Case with the Immersion Cub...Describing and Interpreting an Immersive Learning Case with the Immersion Cub...
Describing and Interpreting an Immersive Learning Case with the Immersion Cub...
 
Randomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNERandomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNE
 
快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样
快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样
快速办理(UAM毕业证书)马德里自治大学毕业证学位证一模一样
 
Immersive Learning That Works: Research Grounding and Paths Forward
Immersive Learning That Works: Research Grounding and Paths ForwardImmersive Learning That Works: Research Grounding and Paths Forward
Immersive Learning That Works: Research Grounding and Paths Forward
 
23PH301 - Optics - Optical Lenses.pptx
23PH301 - Optics  -  Optical Lenses.pptx23PH301 - Optics  -  Optical Lenses.pptx
23PH301 - Optics - Optical Lenses.pptx
 
ESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptxESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptx
 
The binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defectsThe binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defects
 
Farming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptxFarming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptx
 
Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...
 
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of ProteinsGBSN - Biochemistry (Unit 6) Chemistry of Proteins
GBSN - Biochemistry (Unit 6) Chemistry of Proteins
 
11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf
 

Strategies for setting futility analyses at multiple time points in clinical trials

  • 1. at multiple time points in clinical trials Lu Mao, PhD Student University of North Carolina at Chapel Hill Supervisor: Paul Gallo, PhD Strategies for Futility Analyses
  • 2. OUTLINE | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 2 Background for Futility Analysis •Motivation •B-value theory •Conditional Power •Predictive Power Methods and Results •Two futility looks •Three futility looks Software (R) Demonstration Conclusion Stop for Futility
  • 3. BACKGROUND - MOTIVATION | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 3 Traditional Design: fix the sample size and perform analyses after all subjects have been recruited. Group Sequential Design: several interim analyses are conducted in the course of patient enrollment •Safety •Early determination of (in)efficacy •Ethics •Cost reduction Futility Analysis: a group sequential design to allow for early termination of the trial when the likelihood of establishing efficacy at final stage is decide to be low.
  • 4. BACKGROUND - MOTIVATION An ideal futility analysis design: 1.Considerably curtails the length of the trial when there is no/negative effect 2.Do not substantially affect the operating characteristics (Type I and Type II errors) of the original fixed-sample design. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 4 H0 HA Futility rules Final Analysis STOP
  • 5. BACKGROUND - STATISTICAL TOOLS 5 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only  Group sequential trials – multiple testing on several statistics based on the first m observations, m=n1,...,nk  Let the test statistics, e.g. z scores, be . The rejection region R={|Zn|>z1-α/2}  If Z>0 means a positive effect for the treatment, we wish to stop the trial for futility when the interim Z scores are very low, e.g. set (zn1 ,...,znk ), and at the i’th interim,  Needs to find the joint distribution of ( ,..., , ) n1 n n Z Z Z k ( ,..., , ) n1 n n Z Z Z k ni ni Z  z Stop to Accept H0
  • 6. BACKGROUND - STATISTICAL TOOLS  Typically, the Z statistic calculated from the first m obs can be expressed in terms of a partial sum statistic in the form where is an estimate of , the variance of .  Example: Two sample normal test 6 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only    m m i i S 1  mˆ S Z m m  0 1 0 2 ˆ ( ) 2 ˆ ( )  m  X Y m X Y Z m i i i m      2 ˆ i  2 
  • 7. BACKGROUND - B-VALUE THEORY  Easier to find the joint distribution of by the well known Brownian motion approximation to the partial sums.  Indeed, if is the effect size (true state of nature), where is related to the true state of nature and the overall sample size, and is the fraction of sample size at the j’th interim, call it information time. Obviously 7 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only ( ,..., , ) n1 n n S S S k                                                                               1 , 1 1 1 1 1 1 1 1 1 1 1 1 1 1 k k k k d n i i n i i n i i n n n t t t t t t t t t t N n S S S n k k                 i   E   n  t n n j k j j  , 1,, 0  1. j t
  • 8. BACKGROUND - B-VALUE THEORY  Call the left side of the previous formula the B values, denoted .  We have seen that the B scores follow the marginal distribution of a standard Brownian motion with drift term θ  The relationship between the Z scores and the B scores is simple:  Or: 8 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only ( ( ), , ( ), (1)) 1 B t B t B k  ( ) 1 ˆ ˆ j j n j j n n B t n t S n n n S Z j j j      j j n j B(t )  t Z
  • 9. BACKGROUND - CONDITIONAL POWER  Conditional Power: quantifies the notion of likelihood of success given the current data  Futility rule: stop at the j’th interim if  Since , we have  Therefore 9 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only CP b P B z B t b j k j j j ( ; ) ( (1) | ( ) ; ), 1, , 1 / 2          ( ( ); ) . j j CP B t                                    1 , (1) 1 ( ) j j j j j t t t t N B B t    j j j j B(1) B(t ), N B(t )  (1t ) , 1t                j j j j t z t b CP b 1 (1 ) ( ; ) 1 1 / 2   
  • 10. BACKGROUND - CONDITIONAL POWER Conditional power based on the hypothesized 휃 does not allow the state of nature to adapt to the observed data. Conditional power based on the current estimate 휃 푗 퐶푃푏푗;휃 푗≡푃퐵1>푧1− 훼 2 퐵푡푗=푏푗;휃 푗 Recall that 퐵푡푗∼푁푡푗휃,푡푗, we obtain 휃 푗=퐵(푡푗)/푡푗 퐶푃푏푗;휃 푗=1−Φ 푧1−훼2 −휃 푗1−푡푗−푏푗 1−푡푗 =1−Φ 푧1−훼2 −푏푗푡푗 1−푡푗 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 10
  • 11. BACKGROUND - PREDICTIVE POWER Another way of allowing the state of nature to adapt to the observed data is (Bayesian) predictive power. Predictive power – conditional power averaged over the posterior distribution of the state of nature: 푃푃푏푗;휋⋅=∫퐶푃푏푗;휃휋휃푏푗푑휃 Typically, we take the uniform prior: 휋휃∝1 After some derivation, we obtain 푃푃푏푗;휋⋅≡1=1−Φ 푡푗푧1−훼2 −푏푗푡푗 1−푡푗 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 11
  • 12. BACKGROUND - POWER FUNCTION We have seen three ways of choosing the futility boundary (푏1,⋯,푏푘) 1.Conditional power: 퐶푃 2.Conditional power based on current estimate: 퐶푃(휃 ) 3.Predictive power: 푃푃 Given the boundary, the power function is given by Ψ휃≡푃Reject 퐻0휃 =P휃Btj≥푏푗,푗=1,⋯,푘,B1>z1−훼2 <푃휃퐵(1)>z1−훼2 | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 12
  • 13. BACKGROUND - ERROR PROBABILITIES Thus the power function is reduced when either 휃=0 or equals hypothesized effect size In statistical language, both the type I error rate and power are decreased. Power loss: a fraction of successful trials are terminated by the futility rule. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 13
  • 14. BACKGROUND - PREVIOUS STUDIES Chang et al. 2004 •Assesses one futility look at midway of trial based on conditional power; •Shows that most power loss can be “reclaimed” by lowering the (final) critical value to achieve type I error rate exactly 훼; (recall that for critical value 푧1−훼2 the overall type I error < 훼) •Provides a graphical method of doing this Lachin et al. 2005 •One futility look based on conditional power at midway •Suggests an iterative algorithm of determining the critical value that achieves 훼 to regain power. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 14
  • 15. BACKGROUND - PREVIOUS STUDIES Snapinn et al. 2006 •Reviews conditional power approach to futility rule; •Note the problem of reclaiming 훼: rules become binding, while it crosses the boundary it has to stop. Emerson et al. 2005 •Considers CP, CP(휃 ), PP and various other scales in determining futility rules •Argues that the scale used is less important than the resulted operating characteristics. V. Shih & P. Gallo 2010 •Investigate power loss vs sample size reduction for one futility rule at midway based on CP, CP(휃 ), PP. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 15
  • 16. METHODS – GENERAL Rationale •Nonbinding futility rule: no type I error reclaiming; •ASN (average sample number) under 퐻0 is the optimization target, provided that other factors are controlled; •To regain power, we may enlarge the sample size – sample size inflation (SI) is the control target; •No power regaining – power loss is the control target. Setting •Multiple futility looks at arbitrary time points – For simplicity and practicality we only consider two and three looks at evenly spaced interims; •Different scales (CP, CP(휃 ), PP, etc ) of setting futility rules. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 16
  • 17. METHODS – GENERAL Aim •Setting up a theoretical framework for nonbinding futility rules; •Comparing numerical results using different scales CP, CP(휃 ) and PP; •Develop easy-to-implement program, based on efficient numerical algorithms, to allow the user to choose different scales, values of the scale, and setup of interim time points. Facilities •Most numerical analyses are done in R; a few in MatLab; •We make use of the R package mvtnorm for multivariate normal distribution function evaluations. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 17
  • 18. METHODS – SAMPLE SIZE INFLATION Sample size inflation (SI) approach •The trade-off is between SI factor and ASN (average sample size under 퐻0). •Recall that 휃=푛훿/휎; if we find the 휃푆퐼 that achieves power 1−훽, then since the hypothesized effect size is known, we obtain 푛; •Since in the fixed sample test 휃0=푧1−훼2 +푧1−훽, we have the sample size inflation factor: 푅= 휃푆퐼 푧1−훼2 +푧1−훽 2 •By definition 휃푆퐼=Ψ−1(1−훽); by monotonicity, use bisection method to obtain 휃푆퐼; •Note that the inflation factor 푅 is NOT dependent on the hypothesized effect size. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 18
  • 19. METHODS – POWER LOSS Power loss approach: •Without SI to regain power; •We look at equal -CP -CP(휃 ) -PP -Equal power loss (Power loss at tj: 푙푗휃=푃휃(퐵푡푖≥푏푖,푖=1⋯,푗− 1,퐵푡푗<푏푗,퐵1>푧1−훼2 )) at two and three evenly spaced interims. •The trade-off is between power loss and ASN •We also assess the optimal rules (the rules that result in minimum ASN given certain power loss) -Optimization done by grid search | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 19
  • 20. RESULTS – SAMPLE SIZE INFLATION | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 20 SI factor •To achieve 1−훽=0.9 based on equal conditional power 훾 at 푘=1,⋯,5 evenly spaced interims •when 훾≤0.5, inflation is in fact negligible (≤1.1), indicating great practicality in applying these non-binding futility rules through SI. SI Conditional power 훾 k 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 1.002 1.007 1.018 1.038 1.073 1.129 1.221 2 1.006 1.016 1.033 1.063 1.106 1.175 1.282 3 1.009 1.022 1.045 1.078 1.130 1.201 1.314 4 1.011 1.029 1.054 1.092 1.145 1.222 1.337 5 1.015 1.033 1.061 1.102 1.159 1.237 1.355
  • 21. RESULTS – SAMPLE SIZE INFLATION ASN •Same setting as the previous table •There is substantial reduction of sample size for k=2, 3, 훾≤0.5 , a practical range of futility looks with tolerable SI. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 21 ASN Conditional power 훾 k 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 0.823 0.766 0.722 0.691 0.675 0.674 0.696 2 0.756 0.715 0.678 0.647 0.622 0.609 0.609 3 0.716 0.677 0.648 0.622 0.604 0.588 0.582 4 0.693 0.657 0.628 0.606 0.589 0.578 0.571 5 0.680 0.643 0.615 0.595 0.580 0.569 0.563
  • 22. RESULTS – SAMPLE SIZE INFLATION | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 22 Additional issue with SI •Power=1−훽 at hypothesized effect size, by construction; •Need to assess the global power behavior, especially those near the hypothesized 훿 •The right figure shows that the power curves for the “SI”ed futility design are almost indistinguishable from those of the fixed sample (reference) design when 훿≥0.5∗(designed 훿)
  • 23. RESULTS – POWER LOSS Two futility looks | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 23 Power loss Equal CP Equal CP(휃 ) Equal PP Equal power loss Optimal 0.002 0.751 0.769 0.724 0.722 0.717 0.003 0.726 0.727 0.696 0.692 0.689 0.005 0.700 0.692 0.663 0.665 0.661 0.007 0.674 0.658 0.636 0.640 0.635 0.010 0.648 0.627 0.611 0.615 0.610 0.015 0.620 0.596 0.584 0.591 0.584 0.021 0.591 0.565 0.557 0.567 0.557 0.029 0.562 0.537 0.532 0.544 0.532 0.041 0.532 0.508 0.506 0.520 0.505
  • 24. RESULTS – POWER LOSS Equal PP is the closest to the optimal bound over all; Equal power loss approximates the optimal bound only when power loss is very small. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 24
  • 25. RESULTS – POWER LOSS Three futility looks | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 25 Power loss Equal CP Equal CP(휃 ) Equal PP Equal power loss Optimal 0.003 0.708 0.738 0.680 0.673 0.661 0.005 0.683 0.708 0.644 0.641 0.631 0.007 0.659 0.678 0.618 0.620 0.611 0.010 0.635 0.637 0.593 0.594 0.584 0.014 0.610 0.600 0.563 0.569 0.559 0.020 0.584 0.564 0.536 0.548 0.534 0.028 0.556 0.528 0.508 0.522 0.507 0.037 0.527 0.495 0.482 0.500 0.481 0.051 0.495 0.462 0.454 0.476 0.453
  • 26. RESULTS – POWER LOSS Overall, equal PP performs best of all, again; less satisfactory for small power loss; Similarly, equal power loss is doing well for small power losses but not so for greater ones. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 26
  • 27. RESULTS – OPTIMAL BOUNDS Optimal bounds for two futility looks (precision 0.001) | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 27 Power loss 푧1(훾1) 푧2(훾2) 푧3 ASN 0.002 -0.852 (0.361) 0.371 (0.159) 1.960 0.717 0.003 -0.674 (0.409) 0.449 (0.187) 1.960 0.689 0.005 -0.518 (0.452) 0.527 (0.218) 1.960 0.661 0.007 -0.383 (0.490) 0.615 (0.257) 1.960 0.635 0.010 -0.259 (0.525) 0.693 (0.294) 1.960 0.610 0.015 -0.103 (0.569) 0.732 (0.313) 1.960 0.584 0.021 0.032 (0.606) 0.815 (0.355) 1.960 0.557 0.029 0.170 (0.643) 0.884 (0.392) 1.960 0.532 0.041 0.314 (0.680) 0.966 (0.437) 1.960 0.505
  • 28. RESULTS – OPTIMAL BOUNDS Optimal bounds for three futility looks (precision 0.01) | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 28 Power loss 푧1(훾1) 푧2(훾2) 푧3(훾3) 푧4 ASN 0.003 -1.13(0.457) -0.09(0.284) 0.63(0.112) 1.96 0.661 0.005 -0.98(0.491) 0.00(0.316) 0.79(0.174) 1.96 0.631 0.007 -0.86(0.519) 0.08(0.345) 0.84(0.201) 1.96 0.611 0.010 -0.66(0.565) 0.16(0.375) 0.86(0.209) 1.96 0.584 0.014 -0.49(0.603) 0.21(0.394) 0.90(0.229) 1.96 0.559 0.020 -0.38(0.627) 0.34(0.444) 0.94(0.250) 1.96 0.534 0.028 -0.22(0.662) 0.43(0.480) 0.94(0.251) 1.96 0.507 0.037 -0.08(0.691) 0.48(0.500) 1.05(0.312) 1.96 0.481 0.051 0.02(0.711) 0.67(0.575) 1.10(0.346) 1.96 0.453
  • 29. RESULTS – DISCUSSION Equal conditional power is not a good idea for futility rule at multiple time points; Intuitively, the conditional power does not adapt to the observed data as it moves along; The same conditional at a later point means drastically different things comparing to an earlier point, if the early data are already contradicting the hypothesized 휃. Allowing the state of nature to adapt is probably the reason for the success of equal PP. Note that our findings SHOULD NOT be taken to mean that the idea of conditional power is bad. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 29
  • 30. RESULTS – DISCUSSION Compare the bounds: •This is the futility bounds for power loss 0.01; •Equal PP and optimal bounds coincide very well; •Comparing to the optimal, equal CP is conservative in the beginning and aggressive in the end; equal CP(휃 ) is aggressive in the beginning and conservative in the end. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 30
  • 31. DEMONSTRATION OF fut() | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 31 In addition to 훼 and 훽, allows the user to choose interim time points, the corresponding conditional (predictive) power, the scale (CP, CP(휃 ), PP), and whether you want SI. Example: •훼=0.05,훽=0.1 •Two futility looks at one third and one half of the sample size •Use predictive power 0.2 and 0.3 respectively •No sample size inflation to regain power
  • 32. DEMONSTRATION OF fut() •Use the summary() function to print out the details: | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 32
  • 33. DEMONSTRATION OF fut() •Plot the boundary: >plot(D) | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 33 •Plot the power function >powerplot(D)
  • 34. CONCLUSIONS We have established an SI framework to non-binding futility rule with uncompromised power; We have shown that in realistic situations (푘=2,3), equal PP across the time points results in approximately optimal (in terms of ASN) bounds. We have developed an easy-to-use R program fut() for design of nonbinding futility rules. | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 34
  • 35. THANK YOU FOR YOUR ATTENTION! QUESTIONS | Lu Mao | 08/14/2012 | Futility analyses | Business Use Only 35