Sample Size Re-Estimation: Controlling the Type-1 Error Yannis - - PowerPoint PPT Presentation

Yannis Jemiai, Ph.D.

26 September 2017 ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop

Sample Size Re-Estimation: Controlling the Type-1 Error

2

unblinded sample size re-estimation is an essential design tool

Y Jemiai – 26 Sep 2017

Addresses uncertainty in trial design assumptions One of the most popular adaptations, especially when using a Promising Zone approach 21st Century Cures Act, PDUFA VI, encourage the use of adaptive designs Regulatory guidance documents exist from EMA (2007), FDA CDER / CBER (2010), and CDRH (2016) Increasingly many examples of regulatory acceptance

Regulatory-Industry Statistics

3

So what are some of the issues concerning uSSR designs?

Y Jemiai – 26 Sep 2017

Can type-1 error be controlled? Can sound adaptive decision rules be developed? How do we get a point estimate and confidence intervals for the treatment effect? How do we avoid operational bias during trial conduct?

We focus here on type-1 error control

Regulatory-Industry Statistics

4

Why does type-1 error get inflated?

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

Consider a two-stage design without sample size increase Suppose now that we increase the sample size in stage II from n(2) to n*(2), but we do not change the critical value This will lead to type-1 error inflation

5

How can we control type-1 error then?

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

6

Chronology of development (partial list)

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

7

• 1. Use a weighted statistic with pre-specified weights

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

8

Also called the p-value combination approach

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

9

• 2. Use the Conventional Wald Statistic

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

10

Extended CDL Method

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

11

Why does it work?

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

12

… and what are the concerns?

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

13

• 3. Preserve Conditional type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

14

Preserving the overall type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

15

Conditional type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

16

Conditional type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

17

Conditional type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

18

Conditional type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

19

Conditional type-1 error rate

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

20

Points to consider

Y Jemiai – 26 Sep 2017

Handling survival endpoints Usable information at interim analysis Non-inferiority & equivalence settings Independent increments Small samples

Regulatory-Industry Statistics

21

Recap: challenges in unblinded SSR trials

Y Jemiai – 26 Sep 2017

Type-1 error control is not an obstacle. Methods exist to ensure strong control Inference remains a challenge, but making some progress Decision-making algorithm can be optimized using simulations and latest research Operational bias can be addressed/minimized by using iDMCs, putting in place proper processes, and making use of technology

Regulatory-Industry Statistics

22

“By failing to prepare, you are preparing to fail.”

• Benjamin Franklin

22

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics

23

Thank you

Y Jemiai – 26 Sep 2017 Regulatory-Industry Statistics