Sample Size Considerations for Japanese Patients in a Multi-Regional - PowerPoint PPT Presentation

Sample Size Considerations for Japanese Patients in a Multi-Regional Trial Based on MHLW Guidance Hui Quan, Peng-Liang Zhao, Ji Zhang, Martin Roessner and Kyo Aizawa Sanofi-Aventis Presented at 2009 Rutgers Biostatistics Day April 3, 2009

Outline Bridging study to Multi-regional clinical trial (MRCT) PMDA guidance Normal endpoint Survival endpoint Simulation Results Examples Discussion

From Bridging study to MRCT Differences in ethnicity, culture and clinical practice may have impact on efficacy, safety and dose regimen Duplications of large clinical trials in all regions demand resources and delay the approvals of new drugs. ICH E5 issued in 1998 recommends a framework for evaluating ethnical impact Conduct Bridging study to show evidence of similarity Extrapolate data from the original region to a new region

From Bridging study to MRCT (2) No standard for bridging studies: no statistical criteria to assess similarity of two populations Shih (2001): predictive probability of new data falling within the previous experience Chow et al (2002): sensitivity index and bioequivalence approach Hsiao et al. (2003): GS technique for internal validity assuming sequential data availability In Japan: similarity criteria to be set on a case-by- case basis through a PMDA consultation

From Bridging study to MRCT (3) Since ICH E5, new drug approvals in Japan based on bridging strategy increased from 3.2% in 1999 to 25% in 2003 However, bridging studies were often after new drug’s approval in the original region Availabilities of new drugs to Japanese patients were delayed

“Drug lag” in Japan Days from first approval in the world to launch in each country (average of top 100 products) 1417 = app. 4 years 915 2.5 years 757 620 583 538 512 505 Japan France Denmark Germany Sweden Switzerland UK USA

Bridging Strategy The bridging study is often a dose ranging study NDA (1Y review) Phase 1 dose-finding Pivotal trials US/EU: Drug lag Phase 1 Dose-finding / Bridging Japan: J-NDA (2Y review)

PMDA dual approach: PMDA issued a new guidance in September 2007 To Promote Japan’s participation in multi -regional (Global) clinical trials to shorten sponsor’s drug development time in Japan In a Q&A format Q6 is specifically for assessing consistency of treatment effects PMDA planned to shorten the review time Increase PMDA reviewers from 90 to 300 by 2011 Decrease review time from 21 months to 12 months by 2011 Reviewers 9 months / sponsor 3 months Overall, reduce drug lag (time between overseas and Japan approvals) from 4.3 years to 1.5 years by 2011

MRCT towards Simultaneous submissions NDA (1Y review) US/EU: Phase 1 Multi- Intern. regional Simultaneous dose- clinical approvals finding Japan: Phase 1 trials Stand alone Japan study J-NDA (1Y review) Start Japan development earlier Reduce J-NDA review timelines

Key decision point about MRCT Is there a difference in US/EU: Phase 1 Intern. dose-response between the dose- J-population and the other finding study population? Japan: Phase 1 Yes No Analyze other If selected Include J-patients variables, e.g. PK/PD doses are in Multi-Regional the same Pivotal Clinical Confirmation of interaction Trials Conduct parallel clinical trials

PMDA Guidance on MRCT No recommendation of any definitions for consistency, but two methods were provided as examples (superiority trials, Non-inferiority trials?) Method 1: Enough Japanese patients for     1 ' Pr(D Japan /D all > )= ≥0.8 and ≥ 0.5 Observed non-inferiority   Not H 0 : δ JP < δ vs H a : δ JP ≥ δ Method 1: Sekiguchi et al. (JSM, 2007) using simulation for a MR oncology trial.

PMDA Guidance on MRCT (2) Method 2: Enough patients in all regions for   Pr(D 1 >0, D 2 >0, D 3 >0)= ≥0.8 1 ' lack of observed qualitative interaction Method 2: Kawai et al. (DIJ, 2007) The focus here: Method 1 A systematic and comprehensive discussion on sample size calculations Closed form formulas for normal, binary and survival endpoints.

Normal Endpoint   For power and α level two-sided test, the overall 1   2 2 2 ( ) z z    / 2 N  2 Then ˆ ˆ ˆ      ( ) / N N N all J J NJ NJ   u  f Suppose treatment effects and is the J NJ u uJ  fraction of Japanese patients ( ) N f N u

Normal Endpoint (2) For ˆ ˆ          Pr( | , ) 1 ' J all J NJ We have       ( ) ( ( 1 ) ) z z f u u f    / 2 u u z        ' 2 ( 1 ( 1 ) ) 1 ( 2 ) u f f u u

Normal Endpoint (3)    If u =1 or , a closed form solution J NJ 2 z     ' of f 1        2 2 2 2 ( ) ( 1 ) ( 2 ) z z z    / 2 '  ˆ  Treating as a fixed all ˆ            Pr( | ) 1 ' J J NJ We have  2 2 2 2 z z N       ' ' ' N J f N f N       1 1 2 2 2 2 ( 1 ) ( ) ( 1 ) z z   / 2

Normal Endpoint (4)

Normal Endpoint (5) To have a positive trial and satisfy MHLW requirement, consider ˆ ˆ ˆ                 Pr( 0 , / / 2 0 | ) z N  J all all J NJ z  Correlation    ' 0        ( 1 )( 1 ' ) z z   / 2 , The conditional probability ˆ ˆ                  Pr( 0 | / / 2 0 , ) 1 ' z N  J all all J NJ

Normal Endpoint (6)

Normal Endpoint (7) For imbalanced design, N for placebo and kN for active treatment, replace k  1  2 2  2 by k ' f f Actually, and are independent with k 1 u For binary endpoint, replace    2  2 ( 1 ) ( 1 ) p p p p by 1 1 0 0

Survival Endpoint Consider Proportional Hazards model     ( ) ( ) t t e 1 0 The power is often based on log rank test    / 2  E ~ ( , 1 ) T N and where E is the expected total number of events of 2 groups.    ˆ 2 / ~ ( , 4 / ) T E N E   1 For power ,  2 4 ( ) z z    / 2 E  2

Survival Endpoint (2) There are 4 approaches depending on what asymptotic distributions are used for  ˆ  1 J e        Pr( | , ) 1 ' (*)  ˆ  J all 1 all e   ˆ ˆ Difficult to calculate the correlation between & J all  ˆ if pooled data are used for all       ˆ ˆ ˆ  w  ( 1 ) w w ( 0 1 ) Consider all J NJ Note that, this is for design not for analysis

Survival Endpoint (3)  J / When , weight=inverse of the variance and w E E   ˆ ( ) 4 / Var E all is same as the one from the pooled analysis. Consider the asymptotic distribution for   ˆ ˆ     1 J ( 1 all ) e e   u  uJ  g Suppose and . should satisfy E g E J NJ u u

Survival Endpoint (4)    When , a closed form solution J NJ  2 2 4 e z   ' g   1        2 2 2 2 2 ( 1 ) ( 1 ) 4 ( 2 ) E e e z  ' The number of events for Japanese patients 1  E J g E 1

Survival Endpoint (5)  ˆ  Replace by . For all all  ˆ  1 J e           Pr( | ) 1 '   J all 1 all e the number of events for Japanese patients 2 4 z    ' E E       2 J J 2 ( log( 1 ( 1 ))) e

Survival Endpoint (6) As Hung et al. (SIM, 2003), we can also consider asymptotic distribution for   ˆ 1 J e     ˆ log( )( log )   ˆ 1 all e      ˆ J / w E E Then, when in and J NJ all  2 2 4 e z E   ' E J   3    2 2 2 2 (log ) ( 1 ) 4 E e e z  '     ˆ ˆ  2 2 4 Or if set in , e z    all ' E E  4   3 J J 2 2 (log ) ( 1 ) e

Survival Endpoint (7)

Survival Endpoint (8) Different approaches give very different required number of events 1  2  For the first case, 18 . 5 % 24 . 2 % E J E E J E 4  3  11 . 1 % E J E 10 . 1 % E J E Simulation is used to check the coverage