Efficacy Data in regulatory settings, DSBS January, May 2013

H. Lundbeck A/S

Efficacy Data in regulatory settings, DSBS January, May 2013

H. Lundbeck A/S 2

Outline

Part 1: Objectives and Endpoints in test strategies

Part 2:

– Integrated Data Analysis: Purpose, Requirements, Terminology

– Methodology for Pooled and Meta Analysis

– Applications to filing of Vortioxetine

H. Lundbeck A/S 3

Part 1: Endpoints in RCTs

Secondary Endpoints are Increasingly importantfor differentiation of products

• highly competitive markets• demands from authorities• Publishing on clinicaltrials.gov

H. Lundbeck A/S 4

Definitions of Endpoints in RCTs:

’Good old Days’: Primary, Secondary and Exploratory

Now:

Primary: More or less as before

Secondary: Key SecondariesOther Secondaries

Exploratory: perhaps bigger than before

H. Lundbeck A/S 5

Regulatory view

Primary Endpoint: Multiplicity control in case of e.g. several doses

Key Secondary Endpoints should be under proper multiplicity control together with the primary and can potentially be included in labelling text and promotional material. Will normally require significant primary

Other Secondary Endpoints can (normally) not be included in labelling text but have to go on ’www.clinicaltrials.gov’

Exploratory endpoints can (normally) not be included in labelling text but does not have to go on ’www.clinicaltrials.gov’

- Unclear whether secondary analyses have to go on .gov

H. Lundbeck A/S 6

Authority Requirements to protocols and SAPs (EMA+FDA)

• Clinical formulation of objectives

• Clear correspondence between objectives and endpoints

Testing Strategy• Primary and Key secondaries should be selected based

on ‘Objectives’• Only one endpoint per objective. No redundancy• Only one analysis method (population) per endpoint

H. Lundbeck A/S 7

Objectives and Endpoints

Objective Endpoint Analysis Methodology

Similar for other objectives. Select one row within each objectiveOften a mix is seen in protocols

H. Lundbeck A/S 8

Primary Analysis

Objective Endpoint Analysis Methodology

: Secondary analysis method of primary endpoint adressing primary objective Can not be used as key secondary

H. Lundbeck A/S 9

Key Secondary Analysis I

Objective Endpoint Analysis Methodology

H. Lundbeck A/S 10

Key Secondary Analysis II

Objective Endpoint Analysis Methodology

H. Lundbeck A/S 11

Example: Depression Study

Primary Objective:

Evaluate the efficacy of LuAA21004 compared to Placebo on depressive symptoms (in patients with MDD).

Key Secondary Objectives:

Evaluate the efficacy of LuAA21004 compared to Placebo on1. Global Status2. Functioning3. Anxiety

Assessments/endpoints MADRS, HAM-D (Response,Remission)adressing objective CGI-S, CGI-I (Response, Remission) SDS, work/family/social/total HAM-A, HAM-D Subscale

H. Lundbeck A/S 12

Hierarchical Testing

MADRS

HAM-A

SDS

CGI-I

Depression

Global Status

Functionalitywork/social/family

Anxiety

One endpoint per objectiveTwo doses: alfa=2.5% in each sequence

H. Lundbeck A/S 13

Primary Objective

Objective Endpoint Analysis Methodology

Response/Remission considered redundant, not a separate objective.However, special interest in EU

Depression

MADRS

OC

Response

HAM-D

Remission

LOCF

Non-par

MMRM

H. Lundbeck A/S 14

Response and Remission

EMA: Particular Clinical Relevance + Redundant

FDA: Arbitrary and Inadequate Definition

+ Redundant

Response and Remission:– attractive for profiling– attractive for pricing – difficult to formulate as separate objective

H. Lundbeck A/S 15

EMA: Responders

MADRS

HAM-A

SDS

CGI-I

- proceed as long as p<0.05

MADRS 50% Response

MADRS Remission

”Branching”, overall α>5%

Confirming Clinical Relevance

H. Lundbeck A/S 16

Number of Key Secondary Endpoints

– no formal requirement or limitations– limited through non-redundancy within and

between objectives– ’Rule of thumb’: 4-5 tests within each dose– chose hierarchi according importance and

’hit-likelihood’– status of non-tested endpoints can be unclear

H. Lundbeck A/S 17

Testing Strategy

– How to report p-values outside testing strategy or after stop within sequence ?

– Phrasing ’Statistical Significance’ should be reserved for results from testing strategy

Tip: Phrasing ’seperated from placebo’ has been introducedin accepted Lundbeck publications and in filing documents.

Other possible phrasings: Nominal significanceNominal p<0.05Nominal evidencePotential significance

H. Lundbeck A/S 18

Part II

Integrated Data Analyses in Regulatory Setting

H. Lundbeck A/S 19

Integrated Data Analyses

When a clinical development program entersregistration phase a need for integrated analyses arises:

• Regulatory requirements• Questions during approval phase• Profiling after approval

H. Lundbeck A/S 20

Terminology

Meta Analysis

Pooled Analysis

Integrated Data Analysis

H. Lundbeck A/S 21

Terminology

Definitions of Meta-Analysis:

FDA: Meta-analysis refers to the analysis of analyses...the statistical analysis of a large collection of analysis results from individual studies for the purpose of integrating the findings. (Glass, 1976) Examples of related terms used in literature include: analysis of combined data, combined analysis, analysis of pooled data, and pooled analysis. No matter what term is used, the objective is to use appropriately sound methods when formulating an integrated analysis.

ICH E9 + EMA: The formal evaluation of the quantitative evidence from two or more trials bearing on the same question. This most commonly involves the statistical combination of summary statistics from the various trials, but the term is sometimes also used to refer to the combination of the raw data.

H. Lundbeck A/S 22

Meta-analysis Definition (google)

• Statistical solutions Software: Meta-analysis is a statistical technique in which the results of two or more studies are mathematically combined in order to improve the reliability of the results. Studies chosen for inclusion in a meta-analysis must be sufficiently similar in a number of characteristics in order to accurately combine their results. When the treatment effect (or effect size) is consistent from one study to the next, meta-analysis can be used to identify this common effect.  When the effect varies from one study to the next, meta-analysis may be used to identify the reason for the variation

• Wikipaedia: In statisitics, a meta-analysis refers to methods focused on contrasting and combining results from different studies, in the hope of identifying patterns among study results, sources of disagreement among those results, or other interesting relationships that may come to light in the context of multiple studies

H. Lundbeck A/S 23

Terminology

Meta Analysis

Pooled Analysis

Integrated Data Analysis

Meta Analysis

H. Lundbeck A/S 24

Terminology

Meta Analysis

Pooled Analysis

Integrated Data Analysis

Pooled Analysis

H. Lundbeck A/S 25

Terminology in this presentation

Meta Analysis

Pooled Analysis

Integrated Data Analysis

H. Lundbeck A/S 26

Terminology

Meta Analysis (AD):

A specific statistical methodology based on summary statistics or aggregate data from each trial (AD Meta Analysis)

Pooled Analysis (IPD):

Statistical analysis based on data pooling at individual patient data level, that is, combination of raw data. (IPD Meta Analysis)

H. Lundbeck A/S 27

Pre-requisities for Integrated analyses

Similarity of Studies with respect to

• Clinical endpoints• Study designs• Populations

H. Lundbeck A/S 28

FDA: ISE requirements

• Integrated summary demonstating substantial eveidence of effectivenes

• Evidence to support recommended dosing in labelling

• Analyses in subgroups: Sex, age, race• Dosing in specific subgroups

- So, actually no specific demand for integrated analysis, could just be side by side presentation

H. Lundbeck A/S 29

EMA: Points to consider on Meta-Analysis

• Not a requirement

• Cannot normally serve a primary

• Cannot save individual negative studies

• Needs prespecification

H. Lundbeck A/S 30

EMA: Pre-requisits for acceptance of results from Meta-analysis as pivotal evidence

Pre-specification

• Statistical Methodology• Arguments for Inclusion and exclusion of

studies• Plan for evaluation of robustness of

results: subgroup, subsets of studies etc..• Populations

H. Lundbeck A/S 31

EMA: Accepted Purposes of Meta-Analysesfor supportive evidence

• Precise estimate of treatment effects

• Confirm effect in subgroups

• Secondary outcomes requiring more power

• Evaluate dose-response

• Evaluate conflicting study results

- similar to FDA ISE

H. Lundbeck A/S 32

Pooled Analysis

H. Lundbeck A/S 33

Properties of Pooled Analysis

Intuitively attractive using individual patient dataFlexibility in having original data (subgroups, outliers etc.)

Complex statistical modelling possible/necessaryAssumptions on variability, baseline dependence, sites etc. Heterogeneity not straightforward

Risk of getting meaningless comparisonsDesign and convergence issues when using MMRM

Not really recommended by FDA?: Correspondence in Relation to AA21004: ’pooling on patient level is in general not recommended’

H. Lundbeck A/S 35

AA21004 Data Package for MDD

8 Studies for Major Depressive Disorder

Differences between Studies:

1. Duration, 6-week 8-week2. Doses: 1, 2.5, 5, 10, 15 ,20 3. Primary endpoint: MADRS, HAMD-244. Method for primary (ANCOVA LOCF, MMRM, nominal/window)5. Test Strategy (step-down/alfa-split)6. Differences in key secondaries: SDS, Response, CGI7. Region8. Results

H. Lundbeck A/S 36

AA21004/Vortioxetine Studies for MDD

Short-Term Long-Term

HLu 11492,PoC

HLu 11984,DF

TAK305

HLu 13267

TAK315

TAK316

TAK303

TAK304

Hlu 12541, Elderly

HLu 11985, Relapse prevention

6 weeks 8 weeks 8 weeks 8 weeks 8 weeks 8 weeks 6 weeks 8 weeks 8 weeks OL: 12 weeksDB: 24-64 weeks

PBO

5 mg10 mg

225 mg VEN (AR)

PBO2.5 mg5 mg10 mg

60 mgDUL (AR)

PBO1 mg5 mg10 mg

PBO

15mg20mg

60 mg DUL (AR)

PBO

15mg20mg

60 mg DUL (AR)

PBO

10mg

20mg

PBO

5 mg

PBO2.5 mg5 mg

60 mg DUL (AR)

PBO

5 mg

60 mg DUL (AR)

PBO

5 mg10 mg

EU/Asia/CA

EU/Asia/CA

EU/ZA/Asia

EU/ZA US US US US EU/CA/US

EU/CA/Asia

Positive study

(MADRS, LOCF)

Failed study, but supportive

(MADRS, LOCF)

Positive study

(HAM-D24, MMRM)

Positive study

(MADRS, MMRM)

Positive study

(MADRS, MMRM)

Positive study

(MADRS, MMRM)

Failed/nega-tive study

(HAM-D24, LOCF)

Negative study

(HAM-D24, LOCF)

Positive study

(HAM-D24, LOCF)

Positive study

(Time to relapse)

H. Lundbeck A/S 37

Methodology for Pooled Analysis

Example: Standard ANCOVA model

MADRS, LOCF, Week 8

Alternatives (SAS):

1. model MADRS_DL = MADRS_BL ARMCD; (Naive)

2. model MADRS_DL = MADRS_BL ARMCD STUDY;

3. model MADRS_DL = MADRS_BL ARMCD SITEID(STUDY);

4. model MADRS_DL = MADRS_BL*STUDY ARMCD SITEID(STUDY);

Further Modelling: Random STUDY*ARMCD; Random Treatment*Study Effect Repeated group=study; Heterogeneous Variability

H. Lundbeck A/S 38

Example: Treatment versus Placebo

Estimate S.E P-value

Original StudyN: 112 versus 105

-6.42 1.36 <0.0001

Pooled Analysis N: 112 versus 1290

-4.60 1.11 <0.0001

- substantial difference in estimate

TTreatment Arm only in one Study

H. Lundbeck A/S 39

Interpretation of Pooled Methodology

Misleading Estimates

MMRM design and convergence problems

Modelling does not seem to account for all study differences

A lot of effort can be done to make the pooled analysis do what the meta-analysis seems to do automatically

Seems not to be the best choice for AA21, but was used for small subgroups

H. Lundbeck A/S 40

Meta Analysis

H. Lundbeck A/S 41

Properties of Meta Analysis

Analysis of analyses

Original data not needed(survey setting not so relevant for AA21)

Only relevant comparisons are retained

Works on any treatment estimate (+/-SE)logistic regression, Cox, ANCOVA, SES

Well-established method for heterogeneity

Less powerfull ?

Pairwise Comparisons mainly

H. Lundbeck A/S 42

Methodology for Meta Analysis

Trials: i=1,…,k

Fixed Effects Modelling:

ai = true treatment effect in trial iâi= estimated treatment effect in trial i

vi = variance of âi

wi = 1/vi, weights

Estimated effect: â = Σwiâi/Σwi

Variance of estimate: 1/Σwi

Test H0: ai=0: (Σwiâi)2/Σwi ~ ϰ2(1)

ref: Encyclopaedia of Biostats. page 2570-2578Der Simonian (1976)

H. Lundbeck A/S 43

Methodology for Meta Analysis with heterogeneity: Random Effects

Test for heterogeneity of effects

Q = Σwi(âi – â)2 ~ ϰ2(k-1) I2=max(0,100*(Q-k)/Q)

I2 describes the percentage of total variation across studies that is due to heterogeneity rather than chance (ref: Higgins, 2007)>50% considered problematic

Random effects in case of Heterogeneity: ai ~ N(a*,σ2) , σ2 estimated using Q (ref: Der Simonian)

wi*= 1/(vi+σ2)

â*= Σwi*âi/Σwi

*

Variance of estimate: 1/Σwi*

Test: (Σwi

*âi)2/Σwi* ~ ϰ2(1)

H. Lundbeck A/S 44

Meta Analyses in SAS

PROC MEANS;

H. Lundbeck A/S 45

Plan for Meta Analyses on AA21004 for regulatory purposes

• ‘Prespecification’ in separate SAP

• To be shown in 2.7.3

• Applied for sub-groups: gender, baseline severity

• Pooled analyses for small subgroups

- not all studies finalised at planning stage

H. Lundbeck A/S 46

Preliminary Meta Analysis without two non-finalised studies. Differences to Placebo

Dose Response ?

10 better than 5 ?

Fixed or Random ?

Removed for confidentiality reasons

H. Lundbeck A/S 47

Meta Analysis with all studies

Dose Response ?

15 mg ?

Removed for confidentiality reasons

H. Lundbeck A/S 48

Meta Analysis SummaryDifferences to Placebo

Removed for confidentiality reasons

H. Lundbeck A/S 49

Pooled versus Meta Analysis of AA21004

• Severe heterogeneity complicates interpretations

• Confounding with Region US/Non-US

• For both analysis types it is mandatory that interpretations involving comparions across treament arms take the individual study results into account.

• The random effects model has less power in the presence of heterogeneity but estimated treament differences change only slightly. Does not solve all heterogeneity problems.

• Random effects not feasible in pooled MMRM, but gets close to Meta results for LOCF

• Neither method completely satisfactory

• Mixed treatment comparison (MTC) meta-analysis allows several treatments (doses) to be compared in a single analysis while utilising direct and indirect evidence

H. Lundbeck A/S 50

Meta Analysis in the filing documents

• Need to downplay due to severe heterogeneity

• Demonstrate Region issue US/Non-US

• Results across subgroups: age, bmi, gender, severity

• Argumentation for dose

H. Lundbeck A/S 51

Planned talk at 8 January 2013

…… 4 Months Later ……..

H. Lundbeck A/S 52

Lu AA21004, Vortioxetine, Brintellix

August 2012: Filing EUSeptember 2012: Filing US

January 8: Planned Talk at DSBSJanuary 17: Day 120 Q’s received

April 15: Day 120 Q’s answered

June 7: Day 180 Q’s

H. Lundbeck A/S 53

Day 120 Questions

•Focus on US versus non-US

•No focus on pool- versus meta- approach

•Some value of meta-analyses in terms of dosing and subgroup arguments

•Testing strategy issues only in relation to PROs

H. Lundbeck A/S 54

Guidelines

EMA: Points to consider on applications with meta-analysis, (2001)

FDA: Guidance for industry. Integrated Summary of Effectivenesss ISE (2008)

ICH E9: Statistical Principles for Clinical Trials