d design and analysis approaches i d a l i a h to
play

D Design and Analysis Approaches i d A l i A h to Evaluate - PowerPoint PPT Presentation

Cytel East Users Group Meeting Cambridge, Massachusetts Cambridge, Massachusetts D Design and Analysis Approaches i d A l i A h to Evaluate Cardiovascular Risk October 12, 2012 11:45-12:15 Brenda Gaydos, Ph.D. Research Fellow Outline


  1. Cytel East Users Group Meeting Cambridge, Massachusetts Cambridge, Massachusetts D Design and Analysis Approaches i d A l i A h to Evaluate Cardiovascular Risk October 12, 2012 11:45-12:15 Brenda Gaydos, Ph.D. Research Fellow

  2. Outline � Background g � Statistical Methods (tools) � Development Options � Single CV Trial g � Two CV Trials � Single Large Development Study Si l L D l t St d 2

  3. Background • CV disease remains the leading cause of morbidity and mortality in patients with diabetes • In light of the potentially harmful CV effects raised with rosiglitazone, regulatory agencies now require Sponsors to show that a new therapy for T2DM is not associated with an unacceptable increase in CV risk Primary – Hazard Ratio (HR) – Time to first occurrence of any of the following adjudicated components: • MACE (or 3-point MACE): CV death, non-fatal MI, non-fatal stroke • MACE +: typically 4 th component hospitalization for unstable angina – Cox proportional hazards model – Non-inferiority to standard of care – ITT population 3

  4. FDA guidance: CI for CV Meta Analysis FDA guidance: CI for CV Meta-Analysis Upper bound of a 2- Upper bound of a 2 sided 95% CI for Conclusion estimated CV risk Ratio Data are inadequate to support approval. >1.8 A large safety trial should be conducted The potential for CV harm may still exist. 1.3 – 1.8* An adequately powered and designed post-marketing trial is needed to show an upper bound < 1.3 <1.3* 3 Post-marketing CV trial is generally not needed g g y *with a reassuring point estimate CI = confidence interval 2008 � FDA � Guidance � for � Industry: � Diabetes � Mellitus � – Evaluating � CV � risk � in � new � antidiabetic � 4 therapies � to � treat � type � 2 � diabetes. �� www.fda.gov

  5. Cardiac Safety Research Consortium White Paper Working title Working title: Designs and statistical approaches to assess CV Designs and statistical approaches to assess CV • risk of new type 2 diabetes therapies in development Target journal: American Heart Journal • Objectives Increase the quality and efficiency of CV risk assessment of new • therapies to treat T2DM therapies to treat T2DM Propose study designs and statistical analysis methods to meet • current CV safety regulatory requirements Discuss operational considerations (e g processes for interim Discuss operational considerations (e.g. processes for interim • • analyses) Use simulation to provide examples and discuss impact of • decisions decisions 5

  6. Typical Development Program Efficacy Studies – 3-5 Phase 3 studies (HbA1c is primary) – 1-2 Phase 2 studies 1 2 Phase 2 studies Discharge 1.8 and 1.3 based on meta-analysis – Independent, blinded, adjudication of all CV events – Prospectively planned meta-analysis at end of phase 3 – Sufficient events to allow a meaningful estimate of risk – Include patients at higher risk of CV events (e.g. relatively advanced disease elderly patients some degree of renal impairment) disease, elderly patients, some degree of renal impairment) – Controlled trials of longer duration needed (minimum 2 years) Challenges g – Few events – Typically lower risk population – Relatively short duration – Can meet statistical significance, but be inconsistent across sensitivity analyses 6

  7. No Dedicated CV Trial: Challenging Assume: – All trials start in parallel; Fixed duration follow-up – 1 year to fully enroll a trial; 1% lost to follow up 1 year to fully enroll a trial; 1% lost to follow-up – 90% power for non-inferiority (1.3) True � HR Fixed � Sample � Size Sample Size � Duration (2% � event � rate � (1% � event � rate � on � control) on � control) 0.80 18 � months 10,058 20,017 (178 events) (178 events) 2 � years 6,750 13,405 3 � years 4,106 8,118 1 18 � months 31,028 61,722 (611 � events) 2 years 2 � years 20 831 20,831 41 342 41,342 3 � years 12,681 25,047 7

  8. Some Challenges Initiating a CV Study Initiate during phase 3 development – Benefit: Insure timely discharge of 1.8 – Need CV study prior to knowing dose/effect – If continue the CV study, need to maintain appropriate blind for interim – True HR unknown (assume equivalent for powering) True HR unknown (assume equivalent for powering) – Rate of events unknown (over/under estimate N needed to maintain acceptable duration) Initiate after phase 3 development – Risk not meeting 1.8 – Same uncertainty in unknown HR and rate of events Same uncertainty in unknown HR and rate of events 8

  9. Statistical Methods Setting • Desirable to initiate a CV study in phase 3 development • Desirable to leverage accruing information to mitigate risk in the presence of so much uncertainty • Focus on methods that are well understood Focus on methods that are well understood Methods Methods – Meta-analysis – Group Sequential Designs – Re-estimating #events R ti ti # t – Sample-size re-estimation 9

  10. Statistical Methods Meta-Analysis : Reduce patient exposure by efficiently utilizing events – Acceptable for 1.8 (phase 2,3 & possibly CV trial) p (p , p y ) – Acceptable for 1.3 (CV trials & possibly phase 2,3 trials) – ? Acceptable for 1 (CV trials) • Does superiority need to be demonstrate in a single CV Outcomes trial? • Does superiority need to be demonstrate in a single CV Outcomes trial? • Typically seeing gated hypothesis testing within meta-analysis: 1 st test HR < 1.3, then test HR < 1 • If an interim analysis is utilized for assessing 1.8: – Need acceptable process to maintain blind of ongoing studies – Completely blind the sponsor (CRO or some other body) – Blind the study team, but not the sponsor (e.g. team internal to sponsor, but firewalled from study team; internal steering committee with CRO) • What will be published in SBA? [Transparency / Data Confidentiality] 10

  11. Statistical Methods Group Sequential Designs : Opportunity to stop early for success (1.8, 1.3 or 1) (1.8, 1.3 or 1) – Opportunity to answer the question sooner & reduce patient exposure – Can be combined with meta-analysis to further reduce patient exposure • Determine in advance the maximum number of events and alpha Determine in advance the maximum number of events, and alpha spend – Allows for multiple interims to avoid looking too early or too late – Need to establish minimum clinically meaningful exposure (not Need to establish minimum clinically meaningful exposure (not just about statistical significance on MACE) Current recommendations: � Encouraging of group sequential designs � Determine a-priori alpha spend and number of events at each analysis � Alpha spend is sponsor’s choice (preference for O’Brien-Fleming) p p p (p g) � Report adjusted point estimator 11

  12. Group Sequential Design (GSD) Approaches Assume single CV study to demonstrate HR <1.3, non-inferiority – O’Brien-Fleming spending function, 3 look design for early stopping, 90% power – Fixed design requires 611 events if true HR = 1 True HR Average � # � Events � Average � # � Events � (400, � 513, � 626) (500, � 565, � 629) 1.00 480 527 0.90 418 503 0.85 406 501 0.80 401 500 0.75 400 500 If the true HR is 1 Design Pr Stop � at � Interim � 1 Pr � Stop � By � Interim � 2 Pr � Success �� By Final Analysis 400, � 513, � 626 0.52 0.767 0.90 500, � 565, � 629 0.75 0.838 0.90 12

  13. Statistical Methods Sample-size re-estimation (Duration) : Right-size the study – Sample-size drives study duration, NOT power p y , p – Opportunity to increase sample size if needed to maintain reasonable study duration once more information is gathered on event rate – Analysis can be done using blinded data (observed event rate) Analysis can be done using blinded data (observed event rate) Re-estimating # Events (Power) : Minimize patient years – # events drive power – # events drive power – Delay upfront investment to power for superiority given initial uncertainty in true HR – Size initially for non-inferiority with the option to increase # events if Si i iti ll f i f i it ith th ti t i # t if superiority is likely (e.g. utilize estimate of HR at ~400 events) – Analysis likely will require unblinded data 13

  14. Click to edit Master title style DEVELOPMENT OPTIONS 14

  15. Single CV Trial: Approaches A. Fixed Design: Assessing 1.3 only (or 1) • 1.8 assessed only from phase 2 & 3 via meta-analysis y p y Pro: No interim analysis needed Con: Cannot be used to discharge 1.8 if insufficient events observed (even if initiated prior to end of phase 3) To utilize the CV trial as back-up to discharge 1.8 • Group Sequential Design approach would be needed (alpha Group Sequential Design approach would be needed (alpha spending 1.8) • Needs to be pre-specified in meta-analysis plan PRIOR to unblinding Phase 3 unblinding Phase 3 • CV Trial needs to incorporate an interim analysis based on timing relative to the total #events needed for the meta-analysis 15

Recommend


More recommend