220 likes | 468 Views
Statistical Considerations for Implementing the FDA CV Guidance for T2DM. Craig Wilson, PhD NIC-ASA Fall Meeting October 15, 2009. Regulatory History (1). Rosiglitazone (Avandia; GSK) TZD for treatment of T2DM Nissen and Wolski (May 2007) Meta-analysis of 42 studies (Ph2/3/4)
E N D
Statistical Considerations for Implementing the FDA CV Guidance for T2DM Craig Wilson, PhD NIC-ASA Fall Meeting October 15, 2009
Regulatory History (1) • Rosiglitazone (Avandia; GSK) • TZD for treatment of T2DM • Nissen and Wolski (May 2007) • Meta-analysis of 42 studies (Ph2/3/4) • Increased risk of MI (OR=1.43; 95%CI [1.03, 1.98]) • Possible increased risk of CV death (OR=1.64; 95%CI [0.98, 2.74]) • Resulted in FDA black box warning • Questions regarding validity of analysis • 6 of 48 studies excluded with no events
Regulatory History (2) • FDA Advisory Committee Meeting (July 2008) • Need for data to assess CV risk in T2DM • Endpoints for consideration • Reasonable NI margin for risk ratio to rule out excess CV risk
FDA CV Guidance for T2DM (1) • Final guidance issued December 2008 • Largely influenced by June 2008 advisory committee meeting • Will be folded into draft FDA DM guidance (February 2008) by end of 2009 • Key message: All development programs for T2DM should rule out “unacceptable increase in CV risk”
FDA CV Guidance for T2DM (2) • Programs should analyze “important CV events” • Suggests CV mortality, MI, stroke (core MACE) • Suggests hospitalization for ACS, urgent revascularization, “other endpoints” should be adjudicated • Only core MACE + hospitalization for unstable angina likely to be accepted as primary endpoint per DIA meeting (September 2009) • Offers guidance for new and existing development programs • Investigational drugs must not increase CV risk by more than 80% vs. control to initially market; definitely rule out 30% increase to continue on market
New Studies/Programs • Prospective adjudication • High-risk population • Advanced disease (eg, recent event or long duration of diabetes) • Elderly • Renally impaired • Meta-Analysis (Pooled Analysis) • Should include Ph2/3 studies • Comparison vs. control (placebo and/or active) • Long-term data (eg, ≥2 years) • Stand-alone trial • Might also pool with other analyses
Existing Studies/Programs • “Meta-analysis” of Ph2/3 data • “Unacceptable risk” margins discussed in context of upper bound of 2-sided 95% CI for risk ratio vs. control • >1.8 Safety trial needed (stand-alone or pool with Ph2/3 data) to market • Between 1.3 and 1.8 + favorable risk-benefit market + post-market trial to rule out 1.3 • <1.3 + favorable risk-benefit market; post-market trial “generally may not be necessary”
Stand-Alone Trial Considerations • Stand-alone trial: “…if the data from all the studies that are part of the meta-analysis will not by itself be able to show that the upper bound of the 2-sided 95% CI is <1.8, then an additional single, large safety trial should be conducted that alone, or added to other trials, would be able to satisfy this upper bound before NDA/BLA submission.” • Stand-alone or pooled approach can also be used if between 1.3 and 1.8 • Guidance doesn’t address combining 1.8 and 1.3 analyses in same trial • Guidance introduces possibility of approval for <1.8 when true risk ratio is 1.3
Advisory Committee Meetings • Conducted in April 2009 • SMQ MACE: CV-related events (serious or non-serious) based on MedDRA SMQs • Custom MACE: CV-related events (serious or non-serious) identified by FDA
Saxagliptin • Onglyza; DPP-4 inhibitor; developed by BMS • Presentation included controlled data up to 1 year • Had favorable risk ratios • SMQ MACE (RR=0.85; 95% CI [0.52, 1.42]) • Custom MACE (RR=0.2; 95% CI [0.04, 0.79]) • Approved by FDA in August 2009
Liraglutide • Victoza; GLP-1; developed by Novo Nordisk • Presentation included controlled data up to 1 year • Had less favorable risk ratios • SMQ MACE (RR=0.9; 95% CI [0.56, 1.31]) • Custom MACE (RR=0.7; 95% CI [0.32, 1.57]) • Data comparing to placebo alone not as favorable • Approved by EMEA in June 2009; still awaiting FDA approval
DIA Meeting • Held September 2009 • Core MACE (CV death, MI, stroke) only agreed primary endpoint • Concerns regarding unstable angina adjudication (ability to capture only serious/significant cases) • Some flexibility may exist for UA in the future • Traditional statistical approaches encouraged (non-adaptive) • Concerns about Type 1 error protection • Total of 18 CV study designs submitted to FDA so far • Suggestions that EMEA may follow a more “holistic” approach to assessing CV risk than FDA
Possible Stand-Alone Trial Designs (1) • Stand-alone trial most robust way to rule out CV risk • FDA prefers placebo comparison per guidance • Few opportunities to determine scope of possible analyses acceptable to FDA • Must rely on reasonable available methods
Possible Stand-Alone Trial Designs (2) • Time-to-event analysis (Cox regression) seemingly most appropriate analysis method • Reasonably simple and widely understood • Easily generates required hazard margin • Specific methods not stated in FDA guidance; however, method should account for potential non-constant hazard • Intent to treat • Survival function specification may rely on completed trial • Proportional hazards assumption? • Implications of interim analyses?
Possible Stand-Alone Trial Designs (3) • Traditional analysis • Single analysis after 611 events to rule out 1.3 • Pros: Widely accepted; no concern about Type 1 error control • Cons: One-and-done scenario; time consideration if needed to market
Possible Stand-Alone Trial Designs (4) • Separate trials (Fleming; DIA) • Conduct one trial to rule out 1.8 (122 events) • Conduct a separate trial (611 events) to rule out 1.3 • Pros: Widely accepted; no concern about Type 1 error control • Cons: Still one-and-done scenarios; what if margin barely missed? What if trials give conflicting results? Cost considerations
Possible Stand-Alone Trial Designs (5) • Traditional analysis with single step-down interim look • Analyze after 122 events to rule out 1.8 • If successful, continue to 611 events to rule out 1.3; otherwise, declare futility and stop the trial • Pros: Efficient use of study population • Cons: Prevents assessment of 1.3 if 1.8 can’t be ruled out; power reduction to rule out 1.3 without more events
Possible Stand-Alone Trial Designs (6) • Adaptive monitoring (Connor and Berry; January 2009) • Currently being used for CV trial for Libigel (BioSante) to rule out an upper bound of 2.0 • Could be modified to current CV guidance • Use predictive probabilities to monitor accrual and likelihood of trial success (<1.3) • Conduct periodic pre-planned analyses • Stop accrual for high predictive probability of trial success • Declare futility for low predictive probability of trial success • Stop trial for success when upper bound of CI is <1.3 • Type 1 error and power confirmed via simulation • Pros: Allows frequent monitoring of trial; potentially reduces total subjects enrolled; may allow interim monitoring to rule out 1.8 • Cons: Potential for regulatory agencies to be skeptical, given strong interest in Type 1 error protection (DIA meeting)
Possible Stand-Alone Trial Designs (7) • Group sequential design (Fleming; DIA) • Analyze after 122 events to rule out 1.8 • If successful, continue to 611 events to rule out 1.3; use alpha-spending function (OBF) to control Type 1 error for 2 analyses • Pros: Efficient use of study population; still allows assessment of 1.3 even if 1.8 can’t be ruled out • Cons: Would potentially require more time to market (if 1.8 can’t be ruled out); difficult to rule out 1.8 at interim unless true hazard highly favorable; may require more events than stated to maintain desired power
Possible Stand-Alone Trial Designs (8) • Group sequential design with step-down • Use alpha-spending function (OBF) and repeated CIs to rule out 1.8 (Durrleman and Simon [1990]) • If successful, use separate alpha-spending function (OBF) and repeated CIs to rule out 1.3; otherwise, stop the trial for futility • Pros: Efficient use of study population; provides more power to rule out 1.8 if true hazard is close to 1.0; provides early assessment to rule out 1.8 • Cons: Prevents assessment of 1.3 if 1.8 can’t be ruled out; potential power reduction to rule out 1.3 without adding more events
Superiority? • Could design from beginning as superiority trial using traditional approach • Would likely require substantial data • Unlikely to get label claim • Step-down analysis once 1.3 ruled out? • Dependent on proximity of upper 95% CI bound to 1.3 • Adaptive design likely best suited for task • How long do you look?