490 likes | 571 Views
CV Risk Assessment for T2DM Therapies. Reliable Evaluation of Benefit-to-Risk in T2DM: Some Statistical Considerations July 1, 2008 Thomas R. Fleming, Ph.D. Professor, Dept of Biostatistics University of Washington
E N D
CV Risk Assessment for T2DM Therapies Reliable Evaluation of Benefit-to-Risk in T2DM: Some Statistical Considerations July 1, 2008 Thomas R. Fleming, Ph.D. Professor, Dept of Biostatistics University of Washington tfleming@u.washington.edu
Reliable Evaluation of Benefit-to-Risk • Surrogate Endpoints • Validation of Surrogates • Evaluation of Safety • Conclusions
Biomarkers (Surrogates) in Diabetes Treatment effects on Biomarkers ~ HbA1c , C-Peptide • Establish Biological Activity • But not necessarily overallClinical Efficacy ~Retinopathy, Nephropathy, CVD, Stroke, MI ~ Major hypoglycemic events: Coma/Seizure
Multiple Pathways of the Disease Process What magnitude and what duration is needed? Intervention HbA1c End-Organ Glycemic Control Diabetic Complications Disease Intervention End-Organ Diabetic Complications Disease HbA1c Glycemic Control
Interventions having Mechanisms of Action Independent of the Disease Process Intervention • HbA1c End Organ • Glycemic Control Diabetic • Complications • Unintended negative effects • e.g., Hypoglycemic Events • Alternative beneficial effects Disease
Recent Experiences: Misleading Surrogates • Strategy for HbA1c 6.0 vs. 7-7.9 Target: ACCORD • ↓ HbA1c vs. Suggested ↑ Mortality • Erythropoietin : Renal, Oncology • ↑ Hemoglobin vs. ↑ Thrombosis & ↑ Mortality • Troglitazone : T2DM Prevention Program (DPP) • ↓ HbA1c vs. ↑ Serious Hepatic Risks • Torcetrapid / Atorvastatin in CHD: ILLUMINATE • ↑ HDL ↓ LDL vs. ↑ Death & ↑ CHD Death/MI/Stroke • Muraglitazar & Rosiglitazone : Type 2 Diabetes • ↓ HbA1c vs. Suggested ↑ Death / Stroke / MI or ↑ MI
End Stage Renal Disease Standard Dose Epogen Hematocrit 30% High Dose Epogen Hematocrit 42% R Goal:Normalize Hematocrit Values reduce Death and MI
Patient Distribution & Percent Deaths by Hematocrit % STANDARD DOSE EPOGEN 60% 45% 30% 15% 0% 27-30 30-33 33-36 36-39 39-42
Patient Distribution & Percent Deaths by Hematocrit % STANDARD DOSE EPOGEN • 30% death RR for 10 pt in hem. 60% 45% 30% 15% 0% 27-30 30-33 33-36 36-39 39-42 60% 45% 30% 15% 0% HIGH DOSE EPOGEN 27-30 30-33 33-36 36-39 39-42
Patient Distribution & Percent Deaths by Hematocrit % STANDARD DOSE EPOGEN • 30% death RR for 10 pt in hem. 60% 45% 30% 15% 0% 27-30 30-33 33-36 36-39 39-42 60% 45% 30% 15% 0% HIGH DOSE EPOGEN 27-30 30-33 33-36 36-39 39-42
Patient Distribution & Percent Deaths by Hematocrit % STANDARD DOSE EPOGEN • 30% death RR for 10 pt in hem. • in hematocrit 60% 45% 30% 15% 0% 27-30 30-33 33-36 36-39 39-42 60% 45% 30% 15% 0% HIGH DOSE EPOGEN 27-30 30-33 33-36 36-39 39-42
Patient Distribution & Percent Deaths by Hematocrit % STANDARD DOSE EPOGEN • 30% death RR for 10 pt in hem. • in hematocrit • 30% in death RR 60% 45% 30% 15% 0% 27-30 30-33 33-36 36-39 39-42 60% 45% 30% 15% 0% HIGH DOSE EPOGEN 27-30 30-33 33-36 36-39 39-42
End Stage Renal Disease Standard Dose EPOGEN High Dose EPOGEN R Results (Interim at 1/2 planned endpoints) n Death/MIDeath Standard Dose 615 164160 High Dose 618 202195 Death / MI relative risk: 1.30 (0.94, 1.79) Besarab et al, NEJM 339:584-590, 1998: “ in incidence of thrombosis of vascular access sites”
Recent Experiences: Misleading Surrogates • Strategy for HbA1c 6.0 vs. 7-7.9 Target: ACCORD • ↓ HbA1c vs. Suggested ↑ Mortality • Erythropoietin : Renal, Oncology • ↑ Hemoglobin vs. ↑ Thrombosis & ↑ Mortality • Troglitazone : T2DM Prevention Program (DPP) • ↓ HbA1c vs. ↑ Serious Hepatic Risks • Torcetrapid / Atorvastatin in CHD: ILLUMINATE • ↑ HDL ↓ LDL vs. ↑ Death & ↑ CHD Death/MI/Stroke • Muraglitazar & Rosiglitazone : Type 2 Diabetes • ↓ HbA1c vs. Suggested ↑ Death / Stroke / MI or ↑ MI
Reliability of Surrogate Endpoints Intervention • Surrogate True Clinical • Endpoint Endpoint • Unintended negative effects • Alternative beneficial effects Disease
Reliable Evaluation of Benefit-to-Risk • Surrogate Endpoints • Validation of Surrogates • Evaluation of Safety • Conclusions
Validation of Surrogate Endpoints Property of a Valid Surrogate ·Effect of the Intervention on the Clinical Endpoint is reliably predicted by the Effect of the Intervention on the Surrogate Endpoint
Z = 0 : target 7-7.9; Z = 1 : target 6.0 S(t) : HbA1cat time t is the CVD / MI / Stroke rate (t | Z) = 0(t) eZ (1) i.e., 1(t) = 0(t) e In ACCORD, to detecte = 0.85 with 89% power when preserving 0.025 false pos error rate, the trial requires L = 1540 events. In turn, this requires approximately 10,000 patients, followed for approximately 5-6 years.
Prentice’s Sufficient Conditions to Validate a Surrogate Endpoint 1.The surrogate endpoint must be correlated with the clinical outcome 2.The surrogate endpoint must fully capture the net effect of treatment on the clinical outcome
Z = 0 : target 7-7.9; Z = 1 : target 6.0 S(t) : HbA1cat time t is the CVD / MI / Stroke rate (t | Z) = 0(t) eZ (1) (2) (t | Z, S(t) ) = 0(t) eZ + S(t) Proportion of net treatment effect explained by the surrogate endpoint: p = 1 -
Meta-analyses are required to explore the validity of surrogate endpoints
Z = 0 : target 7-7.9; Z = 1 : target 6.0 S(t) : HbA1cat time t (1) (t | Z) = 0(t) eZ (2) (t | Z, S(t) ) = 0(t) eZ + S(t) Proportion of net treatment effect explained by the surrogate endpoint: DeGruttola et al, J Infectious Diseases 175:237-246, 1997 p = 1 -
Validation of Surrogate Endpoints Intervention • HbA1c Major Clinical • Glycemic Control Events • Unintended negative effects • e.g., Hypoglycemic Events • Alternative beneficial effects Disease
Z = 0 : target 7-7.9; Z = 1 : target 6.0 S(t) : Hypoglycemic Event at time t (1) (t | Z) = 0(t) eZ (2) (t | Z, S(t) ) = 0(t) eZ + S(t) Proportion of net treatment effect explained by the surrogate endpoint: DeGruttola et al, J Infectious Diseases 175:237-246, 1997 p = 1 -
Validation of Surrogate Endpoints Statistical ·Meta-analyses of clinical trials data Clinical ·Comprehensive understanding of the ~Causal pathways of the disease process ~Intervention’s intended and unintended mechanisms of action
Effects on HbA1c vs. Effects on CVD / Stroke / MI Hypothetical Illustration
Validation across multiple Clinical Outcomes • T2DM(For an array of interventions) • Effects on HbA1c predicting effects on: • Microvascular composite outcomes • Macrovascular composite outcomes • Other composites of clinical efficacy measures • Anti-Hypertensives(>500,000 patients from rand trials) • FDA Cardio-Renal Advisory Committee: June 15, 2005 • Effects on Blood Pressure predicting effects on • each of the following, considered individually: • Stroke, MI, CVD, Overall Mortality, Heart Failure
1.50 Recent trials Recent Older AASK L vs. H ALLHAT/Dox Older trials placebo ABCD/NT L vs. H ATMH 1.25 ALLHAT/Aml EWPHE Older trials active ALLHAT/Lis HEP ALLHAT/Lis 65 HOPE P<.0001 ALLHAT/Lis Blacks HOT ANBP2 HOT M vs. H 1.00 CONVINCE INSIGHT DIABHYCAR MIDAS/NICS/VHAS Odds Ratio (experimental/reference) ELSA L vs. H IDNT2 MRC LIFE/ALL MRC2 0.75 LIFE/DM PART2/SCAT NICOLE PATS PREVENT PROGRESS/Per SCOPE PROGRESSION/Com RCT70-80 0.50 RENAAL SHEP STONE STOP 1 STOP2/CCBs 0.25 STOP2/ACEIs Syst-China Syst-Eur -5 0 5 10 15 20 25 UKPDS C vs. A Difference (reference minus experimental) in Systolic BP (mm Hg) UKPDS L vs. H Odds Ratio for CV Events and Systolic BP Difference: Recent and Older Trials Slide: Henry Black’s lecture Staessen et al. J Hypertens. 2003;21:1055-1076.
Reliable Evaluation of Benefit-to-Risk • Surrogate Endpoints • Validation of Surrogates • Evaluation of Safety • Conclusions
Primary Goal Identifying effective interventions that are safe An important objective: Obtaining a Timely & Reliable Assessment of Benefit-to-Risk … Detecting long & short term risks… … Detecting rare & frequent risks… Enabling one to rule out unacceptable increases in safety risks
Approaches to Pre- and Post-Marketing Evaluation of Safety • I Passive Surveillance Systems • II Active Surveillance Systems • III Large Randomized Clinical Trials • Surveillance for new safety signals • Exploration of existing signals
“Post Marketing” Passive Surveillance Systems AERS: A single system for collection & analysis of reports on all serious AEs (based on voluntary submission of MedWatch forms for AEs that caregivers believe might be drug related) … for Safety Rare events: < 1/1000 “signal detection” … hypothesis generation requires Large RR, or Temporal relationship
Passive Surveillance Systems + Timely information + Uniformity of reporting procedure – Not randomized; no comparator group – Lack of denominator – Voluntary submission underreporting
“Post Marketing” Active Surveillance Systems Large Prospective Cohorts & Large Linked Databases Linking Automated Databases from HMOs 500,000 patients (e.g., Northern California Kaiser) … for Safety • Rare events: < 1/1000
Active Surveillance Systems + Information on a defined population + Complete numerators & denominators + Fast & inexpensive – Not randomized – Confounder information often unavailable – Outcome specificity … is stated event truly an event? – Outcome sensitivity
Safety Assessments: Advantages provided by Randomized Prospective Cohorts • Randomization removes systematically occurring imbalances in baseline characteristics ~Pts/caregivers don’t start/stop treatments “at random” ~ Known & recorded covariates are the “tip of iceburg” • Need to conduct an ITT evaluation Need ability to define a time 0 cohort ~Assess risk over specified time interval, even if intervention is stopped earlier in time ~Risk cannot be assumed to be independent of duration of exposure
Safety Assessments: Advantages provided by (Randomized) Prospective Cohorts • Sensitivity & Specificity ~Systematic and reliable capture of all events ~Timely adjudication, (esp. in open label settings) • Retention ~Maintaining follow-up for targeted durations • Adherence ~ Achieving proper dose and duration of use
TypicallySized (2N = 1K – 20K) • “Pre- & Post-Marketing” Randomized Trials • Moderate term & moderate/high frequency • Annual Rate of Events • 10 vs. 30 /1,000 2N = 2K p.y.* • 1 vs. 3 /1,000 2N = 20K p.y.* • 10 vs. 15 /1,000 2N = 20K p.y.* • Cox 2 Inhibitors, ADHD adults • 20 vs. 25 /1,000 2N = 40K p.y.* • Type 2 Diabetes • + Causal Inference • * 90% power to rule out targeted ↑; = 0.025
The PRECISION Trial: Ruling out Excess Rates of CV Death / Stroke / MI Pain Medications in Patients with Osteoarthritis & Rheumatoid Arthritis With or at Hi Risk for CV Disease Celecoxib Ibuprofen Naproxen R
PRECISION Trial Using the Proportional Hazards Model for the rate of “CV Death/Stroke/MI” Naproxen : Celecoxib : If one sets with 1α=0.025 & power 0.90, then L=508 events are required.
“CV Death / MI / Stroke” Events Celecoxib compared with Naproxen Celecoxib better Naproxen better A B C D CEL: superior CEL: ruling out unacceptably inferior CEL: neither ruling out to be unacceptably inferior, nor establishing as inferior CEL: inferior 0.841.001.12 1.191.333 Hazard Ratio (Celecoxib / Naproxen) Based on analysis at L=508 Events
Performance Standards in Non-inferiority Safety Trials • Enrollment Rate • need timely result • Target Population / Ineligibility Rate / Event Rate • need to address settings where excess risk is most plausible • need sufficiently high risk to achieve targeted number of events • Adherence • must at least match adherence in prior trials with safety signal • include frequency/timing of withdrawal from rand. treatment • Cross-ins • minimize by: careful screening; educating caregivers & patients • …Very challenging in a post-marketing setting… • Retention • critical to maintaining integrity of randomization
Reliable Evaluation of Benefit-to-Risk • Surrogate Endpoints • Validation of Surrogates • Evaluation of Safety • Conclusions
Consequences of Reliance on Surrogate Endpoints For Accelerated or Full Regulatory Approval • Less reliable evidence regarding Efficacy • Less reliable evidence regarding Safety • …The stronger the efficacy evidence , the greater the • resilience regarding uncertainties about safety… • Recent Experiences: • Tysabri : PML in Crohns Disease & Multiple Sclerosis • Erythropoiesis Stimulating Agents : • Chemo-Induced Amemia & Hemodialysis in CHF • Muraglitazar & Rosiglitazone : Type 2 Diabetes
Goals in Development of Interventions in Diabetes • Achieving an efficient evaluation • of experimental interventions… • possibly with the use of biomarkers • Achieving reliable as well as timely • evaluation of both safety and efficacy • of new interventions • Providing patients and care-givers • not only a choice, but an informed choice