What Is the Role for Analyses of Administrative Data in Assessing Drug Safety in Post-Market Commitment (PMC) Studies?

1. What Is the Role for Analyses of Administrative Data in AssessingDrug Safety in Post-MarketCommitment (PMC) Studies? Cathy W. Critchlow, PhD Global Epidemiology, Amgen, Inc. September 29, 2006

2. Outline • Why should we consider additional approaches (e.g., analyses of administrative data) to post-market commitment studies? • What are situations where analyses of administrative data can be used to supplement, or even replace, clinical post-market commitment studies? • Study design issues • Strengths and limitations of administrative data analyses as a component of post-market surveillance

3. Post-Market Surveillance –Continuing Assessments of Safety or Efficacy Post-market surveillance Post-market commitment studies Routine surveillance • Spontaneous reports • Unexpected or rare AEs Registries or studies conducted to ‘complete’ pre-market assessments Observational studies • Rate of expected/unexpected AEs • Relative rate of ‘hard’ endpoints • Dynamic response to emerging issues, hypotheses Many examples ??? Can analyses using automated databases help meet these needs?

4. Need for Reevaluation of Post-MarketAssessment Strategies • Withdrawal of Cox-2 inhibitors after several years and several million patient exposures contributes to “perception of crisis that has compromised the credibility of FDA and the pharmaceutical industry”* • Public’s loss of faith in the ability of industry to deliver safe and effective drugs† • 13% think pharmaceutical companies are “generally honest and trustworthy” • 60% not confident that drug companies will publicly disclose safety data in a timely manner *IOM Report: The Future of Drug Safety: Promoting and Protecting the Health of the Public †Harris Poll, 2004

5. Need for Reevaluation of Post-MarketAssessments Strategies (2) • Reliance on regulation alone to demonstrate long-term safety has not worked • Unmet phase 4 commitments • 114 (9.6%) of 1,191 open PMCs met* • Confirmation of efficacy using hard endpoints in phase 4 commitment studies for drugs receiving “fast-track” approval based on surrogate measures • FDA Critical Path Initiative • Use of database registries and electronic medical record systems to compare outcomes among relevant patient groups in post-market drug evaluations * Federal Register 2005;70:8379-81.

6. Crux of the Issue…..Post-Approval Drug Safety • Typically, 500 - 3000 patients exposed to drug in phase 3 testing • Drug effects detectable with an incidence ~1–6 per 1000 • To quantify the risk of an event with incidence of 2 per 10,000/year (precision ±1 per 10,000) with 95% probability, need ~80,000 subjects followed for 1 year • Difficult to conduct studies this large in a timely fashion

7. Opportunities • Consider additional or alternative strategies to demonstrate long-term drug safety or efficacy • Analyses of administrative data* in post-market surveillance • Collaboratively establish high standards for the conduct of observational data analyses conducted as part of post-market commitments • Demonstrate safety and effectiveness of drugs in ‘real-world’ settings *Claims data (commercial insurance, Medicare, Medicaid), medical record data, national databases

8. Qualities of the Ideal Database • Comprehensive • Inpatient & outpatient care; ER visits; lab & radiological tests; prescribed & OTC drugs; mental health care; alternative therapy • Large, stable population • Unique identifiers for linkage • Regular, frequent updates • Verifiable, reliable • Capacity for chart review or patient interviews • Confounder data • Compliance

9. But, few databases are ideal….(some are better than others) • Problematic situations • Illnesses that do not reliably come to medical attention • Inpatient drug exposures • Outcomes poorly defined by ICD-9 coding • When necessary confounder data cannot be obtained • Very long latency events • Need to understand the limitations of any database • Purpose for which database was created • Data quality, validity, completeness • Availability of confounder data • Patient follow-up • Access to source information

10. Issues to Consider in the Design of PMCs • What is the objective? • Hypothesis generation (descriptive or exploratory) vs. confirmation? • Evaluating expected vs. unexpected events • Apriori specification of events of interest • Timing of events of interest • Short-term vs. long latency outcomes

11. Most PMCs are Observational…..Issues of Study Validity • Bias • Selection bias • Information bias • Misclassification of covariates, exposure, outcome • Data validity • Confounding • By disease severity, treatment indication, comorbid conditions • Unmeasured covariates • Time-dependent confounding • Physician prescription patterns (‘channeling’) • Dosing variability according to patient responsiveness • What are appropriate comparator groups?

12. Other Considerations…… • Urgency of need for data • Drug first in class or are other relevant data available? • Risk vs. benefit profile • Numbers of persons to be exposed • Expected AE incidence rate • Signal detection – what constitutes a safety signal? Implications for study design, sample size, comparators, interim analyses, scientific rigor required

13. What are situations where analyses of administrative data can be used to supplement, or even replace, clinical PMC studies?

14. PMC Scenario 1: Single-Arm Prospective, Observational Clinical Registry • Characterize long-term safety profile of approved drug • Incidence of various adverse events • Drug utilization in the ‘real-world’ • Special populations, e.g., children • Effect of comorbid conditions • Drug-drug interactions

15. Clinical Registry Hypothesis generating Pre-specified outcomes; unexpected events Modest sample size Will not observe rare events Effect measure Absolute incidence rate SIR (external comparator) Prospective data collection Difficult to assess long latency events Virtual (Database) Registry Hypothesis confirmation Post-hoc analyses of unexpected events Large sample size Study rare events Effect measure Absolute risk in select population Relative risk (internal comparator) Retrospective study design Potential for answers sooner PMC Scenario 1: Single-Arm Prospective, Observational Registry Objective: Characterize drug safety post-approval

16. Clinical Registry Data quality Outcome adjudication Covariate data can be obtained Regulatory definitions of AEs Sources of bias, confounding Potential selection, recall or information bias Differential loss-to-followup with respect to risk of outcomes? Virtual (Database) Registry Data quality Validity of algorithms assessing drug exposure, disease severity, outcomes Comparable ascertainment of data from exposed and comparator groups Data from all medical care providers Sources of bias, confounding Relevant covariate data available? Confounding by indication for treatment, comorbidities Stability of population PMC Scenario 1: Single-Arm Prospective, Observational Registry (2) Objective: Characterize drug safety post-drug approval Objective: Characterize drug safety post-drug approval

17. When/What Could Analyses of Administrative Data Contribute to this PMC Scenario? Objective: Characterize drug safety post-drug approval • When… • Large sample size needed to assess rare events • Events specified apriori • Objective lab-driven diagnoses • Confounder data available • Denominator needed to calculate population incidence rates • What….. • Background incidence rates • Attributable risk of events

18. PMC Scenario 2: Controlled Studies Further Assessing Efficacy • Obtain additional data regarding meaningful clinical endpoints • Confirm estimates of efficacy of drugs receiving “fast-track” approval based on surrogate measures • Head-to-head comparisons • New vs. existing drug

19. Clinical Study Modest sample size Expense of large study of infrequent outcomes Effect measure Relative risk compared to placebo or standard of care Prospective data collection Potential ethical issues in randomized trial with placebo control and/or long follow-up Database Study Large sample size Study rare events Effect measure Relative risk More easily do ‘head-to-head’ comparisons Retrospective cohort study Potential for answers sooner PMC Scenario 2: Controlled StudiesFurther Assessing Efficacy Objective: Compare incidence of clinical (efficacy) endpoints among exposed and unexposed groups

20. Clinical Study Data quality Outcome adjudication Covariate data can be obtained Sources of bias, confounding Potential selection, recall or information bias Differential loss-to-followup with respect to risk of outcome? Database Study Data quality Validity of algorithms assessing drug exposure, disease severity, outcomes Sources of bias, confounding Unmeasured confounders? Confounding by indication for treatment, comorbidities Stability of population PMC Scenario 2: Controlled StudiesFurther Assessing Efficacy (2) Objective: Compare incidence of clinical (efficacy) endpoints among exposed and unexposed groups

21. When/What Could Analyses of Administrative Data Contribute to this PMC Scenario? Objective: Compare incidence of clinical (efficacy) endpoints among exposed and unexposed groups • When… • Large sample size needed to assess rare outcomes, long-latency outcomes • Valid ascertainment of outcome • Short-term effects • Recall or interviewer bias could effect association • What….. • Timely validation of surrogate markers • Head-to-head comparison of outcome incidence

22. Opportunities - Revisited • Potential role for analyses of administrative data in post-market surveillance • Collaborative establishment of regulatory thresholds for the conduct, analysis and interpretation of observational data analyses conducted as part of post-market commitments • Demonstrate safety and effectiveness of drugs in timely fashion

23. Thank you!