1. LEADING RESEARCH… MEASURES THAT COUNT Challenges of Studying Cardiovascular Outcomes in ADHD Elizabeth B. Andrews, MPH, PhD, VP, Pharmacoepidemiology and Risk Management Drug Safety and Risk Management Advisory Committee Meeting February 9, 2006

2. © 2005page 2 Key Questions for Today What is the absolute risk of MI, stroke, and sudden death in users of ADHD medicines? Is the risk of MI, stroke, and sudden death increased in users of ADHD medicines compared with the risk in the general population? Is the risk of MI, stroke, and sudden death increased in users of ADHD medicines compared with the risk in individuals with ADHD and not treated with these medications? Do the risks differ across drugs for ADHD? What is the upper limit of the potential increased risk, given limits of study feasibility? What are the consequences of the answers to these questions? ADHD = attention deficit hyperactivity disorder MI = myocardial infarction

3. © 2005page 3 Start With the Goal We should measure the risk difference, not the relative risk. What added risk has public health or policy significance, given the benefits of treatment? What level of increased risk would be acceptable to patients and their families?

4. © 2005page 4 Ideal Study Subject Selection Individuals with ADHD Individuals with ADHD who use ADHD medicines General population comparison Exposure Measurement Exact dose and duration of treatment Outcome Measurement Identify and characterize 100% of cases of MI, stroke, and sudden death Risk Factor Measurement Characterize all study members on prior medical history, including cardiac abnormalities Capture information on all potential confounders, including substance abuse Follow-up Complete follow-up of all patients for multiple years Analysis Unbiased comparisons of differences across all groups of interest Power or Precision Sufficient size to rule out relatively small increases in risk

5. © 2005page 5 Subject Selection Individuals treated with ADHD medicines Individuals with documented ADHD only? Comparison group not treated with ADHD medicines Pediatrics and adults Lower and upper age limits? Exclusions Underlying cardiac abnormality, if known? History of cardiovascular disease Other (e.g., cancer)

6. © 2005page 6 Exposure Medications All drugs for ADHD? Specific medications Stimulants Methylphenidate Amphetamine Dexmethylphenidate Dextroamphetamine Methamphetamine Atomoxetine Other drugs used in ADHD? Incident use vs. prevalent use Select users with no prior recorded use Multiple medicines for same patient Capture all use, record start date of all new uses, estimate end date

7. © 2005page 7 Outcomes Outcomes of interest MI Stroke Sudden death Common Features Rates show similar patterns by age Rare in children in age-range for ADHD treatment (<3/100,000 p-y) More common in adults (1/1000 p-y) Risk factors show similar patterns by age

8. © 2005page 8 Myocardial Infarction Most cases result in medical attention. Cases can be identified through medical claims. Prior studies have demonstrated high positive predictive value (PPV) for claims algorithms in adults. There is no need to validate diagnoses through chart abstraction. Cases not resulting in medical attention will be missed in most claims databases.

9. © 2005page 9 Stroke Most cases result in medical attention. Cases can be identified through medical claims. Prior studies have demonstrated high positive predictive value (PPV) for claims algorithms in adults. Claims alone may not be sufficient to distinguish between ischemic and hemorrhagic stroke. Important to distinguish between new and repeat stroke (>20% of hospitalizations for stroke are repeat). Cases not resulting in medical attention will be missed in most claims databases.

10. © 2005page 10 Sudden Death Many cases do not result in medical attention. If exposure cohorts are followed via health records, need to link with vital records to ascertain deaths. Death certificates may not have sufficient accuracy to differentiate between deaths of cardiovascular origin from other deaths. Death certificate cause of death will be more accurate if it mentions that an autopsy was performed. Coroner’s reports are ideal for determining circumstances of death (Autopsies will be performed for most cases of sudden death in children).

11. © 2005page 11 Risk Factors - Partial List Age Diabetes Gender Coagulation disorders Race Medications Smoking Hyperlipidemia Substance abuse Epilepsy Hypertension Obesity Cardiac abnormalities

12. © 2005page 12 Source: Epidemiology: An Introduction, Kenneth J. Rothman, 2002

13. © 2005page 13 Design Retrospective cohort study (MI, stroke, sudden death) Include all relevant incident exposures Create operational definitions of risk periods consistent with use and prescribing patterns Drug A Drug B No ADHD Drug Pre-exposure time Exposure time Follow up until event or end of study Drug A Drug B Stroke

14. © 2005page 14 Design Alternative (Case-Control Study) Identify all cases within defined population Sample controls from same population Create operational definitions of exposure periods relevant to outcomes Drug A Drug B No ADHD Drug Drug B Stroke No Stroke Drug A

15. © 2005page 15 Data Collection Strategy Data available in electronic claims Age, sex Exposures (ADHD medications) Outcomes (MI, stroke) Combination of diagnosis codes, procedures, prescriptions Many potential confounders (conditions, medications) Observation time before and after incident exposure Data not available in electronic claims Death, cause of death Smoking BMI Over-the-counter aspirin use Substance abuse Race BMI = body mass index

16. © 2005page 16 Supplemental Data Collection Death Linkage with vital records to determine death Coroner’s report for cause and circumstances of death, cardiac abnormalities Medical record abstraction? Classification of stroke type when not clear from claims (i.e., no Rx for anticoagulants) if stroke type is important distinguish

17. © 2005page 17 Supplemental Data Collection Patient-based survey for race, BMI, smoking, aspirin use, substance abuse Not captured in claims Not captured in medical records Accuracy greatest if patient self-report

18. © 2005page 18 How Many Studies or Analyses? Risks and risk factors may differ between pediatric and adult populations. Example: cardiac abnormality in pediatrics, cigarette smoking in adults Risk factors may differ across the three outcomes of interest. Example: epilepsy for sudden death Six distinct sets of analyses will be required.

19. © 2005page 19 Analysis Plan Measure: incidence of each event (#cases/person-time) Compare incidence by exposure category Stratify on age, gender, covariates if possible In adults, model incidence ratio for ADHD drugs vs. general population using multivariable model, controlling for measured confounders Compare drugs if appropriate

20. © 2005page 20 Adjust for Important Unmeasured Confounders Use external adjustment to control for for confounders measured in survey Obtain prevalence of confounders (e.g., smoking) in ADHD from patient-based survey Obtain prevalence of confounders in general population from literature Obtain estimates of association between confounders and outcomes (e.g., smoking and MI) from literature Adjust and conduct sensitivity analyses

21. © 2005page 21 Study Size Study size needed will be driven by outcome rates, but restricted by availability of data. Outcomes are extremely rare in young populations (<2/100,000). Study may be able to establish upper bound of the potential increased risk.

22. © 2005page 22 For Event With 3/100,000 Over 3 Years Probability that the upper limit of the confidence limit will be less than 3 95% Confidence level Expected Relative Risk = 1 Allocation Ratio: 1 unexposed for each 1 exposed

23. © 2005page 23 Study Size Outcomes more common in adults – study may be able to compare risks between drugs and general population and among drugs.

24. © 2005page 24 For Event With 3/1,000 Over 3 Years Probability that the upper limit of the confidence limit will be less than 2 95% Confidence level Expected Relative Risk = 1 Allocation Ratio: 1 unexposed for each 1 exposed

25. © 2005page 25 Matching the Ideal With the Possible Designs Many differences exist between the ideal and the possible! Anticipated problems Sudden death will not be completely captured. Race information cannot be measured directly. Substance abuse information will not be complete.

26. © 2005page 26 Conclusions - Study Design Data collection and analyses need to be tailored to the three different outcomes. Analyses will differ for children and adults. No single database will be sufficient. Large medical databases: are essential to this type of research question, are limited to medical encounters and associated data, and may be supplemented through linkages and patient-based surveys.

27. © 2005page 27 Conclusions – Next Steps 1. Policy decision is key: determine level of increased risk that would dictate change in clinical guidance – in children and adults. 2. When feasible and appropriate, design study to address these levels of concern. 3. When such a study is not feasible, continue active surveillance study to reduce level of uncertainty around the potential increased risk.