RTI Health Solutions Practical and Methodological Issues in Long-Term Follow-Up Studies Elizabeth B. Andrews, Ph.D. Vice President, RTI Health Solutions

Methodologic Issues

3 Why Conduct Long-Term Follow-Up Studies?  Adverse events may not manifest until months or years after treatment  Adverse events which were ambiguous during short-term courses of therapy may manifest clearly with long-term therapy  Adverse events may be infrequent and require larger sample size than possible in clinical trials

4 Start with the Goal in Mind  Key questions for today  Are these drugs associated with cancer at a level that would warrant modification of current prescribing recommendations?  What is the baseline level of risk of skin cancer and lymphoma in the pediatric population?  What is the estimated increase in risk that must be detected for safety assurance?

5 Start with the Goal in Mind  Should the potential increase in risk for the adverse event, as compared to background, be measured by relative risk or risk difference? If baseline 10-year risk is 2/10,000 and observed risk is 10/10,000  Relative risk is 5  Risk difference is 8/10,000 (roughly 1 new case per 1,000 exposed, over 10 years)  What potential increase has public health/policy significance?  What level of increased risk would be acceptable to patients/families?

6 Surveillance Study vs. Etiologic Study  What is the goal of the study?  Reduce the uncertainty about a possible increased risk? (Surveillance approach)  Use standard study designs, in general  Consider analytic methods to evaluate information as it emerges  Detect or rule out small increase in risk? (Etiologic approach)  Use standard study design  Power study to achieve predicted outcome

7 Surveillance Study – Example  Acyclovir Pregnancy Registry  Patient exposure to acyclovir during pregnancy registered and followed  Outcomes captured (e.g., pregnancy outcomes, birth defects)  Frequency of birth defects compared to population expected data (collected using similar methods)  Concluded that overall frequency of birth defects was similar in acyclovir and general population (3–4%)  Determined that study had ability to detect a 7-fold increase in risk of events that occur 1/1,000.

8 Comparison Groups – Points to Consider  What is the goal of the study?  Detect possible signal?  Reduce uncertainty relating to possible increased risk?  Single arm registry  Can identify incidence of events over follow-up period in exposed patients  Can identify if and when event rate exceeds threshold of “expected” risk  Need well-defined data on “expected” risk  Study with concurrent comparison group  Can establish whether the incidence of events is similar between exposed and comparison groups  Can explore role of potential confounders  Can help assess signal from exposed group (e.g., 2 cases out of 5,000 over 3 years)

9 Study with Comparison Group – Example  Rheumatoid Arthritis Azathioprine Registry (RAAR)  Enrollment over 10 years through rheumatologists of patients starting AZA (n=420) or other DMARD (n=1006) therapy  Follow-up for exposures and serious events (e.g., lymphoma, all cancers) for 5+ years per patient  Follow-up excluded basal cell and squamous cell carcinomas because of potential detection bias and underascertainment  Sample size designed to enable detection of increased risk of 2.5–3-fold with full follow-up

10 Potential Study Designs  Longitudinal Follow-Up Study  Case-Control Study  Variations on either

11 Longitudinal Follow-Up Study Cohort study: Two or more groups identified based on exposure, followed over time, and compared for events of interest Measures of frequency of event: incidence, risk Measure of comparative risk: risk or rate ratio Exposed Patients Event Comparison Patients No event Time Event No event Examples: RAAR, patient registries, large simple trials

12 Case-Control Study Case-control study: Groups identified based on events and compared for antecedent factors Measure of exposure: Exposure odds or rate Measure of effect: Odds ratio Exposure Cases Exposure Controls No Exposure Time No Exposure Examples: Studies of Agranulocytosis, Severe Cutaneous Reactions, Vaginal Cancer (DES), Neural Tube Defects (Folic Acid)

13 Study Population – Points to Consider  Methods for identifying patients (e.g., referral centers, direct patient recruitment)  Will methods select typical patients or highly skewed cohort?  Will patients be newly treated or already on therapy?  Inclusion criteria (e.g., indication, severity, exposure level)  Will study maximize ability to detect risk if true risk exists?  Should study represent “typical” use patterns?  Comparison group  Will this group have same baseline risk as exposed group? If not, how will they differ? What analytic methods will be used in comparison?

14 Exposure Measurement – Points to Consider  What minimum exposure dose and duration are sufficient for inclusion?  What level of ongoing exposure information is necessary?  Drug  Dose  Duration  Site  What periodicity of follow-up is necessary?

15 Outcome Measurement – Points to Consider  How would outcomes be identified?  Patient self-report  Record abstraction from treating physician  Required physical exams  Link with cancer registry and/or other files  What level of detail is required?  What biases might be expected?  Greater detection in calcineurin inhibitor group if drugs are suspected to be associated with outcome

16 Potential Confounders – Points to Consider  Other treatments for atopic dermatitis  Other conditions related to atopic dermatitis (e.g., asthma) and treatment  Other variables not yet understood at time of study planning

17 Analysis Issues  Analytic methods need to handle time- dependent measures  Patient characteristics at enrollment  Medication exposures  Potential confounders  Unanticipated practice pattern changes

18 Ideal Design  Subject Selection: Exposed and unexposed group with same baseline risk  Exposure Measurement: Dose and duration of all relevant treatments and potential confounders  Outcome Measurement: Complete ascertainment of outcomes in both groups  Follow-up: Complete follow-up for sufficient time to observe outcomes (10 years?)  Power: Ability to detect or rule out an increased risk of X over the expected or observed in the unexposed group But … is the ideal practical?

Practical Considerations

20 Selecting the General Approach Complexity Size Small (1,500) Large (10,000+) Simple (e.g., mail) Highly Complex (e.g., Routine physician exams) Major Public Health Trials (e.g.,WHI) Pivotal RCTs Large Safety Studies ** *

21 General Considerations  Cost of study  Opportunity costs (to regulators, sponsors, physicians, patients)  Indirect impact of study on treatment choices (physicians, patients)  When is it reasonable to do such a study?  What are benchmarks?  What is standard practice in these circumstances?

22 Follow-Up Methods  High retention rate over multiple years is essential  Tools to help maximize follow-up  Enrollment of child and family  Routine contact with child/family to update contact details  Tracking  Incentives  Minimize study burden

23 Study Retention  Retention includes 2 components  Tracking  Can patient be located?  Participation  Will patients voluntarily continue in the study?

24 Patient Tracking  Tracking can locate patients when they move, change status, change address  Some examples of studies using tracking show >90% of study patients can be successfully located over periods of 10–20 years (e.g., Piedmont Health Survey of the Elderly, with 99% location rate at 10 years)  Special considerations exist in following pediatric patients into adulthood

25 Study Participation  Participation varies with  Mode of data collection  Periodicity of contact  Interest of patient (family) in study objectives  Incentives  Burden to participant(s)  Special considerations in pediatrics  Patient and parent participation  Changes in consent and data collection over time  Plan for annual attrition based on study methods selected

26 Other Practical Considerations  IRB/HIPAA privacy issues  Treatment cannot ethically be conditioned on participation in research (e.g, mandatory registry is probably not an option)  Who will give assent/consent, when, how often?  What IRB approvals will be needed?  Will HIPAA waivers be needed for access to records?

27 Conclusions  Study requires epidemiologic expertise in design and analytic methods  Key focus must be on long-term retention  Study must minimize burden on participants  Successful design will be a compromise between the ideal and the practical  Study design must be tailored to the ultimate goal of the study