1. RTI International RTI International is a trade name of Research Triangle Institute. www.rti.org Selecting Endpoints for Clinical Trials

2. RTI International Issues to Consider  Criteria for endpoint selection  Single vs multiple endpoints  Types of endpoints  Tradeoffs between types of endpoints  Some issues in endpoint processing 2

3. RTI International Criteria for Endpoint Selection  Clinical relevance o Is the endpoint a direct measure of the clinical outcome of ultimate interest? o Will changes in the endpoint alter treatment?  Statistical properties o Power and sample size for alternate endpoints o Can the endpoint be measured reliably and reproducibly? 3

4. RTI International Reliability and Reproducibility 4 Less Accurate More Accurate More Precise Less Precise

5. RTI International Can a Trial Have 2 ‘Primary’ Endpoints?  Most trials have one primary endpoint o Sample size and power are based on this endpoint  Some trials have 2 endpoints of equal interest o Example: PTSD and depression scores in a trial of treatment of military personnel with combat experience o Usually, the two endpoints are analyzed separately  Adjust critical p value for multiple testing (p<0.025)  Determine sample size separately for each endpoint  Use the larger sample size for the trial  Actual power for the trial depends on the correlation between outcomes o Define success: p<0.025 on either or both endpoints? 5

6. RTI International Types of Endpoints 6

7. RTI International Types of Endpoints Binary – presence/absence of disease, often in a defined period, such as 30-day mortality o Chi square tests, exact tests, logistic regression Time to event – time to occurrence of a fixed event, such as heart attack or death o Kapan Meier survival curves, proportional hazards regression 7

8. RTI International Types of Endpoints Ordinal categorical – more than 2 categories with intrinsic order, such as a pain scale o Logistic regression for ordinal outcomes o Use linear models if the number of categories is large Frequency – count the number of recurrent events a subject experiences in a defined period, such as headaches or pain crises in sickle cell disease o Poisson regression o Linear models if counts are high enough 8

9. RTI International Tradeoffs Between Types of Endpoints >1 type of endpoint may be appropriate for a trial Consider a trial of a new treatment to reduce blood pressure Do we measure change in BP (continuous) or determine whether a subject’s BP is reduced below a threshold (binary)? The continuous measure may offer greater statistical power but the binary measure may be more clinically meaningful In practice, both outcomes are often reported in a trial Select one as primary and power accordingly 9

10. RTI International Tradeoffs Between Types of Endpoints Consider a trial with an outcome such as one-year survival Should we use methods for a binary outcome or survival analysis? Use survival analysis if duration of follow-up varies among subjects Can use either one if all subjects followed for the same length of time Survival analysis may offer greater statistical power However, what is more important, the rate of outcomes or the number of patients having outcomes? 10

11. RTI International Continuous Outcomes May Be Surrogate Endpoints Surrogate endpoints - Used in place of clinical outcomes when time to the outcome is very long Example: HIV RNA level instead of time to AIDS or death o Time to AIDS or death varies inversely with HIV RNA level (Mellors et al. Science 1996) o Reducing HIV RNA level leads to prolonged survival Another example: GFR instead of time to kidney failure Caution: HCV RNA levels in blood do not predict rate of fibrosis in the liver Choose surrogate endpoints carefully Should be directly related to disease progression – altering the surrogate will alter the ultimate outcome 11

12. RTI International Assay Results as Endpoints Assay results are the endpoints in many trials (HIV RNA titer, antibody titer in vaccine trials, etc) Suppose that assay methods vary among study sites o Differences in sensitivity, accuracy and/or variation may complicate interpretation of results Stratifying on study site may or may not help o Common solution - use a central laboratory Another problem – inter-assay variation may reduce power in a longitudinal study o Solution – run all samples from a subject in one batch Standardization of assay results is very important 12

13. RTI International Endpoint Adjudication Some endpoints require interpretation of source material (MRI, x-ray, biopsy, etc.) by knowledgeable individuals Systematic differences in interpretation among observers can cause problems in trial results Bias is also possible if the observer knows a subject’s treatment assignment – a problem in unblinded trials Solution: use a central adjudication panel of experts who do not know the treatment assignments Common approach – 3 panel members Need to specify review procedures in advance (e.g. consensus vs 2 reads plus a tiebreaker) 13

14. RTI International Endpoint Selection Endpoint selection can depend on both the clinical meaning and statistical properties of candidate endpoints o Clinical: will a successful trial using that endpoint alter practice? o Statistical: power, reproducibility and reliability, and (for surrogate endpoints) relationship to the outcome of ultimate interest. Selection requires collaboration between the clinical researcher and the statistician Start the collaboration early 14

15. RTI International Endpoint Selection Don Corleone, famous Sicilian statistician: why didn’t you come to me sooner? 15