ABSTRACT
Trials in Schizophernia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .511 19.5 Non-Gaussian Endpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
19.5.1 Binary Endpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512 19.5.2 Two Failure-Time Endpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513 19.5.3 Ordinal Surrogate and a Survival Endpoint . . . . . . . . . . . . . . . . . . . . 514 19.5.4 Methods for Combined Binary and Normally Distributed
Endpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514 19.5.5 Longitudinal Endpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
19.6 Unified Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518 19.6.1 Likelihood Reduction Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518 19.6.2 Information-Theoretic Unification. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519 19.6.3 Fano’s Inequality and the Theoretical Plausibility of
Finding a Good Surrogate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522 19.6.4 Application to the Meta-Analysis of Five Clinical Trials in
Schizophrenia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522 19.7 Alternatives and Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525 19.8 Prediction and Design Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
19.8.1 Surrogate Threshold Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
19.9 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531 Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
The rising costs of drug development and the challenges of new and reemerging diseases are putting considerable demands on efficiency in the drug candidates selection process. A very important factor influencing duration and complexity of this process is the choice of endpoint used to assess drug efficacy. Often, themost sensitive and relevant clinical endpoint might be difficult to use in a trial. This happens if measurement of the clinical endpoint (1) is costly (e.g., to diagnose cachexia, a condition associated with malnutrition and involving loss of muscle and fat tissue, expensive equipment measuring content of nitrogen, potassium, and water in patients’ body is required); (2) is difficult (e.g., involving compoundmeasures such as encountered in quality-of-life or pain assessment); (3) requires a long follow-up time (e.g., survival in early stage cancers); or (4) requires a large sample size because of low event incidence (e.g., short-term mortality in patients with suspected acute myocardial infarction). An effective strategy is then proper selection and application of biomarkers for efficacy, replacing the clinical endpoint by a biomarker that ismeasuredmore cheaply, more conveniently, more frequently, or earlier. From a regulatory perspective, a biomarker is considered acceptable for efficacy determination only after its establishment as a valid indicator of clinical benefit, that is, after its validation as a surrogate marker (Burzykowski et al. 2005).