Concluding Remarks

Longitudinal measurements contain the full information about the dynamics of a continuous biomarker, including the magnitude and timing of changes. Hence, it could be argued that such a biomarker could offer the greatest potential for becoming a surrogate for a clinical endpoint that may be distant in time.

Longitudinal measurements of PSA were evaluated as a candidate surrogate for OS by several authors. In particular, Buyse et al. (2003), Renard et al. (2003), and Collette et al. (2005) applied the meta-analytic approach, presented in the current chapter, to various sets of data from prostate cancer clinical trials. In none of the analyses was PSA found to be an acceptable surrogate for OS. Given this result, it is not surprising that none of the other PSA-based endpoints (such as, e.g., PSA response, time to PSA progression, etc.) was found to be a valid surrogate either (Buyse et al., 2003; Collette et al., 2005).

As illustrated in this chapter, evaluation of a continuous, normally distributed, longitudinal surrogate for a failure-time true endpoint is not trivial due to the need to use a joint model for the two longitudinal and failuretime processes. The model is needed if one wants to assess the strength of both the individual- and trial-level associations. However, if the former is not of interest, then the marginal linear mixed-effects and proportional-hazards models can be used to evaluate the strength of the trial-level association. The marginal analysis can be easily implemented with existing software tools.

 
Source
< Prev   CONTENTS   Source   Next >