Statistical conditions for the validation of surrogate markers: A simulation study
Description
The primary objective in the analysis of Clinical Trials data is to assess the treatment's clinical efficacy through outcome measures that unequivocally reflect tangible benefit to the patient. However, the endpoint of greatest relevance to inferences concerning therapeutic efficacy is frequently not practical or even feasible to measure. In these cases we must rely on alternative, or surrogate, endpoints. These surrogate markers are outcome measures that belong in the causative pathway of the disease process and partially or completely reflect the 'true' endpoint. The benefit of using surrogate markers lies in the fact that one may design clinical trials that have a reduced sample size and/or duration. A detailed procedure to validate a surrogate marker has been outlined in this dissertation, and results of the simulation study show that if followed, a complete surrogate can be detected. The simulation also shows that surrogacy is only present under certain (narrow) limits of intercept and slope parameter variability of the surrogate. Type II Censoring is not an important factor in determining whether a potential surrogate is adequate (provided an adequate sample size). Application of the simulation results to concrete examples in Clinical Trials of AIDS Therapy indicate that: CD4 Count is an incomplete surrogate marker for clinical trials of AIDS therapy, and that HIV RNA PCR is an excellent candidate to become a surrogate for clinical trials of AIDS therapy