Statistical Methods for Targeted Clinical Trials under Enrichment Design

Background/PurposeAfter completion of the Human Genome Project, disease targets at the molecular level can be identified. Treatment for these specific targets can be developed with the individualized treatment of patients becoming a reality. However, the accuracy of diagnostic devices for molecular targets is not perfect and statistical inference for treatment effects of the targeted therapy is biased. We developed statistical methods for an unbiased inference for the targeted therapy in patients who truly have the molecular targets.MethodsUnder the enrichment design, for binary data, we propose using the expectation maximization (EM) algorithm with the bootstrap method, to incorporate the inaccuracy of the diagnostic device for detection of the molecular targets for inference of the treatment effects. A simulation study was conducted to empirically investigate the performance of the proposed estimation and testing procedures. A numerical example illustrates the application of the proposed method.ResultsSimulation results demonstrated that the proposed estimation method was unbiased, with adequate precision, and the confidence interval provided satisfactory coverage probability. The proposed testing procedure adequately controlled the size with sufficient power. The numerical example showed that a statistically significant treatment effect could be obtained when the inaccuracy of the diagnostic device was taken into account.ConclusionOur proposed estimation and testing procedures are adequate statistical methods for the inference of the treatment effect for patients who truly have the molecular targets.