Abstract B84: Can propensity score analyses be alternatives to randomized trials.

Abstract

Abstract Introduction: The use of propensity Score (PS) in oncology is increasing. Previous reports have, however, shown that, in the case of a normal or binary response, the inclusion of variables which affect treatment choice but do not correlate with the response into the statistical model for calculating PS needs a caution because the statistical power to detect the association between treatment and response gets decreased. In this study, we evaluated the PS method for survival response in terms of power loss under various scenarios. Method: We conducted simulation studies to compare statistical power between the Cox models with and without PS. We assumed 4 strata, each having a unique PS (a probability of choosing test drug) as in Table 1. Patients belong to one of the 4 strata and received either test drug or control drug according to his/her PS. Every stratum has an exponential survival distribution with a mean of 1.25 years for the test drug and 1 year for the control drug. Of the 3 scenarios we considered, Scenario 1 is closest to the condition of "clinical equipoise," while Scenario 3 reflects the situation that variables unrelated to survival strongly affect treatment choice. Result: See Table 2. Statistical power was decreased in any scenario and Scenario 3 was most serious. Conclusion: The indiscriminate inclusion of variables into PS will lead to a power loss. It would be inevitable in an evaluation for anti-cancer drugs where it is not easy to exclude variables which are unrelated to survival because the choice of drug depends on risk as well as benefit for each patient. Citation Information: Mol Cancer Ther 2013;12(11 Suppl):B84. Citation Format: Naoki Ishizuka, Takeharu Yamanaka. Can propensity score analyses be alternatives to randomized trials. [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2013 Oct 19-23; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2013;12(11 Suppl):Abstract nr B84.