In new drug development which could be effective only for subpopulation, we need to evaluate treatment efficacy based on predictive biomarkers that could identify patients who benefit from the treatment. In all-comers, biomarker-stratified clinical trials, adaptive analysis plans across the overall population and biomarker-defined subpopulations have been frequently employed. In this paper, we consider bias correction in estimating treatment effects in the biomarker subpopulation when treatment efficacy in the subpopulation is indicated by a preliminary hypothesis test in the overall population. The problem is the difficulty in estimation when the test statistic in the overall population approaches to the truncated significant point, which reflects a fundamental problem in considering estimation in an unbound parameter space of the test statistic that is actually bounded on the basis of its rejection region. To address this problem, we propose a new approach based on application of randomized test with smoothing function on the probability of rejection to the preliminary overall test. Simulation on the proposed method, including comparison with existing method will be provided.