A Model-Free Approach for Testing Association

Abstract The question of association between outcome and feature is generally framed in the context a model based on functional distributional forms. Our motivating application that identifying serum biomarkers angiogenesis, energy metabolism, apoptosis inflammation, predictive recurrence after lung resection node-negative non-small cell cancer patients with tumour stage T2a or less. We propose an omnibus approach for testing free assumptions forms distributions can be used as general method. This proposed maximal permutation test idea thresholding, readily implementable computationally efficient. demonstrate tests maintain their levels have strong power detecting linear, nonlinear quantile-based associations, even outlier-prone heavy-tailed error under nonparametric setting. additionally illustrate use this model-free screening further examine level these binary outcome. compare performance comparator methods our to identify preoperative associated early patients.