Each week, we will be publishing layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.

The article featured today is from Canadian Journal of Statistics, with the full article now available to read in Early View here.

Liu, Y., Xu, J. and Li, G. (2020), Sure joint feature screening in nonparametric transformation model for right censored data. Can J Statistics. https://doi.org/10.1002/cjs.11575

Feature screening, which filters out a majority of noise variables when there are far more features than observations, is important in many biomedical studies.  For example, in cancer genomics, a primary goal is to identify important features and build a model that predicts the survival outcomes of future patients. However, with the number of genes greatly exceeding the sample size and the expression levels of some genes often being highly correlated, this task is challenging and can sometimes only be accomplished upon the availability of an effective dimensional reduction procedure such as feature screening.  This paper considers a survival outcome that is subject to right censoring, and develops a new statistical method for joint feature screening of high dimensional covariates with theoretically justified statistical guarantee of retaining important covariates.  The method is developed under a nonparametric transformation model that imposes minimal assumptions on the functional relationship between the survival outcome and the covariates, which makes it more robust and widely applicable than other model-based screening methods. It can help medical researchers identify genes that are relevant to disease progression, draw statistical inference, and devise a predictive model for the survival outcome of a patient using the genetic profile of a tumor. As an illustration, the method is used to analyze the survival times of the diffuse large-B-cell lymphoma patients.