A0238
Title: Feature screening and selection in competing risks models
Authors: Marialuisa Restaino - University of Salerno (Italy) [presenting]
Abstract: In the analysis of time-to-event data, competing risks data are encountered when individuals may fail from multiple causes (for example, $K>2$), and the occurrence of one failure event precludes the others from happening. Different (un)correlated features should influence the events, and the same feature should affect more than one event. Moreover, the number of covariates ($p$) should be very large and sometimes should be greater than the sample size ($n$). Thus, a reduction of variables is crucial. In the literature, some authors focused on screening and selecting the variables under the assumption that a) the number of events $K$ is 2, and b) one event is of the main interest, while the other can be neglected. Thus, the aims are to i) compare the performance of some existing methods for screening and selecting the most significant variables, ii) highlight their main advantages and disadvantages, and iii) propose a new procedure able to identify the relevant covariates in the framework of high and ultra-high dimensions, in the presence of highly correlated variables, and when the number of events $K$ is larger than two.
