A0398
Title: Derivation of optimal solution by full enumeration for subgroup identification
Authors: Masahiro Mizuta - The Institute of Statistical Mathematics (Japan) [presenting]
Abstract: Subgroup identification methods are techniques for identifying subsets for which a particular treatment, etc., is effective. For example, a widely used expression in relation to COVID-19 is "recommend vaccination for people of a certain age". Identifying valid subsets is a challenging task, and several approaches have been proposed (Pmtree, glmtree, QUINT, etc.). However, these methods do not guarantee that the optimal subgroup with the maximum validity evaluation function is obtained. Consequently, techniques that determine the "optimal subgroups in the strict sense" by enumerating all subgroups and computing the effectiveness of each are effective. For instance, the algorithm that is developed enumerates 43,199,128,758 (43.2 billion) subgroups in artificial data with p=5 and n=100. The optimal subgroups can then be derived. Furthermore, realistic modifications of the algorithm have also been implemented. Subgroup identification is the foundation of subgroup analysis and is a crucial technique in personalized medicine. On the other hand, the treatment of subgroups must be meticulously considered when implementing statistical decisions. The findings can be utilized to contribute to subgroup identification and analysis.
