Application of rule induction to discover survival factors of patients after bone marrow transplantation

Decision rules are commonly used tool for classification and knowledge discovery in data. The aim of this paper is to provide decision rule-based framework for analysis of survival data and apply it in mining of data describing patients after bone marrow transplantation. The paper presents a rule induction algorithm which uses sequential covering strategy and rule quality measures. An extended version of the algorithm gives the possibility of taking into account user’s requirements in the form of predefined rules and attributes which should be included in the final rule set. Additionally, in order to summarize the knowledge expressed by rule-based model, we propose the rule filtration algorithm which consists in selection of statistically significant rules describing the most disjoint parts of the entire data set. Selected rules are identified with so-called survival patterns. The survival patterns are rules which conclusions contain Kaplan-Meier estimates of survival function. In this way, the paper combines rule-based data classification and description with survival analysis. The efficiency of our method is illustrated with the analysis of data describing patients after bone marrow transplantation.