Exploring Generalized Association Rule Mining for Disease Co-Occurrences

Association rule mining offers an automated approach for discovering new knowledge about diseases. A known challenge is how to constrain the search space to prevent an exponential explosion of rules while minimizing information loss. In this study, generalized association rule mining techniques were used to identify disease cooccurrences based on ICD-9-CM codes in a statewide hospital discharge data set. The Clinical Classifications Software (CCS) categorization scheme and the numerical hierarchy of ICD-9-CM were used to generalize the codes and produce generalized associations for comparison with associations generated from the raw data. By maintaining links between the raw and generalized data, associations lost in the generalization process, overlapping associations, and new associations were identified. In addition, preliminary results indicate that the concept hierarchy used for generalization may influence the associations found.