Knowledge Discovery and Data Mining of AHRQ / HCUP Administrative Data

The 2005 Nationwide Inpatient Sample (NIS) contains almost 8 million transactions. Datasets of this size can be under-utilized due to their complexity and the difficulty in comprehending and exploring the relationships among variables. To exploit this rich data set, we apply a customized, semidescriptive data mining approach called DomainConcept Mining (DCM). The DCM approach partitions and analyzes the collection of data, and discovers associations among attributes. Using this approach, about 8.9 million frequent itemsets were discovered in our 149 partitions of the 2005 NIS. To facilitate the navigation of the results, we develop a web application, the DCMiner, which assists the research community in identifying clinically meaningful patterns in the NIS dataset for further examination and analysis. The DCMiner demonstrates the potential for using computational methods to provide an efficient, robust, and flexible tool to healthcare researchers for knowledge discovery, which may lead to further clinical studies.