Zero-suppressed decision diagrams (ZDDs) are a data structure for representing combinations over item sets. They have been applied to many areas such as data mining. When ZDDs represent large-scale sparse datasets, they tend to obtain an unbalanced form, which results performance degradation. In this paper, we propose a new data structure three-way indexing ZDD, as a variant of ZDDs. We furthermore present algorithms to convert between three-way indexing ZDDs and ordinary ZDDs. Experimental results show the effectiveness of our data structure and algorithms.