Text Mining on Reports of Near Miss and Accidents about Wheelchair

To prevent accidents involving wheelchairs, both accidents and near-miss events should be investigated, as near misses are related to future accidents according to Heinrich’s law. Although previous studies have mainly focused on accident cases, near-miss events have not been sufficiently analyzed due to the difficulty of handling text-based reports systematically. Therefore, this study aimed to analyze reports of near misses and accidents involving wheelchairs using text-mining techniques. A total of 79 reports published by the Association for Technical Aids in Japan were analyzed. The analysis included word frequency, keyword distribution patterns, and co-occurrence network analysis based on the Jaccard coefficient to reveal structural relationships among terms. The results showed that key terms such as “use,” “assistance,” “foot,” “brake,” “fall,” “lever,” “step,” and “balance” frequently appeared, indicating human, operational, and environmental factors. Furthermore, co-occurrence analysis revealed several characteristic event structures, such as “lever–push–brake,” “balance–loss,” and “step–fall,” suggesting that incidents often occur during operation, posture instability, and environmental interactions. Central clusters consisting of “use,” “assistance,” and “foot” indicate that many risks are associated with user–caregiver interactions and physical support conditions. These findings suggest that preventing wheelchair-related incidents requires not only improving device operability and reliability but also addressing environmental barriers such as steps and enhancing support for user balance and caregiver interaction.