RG Price*, YJ Huang, BJ Salter, University Utah, Salt Lake City, UT

• 
  

SU-D-TRACK 6-7 (Sunday, 7/25/2021) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Purpose: Incident-learning is a powerful tool necessary for continuous improvement of quality and safety within radiation oncology. However, as the percentage of captured incidents increases, so does the difficulty of identifying important trends/topics meriting the highest attention of the internal safety committee. This study utilized automated text-mining procedures to identify important topics for review.

Methods: Data was exported to CSV for 133 incident reports from our institutional RO-ILS database. Narratives were extracted, and data was imported to Orange, an open-source data-mining tool, for processing and analysis. Text was converted to lower case before using regexp for tokenization(breaking into single-word pieces). Punctuation and unimportant words were filtered using a list of stopwords(words such as “the”). Finally, tokens were used to create a range of N-gram phrases (phrases with N number of words) which were counted and regularized to relative weighted “importance” before visualization using word clouds. Phrases with largest weightings were further investigated to determine if this was indicative of a topic that needed additional action.

Results: The N-gram phrase “v-sim” was observed to receive a high weighting. This represented a previously unidentified issue and, upon further investigation, was recognized as being indicative of both the number of events identified at the patient’s v-sim appointment, as well as events that occurred due to shortened sim-to-treatment timelines. “Setup instructions” also received a high weighting, and was identified as being indicative of a previously known trend caused by incorrect or unclear setup instructions. This was encouraging, as it validated the potential of the tool. Other notable findings included “iso-center”, which was tagged in many events with various topics. This was ascertained to be indicative of the large number of error pathways that can result in isocenter related problems.

Conclusion: Preliminary use of text-mining tools shows promise for identifying relevant safety-related topics for review.

Handouts

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email