Extractive summarization of clinical trial descriptions

Area of Science:

• Biomedical Informatics
• Natural Language Processing

Background:

• Clinical trial descriptions are lengthy and time-consuming to review.
• Efficient summarization is needed to accelerate research comprehension.

Purpose of the Study:

• To evaluate the effectiveness of extractive text summarization for clinical trial descriptions.
• To assess the quality of automatically generated summaries using ROUGE metrics and human evaluation.

Main Methods:

• A novel dataset was created from clinicaltrials.gov records.
• Several text summarization algorithms were applied to trial descriptions.
• ROUGE metrics and human reviewer assessments (Likert scale) were used for evaluation.

Main Results:

• The dataset was refined to 101,016 usable records.
• TextRank algorithm achieved the best performance with ROUGE-L F1 score of 0.3003.
• Human evaluations correlated with ROUGE scores, indicating summary usefulness.

Conclusions:

• Extractive summarization is a viable method for creating concise clinical trial synopses.
• ROUGE-L F1 scores can automate the assessment of summary quality.