Extractive Clinical Question-Answering With Multianswer and Multifocus Questions: Data Set Development and Evaluation...

Area of Science:

• Natural Language Processing (NLP)
• Artificial Intelligence (AI) in Healthcare
• Clinical Data Analysis

Background:

• Extractive Question-Answering (EQA) aids in answering patient questions using clinical notes.
• Existing datasets lack support for multi-answer and multi-focus questions common in clinical settings.
• Developing AI for realistic clinical EQA requires specialized datasets.

Purpose of the Study:

• To create a novel dataset for developing and evaluating clinical EQA systems.
• To address the limitations of existing datasets by incorporating multi-answer and multi-focus question capabilities.
• To facilitate the creation of AI solutions that handle complex, natural clinical queries.

Main Methods:

• Leveraged annotated relations from the 2018 National NLP Clinical Challenges corpus.
• Generated an EQA dataset including 1-to-N, M-to-1, and M-to-N drug-reason relations.
• Developed and tested a baseline EQA solution on the newly created dataset.

Main Results:

• The RxWhyQA dataset comprises 96,939 question-answering entries.
• 25% of answerable questions required multiple answers, and 2% involved multiple drugs.
• Baseline EQA achieved an F1-score of 0.72, with notable performance differences for multi-answer and multi-drug questions.

Conclusions:

• The RxWhyQA dataset is suitable for training and evaluating EQA systems for multi-answer and multi-focus questions.
• Multi-answer EQA presents significant challenges, necessitating further research and investment.
• The shared dataset promotes research towards more realistic clinical EQA scenarios.