A Deep Learning Method to Automatically Identify Reports of Scientifically Rigorous Clinical Research from the...

Area of Science:

• Biomedical informatics
• Artificial intelligence in medicine
• Information retrieval

Background:

• Efficiently locating scientifically sound studies is a significant challenge in practicing evidence-based medicine.
• Current search methods may not always be optimal for identifying high-quality clinical research.

Purpose of the Study:

• To evaluate a deep learning approach for retrieving scientifically sound treatment studies from biomedical literature.
• To compare the performance of a deep learning model against established search filters.

Main Methods:

• A Convolutional Neural Network was trained on a large dataset of PubMed citations (title and abstract).
• The deep learning model's performance was benchmarked against PubMed's Clinical Query filters and McMaster's textword search.
• The Clinical Hedges dataset served as the gold standard for evaluation.

Main Results:

• Deep learning demonstrated high precision, outperforming the Clinical Queries Broad filter and McMaster's textword search.
• Recall was comparable to existing state-of-the-art filters, with slight variations.
• The model performed particularly well on citations not yet indexed with controlled vocabulary.

Conclusions:

• Deep learning offers a promising, feature-engineering-free method for identifying rigorous clinical research.
• This approach enhances the efficiency of evidence-based medicine by improving study retrieval.
• Further research is recommended to optimize the model and explore its applicability to other clinical research areas.