[Artificial intelligence based Chinese clinical trials eligibility criteria classification].

Area of Science:

• Artificial Intelligence in Clinical Research
• Natural Language Processing for Medical Text
• Clinical Trial Optimization

Background:

• Subject recruitment is crucial for clinical trials, relying on eligibility criteria.
• Semantic analysis of these criteria can enhance trial design and automate patient recruitment.
• Automated classification of Chinese eligibility criteria is needed.

Purpose of the Study:

• To explore automatic semantic category classification of Chinese eligibility criteria using AI.
• To develop and evaluate AI models for this classification task.
• To provide a valuable dataset and benchmark for future research.

Main Methods:

• Collected and annotated 38,341 Chinese eligibility criteria sentences across 44 semantic categories.
• Organized an academic shared task with 75 participating teams.
• Utilized pre-trained language models (BERT) combined with neural networks and ensemble modeling.

Main Results:

• The best system achieved a macro F1 score of 0.81 using a BERT-based ensemble model.
• Mixed models combining pre-trained language models and neural networks were mainstream.
• Data preprocessing/postprocessing and data volume significantly impacted classification performance.

Conclusions:

• AI, particularly BERT and ensemble methods, shows strong potential for classifying Chinese eligibility criteria.
• This work provides a valuable dataset and benchmark for medical short text classification.
• Optimizing data processing and addressing data imbalance are key for future improvements.