Matching patients to clinical trials using semantically enriched document representation.

Area of Science:

• Biomedical Informatics
• Clinical Research Informatics
• Artificial Intelligence in Healthcare

Background:

• Clinical trial patient recruitment is a critical bottleneck in medical research and drug development.
• Manual patient recruitment is time-consuming and prone to errors.
• Automating this process is essential for accelerating clinical research.

Purpose of the Study:

• To develop and evaluate an automated approach for patient recruitment in clinical trials.
• To leverage natural language processing (NLP) and machine learning (ML) for extracting patient eligibility information from clinical documents.
• To improve the efficiency and accuracy of identifying suitable candidates for clinical trials.

Main Methods:

• Utilized NLP and ML techniques to extract key information from narrative clinical documents.
• Employed domain ontologies to enrich clinical documents with semantic vector representations, addressing variations in reporting styles.
• Developed and compared a Multi-Layer Perceptron (MLP) model against other neural networks and conventional ML models for patient eligibility determination.

Main Results:

• The proposed ML approach achieved a micro-F1-Score of 84% across 13 different eligibility criteria.
• The Multi-Layer Perceptron (MLP) model outperformed other evaluated ML and neural network models.
• Semantically enriched clinical documents led to more effective patient cohort selection compared to original documents.

Conclusions:

• An end-to-end ML-based solution can effectively automate clinical trial patient recruitment.
• The developed system achieves performance comparable to state-of-the-art methods relying on manual rules or engineered features.
• Semantic enrichment of clinical data enhances the performance of ML models in patient cohort selection.