Cohort selection for clinical trials using hierarchical neural network.

Area of Science:

• Clinical Research Informatics
• Natural Language Processing
• Machine Learning

Background:

• Accurate patient cohort selection is crucial for successful clinical trials.
• Traditional methods for cohort selection can be time-consuming and prone to errors.

Purpose of the Study:

• To develop and evaluate a hierarchical neural network for automated patient cohort selection.
• To improve the efficiency and accuracy of identifying eligible patients for clinical trials.

Main Methods:

• A hierarchical neural network combining Convolutional Neural Networks (CNN) or Long Short-Term Memory (LSTM) with highway networks and self-attention was designed.
• The model processed sentence representations, adjusted information flow, reweighted sentences, and generated document representations for classification.
• Performance was evaluated using micro-averaged precision, recall, and F1 score against baseline models.

Main Results:

• The proposed LSTM-Highway-LSTM model achieved the highest micro-averaged F1 score of 90.21%, outperforming baseline models.
• The hierarchical approach significantly improved cohort selection accuracy compared to simpler models.
• The results demonstrated the effectiveness of the proposed method in the n2c2 clinical challenge.

Conclusions:

• The hierarchical neural network is a highly effective tool for automated cohort selection in clinical trials.
• Future work should address challenges such as word ambiguity, negation, number analysis, and imbalanced data.