Cohort selection for clinical trials using deep learning models.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Clinical Trial Management

Background:

• Automated patient identification for clinical trials is crucial for efficient research.
• Natural language processing (NLP) and deep learning offer potential solutions for complex cohort selection tasks.
• The 2018 n2c2 shared task focused on evaluating deep learning for clinical trial cohort selection.

Purpose of the Study:

• To evaluate the effectiveness of various deep learning architectures for automated clinical trial cohort selection.
• To assess the contribution of NLP and deep learning in identifying patients meeting clinical trial criteria.
• To investigate the impact of a fully connected feedforward layer on deep learning model performance.

Main Methods:

• Formulated cohort selection as a multilabeling problem.
• Explored deep learning architectures: Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and hybrid CNN-RNN models.
• Investigated the effect of a fully connected feedforward layer on model performance.

Main Results:

• Recurrent Neural Network (RNN) and hybrid CNN-RNN models achieved the best performance.
• The addition of a fully connected feedforward layer generally improved results across architectures, with an exception in the hybrid model.
• Performance gains were observed despite the limited dataset size.

Conclusions:

• Deep learning methods demonstrate potential for effective feature learning in cohort selection.
• These models can serve as a preliminary filter for clinical trial cohort identification.
• Implementing deep learning can substantially decrease the time and cost associated with clinical trials.