Clinical Staff Using Natural Language Processing Model Enhances Accuracy of Clinical Trial Prescreening Process

Objective:

To assess the impact of a pretrained natural language processing model on the accuracy and efficiency of clinical trial prescreening, specifically focusing on chart-level accuracy and review times.

Key Findings:

• AI-assisted prescreening achieved a chart-level accuracy of 76.1%, compared to 71.5% for human-alone assessments (P for superiority = .002).
• AI algorithms improved accuracy in seven eligibility criteria areas, while human review excelled in assessing ECOG status.
• Mean review times were similar between human-alone and human-plus-AI assessments (34.7 vs 37.8 minutes; P = .513), indicating no significant increase in review time.

Interpretation:

Integrating AI into clinical trial prescreening processes enhances accuracy without significantly increasing review time, potentially increasing patient enrollment in trials by streamlining the process.

Limitations:

• The study was conducted at a single cancer center, which may limit the applicability of the findings to other settings.
• Efficiency analysis was not the primary endpoint, and results should be interpreted with caution due to potential biases.

Conclusion:

The integration of AI in clinical trial prescreening can improve accuracy and efficiency, suggesting a scalable model for other health systems to enhance patient enrollment in clinical trials.