Natural Language Processing to Adjudicate Heart Failure Hospitalizations in Global Clinical Trials

Area of Science:

• Cardiovascular Medicine
• Artificial Intelligence in Healthcare
• Clinical Trial Methodology

Background:

• Physician-led clinical events committee review is the standard for cardiovascular outcome identification in trials but is resource-intensive and lacks reproducibility.
• Automated adjudication using artificial intelligence (AI) offers potential for larger, more cost-effective trials but requires validation in global studies.

Purpose of the Study:

• To develop and validate a novel AI model for automated heart failure adjudication.
• To assess the accuracy and efficiency of AI-based adjudication compared to traditional methods in international clinical trials.

Main Methods:

• Developed the Heart Failure Natural Language Processing (HF-NLP) AI model using hospitalization data from three international clinical outcomes trials.
• Tested the HF-NLP model on heart failure hospitalizations from the DELIVER trial, comparing AI adjudications with those from a physician clinical events committee.
• Utilized Food and Drug Administration-based criteria for adjudication by both AI and the clinical events committee.

Main Results:

• The AI-based adjudication demonstrated 83% agreement with the clinical events committee.
• A hybrid approach (AI with human review for uncertain cases) achieved 91% agreement, reducing adjudication workload by 84%.
• The estimated treatment effect of dapagliflozin on heart failure hospitalization was consistent between AI and clinical events committee adjudications.

Conclusions:

• AI-based adjudication of clinical outcomes can significantly improve the efficiency of global clinical trials.
• The developed AI model maintains accuracy and interpretability, offering a viable alternative to manual review.
• Automated adjudication holds promise for streamlining large-scale cardiovascular outcome studies.