Inferring ADR causality by predicting the Naranjo Score from Clinical Notes.

Area of Science:

• Pharmacovigilance and Drug Safety
• Artificial Intelligence in Healthcare
• Clinical Informatics

Background:

• Clinical judgment studies are crucial for drug safety surveillance and quantifying medication-adverse drug reaction (ADR) causality.
• Current methods require manual chart review by physicians to complete the Naranjo questionnaire, which is time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate an automated methodology for inferring causal relationships between medications and ADRs.
• To leverage natural language processing (NLP) and machine learning to analyze patient discharge summaries.

Main Methods:

• Utilized Bidirectional Encoder Representations from Transformers (BERT) for automated extraction of relevant patient data pertinent to Naranjo questionnaire criteria.
• Employed logistic regression, a statistical learning model, to predict Naranjo scores and determine causality between drugs and ADRs based on extracted information.

Main Results:

• The proposed methodology achieved a macro-averaged F1-score of 0.50 and a weighted F1-score of 0.63 in inferring causal relationships.
• Demonstrated the feasibility of automating aspects of clinical judgment for pharmacovigilance using AI.

Conclusions:

• The developed automated approach shows promise in streamlining the process of causality assessment in pharmacovigilance.
• This methodology can potentially reduce the manual burden on healthcare professionals and enhance the efficiency of drug safety monitoring.