Generative Transformers for Pharmacovigilance Signal Detection using Electronic Health Records.

Area of Science:

• Pharmacovigilance and Drug Safety
• Health Informatics
• Artificial Intelligence in Medicine

Background:

• Adverse drug reactions (ADRs) pose significant risks to patient safety and healthcare systems.
• Current post-market surveillance methods, like disproportionality metrics, struggle with temporal causality and electronic health record (EHR) data.
• Existing methods fail to fully leverage the potential of EHR data, leading to under-reporting and bias in ADR detection.

Purpose of the Study:

• To develop and evaluate novel methods using generative pre-trained transformers (GPT) for improved ADR signal detection in EHR data.
• To address the limitations of traditional disproportionality metrics in modeling temporal relationships and handling complex EHR data.
• To enhance the accuracy and efficiency of pharmacovigilance systems through advanced AI techniques.

Main Methods:

• Implementation of generative pre-trained transformer (GPT) models for analyzing EHR data.
• Comparative analysis against established baseline methods using disproportionality metrics.
• Evaluation of model performance on data from two distinct healthcare systems.

Main Results:

• GPT models demonstrated superior performance in ADR signal detection compared to traditional methods.
• An absolute gain of 6-16% in overall AUROC was achieved by the GPT models.
• The study confirmed the effectiveness of transformer-based models in integrating comprehensive clinical data for pharmacovigilance.

Conclusions:

• Generative pre-trained transformers offer a promising advancement for ADR signal detection in EHR data.
• Transformer-based models can significantly enhance pharmacovigilance by overcoming limitations of existing methods.
• This approach holds potential for improving patient safety and reducing healthcare burdens associated with ADRs.