Bayer

Bayer's pharmacovigilance teams manually reviewed hundreds of thousands of adverse event (AE) reports annually across phone, email, CRM, and other channels. Traditional keyword-based rule searches generated excessive false positives — one team scanned 10,000 records daily, with up to 800 requiring deeper manual review. The process was time-consuming, labor-intensive, and inconsistent across geographies.

Bayer built a centralized AE-detection engine on AWS using Amazon SageMaker to train ML models on high-quality internal GxP-compliant data. The engine ingests data from disparate sources, runs batch inference for high-throughput processing, and exposes a low-latency API endpoint for near real-time use cases like chatbots. Protected health data is stored in Amazon S3 with encryption and access controls, and results are visualized via Amazon QuickSight dashboards.

The engine significantly reduced false positives compared to keyword-based approaches and saved substantial time previously spent on manual case review. It standardized and harmonized PV processes across multiple channels and geographies through a single centralized engine. The real-time API endpoint responds within an average of 170 milliseconds, enabling integration with chatbots and other interactive systems.

• Centralizing AE detection into one ML engine eliminates inconsistency from distributed manual review across regions and channels.
• Running ML on a GxP-validated AWS infrastructure allows pharmaceutical companies to meet compliance requirements without sacrificing innovation speed.
• Designing dual inference modes (batch and real-time API) makes the same model reusable across high-throughput and low-latency use cases.