AI-Enhanced Patient Transfer System
Use Case: Amazon Sagemaker AI Proof of Concept and Pilot
Business Challenge
A leading healthcare emergency consultation service providing systems to acute care hospitals was struggling with data integrity issues in their patient transfer processes. The organization manages nearly 2 million acute care patients annually across over 100 hospitals, generating tens of millions of HL7 messages.
Their mission-critical operations depend on real-time bed availability tracking, but they faced significant challenges with data quality. False data transactions occurred due to unsupported special business workflows at various hospitals.
For example, when the HL7 ADT A02 message (typically used for patient transfers) was used to temporarily move ICU patients to operating rooms, patients were falsely registered as discharged from the ICU. Additionally, limited data field automation resulted from compatibility issues across different hospital EMR systems like Meditech, Epic, and Cerner, each with distinct data attributes and formats. This diversity prevented certain data fields from being directly mapped within hospital systems, necessitating continued manual data entry and increasing the risk of errors.
Datavail’s Solution
We developed a comprehensive AI solution to address these challenges, implementing a phased approach to ensure successful integration. We began with a feasibility assessment and POC design, creating an analysis document that outlined all feasible solutions and estimates.
This was followed by POC development to confirm feasibility and establish estimates for future phases. The implementation started with a single use case pilot in selected healthcare organizations to validate interoperability, performance, and utility before expanding deployment to additional facilities.
The technology stack used multiple AWS services, including Amazon RDS for MySQL, AWS Lambda, AWS Glue, Amazon S3, Amazon SageMaker (Processing, Wrangler, Training, Endpoint), Amazon API Gateway, Amazon CloudWatch, SageMaker Model Monitor, and Amazon SNS.
Our solution was specifically designed to recognize pattern-based exceptions – when A02 messages were used for temporary location transfers – while maintaining the patient in the ICU list instead of discharging them.
The AI system learned from human corrections to build intelligent rules that would prevent future errors, ensuring bed numbers remained accurate with each transaction and maintaining continuity of care throughout patient movements.
Business Impacts
- Improved Data Accuracy: Enhanced patient transfer tracking significantly reduced false discharges and transfer errors across the hospital network
- Lower Operational Costs: Reduced the need for manual corrections and data reconciliation efforts across the healthcare network
- Reduced Manual Workload: Maximized automation capabilities, minimized human data entry requirements, and reduced the operational burden on hospital staff
- Real-Time Critical Care Data Access: Provided reliable access to accurate, real-time critical care information, improving decision-making capabilities
- Sustainable Future System: Established an adaptive AI system that continuously learns from corrections, improving over time