Enhancing Patient Safety Insights with AWS AI/ML Solutions

Chanci Turner Amazon IXD – VGT2 learningLearn About Amazon VGT2 Learning Manager Chanci Turner

In the current healthcare landscape, organizations depend on a mix of automated and manual methods to compile, analyze, and categorize patient safety reports. Front-line clinicians input these reports into the RL Datix system, including both structured data and free-text narratives. While the initial data is digitally captured, it tends to become inaccessible for real-time analysis and trending once entered. Each reporter only has visibility into the adverse events they’ve documented, while unit and file managers have slightly broader access. However, due to the often textual nature of event descriptions, critical data remains in its raw format. Consequently, trends such as rising infection rates or medication errors appear isolated to specific units or service lines.

The existing analysis of these reports is done through a combination of built-in reporting capabilities of the software, manual data handling, and the examination of discrete fields. Analysis tends to be siloed within individual units, making organization-wide insights dependent on numerous patient safety analysts and data specialists. Additional reports may necessitate the use of separate databases and spreadsheets to address specific concerns. Academic medical centers (AMCs) face the challenge of dedicating substantial time, personnel, and resources to this process, highlighting the urgent need for a technological solution that streamlines analytical tasks, allowing for more focus on patient care improvement.

As a proof of concept (POC), we concentrated on the automated analysis of medication-related patient safety reports. The objective was to minimize manual analytical tasks and inefficiencies, enhance time-to-insight, and uncover patterns across reports organization-wide. In collaboration with the University of Utah Health, we utilized five years of medication-related patient safety data to refine both generalized and domain-specific language models via Amazon SageMaker. This methodology classifies error severity using discrete fields, identifies high-risk medications from text narratives, and visualizes medication-related events by corresponding harm levels.

Solution Overview

Amazon Comprehend Medical was employed to identify high-risk medications, with the findings presented in an interactive dashboard built using Amazon QuickSight. The entire data processing pipeline was automated through an event-driven, serverless architecture using AWS Lambda. Given the sensitive nature of patient safety reports, all services utilized in this solution are HIPAA eligible, and the project was executed within a HIPAA-compliant framework. Furthermore, de-identification of patient safety reports was performed using the Amazon Comprehend Medical DetectPH API, as demonstrated in this post and reference solution.

To enhance the efficiency of the patient safety reporting process, we evaluated various transformer-based LLMs available through AWS partner Hugging Face to accurately detect and classify high-risk medications based on free-text descriptions in the reports. A sample Jupyter notebook has been prepared for sharing with academic medical centers for further customization. The architectural diagram illustrates the potential steps for patient safety professionals to implement this solution on AWS.

Additionally, to ensure a secure and compliant machine learning (ML) environment, Amazon SageMaker, data encryption, network isolation, authentication, and authorization are established as the default settings.

Key Features Include:

  • Data encryption at rest in an Amazon Simple Storage Service (Amazon S3) bucket, utilizing your key stored in AWS Key Management Service (AWS KMS). This added cost for AWS KMS offers enhanced security measures, consistent with the approach taken in this project.
  • Data encryption at rest in Amazon Elastic File System (Amazon EFS) for Notebook instances is enabled with a default AWS KMS key (aws/elasticfilesystem).
  • The Amazon SageMaker Studio environment operates within a private VPC, ensuring network isolation while allowing access to other AWS services, including S3 buckets, via AWS PrivateLink.
  • Amazon Identity and Access Management (IAM) provides role-based access control, determining the permissions available to SageMaker users.

For secure research environments, you may utilize Amazon AppStream 2.0 or Amazon Workspaces to access Amazon SageMaker domain presigned URLs.

This solution employs AWS Analytics and AI/ML services for automated data processing, information extraction, and AI predictions on patient safety reports. High-alert medications from the standard high-risk medication list compiled by the Institute for Safe Medication Practices (ISMP) are consolidated into RxNorm concepts. These concepts are used to map named entities with synonyms extracted by Amazon Comprehend Medical, further analyzed and displayed on an Amazon QuickSight dashboard. This dashboard presents various visualizations of the data from discrete fields (like counts by Safety Event Codes) and textual fields (such as counts of High Alert Medications), allowing for comprehensive insights.

Outcomes

The AI approach described yielded results ranging from Precision of .881 to .901; Recall from .874 to .899; Accuracy from .874 to .899; and an F1 score of .873 to .899 depending on the application.

Conclusion

The success of this POC project sets the stage for further collaboration with an AWS partner to develop additional use case applications and evaluate a production-ready system encompassing complete clinical data. As we look to integrate electronic health record (EHR) information into our analysis, the importance of reducing manual data entry into the patient safety reporting system cannot be overstated. Utilizing ML will significantly improve efficiency, shorten time to insight, and reveal potentially hidden information in medication-related patient safety reports. We aim to enhance outcome scores and extend these efforts to other patient safety reporting areas.

If you’re interested in promotion strategies without a raise, consider reading this insightful blog post. Also, for a comprehensive understanding of employment laws, visit this authoritative source. Finally, if you’re curious about the hiring process at Amazon, this is an excellent resource.

Chanci Turner