Early detection of diseases and understanding disease progression patterns are critical for ensuring an accurate, timely diagnosis and accelerating access to the best treatments. A comprehensive view of the patient journey can improve health outcomes, whether underlining conditions are rare or common. However, gaining these insights is challenging due to data aggregation, analysis, and communication challenges.
Conventional approaches that are using traditional market research or retrospective prescription data to capture insights about the patient journey often fail to decipher the increasingly complex and intricate care management pathways. They provide insufficient insights across the continuum of care making it challenging to identify eligible patients for treatments and to recommend early interventions.
Relying on traditional data sources often leads to inefficient targeting of healthcare professionals (HCPs) who might have patients progressing in their disease. Suboptimal resource allocation and messaging can create delays in product adoption and patient’s access to needed treatments.
The increasing sophistication and volume of collected longitudinal healthcare data, along with advances in artifical intelligence (AI)/machine learning (ML), and next-generation analytics, can help life sciences companies to analyze the complexity of real-world patient journeys. These datasets reveal patient outcome patterns, such as disease progression, and help to improve the companies’ engagement efforts to deliver relevant, timely messages and support to patients and their HCPs so more patients receive proper care and treatments.
Leveraging AI/ML to understand a patient’s journey and unleash actionable insights, that would otherwise go unnoticed with traditional approaches, empowers life sciences companies to optimize their go-to-market strategies in an ever more complex and competitive environment.
Predicting patient outcomes is a critical step in improving the effectiveness of care delivery. Being able to forecast at scale what might happen to patients is key to improving population and individual patient health.
AWS Patient Outcome Prediction (POP) is a portable cloud-native web application designed with architectural and security best practices, validated and in production with multiple customers in various therapeutic areas. POP helps to uncover unique patterns in target patients’ medical history using natural language processing and AI/ML on top of de-identified, longitudinal real-world health data.
The ease of the web-based user interface and extensibility of machine learning powered backend enables life sciences organizations to understand the patient journey comprehensively. POP unleashes actionable insights about patient outcome patterns, such as disease progression, to support the early identification of patients, data-driven care management decisions, and timely interventions.
Deciphering the complexity of the patient journey with the power of AI/ML can provide a competitive edge for organizations who are in the preparation of a new product launch.
Gaining a comprehensive understanding of the patient and provider journey and identifying physicians with a relevant patient base for a new product launch is critical. POP empowers companies to optimize their segmentation, targeting, and opportunity mapping on high-priority physicians who will be critical for product adoption and prescription uptake.
The effort of leveraging AI/ML to predict eligible patients and linking de-identified patient cohorts to their treating physicians helps commercial teams to make better resourcing and go-to-market decisions. Furthermore, commercial teams can compile a list of high-opportunity physicians to optimize their outreach and engagement strategies.
How AWS Patient Outcome Prediction works
The following diagram illustrates the architecture behind the purpose-built application: Patient Outcome Prediction. ML is used on customer health patient data to train models to predict medical patient outcomes, such as disease progression, hospital readmission probability, and more.
Figure 1. AWS Patient Outcome Prediction reference architecture
- The customer accesses POP and inputs patient health data such as medical records, insurance claims, lab reports, doctor’s notes, and more through AWS Transit Gateway.
- AWS Transit Gateway connects Amazon Virtual Private Clouds (VPCs) and on-premises networks through a central hub. It acts as a highly scalable cloud router—each new connection is made only once. Managing incremental connections will no longer be necessary as the network grows in complexity.
- Customer health data is input into Amazon HealthLake (HealthLake), a HIPAA-eligible service that transforms customer health data to be queried and machine learning ingestible.
- Amazon HealthLake supports interoperable standards such as the Fast Healthcare Interoperability Resources (FHIR) format. It can also create a complete and chronological view of patient health data, including prescriptions, procedures, and diagnoses to accelerate patient care through insights.
- Customer datasets with privacy-sensitive data are scanned with Amazon Macie (Macie) via a AWS Step Functions pipeline and Amazon Eventbridge to discover and protect potentially identifiable information (PII) and other custom-defined sensitive data.
- Amazon Macie automates sensitive data discovery at scale using machine learning and pattern matching to detect and protect sensitive data stored in an S3 environment.
- AWS Step Functions automate workflows across AWS services without the need for maintaining code while creating data and ML pipelines.
- Amazon EventBridge is a serverless event bus that lets you receive, filter, transform, route, and deliver events.
- Customer data from HealthLake and Macie is stored in Amazon Simple Storage Service (Amazon S3) and processed by AWS Glue. The data is transformed and prepared to be ingested as training data into a custom-trained Amazon SageMaker (SageMaker) model.
- Amazon Simple Storage Service manages data at any scale with robust access controls, flexible replication tools, and organization-wide visibility with storage classes to reduce costs without upfront investment or hardware refresh cycles.
- AWS Glue is a serverless data integration service that makes it straightforward to discover, prepare, move, and integrate data from multiple sources for analytics, machine learning, and application development. It can support various data processing methods and workloads, including ETL, ELT, batch, and streaming.
- Amazon SageMaker enables building, training, and deploying machine learning models for any use case with fully managed infrastructure, tools, and workflows. It can reduce training time from hours to minutes with optimized infrastructure—boosting team productivity up to 10 times with purpose-built tools.
- Custom SageMaker models are trained to predict patient outcomes such as disease progression for potentially undiagnosed patients, hospital readmission probability, and more. SageMaker model endpoints are then created for model inference.
- Customers access the web client front-end securely to perform model inference through Amazon CloudFront and AWS WAF. This solution enables the delivery of a low-latency, consistent web experience to globally distributed users while maintaining an access list of authorized clients.
- Amazon CloudFront is a content delivery network (CDN) service built for high performance, security, and developer convenience. Cut costs with consolidated requests, customizable pricing options, and zero fees for data transfer out from AWS origins.
- AWS WAF helps protect against common web exploits and bots that can affect availability, compromise security, or consume excessive resources. It can monitor, block, or rate-limit common and pervasive bots.
- When a customer wants to perform model inference from a previously trained model, an AWS Lambda (Lambda) trigger lets the customer pick a SageMaker model endpoint to perform predictions.
- AWS Lambda is a serverless, event-driven compute service that can run code for virtually any type of application or backend service without provisioning or managing servers. It saves costs by paying only for the compute time used—by per-millisecond—instead of provisioning infrastructure upfront for peak capacity.
- When a customer wants to explore model explainability, a Lambda trigger lets the customer look at potential bias and more in their training data and trained models through Amazon SageMaker Clarify.
- Amazon SageMaker Clarify provides machine learning developers with greater visibility into their training data and models so they can identify and limit bias and explain predictions.
- All VPC flow logs, API metrics, and all other AWS resource usage is recorded using Amazon CloudWatch, AWS Config, and AWS CloudTrail (CloudTrail) to ensure operational and cost monitoring.
- Amazon CloudWatch collects and visualizes near real-time logs, metrics, and event data in automated dashboards to streamline infrastructure and application maintenance. It also improves operational performance using alarms and automated actions set to activate at predetermined thresholds.
- AWS Config continually assesses, audits, and evaluates the configurations and relationships of resources. It provides straightforward operational troubleshooting by correlating configuration changes to particular events in the account.
- AWS CloudTrail monitors and records account activity across the AWS infrastructure, providing control over storage, analysis, and remediation actions. It captures and consolidates user activity and API usage across AWS Regions and accounts on a single, centrally controlled platform. Using CloudTrail logs can help protect against penalties by proving compliance with regulations such as SOC, PCI, and HIPAA.
Optionally, a visualization pipeline can be built using Amazon API Gateway (a service to create, maintain, and secure APIs at any scale) allowing for visualizing the predicted results with custom user interface components. It also has no minimum fees or startup costs, which can reduce costs as API usage scales.
As an example, predicted, de-identified patients can be geographically visualized and linked back to their treating physicians. This helps the commercial team of a life sciences organization to uncover providers with a relevant eligible patient pool to optimize segmentation, targeting, and outreach strategies to accelerate product adoption and prescription capture.
Figure 2. Patient Geo Mapping to help segmentation and targeting of HCPs (disease prevalence with treatment eligibility and HPC linkage)
Advances in AI/ML and cutting-edge analytics on longitudinal, real-world health data can help life sciences companies to unleash actionable insights along the patient journey. Gaining a better understanding of patient outcome patterns and making sense of variables such as disease flare-ups, treatment starts, and therapy switches could help life sciences organizations to deliver relevant messages at the right moment of care management. This can support new product launches and accelerate access and adoption of potentially beneficial therapies for patients in need.
For more information on AWS solutions for life sciences, contact an AWS Representative.