Building an AI-Powered Oncology Decision Support Platform

Project background

The healthcare industry – and oncology in particular – faces challenges in identifying suitable laboratory tests so that doctors can make accurate medical diagnoses. Oncologists – like anyone else – strive to find a balance between medical effectiveness and cost efficiency, while doing their best to serve their patients by offering them procedures that are neither excessive nor underutilized.

Our client, a US-based healthcare technology provider, saw this gap in the market as an opportunity to develop an AI-based clinical decision support system for oncology that would simplify the selection of lab tests for diagnostics. The cancer decision support platform aimed to help medical professionals make informed, rapid choices and assist insurance companies in reducing their total expenses.

The project aims to address the distinct needs of healthcare professionals by optimizing oncology lab test choices to eliminate unnecessary tests, expedite clinical decisions, and reduce patient expenses.

Target Audience: The decision support system for oncologists serves cancer centers, testing labs, health plans, and pharmaceutical companies by providing evidence-supported treatment options and enhancing reimbursement strategies.

Business needs

Product features

– Suggested Lab Test Sets: Provides oncologists with optimized lab test recommendations, marked as auto-approved.

– Document Generation: Automatically generates, sends, and receives documents related to lab test orders, decisions, and analytics.

– Regimen Suggestions: Suggests appropriate treatment regimens, with some auto-approved by payers.

– Streamlined Decision-Making Automation: Accelerates the process of selecting and ordering the correct lab tests, ensuring payer approval for faster clinical decision-making.

Solution

1. Microservices Architecture Restructuring

Before, it was 7 to 8 large microservices, but the problem was complexity and a lack of scalability. We changed the architecture by decomposing the services into smaller ones, each consisting of 2 to 3 components, which made it more modular and scalable. This transformation enabled a more modular and scalable architecture, critical for building a robust oncology decision support platform.

For example, report generation, a key part of any oncology diagnostics software development, was resource-intensive. After refactoring, we scaled only the relevant module without affecting others that do not require high performance, thereby optimizing resource allocation.

2. Architecture Review and Optimization

The final version of the system currently consists of approximately 50 microservices, each tailored to handle the complexity of a specific part of the cancer decision support platform. One example is a document-handling system that was initially split into 15 small services, but now consolidated into three stateless components. This restructuring helps maintain performance within the clinical decision support system for oncology.

The overall system is asynchronous and built on messaging queues, allowing both upward and downward scalability. Kubernetes ensures high availability and flexible scaling, aligning the architecture with best practices for modern oncology software for doctors.

3. New Microservices Implementation

As part of our additions to the platform, new services were added to cover external partners such as labs, websites, patients, and insurers. A custom-built authentication service based on Spring Security 6 to manage roles, including doctors, patients, support staff, curators, and developers. This layer of functionality supports secure access within the decision support system for oncologists.

The platform’s communication relies on RabbitMQ, REST APIs, and integrations with Mirth, email, and HL7 protocols. Kotlin is the primary language for development due to its flexibility and licensing advantages over Oracle Java. Deployment is handled via Kubernetes using Terraform scripts prepared by DevOps engineers.

4. Integration with New Laboratories

The system exchanges personal, medical, and order-related data in various formats, including HL7, PDF, CSV, JSON, HTML, and XML, which is critical for real-time interaction within any modern oncology decision support software. Communication protocols include REST, SOAP, and email.

Integration is limited to data exchange rather than specific lab systems, with Mirth handling medical data messaging. Existing workflows remain stable thanks to a flexible architecture and consistent management of tech debt, ensuring seamless integrations that enhance the oncology software for doctors’ experience.

5. Integration with Medical Practices

Our portal supports data redirection from external medical platforms, including multiple authentication mechanisms such as credentials, third-party portals, and custom-built single sign-on solutions similar to Google’s OAuth. These mechanisms are essential for ensuring secure access and maintaining compliance within a clinical decision support system for oncology, enabling smooth collaboration between oncologists and external care providers.

6. Configuration Management System

We implemented a dynamic configuration system that allows curators to define unique behaviors per practice, ranging from UI forms to order processing rules. This flexibility enables us to create custom logic without modifying code, which is particularly important when adapting the oncology decision support software for an environment with a diverse array of clients. Orders from the same practice can be formatted flexibly, ensuring fast onboarding and seamless interaction across platforms.

Our technology stack

• Frontend: JavaScript, TypeScript, HTML, CSS, Angular 11-17, AngularJS, RxJS, Jasmine/Karma
• Databases: PostgreSQL, MariaDB, MongoDB
• Cloud: AWS (EC2, CodeDeploy, S3)
• Backend: Java, Kotlin, Spring 3-5, Spring-Boot 2-3, Spring-Cloud, Spring WebFlux, Hibernate, Liquibase, Swagger, Apache Fop, RabbitMQ
• Other Tools: Jenkins, BitBucket, Jira, Confluence, Zoom

Key achievements delivered