We at CrossAsyst have been working closely with the healthcare industry for well over a decade, building them custom software solutions to make their functioning easier and more efficient by leveraging the latest technologies.
So, when generative AI started making its presence felt as an aid to custom software development, our team hunkered down and worked at how best to leverage this powerful new technology to build tools that will benefit our existing and future clients.
This case study explores how CrossAsyst used generative AI to develop a tool with the potential to transform ICD-10 coding, adding measurable value to healthcare operations.
The Problem: The Inefficiencies and Slow Pace of Manual ICD-10 Coding
Many healthcare organizations today are still relying on manual processes to extract and assign ICD-10 codes from physician notes for insurance and compliance purposes. This is especially true when generating HCFA forms, a required step in billing and claims submission for Workers’ Compensation and other use cases.
Here’s what we saw:
- Each month, providers were processing 1,200 to 1,300 injury intake cases
- The task of manually reviewing physician notes and mapping them to ICD-10 codes took at least 5 minutes per case
- This added up to over 6,000 minutes (100+ hours) per month
- The process required multiple staff members, leaving room for human error, inconsistency, and burnout
Our Solution: A Gen AI-Powered ICD-10 Coding Platform
To address the inefficiencies and complexity of manual ICD-10 coding, our team at CrossAsyst developed a robust, AI-driven solution that fully automates the end-to-end process of extracting, interpreting, and assigning ICD-10 codes, leveraging the combined power of Natural Language Processing (NLP) and Large Language Models (LLMs).
Our solution was designed to handle the unique challenges of unstructured medical data, accurately interpret clinical language, and return relevant ICD-10 codes within seconds, all while integrating smoothly into existing healthcare systems and workflows.
How Our Generative AI-Powered Coding Platform Works
1. Input Through UI or API
Users or systems submit free-text physician notes through a user-friendly web interface or programmatically via API endpoints.
2. Text Processing with NLP Pipelines
Once received, the platform processes the input using our custom-trained NLP pipelines, which clean and structure the text for optimal understanding.
3. ICD-10 Code Generation
This refined input is then passed through our pretrained LLM, hosted on AWS Bedrock, which interprets the clinical context, identifies underlying conditions, and predicts the most relevant ICD-10 codes and their corresponding descriptions.
The result?
Real-Time, High-Accuracy Output
Once the ICD-10 codes are predicted, the results are delivered instantly with a high degree of accuracy, designed to seamlessly fit into the healthcare organization’s existing workflow. The output can be utilized in multiple ways.
1. Direct Insertion into HCFA Forms
The codes and descriptions can be automatically mapped into standardized HCFA (CMS-1500) forms, eliminating the need for manual data entry and significantly reducing the time required for claims preparation.
2. Faster Claims Processing
The increased accuracy of the codes entered also means claims processing can be a lot faster, making it a boon for all stakeholders.
3. Integration with EHR, Billing, or Claims Systems
Through API connectivity, the results can be directly integrated into Electronic Health Records (EHRs), billing platforms, or claims management systems, ensuring continuity across the organization’s technology stack.
3. Optional Validation by Staff
For organizations that require an additional review step, our solution supports an optional validation layer, where staff can quickly review, confirm, or adjust the suggested codes, providing both flexibility and oversight when needed.
Why AWS Bedrock?
Building our platform on AWS Bedrock gave the team at CrossAsyst the ability to scale rapidly, deploy securely, and retain full flexibility in how we select and integrate foundational models. Here’s why AWS Bedrock was the ideal choice for our ICD-10 automation solution.
1. Scalable, Secure Access to Leading Models
AWS Bedrock offers managed access to top-tier foundation models from providers like Anthropic, Meta, and Amazon Titan, allowing us to work with best-in-class AI without the overhead of managing infrastructure, compute, or security concerns.
2. Structured Workflows with Bedrock Agents
With Bedrock Agents, we can design step-by-step, intelligent workflows that handle complex tasks like ICD-10 code reasoning. These agents can invoke AWS Lambda functions, perform intermediate logic, and chain tasks together in a way that reflects how coders think in the real world.
3. Context-Rich Responses Using Retrieval-Augmented Generation (RAG)
Bedrock’s support for Knowledge Bases and Retrieval-Augmented Generation enables our system to reference real-time, domain-specific medical information during inference. This drastically reduces errors and ensures that outputs are accurate, contextually relevant, and verifiable.
Partner With CrossAsyst to Build Gen AI Tools With Real-World Impact
The ICDO-10 coding platform we built is just the tip of the iceberg. The same generative AI architecture can be used to build multiple healthcare related tools, such as:
- Prior authorization documentation
- Auto-generating patient summaries
- Automating claims verification
- Clinical decision support
Our expertise can be leveraged to build generative AI-powered tools across other business verticals as well.
Are you ready to empower your business with generative AI tools that are sure to make a positive impact? Book a meeting with CrossAsyst today, and let us build you future-ready custom software solutions.