Project Synopsis
For healthcare enterprises in the US, having patient data such as medical records, discharge summaries, and insurance records at their fingertips is critical to providing value-based healthcare.
Unfortunately, the major challenge in the healthcare industry is that about 80% of the data is unstructured making timely access to relevant information a major hurdle.
A leading US-based healthcare enterprise leveraged Ideas2IT’s Generative AI program to explore the feasibility of using LLMs to extract insights from medical records at scale.
The intent was to develop an LLM that can scan through extensive collections of medical reports and patient documents and instantly retrieve relevant, accurate data, enabling medical professionals to provide better patient care. All this while ensuring compliance with HIPAA guidelines and complete anonymization of all PII data.
![representing the flow of information from complex healthcare documents to the hands of healthcare professionals, facilitated by state-of-the-art technology.](https://cdn.prod.website-files.com/64f989999025f3e47402a969/66ab35754767af99cb368785_65cafdbd420e9191693b1caa_image-2-65cafd6bc6fcb.png)
The Problem
The high-priority problem statement is treatment delays. These can be life-altering for patients and their loved ones and the entire healthcare ecosystem is leveraging generative AI to improve the speed and accuracy of diagnosis and eventually care by extracting insights and cohesively stitching together all patient records and historical data.
We had to build a fast data retrieval system that could read all this information, have self-learning capabilities, and provide accurate responses to individuals on the field.
![complexity of healthcare documents, with various stakeholders striving to navigate through it to reach the critical information. Overlay this with a clock symbolizing the urgency of timely care.](https://cdn.prod.website-files.com/64f989999025f3e47402a969/66ab35754767af99cb36877f_65cafdd919e728071d2ef563_image-3-65cafd68ec9f3.png)
Solution
- Ideas2IT’s solutions team started by listing out every stakeholder who would interact with the chatbot, mapping out their unique requirements and expectations.In this case, this included everyone from business heads to RCM and Homecare teams.
- As a first step, we carefully redacted PHI from all files. This was thanks to a custom e-PHI handler that we had previously built in-house.
- Once data privacy was airtight, we enhanced document retrieval and indexing. This meant testing out several embedding methods and identifying one that offered the fastest vector processing capability for this use case. We used the open-source FAISS Vector Store to store the data.
- Medical practitioners are busy and need incredibly easy and intuitive input methods to use the model. For this, we turned to the Vicuna-13b model chatbot. For optimal performance, we hosted Vincuna on AWS.
- Finally, we had to create a system by which the model could ingest, understand, and translate the intent of the queries, while accurately extracting the information needed. To do so elegantly, we built out a LangChain pipeline.
- Pursuant steps involved fine-tuning the system and adjusting generic prompts to optimize information retrieval. We also built tailored prompts for different queries until we created a truly precise response mechanism that understood the context.
The Impact
Using the platform, the client’s RCM teams were able to cut down the time taken to extract accurate information by 50%.
Similar results were seen with homecare and tertiary care teams.
The Generative AI solution that we tailored for this healthcare giant, empowered them to retrieve critical patient information speedily and consistently, drastically improving the delivery of value-based care in record time.
![](https://cdn.prod.website-files.com/64f989999025f3e47402a969/66ab35754767af99cb36877b_65cafdf166aab0f61293eba4_image-4-65cafd686603e.png)
"For years, we've been enhancing the patient experience and value of delivery of care in healthcare companies. Our standout projects have been in this field. The swift embrace of Generative AI by practitioners is transforming care delivery. It's improving diagnosis and patient experiences remarkably. We're thrilled to contribute and empower more healthcare providers for better care."
- Murali
Conclusion
Working. Healthcare is a vast, complex industry, and solutions have to go through multiple feasibility checks for actionable business impact.
In conclusion, our collaborative efforts alongside various stakeholders have played a pivotal role in ensuring significant, measurable improvements in how our clients deliver timely, accurate care with better experiences for both practitioners and patients.
Working shoulder to shoulder with different stakeholders to ensure the success of the project has helped us deliver impact and map measurable business objectives.
The expansive and complex nature of the healthcare industry demands thorough feasibility assessments to guarantee actionable business impacts.