Learn to build large language model applications: vector databases, langchain, fine tuning and prompt engineering. Learn more
$2.97B+
$1.64B+
36,000
746 in 47 states
2,686
3.9 M
11000+
The project involved automating the extraction of HIPAA identifiers from a dataset of around 200,000 referral documents.
The document set provided by the client had 500+ different formats with no proper labeling. We started by meticulously reviewing and categorizing the documents to build a document corpus with representative documents.
90% accuracy on referral documents On a representative test set of 929 referral documents, the system was able to extract the desired fields accurately around 90% of the time.
The system's performance was not limited to referral documents alone. It also demonstrated good accuracy on 425 other documents, which were entirely new. The system was able to achieve an overall accuracy of 75% on this new and unseen data, indicating its ability to generalize well.
Overall impact: The client manually processes over 100,000 requests to extract patient information and enter it into the system before engaging the consulting team. By creating an automated HIPAA identifier extraction system, the team helped the client in improving efficiency and reducing errors caused by manual processing. The automated extraction system has reduced the processing time significantly and improved the patient’s look up in the system by providing a combined accuracy of 90% on desired fields. Overall, our achievement of automating extraction of HIPAA identifiers from documents is a remarkable feat that will significantly enhance client’s operations.
