Creating a real-time natural language processing framework in weeks

With as many as 10,000 documents needing extraction per day, the client realized it needed a back-end application to improve processing requests.

Challenge

EXL Health’s client, a leading healthcare provider, struggled with extracting the relevant information from medical referral requests. These requests were in the form of unstructured free-form text, requiring lengthy, manual-intensive reviews to identify any required data.

With as many as 10,000 documents needing extraction per day, the client realized it needed a back-end application to improve processing requests. The company partnered with EXL Health to fully automate the information extraction process.

Human ingenuity in action

Looking and going deeper, EXL Health leveraged an effective clinical data loading infrastructure from its care management application, and customized it to meet the needs of the provider.

In a matter of weeks, EXL Health developed a framework. The system received the documents as PDF images, scanned them using optical character recognition (OCR) to extract the text from the documents, and processed this text using natural language processing (NLP). The results could then be loaded into the downstream care systems in a fully automated process.

EXL Health designed this process with continuous improvement in mind. Real-time monitoring identifies requests that process longer than normal, low-quality extracted data, or failed requests. In any of these cases, the system notified operators to stop the request, and remediate whatever underlying issue was causing these problems.

The quality team performed a manual, retrospective review on a random sample of 3% the daily results to detect any issues. EXL Health communicated issues to the NLP team that incorporated these findings to strengthen the NLP model.

Outcomes

The resulting back-end web application notified reviewers of incoming documents, provided feedback, and successfully routed requests.

Additionally, real-time monitoring prevented bad data from passing on to the client and accounted for the thousands of daily requests. Within a month, EXL successfully instituted an advanced NLP solution for the client enabling them to automate a time-intensive process.