Our work around Claims processing started in 2002 with one of the “Blues”. Improving the workflow, automating as much as possible in order to de-risk frauds, increase the number of processed claims, and optimise the claim handler’s time. One of our favourite projects was Medical Charts Review with a goal to improve quality management and oversight.
- calculate claim risk score
- Duplicate claim detection
- related claims matching
- physician notes processing
- auto assign logic
- Discrepancy check
By leveraging various publicly available and production ready API’s, we were able to create a proof of concept of an NLP solution in less than a week. Our PoC was an NLP application for extracting and categorizing medical and PHI information from physician notes.
By using the Box Cloud Content Management platform with its Skills module and configurable and user friendly Workflows, we were able to quickly utilize Amazon Comprehend Medical solution and custom business logic module.
Claim records selected for review are exported in a suitable file format and uploaded to the Box CCM. Box Skill processes the claim and orders the medical chart and physician note for that claims form the vendor.
When physician note is uploaded to the Box CCM by the vendor, it is processed by Box Skill and Amazon Comprehend Medical to extract all relevant medical information. This information is stored in file’s metadata and compared with the claim’s data.
In case of comparison discrepancies, a task is created in Box Workflow and assigned to the user for manual review and resolution.
High Level Architecture
Once the physician note is uploaded to the Box CCM platform by the chart vendor, it is processed by another Box Skill which invokes Amazon Comprehend Medical service in order to extract the structured information of diagnoses, symptoms, signs, medications, treatments and procedures that apply to a given patient record.
The solution presented above is an NLP application for extracting and categorizing medical and PHI information from physician notes. Learn more about this solution by diving deeper into the Claims Review technical overview document: