Automated reimbursement

Challenge

Until now, the clerks of the Austrian health insurance providers had to process the private medical practitioners’ reimbursements manually in order to extract the relevant information such as practitioner, patient, treatment and costs. In the process, cases are increasing by 8% annually. This has led to an ever-increasing backlog in processing for many insurance companies.

Idea

The clerks should be supported by automatically extracting and validating the information from the invoices. Manual entry is then eliminated, only the verification of the information needs to be done. This should be made possible by:

• Machine text recognition in combination with content recognition through computer vision, natural language understanding and machine learning
• Creation of a highly scalable system for processing large amounts of data

Solution

Picklist, a scalable web-based system, processes the scans of the invoices and presents the results to the clerk in an intuitive view. All he has to do is review the extracted data and correct it if necessary. The system has the following features:

• Complete integration into the existing software landscape
• Acceptance of scanned or photographed fee invoices
• Document pre-processing
• OCR (Optical Character Recognition)
• Document classification (fee invoice, payment document, medical assignment, etc.)
• Entity extraction (patient, insured person, attending practitioner, services billed, diagnoses made, costs)
• Assignment of entities to each other (date to service, name to insurance number)
• Plausibility check
• Recoding of services in free text to the service catalogs of the carriers
• Recoding of remedies to the official reimbursement code of remedies
• Recoding of diagnoses to the official ICD-10 catalog
• Matching of services with diagnoses
• Recognition and reading of handwritten IBANs
• Recognition of handwritten annotations and differentiation of signatures
• Recognition of a balancing note
• Intuitive UI

This is how our modular, automated system works

The clerks are relieved of all steps that can be automated, as the individual components such as text recognition, entity recognition or post-processing communicate with each other in a decoupled manner via queues. The system is both scalable and fault-tolerant. The result of the processing is visible in a modern web interface and the data can still be corrected for special cases before the costs can be reimbursed.

The capabilities of Machine Learning 

The system was taught to find the required content in differently structured medical invoices in order to process them automatically. The machine learning part of the application was continually tested against a wide variety of metrics during development. This ensured that the recognition of future, never-before-seen documents would also work reliably.

Digital archiving as an add-on

In addition to digitalization of the reimbursement process, digital archiving of documents has been introduced. These now only need to be kept digitally. A positive side effect was that employees were enabled to work securely from home for several months during the Covid 19 lockdown starting in March 2020.