Prediction of Diagnosis Related Groups classification for Medical Services

Abstract

The introduction of diagnosis-related groups (DRG) in the medical sector, the cost capping of prices, led to the fact that treatments often have a low margin or are not cost-covering. For this reason, it is extremely important that all services subsequently provided are also charged to the patient. This thesis has been completed in cooperation with the Heuberger Eye Clinic and the software supplier for the daily business MedicalDesktop. Currently, the notes of the medical staff, contained in the MedicalDesktop software, are used to create the patient invoice. This step requires specific medical knowledge to understand both the notes and the DRGs, as a correct allocation has to be made. To support the staff in this task, a machine learning model is developed.The aim of this thesis is to determine whether a recommendation for the DRG allocation can be made on the basis of unstructured notes as well as structured information (e.g., measurement of the sphere). For this purpose, a machine learning model is developed and subsequently optimized. The results show that a recommendation by the model can support the staff helpfully. The results indicate that additional revenue can be generated with the help of the model. On this basis, the hypothesis can be confirmed that unstructured and structured data can improve the accuracy and completeness of invoices.