Simplifying data entry through automatic identification and extraction: a controlled experimental study on Diagnosis-Related Group encoding

This study aims to develop and evaluate an Optical Character Recognition (OCR)-based system to identify and extract Diagnosis-Related Groups (DRG)- related data to facilitate manual entry. This study represents a controlled experimental analysis based on data derived from the Home Page of Medical Records (HPMRs). We developed DRG-AIE (DRG-Automatic Identification and Extraction), a novel system based on PaddleOCR . The control experiment included 20 participants with medical backgrounds. This study reports on the time and accuracy of data identification and extraction, as well as participants’ subjective feedback on DRG-AIE. DRG-AIE reduced entry time by an average of 53.27% per HPMR (71.80 [95% CI: 57.30–99.60] seconds, P < 0.01). The character and field accuracy are 99.36–98.97%, and the corresponding error rates are below the acceptable upper limits. Within different HPMR types, DRG-AIE also achieved significant time savings and an acceptable error rate. Participants’ subjective feedback on DRG-AIE was positive, and all participants indicated that they would recommend it to peers. DRG-AIE significantly accelerates and simplifies manual entry without compromising accuracy. It allows coders to focus on code review, HPMRs quality control, and medical insurance compliance checks. The DRG report it generates improves the verification of medical insurance payment evidence and the identification of payment risk violations.