Document Understanding for Healthcare Referrals

TL;DR

Addressing high costs and errors in healthcare referrals delivered as faxes or scans, the paper presents a hybrid information-extraction pipeline that combines a multi-modal transformer LayoutLMv3 with domain-specific rules to identify patient, referring physician, and exam entities. The method processes OCR text with a custom line-grouping scheme (DBSCAN-based), plus preprocessing and postprocessing rules to improve exact matches. Evaluated on a curated dataset of 3032 training and 112 test pages, the hybrid approach produced meaningful gains in precision and F1 under $MUC-5$ metrics, especially after postprocessing. The work demonstrates that domain knowledge and careful data curation substantially enhance referral-management efficiency in real-world settings.

Abstract

Reliance on scanned documents and fax communication for healthcare referrals leads to high administrative costs and errors that may affect patient care. In this work we propose a hybrid model leveraging LayoutLMv3 along with domain-specific rules to identify key patient, physician, and exam-related entities in faxed referral documents. We explore some of the challenges in applying a document understanding model to referrals, which have formats varying by medical practice, and evaluate model performance using MUC-5 metrics to obtain appropriate metrics for the practical use case. Our analysis shows the addition of domain-specific rules to the transformer model yields greatly increased precision and F1 scores, suggesting a hybrid model trained on a curated dataset can increase efficiency in referral management.

Figures (2)

• Figure 1: Sample of document structures in referrals.
• Figure 2: Grouping of OCR lines in different document structures.