Computer-assisted medical billing information extraction: comparing rule-based and end-to-end transfer learning approaches

Medical billing is important for both healthcare providers and payers, yet filling reimbursement requires tremen-dous effort to process clinical notes, making it labor-intensive and error-prone. This work compares two natural language processing (NLP) approaches to extract patients' history information from clinical notes for billing purposes. A rule-based pipeline built on top of a generic clinical NLP tool CLAMP, is compared against an end-to-end deep neural network architecture. We annotate a gold-standard corpus to evaluate the two approaches. Results show information extraction for medical billing is a challenging problem though NLP has great potential to automate the task. Our work is the first academic study using NLP in Evaluation and Management billing.