Semantic interoperability and price analytics in hospital transparency data: a multi-stage pipeline with NLP and...

Area of Science:

• Health Informatics
• Computational Biology
• Health Services Research

Background:

• The U.S. Hospital Price Transparency mandate mandates public disclosure of machine-readable files (MRFs).
• Significant data heterogeneity in MRFs currently limits their utility for research and consumer price comparison.
• Standardizing diverse MRF data is crucial for enabling meaningful analysis of hospital pricing.

Purpose of the Study:

• To evaluate a novel, multi-stage computational pipeline designed for systematic processing of diverse hospital MRFs.
• To enable robust price analysis and facilitate research and consumer use of healthcare pricing data.
• To address the challenge of data heterogeneity in MRFs for improved healthcare price transparency.

Main Methods:

• Developed a multi-stage computational pipeline integrating a configurable parsing engine and an NLP module.
• Utilized Sentence-BERT embeddings and K-Means clustering for semantic standardization of procedure descriptions.
• Aligned standardized descriptions with CPT codes and applied the pipeline to MRFs from five major U.S. hospitals for five elective procedures.

Main Results:

• Successfully processed 7,449 records from diverse hospital MRFs.
• Identified substantial price variations across hospitals for semantically equivalent healthcare services.
• Predictive modeling using Lasso regression achieved an R-squared of 0.70 (RMSE=$808; MAE=$533).

Conclusions:

• The developed pipeline offers a scalable, semi-automated methodology for MRF research.
• Transforms opaque hospital pricing data into an analytically tractable format.
• Has significant implications for healthcare policy development and consumer-facing healthcare tools.