Machine Learning to Identify Point-of-Care Ultrasound and Evaluate Standardized Documentation: Retrospective...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Ultrasound Technology

Background:

• Point-of-care ultrasound (POCUS) is vital in obstetrics and gynecology (OBGYN) for bedside diagnostics and therapeutics.
• Accurate documentation and billing for POCUS are challenged by inconsistent workflows, variable note quality, and EHR inefficiencies.
• These barriers lead to missed procedural charges, impacting operational, educational, and reimbursement efforts.

Purpose of the Study:

• To leverage machine learning (ML) for automated identification of POCUS procedures in clinical notes.
• To assess the impact of standardized procedure documentation (ProcDoc) templates on billing capture accuracy and efficiency.

Main Methods:

• A retrospective cohort study analyzed 559,029 encounters using EHR data from 11 OBGYN clinics.
• Machine learning models (LightGBM, BioClinBERT) were trained to classify POCUS procedures in clinical notes.
• A standardized ProcDoc smart form was implemented to streamline documentation and CPT code triggering, with pre- and post-intervention periods compared.

Main Results:

• The BioClinBERT model achieved high accuracy (0.97) in identifying POCUS procedures.
• ProcDoc adoption reached 75.1%, and billing recapture dropped from 10.0% to 2.4% post-intervention.
• Standardized documentation significantly improved workflow efficiency and reduced manual audit burden, with most CPT codes originating from ProcDoc templates.

Conclusions:

• ML is effective for extracting POCUS procedures and evaluating workflow interventions.
• Standardized documentation via ProcDoc significantly enhances charge capture accuracy and reduces reliance on manual reviews.
• This ML-driven approach can address documentation inefficiencies and promote healthcare sustainability.