JMIR Publications

ABSTRACT

Background:

Point-of-care ultrasound (POCUS) in obstetrics and gynecology (OBGYN) is essential for diagnosis and treatment. However, documentation and billing of these procedures frequently encounter challenges due to inconsistent practices and inefficiencies in EHR systems.

Objective:

This study employs machine learning (ML) to identify POCUS procedures from clinical notes and evaluates the impact of standardized documentation (ProcDoc) on billing capture.

Methods:

A multi-part retrospective cohort study to (1) develop a machine learning model for identifying POCUS procedures from clinical notes and (2) use the model to evaluate the effectiveness of implementing standardized documentation for these procedures. Machine learning models were developed using clinical encounter data from January 1, 2018, to December 31, 2020. Standardized documentation was introduced in February 2023 and assessed over a period from February 1, 2022, to August 31, 2024. Outcomes included model performance metrics and billing recapture assessments.

Results:

Analysis was conducted on 558,997 clinical encounters. The BioBERT model achieved 0.97 accuracy and 0.88 F1 score. Standardized documentation adoption reached 77.9% within six months. Post-intervention data indicated a reduction in charges identified through recapture from 10.0% to 2.4%, with an overall POCUS usage increase of +0.7%. The odds of recapture were significantly lower post-intervention (OR = 0.22, 95% CI: 0.17 to 0.30, p<.001).

Conclusions:

The study illustrates the effectiveness of using ML to enhance the identification and billing of POCUS procedures in OBGYN settings. The adoption of standardized documentation significantly improved accurate billing capture, highlighting ML's potential in optimizing clinical processes without direct EHR integration.