JMIR Publications

ABSTRACT

Background:

The clinical narrative in the electronic health record (EHR) carries valuable information for predictive analytics, but its free-text form is difficult to mine and analyze for clinical decision support (CDS). Large-scale clinical natural language processing (NLP) pipelines have focused on data warehouse applications for retrospective research efforts. There remains a paucity of evidence for implementing open-source NLP engines to provide interoperable and standardized CDS at the bedside.

Objective:

The clinical protocol describes a reproducible workflow for a cloud service to ingest, process, and store clinical notes as Health Level 7 messages from a major EHR vendor in an elastic cloud computing environment.

Methods:

The technical architecture that enables the real-time, NLP CDS tool incorporates industry-leading and emerging technological capabilities. The overall NLP CDS infrastructure is protocolized with details to export the notes from the EHR, organize them and feed them into an NLP pipeline, input the processed text features into the opioid screener deep learning model, and deliver the resultant scores back to the bedside EHR as a BPA. The final architecture is a real-time NLP CDS tool, and the six components of the architecture are detailed for other health systems to benchmark. We applied the NLP CDS infrastructure to a use-case for hospital-wide opioid misuse screening using an open-source deep learning model that leveraged clinical notes mapped to standardized medical vocabularies. A random sample of 100 adult patient encounters (with an over-sampling of patients with diagnostic codes for substance misuse) in 2021 was extracted and reviewed by an inpatient physician and clinical informatics expert. The protocol also included a human-centered design and an implementation framework with a cost-effectiveness and patient outcomes analysis plan.

Results:

Internal validation of the NLP opioid misuse screener performed similarly to prior published reports for screening opioid misuse with a sensitivity of 93% (95% CI 66%-99%) and specificity of 92% (95% CI 84%-96%). The resultant NLP and deep learning pipeline can process clinical notes and provide decision support to the bedside within minutes of a provider entering a note into the EHR for all hospitalized patients. The longest delay in operational workflow and architecture was receiving cybersecurity approvals, especially with data exchange of protected health information between the Microsoft© and Epic© cloud vendors.

Conclusions:

The deployment of medical artificial intelligence (AI) systems in routine clinical care present an important yet unfulfilled opportunity, and our protocol aims to close the gap in the implementation of AI-driven CDS.