Deployment of Real-time Natural Language Processing and Deep Learning Clinical Decision Support in the Electronic Health Record: Pipeline Implementation for an Opioid Misuse Screener in Hospitalized Adults (Preprint)

• DOI: 10.2196/preprints.44977
• Publication Date: 2023-04-20T14:08:07Z
• Authors: Majid Afshar; Sabrina Adelaine; Felice Resnik; Marlon P Mundt; John Long; Margaret Leaf; Theodore Ampian; Graham J Wills; Benjamin Schnapp; Michael Chao; Randy Brown; Cara Joyce; Brihat Sharma; Dmitriy Dligach; Elizabeth S Burnside; Jane Mahoney; Matthew M Churpek; Brian W Patterson; Frank Liao
• Keywords: Triage; Best practice

<sec> <title>BACKGROUND</title> The clinical narrative in electronic health records (EHRs) carries valuable information for predictive analytics; however, its free-text form is difficult to mine and analyze decision support (CDS). Large-scale natural language processing (NLP) pipelines have focused on data warehouse applications retrospective research efforts. There remains a paucity of evidence implementing NLP at the bedside care delivery. </sec> <title>OBJECTIVE</title> We aimed detail hospital-wide, operational pipeline implement real-time NLP-driven CDS tool describe protocol an implementation framework with user-centered design tool. <title>METHODS</title> integrated previously trained open-source convolutional neural network model screening opioid misuse that leveraged EHR notes mapped standardized medical vocabularies Unified Medical Language System. A sample 100 adult encounters were reviewed by physician informaticist silent testing deep learning algorithm before deployment. An end user interview survey was developed examine acceptability best practice alert (BPA) provide results recommendations. planned also included human-centered feedback BPA, cost-effectiveness, noninferiority patient outcome analysis plan. <title>RESULTS</title> reproducible workflow shared pseudocode cloud service ingest, process, store as Health Level 7 messages from major vendor elastic computing environment. Feature engineering used engine, features fed into algorithm, returned BPA EHR. On-site demonstrated sensitivity 93% (95% CI 66%-99%) specificity 92% 84%-96%), similar published validation studies. Before deployment, approvals received across hospital committees inpatient operations. Five interviews conducted; they informed development educational flyer further modified exclude certain patients allow refusal longest delay because cybersecurity approvals, especially exchange protected between Microsoft (Microsoft Corp) Epic (Epic Systems vendors. In testing, resultant provided within minutes provider entering note <title>CONCLUSIONS</title> components detailed tools other systems benchmark. deployment artificial intelligence routine presents important yet unfulfilled opportunity, our close gap intelligence–driven CDS. <title>CLINICALTRIAL</title> ClinicalTrials.gov NCT05745480; https://www.clinicaltrials.gov/ct2/show/NCT05745480