Desiderata for delivering NLP to accelerate healthcare AI advancement and a Mayo Clinic NLP-as-a-service implementation

Abstract

Data is foundational to high-quality artificial intelligence (AI). Given that a substantial amount of clinically relevant information is embedded in unstructured data, natural language processing (NLP) plays an essential role in extracting valuable information that can benefit decision making, administration reporting, and research. Here, we share several desiderata pertaining to development and usage of NLP systems, derived from two decades of experience implementing clinical NLP at the Mayo Clinic, to inform the healthcare AI community. Using a framework, we developed as an example implementation, the desiderata emphasize the importance of a user-friendly platform, efficient collection of domain expert inputs, seamless integration with clinical data, and a highly scalable computing infrastructure.

An aberration detection-based approach for sentinel syndromic surveillance of COVID-19 and other novel influenza-like illnesses

“…Sign and symptom extraction via natural language processing was accomplished via the MedTagger NLP engine [67] , [68] . The signs and symptoms chosen were selected via a literature review conducted in early March 2020 for known COVID-19 and influenza symptoms [13] , [69] .…”

Natural Language Processing Applications in Business

“…Similarly, a study conducted by researchers from the University of Alabama found that by applying NLP techniques the reportable cancer cases identified was 22.6% more accurate than a manual review of medical records [47]. Further, various other studies have identified the potential benefits, including effective decision-making by physicians; reducing physician burnout (disillusionment among doctors tired of repetitive data entry tasks and administrative duties as well as excessive time spent combing through patient data); addressing accelerating demand for healthcare services; and dealing with increasing numbers of healthcare claims [48][49][50][51], through the application of NLP techniques in the healthcare context.…”

Patient trajectories and risk factors for severe outcomes among persons hospitalized for COVID-19 in the Maryland/DC region

“…Table S4 includes the list of symptom categories and the search terms. We then selected relevant clinical note types for each patient, including H&P, Critical Care Notes, Progress Notes, and ED Notes, focusing specifically on the notes created within 48 hours before and after admission. Next, we pre-processed the note text and extracted only the relevant narrative parts, particularly the chief complaint and history of the present illness sections. We then used a COVID-19-customized version of MedTagger, 2 together with our in-house Python tools to (a) identify phrases and synonyms of particular symptoms within the text narratives, (b) determine if these symptom mentions are negated, possible, or positive in their context, (c) classify symptoms into the predefined 11 categories, and (d) map them to their corresponding UMLS Concept Unique Identifiers (CUIs). These NLP pipelines use a combination of machine learning models, including Conditional random fields (CRFs), 3 and contextual rule-based methods, including regular expressions.…”

“…We then used a COVID-19-customized version of MedTagger, 2 together with our in-house Python tools to (a) identify phrases and synonyms of particular symptoms within the text narratives, (b) determine if these symptom mentions are negated, possible, or positive in their context, (c) classify symptoms into the predefined 11 categories, and (d) map them to their corresponding UMLS Concept Unique Identifiers (CUIs). These NLP pipelines use a combination of machine learning models, including Conditional random fields (CRFs), 3 and contextual rule-based methods, including regular expressions.…”