Utilising Natural Language Processing to Identify Brain Tumor Patients for Clinical Trials: Development and Initial...

Area of Science:

• Medical Informatics
• Clinical Research Informatics
• Natural Language Processing

Background:

• Clinical trial eligibility screening is resource-intensive.
• Natural Language Processing (NLP) models can extract data from Electronic Health Records (EHRs) to improve screening.
• Automating data extraction can streamline the identification of eligible patients.

Purpose of the Study:

• To evaluate an NLP model's ability to extract brain tumor diagnoses from outpatient clinic letters.
• To assess the feasibility of linking extracted diagnoses to ongoing clinical trials.
• To determine the accuracy and relevance of NLP-identified clinical trial matches.

Main Methods:

• Retrospective cohort study of neuro-oncology clinic letters.
• Utilized an NLP model for Named Entity Recognition and Linking to the Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) ontology.
• Human annotators validated extracted concepts and trial relevance; results displayed on an EHR-integrated dashboard.

Main Results:

• The NLP model achieved high performance: macro-precision = 0.994, macro-recall = 0.964, macro-F1 = 0.977.
• Identified 399 medical concepts from 196 clinic letters.
• Linked to 1417 trials, with 755 highly relevant matches for patients meeting eligibility criteria.

Conclusions:

• NLP effectively extracts brain tumor diagnoses from free-text EHR data with minimal training.
• Extracted information can be linked to relevant clinical trials, suggesting integration into clinical workflows.
• Further research is needed to evaluate the impact on clinical outcomes.