OBJECTIVES: Computable operational definitions (CODefs) are essential for identifying patient cohorts in real-world evidence (RWE) studies. However, manual development of these phenotypes is time-consuming and often lacks standardization or validation. The objective is to evaluate the feasibility and effectiveness of using artificial intelligence (AI) software to identify algorithms used in CODefs for population identification in RWE research.

METHODS: We developed a search strategy to identify algorithms for Lung Cancer (LC) in the published literature and executed a ‘living’ search in an AI-driven software platform that utilizes natural language processing and machine learning algorithms to analyze literature from PubMed. The articles were screened for relevance to LC and for the presence of CODef-related terminology or validation statistics. A customized tagging hierarchy was set up to identify therapeutic and coding definitions (e.g., ICD-10-CM, CPT, SNOMED), of LC concepts reported in the literature. AI tagging recommendations were run on the articles the recommendations were highlighted in the text for the reviewer tagging the articles.

RESULTS: The AI-supported search returned 240 studies for screening, of which 94 were excluded for having a <.1 probability of inclusion by the AI model. Twenty-three studies were included and underwent full-text tagging with AI-driven smart tagging recommendations reviewed and applied by team members. The tagging process yielded 31 algorithms for identifying patients with LC which included three algorithms for distinguishing small cell LC and 10 for identifying non-small cell LC within the data sets with varying algorithmic accuracy. The software allowed algorithms to be downloaded to an excel sheet so CODef performance could be compared and referenced for future RWE research.

CONCLUSIONS: AI-assisted identification of algorithms for CODefs is feasible and faster than reviewing articles manually. This approach has the potential to accelerate research timelines and improve reproducibility, as coding methods continue to evolve.