‘E-Noting Nodulomics’: Automating Electronic Clinical Data Mining and Analysis of Pulmonary Nodules Using a Python Algorithm

Abstract

Introduction: Increasing digitalization of clinical care provides an opportunity to improve imaging stratification of pulmonary nodule malignancy risk, data extraction and analysis, with many potential clinical applications. Aim: To present a Python algorithm capable of automating curation and analysis of nodule traits from e-noting records. Methodology and algorithm validation steps will be described. Methods: A software algorithm previously developed in Python that extracts pre-specified keywords was refined to collect audit data from a nodule database to identify traits characteristic of such nodules. Results: The algorithm converted unstructured CT and Multi-Disciplinary team Meeting (MDM) reporting data into spreadsheet format, which included medical history, radiological features, CT dates and reporting data, MDM frequency and discussion outcomes. Keywords were pre-defined to identify nodule traits (e.g. stability or intrapulmonary lymph node) or management plans (e.g. discharge) compared with MDM timing or thoracic specialisation of the reporting radiologist. Conclusion: Such a tool is well suited to MDM decision support to free up admin time (e.g. time spent searching for nodule details to calculate risk scores or scan dates to schedule follow-up scan intervals) and improve governance (e.g. automated reports looking for accuracy or discrepancies in nodule reporting and preventing missed data when searching depended on manual searches prone to human error). Further interrogation by machine learning could give rise to novel informatics biomarkers and cancer prediction tools which will be crucial to evolving image based nodulomics studies.