Automated Python Algorithm Analysis of Benign Pulmonary Nodules Discharged after 2 years Surveillance

Introduction: Clinical data is important to inform nodule malignancy risk as exemplified by Brock and Herder scores. Digitization of routine ‘e-noting’ could improve this. Aim: To present a clinical application of digitization of routine ‘e-noting’ to analyse traits of nodules identified by automated data extraction. Methods: A Python algorithm was used to curate and analyse a database of pulmonary nodules with completion of two years surveillance. CT scan and MDM reports were searched for nodule traits e.g. Intrapulmonary lymph node (ILN), stability, and MDM/CT service burden. Results: 85 patients had data extracted from 213 MDM discussions. Ex-smoking documentation was found in 54 patients, 16 had asbestos exposure, 2 had recorded tuberculosis history. Coronary calcification was recorded in 16 cases (19%) of which 12 were positive and pleural effusion in 8 scans (10%). Nodule traits discussed in the MDM included well-defined (19%), irregular borders (15%) or calcification (20%). 63 patients had data from 110 CT Thorax reports (18 patients excluded if reports not available in MDM database). A thoracic radiologist reported 87 (79%) scans. In 43 (68%) patients, all scans were reported by a thoracic radiologist - these had fewer CT scans (1.6 vs 2.3/patient) and fewer MDM’s prior to discharge (2.6 vs. 3.3/patient). Presence of ILN in CT or MDM was similar, whether all scans were reported by a thoracic radiologist or not (23% vs 25%, OR 0.91). Conclusion: The algorithm could identify key clinical traits and interval timings, allowing curation of data for clinical risk prediction. This would be a valuable adjunct to clinical service (decision or admin support) and nodulomics research.