Abstract Rationale Interstitial lung abnormalities (ILA) and interstitial lung disease (ILD) are progressive conditions, with over half of ILAs advancing within five years and up to 10% annually evolving into clinical ILD. Despite evidence that early detection and antifibrotic therapy alters outcomes, guideline adoption/adherence in pulmonary medicine is suboptimal. The recent ATS ILA statement calls for structured surveillance with annual pulmonary evaluation and pulmonary function testing (PFTs), steps frequently missed. Pulmonary hypertension in ILD (PH-ILD) further worsens prognosis but remains under-recognized despite available treatments. To assess gaps and generate solutions, we applied natural language processing (NLP) to radiology reports to identify unrecognized patients, quantify surveillance lapses, and inform future interventions. Methods All CT reports from 6/1-8/31/2023, at Lahey and Beverly Hospitals (N = 36,856) were screened with the mPower™ NLP platform (Microsoft-Redmond,WA) for ILA-related terms. Flagged CTs were read by a pulmonologist to classify ILA (Fleischner,2020) and UIP pattern (ATS/ERS,2018). 412 total patients were identified; after excluding 75 (66 deceased, 5 stage IV cancer, 3 hospice, 1 other), 337 remained. Group 1 included patients not recently followed by pulmonary (n = 140); Group 2 included those actively followed (n = 197). Demographic, clinical, and diagnostic data—including follow-up, PFTs, echocardiography, and right heart catheterization (RHC)—were abstracted from EMR. Pulmonary artery (PA) and aortic diameters were measured on most recent CT, and PA/A ratio was calculated as described by Wells et al. (2012). Analyses were descriptive; censored 6/12/2025. Results (see Table): The mean age was 73.6 years; 51% male. 181 patients (53.7%) were overdue for a pulmonary visit (365 days): 111 (79.3%) in Group 1; 70 (35.5%) in Group 2. PFTs were overdue in 226 patients (67.1%): 121 (86.4%) in Group 1; 115 (58.4%) in Group 2. Transthoracic echocardiography was performed in 199 patients (59%), with 46 (14%) showing tricuspid regurgitant jet velocity 2.8 m/s: 14 (10.0%) in Group 1; 32 (16.2%) in Group 2. CT PA measurements were available in 323 patients (95.9%); 121 (35.9%) had PA 32 mm: 41 (29.3%) in Group 1; 80 (40.6%) in Group 2. PA/A 1 was present in 79 patients (24.5%): 35 (25.0%) in Group 1; 44 (22.3%) in Group 2. RHC was performed in 19 patients (5.6%): 2 (1.4%) in Group 1 and 17 (8.6%) in Group 2. Conclusion NLP-based mining of CT reports identified a large, underrecognized ILA/ILD population with significant lapses in guideline-based surveillance and highlighted an opportunity to systematically evaluate for PH-ILD using echocardiography, CT, and hemodynamic data. This abstract is funded by: B4 Symptoms
Patel et al. (Fri,) studied this question.