Association of Artificial Intelligence-Screened Interstitial Lung Disease with Radiation Pneumonitis and Mortality in Locally Advanced Non-Small Cell Lung Cancer.
Int J Radiat Oncol Biol Phys 2023;
117:e4-e5. [PMID:
37785334 DOI:
10.1016/j.ijrobp.2023.06.656]
[Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/04/2023]
Abstract
PURPOSE/OBJECTIVE(S)
Radiation pneumonitis (RP) is a common and dose-limiting toxicity following radiotherapy for non-small cell lung cancer (NSCLC). Patients with interstitial lung disease (ILD) are believed to be at increased risk of developing complications including RP, ILD progression, or death. An automated method to identify patients prior to radiotherapy at high risk of developing toxicities or death may allow clinicians to mitigate risk through informed treatment planning and careful patient monitoring.
MATERIALS/METHODS
All locally advanced NSCLC patients treated with definitive radiation from 2006-2021 with a minimum 1 year of follow-up were assessed. RP and mortality data were prospectively collected and retrospectively reviewed. A convolutional neural network (CNN) was previously developed and validated to identify patients with radiographic ILD using planning computed tomography (CT) images, with an accuracy of 0.82. Planning CT scans for the retrospective cohort were used as input to the CNN, with artificial intelligence-screened ILD (AI-ILD) score as an output. AI-ILD scores above our established threshold were labeled as AI-ILD+. The association between AI-ILD score, AI-ILD+/-, mean lung dose (MLD), and the primary outcome of grade ≥2 (G2+) RP or mortality, as well as the secondary outcomes of G2+ RP and mortality were assessed using Wilcoxon rank sum test, univariate and multivariable logistic regression, and Kaplan-Meier survival analysis.
RESULTS
Of 799 patients reviewed, 745 eligible patients were included in the analysis; grade 0-5 RP was reported in 51.3%, 27.1%, 16.9%, 4.0%, 0.1%, and 0.5% of patients respectively. Overall, 22.9% of patients were AI-ILD+, and therefore at high risk (>20% chance) of having true ILD. On UVA, AI-ILD score, AI-ILD+ and MLD were significantly associated with the primary outcome of G2+ RP or mortality, as well as the secondary outcome of mortality. However, only MLD was significantly associated with the secondary outcome of G2+ RP. On MVA, both AI-ILD+ (OR 1.42, 95% CI 1.02-1.97, p = 0.04) and MLD (OR 1.13, 95% 1.05-1.21, p = 0.008) were significantly associated with G2+ RP or mortality. On Kaplan-Meier analysis, the median toxicity-free survival (TFS) time for AI-ILD+ and AI-ILD- patients were 1.7 and 3.4 years respectively, with a 2-year TFS of 48.3% vs. 59.3% (log-rank test: p = 0.02). There was no significant difference in rates of G2+ RP.
CONCLUSION
The AI-ILD algorithm can detect high risk patients with significantly decreased TFS following definitive treatment for NSCLC. AI-ILD classification was not associated with a significant difference in rates of RP when accounting for MLD. Future work will focus on improving the classification algorithm, expert radiologist validation of this dataset, and exploring reasons for the mortality difference in AI-ILD+ patients.
Collapse