Document resource
Introduction Interstitial lung disease (ILD) is a leading cause of death in systemic sclerosis (SSc), especially when extensive ILD is identified through visual scoring of computed tomography (CT). Automated post-processing analysis software can compute ILD extent through lung texture analysis (LTA) of CT images. However, a definition of ILD presence and extensive disease using this technology are lacking. We aimed to identify and validate optimal thresholds of ILD extent quantified through LTA to detect the presence of SSc-ILD and of extensive involvement.Material and Methods SSc patients visiting the Rheumatology Departments of University Hospital Zurich and the Careggi University Hospital Florence were included. Technically suitable images were analyzed through LTA™ (Imbio), quantifying the percentage of lung volume occupied by ILD, as sum of ground glass, reticulation and honeycombing. Two blinded radiologists independently reviewed the CT scans to identify ILD and extensive disease, the latter defined as >20% of lung parenchyma involved, with disagreements resolved by consensus or third reviewer. Patients were randomly split 2:1 into derivation and validation cohorts. Receiver operating characteristic (ROC) curves with area under the curve (AUC) were computed to identify the optimal ILD extent threshold for detecting ILD and extensive disease, using the visual evaluation as the reference standard. Cox regression analysis was performed to determine the impact of ILD and extensive disease by LTA™ on mortality, adjusted for confounders (age, sex, diffuse skin subset, pulmonary hypertension).Results A total of 664/1118 (58%) SSc patients were eligible for the study. Visual analysis identified ILD in 313 (47%) cases, of whom 103 (33%) extensive ILD. In the derivation cohort (433 patients, 206 ILD, 38% extensive), ROC analysis identified the optimal ILD extent threshold for ILD detection at 1% (AUC 0.83), and at 7% for extensive ILD (AUC 0.84). In the validation cohort (231 patients, 104 ILD, 25% extensive), these thresholds achieved 78% sensitivity/71% specificity for ILD presence, and 81% sensitivity/70% specificity for extensive ILD. Over a median 5 years follow-up, 84 (13%) patients died (37 in the ILD group). Adjusted for confounders, both ILD presence and extensive ILD independently predicted mortality, with comparable results by visual and LTA™ assessments.Conclusions Cutoffs for total ILD by LTA™ to detect ILD presence and identify extensive cases were derived and validated, both predicting a negative prognosis, comparably to visual evaluation. Our results lay the foundation for expanding the use of post-processing analysis in SSc-ILD towards automated diagnosis and stratification.