COMBINING RADIOLOGY AND ENVISIA, A MOLECULAR CLASSIFIER, TO IMPROVE USUAL INTERSTITIAL PNEUMONIA (UIP) DIAGNOSISView Session Add to Schedule
Jonathan Chung, Daniel Pankratz, Steven Nathan, Lori Lofaro, Gerard Criner, Fernando Martinez, Kevin Flaherty, David Lynch, Thomas Colby, Jeffrey Myers, Steve Groshong, Mark Steele, Giulia Kennedy, Kevin Brown, Ganesh Raghu, Yoonha Choi, Hannah Neville, Jing Huang, David Lederer, Brandon Larsen, Karel Calero, Lars Hagmeyer, Amy Case, Umair Gauhar, Navdeep Rai, Murali Ramaswamy, John Davis, Neil Barth, Patric WalshPresenting Author(s)
Idiopathic pulmonary fibrosis (IPF) diagnosis guidelines recommend high-resolution computed tomography (HRCT) evaluation. Pathology evaluation for UIP is recommended when HRCT is not definitive. The Envisia Genomic Classifier (EGC) detects molecular UIP in transbronchial biopsy (TBB) samples using a 190-gene machine learning classifier. In prospective validation against reference pathology (UIP vs. non-UIP) in 49 subjects, EGC demonstrated 88% specificity and 70% sensitivity. We report UIP diagnostic performance of radiology in combination with the EGC.
46 of 49 subjects in the validation had available HRCT scans. We defined UIP diagnoses as either 1) HRCT-UIP by central radiology (definite UIP, probable UIP, or possible UIP by Fleischner Society guidelines) or 2) EGC-UIP. We defined non-UIP diagnosis as radiology inconsistent with UIP and a non-UIP EGC result. We evaluated HRCT-UIP plus EGC-UIP against reference standard histopathology.
Central radiology identified 9 of 46 subjects (20%) as HRCT-UIP and 37 of 46 as inconsistent with UIP (PPV of 100% [CI: 66-100%] and NPV of 68% [CI: 50-82%]). The combination of HRCT-UIP plus EGC achieved joint sensitivity of 81% [CI:58-95%] and specificity of 88% [CI:69-97%]. This joint approach identified 20 UIP subjects (43%, PPV of 85% [CI: 62-97%]) and 26 non-UIP subjects (57%, NPV of 85% [CI: 65-96%]).Local radiology identified 21 of 46 subjects (46%) as HRCT-UIP and 25 of 46 as inconsistent with UIP (PPV of 67% [CI: 43-85%] and NPV of 68% [CI: 46-85%]). The combination of local HRCT-UIP plus EGC achieved joint sensitivity of 95% [CI: 77-100%] and specificity of 71% [CI:49-87%]. This joint approach identified 28 UIP subjects (61%, PPV of 75% [CI: 55-89%]) and 18 non-UIP subjects (39%, NPV of 94% [CI: 73-100%]).
Central HRCT-UIP plus EGC-UIP identifies more UIP than radiology alone (20 vs. 9 subjects, p=0.001), improving NPV from 68% to 85% (p=0.006), with a modest decrease in PPV for UIP (100% vs. 85%, p=0.06).
In the local setting, this joint approach also identifies more UIP than radiology alone (28 vs. 21, p= 0.008) with moderately improved PPV (67% to 75%, p=0.03). The improvement in NPV is more substantial (68% to 94%, p=0.002).
HRCT-UIP plus EGC-UIP improves the diagnostic yield for UIP over radiology alone, potentially avoiding the need for pathology diagnosis in patients with suspected IPF.