ORIGINAL ARTICLE

Utility of a Molecular Classifier as a Complement to High-Resolution

Computed Tomography to Identify Usual Interstitial Pneumonia
Luca Richeldi1, Mary Beth Scholand2, David A. Lynch3, Thomas V. Colby4, Jeffrey L. Myers5, Steve D. Groshong6,
Jonathan H. Chung7, Sadia Benzaquen8, Steven D. Nathan9, J. Russell Davis10, Shelley L. Schmidt11,
Lars Hagmeyer12, David Sonetti13, Jurgen Hetzel14, Gerard J. Criner15*, Amy H. Case16, Murali Ramaswamy17,
Karel Calero18, Umair A. Gauhar19, Nina M. Patel20, Lisa Lancaster21, Yoonha Choi22, Daniel G. Pankratz22,
P. Sean Walsh22, Lori R. Lofaro22, Jing Huang22, Sangeeta M. Bhorade22, Giulia C. Kennedy22, Fernando J. Martinez23*,
and Ganesh Raghu24
1
Fondazione Policlinico Universitario A. Gemelli IRCCS, Università Cattolica del Sacro Cuore, Rome, Italy; 2Department of Internal
Medicine, University of Utah, Salt Lake City, Utah; 3Department of Radiology and 6Division of Pathology, National Jewish Health, Denver,
Colorado; 4Department of Laboratory Medicine and Pathology, Mayo Clinic Arizona, Phoenix, Arizona; 5Department of Pathology,
University of Michigan, Ann Arbor, Michigan; 7Department of Radiology, University of Chicago, Chicago, Illinois; 8Department
of Medicine, Einstein Medical Center, Philadelphia, Pennsylvania; 9Inova Heart and Vascular Institute, Inova Fairfax Hospital, Annandale,
Virginia; 10Department of Surgery, St. Luke’s Hospital, Chesterfield, Missouri; 11Department of Medical Specialties, Spectrum Health
Medical Group, Grand Rapids, Michigan; 12Hospital Bethanien Solingen, Institute of Pneumology, University of Cologne, Solingen,
Germany; 13Department of Medicine, University of Wisconsin, Madison, Wisconsin; 14Department of Medicine, University Hospital
Tuebingen, Tuebingen, Germany; 15Department of Thoracic Medicine and Surgery, Temple University, Philadelphia, Pennsylvania;
16
Pulmonary, Critical Care, and Sleep Medicine, Piedmont Healthcare, Atlanta, Georgia; 17Department of Medicine, Pulmonix LLC,
Greensboro, North Carolina; 18Department of Medicine, University of South Florida, Tampa, Florida; 19Department of Medicine,
University of Louisville, Louisville, Kentucky; 20Department of Medicine, College of Physicians and Surgeons, Columbia University,
New York, New York; 21Department of Medicine, Vanderbilt University Medical Center, Vanderbilt University, Nashville, Tennessee;
22
Veracyte, Inc., San Francisco, California; 23Department of Medicine, Weill Cornell Medical College, New York, New York;
and 24Department of Medicine, University of Washington, Seattle, Washington

Abstract Measurements and Main Results: In 96 patients, the Envisia

Rationale: Usual interstitial pneumonia (UIP) is the defining
morphology of idiopathic pulmonary fibrosis (IPF). Guidelines for
IPF diagnosis conditionally recommend surgical lung biopsy for
histopathology diagnosis of UIP when radiology and clinical context
are not definitive. A “molecular diagnosis of UIP” in transbronchial
lung biopsy, the Envisia Genomic Classifier, accurately predicted
histopathologic UIP.
Objectives: We evaluated the combined accuracy of the Envisia
Genomic Classifier and local radiology in the detection of UIP pattern.
Methods: Ninety-six patients who had diagnostic lung pathology as
well as a transbronchial lung biopsy for molecular testing with Envisia
Genomic Classifier were included in this analysis. The classifier
results were scored against reference pathology. UIP identified
on high-resolution computed tomography (HRCT) as documented
by features in local radiologists’ reports was compared with
histopathology.
Classifier achieved a specificity of 92.1% (confidence interval
[CI],78.6–98.3%) and a sensitivity of 60.3% (CI, 46.6–73.0%) for
histology-proven UIP pattern. Local radiologists identified UIP in 18
of 53 patients with UIP histopathology, with a sensitivity of 34.0%
(CI, 21.5–48.3%) and a specificity of 96.9% (CI, 83.8–100%). In
conjunction with HRCT patterns of UIP, the Envisia Classifier results
identified 24 additional patients with UIP (sensitivity 79.2%;
specificity 90.6%).
Conclusions: In 96 patients with suspected interstitial lung disease,
the Envisia Genomic Classifier identified UIP regardless of HRCT
pattern. These results suggest that recognition of a UIP pattern by the
Envisia Genomic Classifier combined with HRCT and clinical factors
in a multidisciplinary discussion may assist clinicians in making an
interstitial lung disease (especially IPF) diagnosis without the need
for a surgical lung biopsy.
Keywords: usual interstitial pneumonia; idiopathic pulmonary
fibrosis; molecular classifier

( Received in original form March 31, 2020; accepted in final form July 28, 2020 )
Correspondence and requests for reprints should be addressed to Ganesh Raghu, M.D., Center for Interstitial Lung Diseases, Department of Medicine and
Laboratory Medicine, University of Washington Medical Center, 1959 NE Pacific Ave, Seattle WA 98195. E-mail: graghu@uw.edu.
This article has a related editorial.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Am J Respir Crit Care Med Vol 203, Iss 2, pp 211–220, Jan 15, 2021
Copyright © 2021 by the American Thoracic Society
Originally Published in Press as DOI: 10.1164/rccm.202003-0877OC on July 28, 2020
Internet address: www.atsjournals.org

Richeldi, Scholand, Lynch, et al.: A Molecular Classifier to Identify UIP 211

 ORIGINAL ARTICLE

At a Glance Commentary
Scientific Knowledge on the
Subject: An early, accurate diagnosis
of interstitial lung disease (ILD) is
important to identify appropriate
therapeutic options, predict the risk
of progression, and assess overall
prognosis. Clinical guidelines
for the diagnosis of idiopathic
pulmonary fibrosis (IPF) conditionally
recommend surgical lung biopsy (SLB)
for histopathologic diagnosis in
patients whose clinical and radiological
context is not definitive for IPF.
However, SLB is associated with
increased morbidity and mortality.
Previously, a biomarker that identified
a molecular usual interstitial
pneumonia (UIP) signature in
transbronchial biopsy samples with
high specificity and adequate
sensitivity was shown to supplement
the clinical evaluation for ILD
diagnosis without the need for SLB.
What This Study Adds to the Field:
This study redemonstrates that the
molecular UIP pattern determined
from transbronchial biopsies using the
Envisia Genomic Classifier has high
reproducibility and sustained accuracy
for the detection of histopathology
features of UIP. Overall, these results
suggest that the recognition of a UIP
pattern by the Envisia Classifier may be
useful as a surrogate histopathology in
combination with high-resolution
computed tomography image
patterns and clinical factors in a
multidisciplinary team discussion to
assist clinicians in making a diagnosis
of ILD, especially IPF.
Interstitial lung diseases (ILDs) are clinically
complex, with various causes, treatment
responses, and clinical outcomes (1).
Chronic ILDs are often associated with
progressive scarring, loss of lung
architecture and function, and poor
prognosis. An early accurate ILD diagnosis
is important to identify appropriate
therapeutic options, predict risk of
progression, and assess overall prognosis
(2).
The hallmark of idiopathic pulmonary
fibrosis (IPF) is the usual interstitial
pneumonia (UIP) pattern of lung fibrosis
identified by radiology and/or
histopathology (3). IPF is a chronic and
progressive ILD of poor outcome and few
effective treatment options (2). Clinical
guidelines for the diagnosis and treatment
of IPF conditionally recommend surgical
lung biopsy (SLB) for histopathologic
diagnosis in patients whose clinical and
radiological context is not definitive for IPF
(4). However, SLBs are associated with
substantial morbidity and mortality (5),
with significant interreader variability of
histopathology for UIP pattern potentially
impacting the ability to confidently identify
UIP (6).
As a result, accurately identifying ILD
diagnosis remains challenging in the current
era of diagnostic testing. In an effort to
address this issue, an international working
group proposed a framework for the
diagnostic classification of ILD based on
clinical and radiological features. This
concept of a working diagnosis of IPF (7–9)
is emerging as a potential option for
informing treatment decisions. In a recent
case-cohort study, a working diagnosis of
IPF was found to lead to an antifibrotic
treatment decision in patients with high-
confidence diagnoses of IPF. However, in
patients with a low confidence (,70%
likelihood) of IPF, physicians were more
likely to recommend an SLB, especially in
those patients at low risk for SLB (10). This
study highlighted the importance of the
diagnostic likelihood of IPF on physicians’
decisions for SLB. Similarly, in patients
with progressive fibrosing ILD regardless
of initial ILD diagnosis and treatment,
nintedanib, an antifibrotic treatment
decreased the rate of FVC decline in
patients with progressive fibrosis defined by
a high-resolution computed tomography
(HRCT) image or evidence of progressive
decline in lung function. Although
approximately 28.5% of these patients had
an SLB and 15% had transbronchial lung
biopsy (TBBx) to make their initial
diagnosis, subsequent behavior of their ILD
deprioritized the need for high diagnostic
confidence with a subsequent SLB (9).
Recently, a biomarker that
distinguishes UIP from non-UIP, the
Envisia Genomic Classifier, showed benefit
in assisting an accurate, confident diagnosis
of IPF in subjects suspected to have IPF
when compared with diagnoses for the same
subjects achieved with the benefit of
pathology results (11–13). The Envisia
Classifier detects molecular UIP in TBBx
samples using a 190-gene machine-learning
classifier. In prospective validation against
reference pathology (UIP vs. non-UIP) in
49 patients, the classifier demonstrated 88%
specificity and 70% sensitivity. Training
and previous validation of the classifier
(Veracyte) have been previously described
(11–13).
We undertook an independent
prospective study in a new cohort of patients
to further validate the genomic classifier. We
demonstrate increased diagnostic yield,
improved sensitivity, and negative
predictive value for UIP when the classifier
is used to complement current HRCT

*G.J.C. is Associate Editor and F.J.M. is Deputy Editor of AJRCCM. Their participation complies with American Thoracic Society requirements for recusal from
review and decisions for authored works.
Supported by Veracyte, Inc. (BRAVE study and development of the Envisia Genomic Classifier).
Author Contributions: L.R., M.B.S., D.A.L., T.V.C., J.L.M., S.D.G., J.H.C., S.B., S.D.N., J.R.D., S.L.S., L.H., D.S., J. Hetzel, G.J.C., A.H.C., M.R., K.C.,
U.A.G., N.M.P., L.L., Y.C., D.G.P., P.S.W., L.R.L., J. Huang, S.M.B., G.C.K., F.J.M., and G.R. participated in the design of the study. L.R., M.B.S., Y.C.,
D.G.P., P.S.W., L.R.L., J. Huang, S.M.B., G.C.K., and G.R. guided the clinical analysis and interpretation of the data. D.A.L. and J.H.C. performed the central
radiology review. T.V.C., J.L.M., and S.D.G. performed the central pathology review. S.D.N., J.R.D., S.L.S., L.H., D.S., J. Hetzel, G.J.C., A.H.C., M.R., K.C.,
S.M.B., and U.A.G. enrolled study participants and collected biopsy specimens. G.C.K., Y.C., D.G.P., P.S.W., L.R.L., and J. Huang designed laboratory
experiments, developed the molecular classification algorithm, and performed the classifier analysis. L.R.L. supervised data collection for the BRAVE study.
L.R., M.B.S., N.M.P., L.L., Y.C., D.G.P., P.S.W., L.R.L., J. Huang, S.M.B., G.C.K., and G.R. interpreted the data analysis, drafted the manuscript, created the
figures and tables, and prepared, revised, and submitted the manuscript. All study authors had access to the study data and reviewed this manuscript. The
corresponding author reviewed and approved the submission of the final manuscript.

212 American Journal of Respiratory and Critical Care Medicine Volume 203 Number 2 | January 15 2021
ORIGINAL ARTICLE
criteria for UIP pattern (4, 8). Some of the
results of these studies have been previously
reported in the form of abstracts (14, 15).
Methods
Study Participants
A total of 447 patients were enrolled in
the BRAVE (Bronchial Sample Collection
for a Novel Genomic Test) study,
which is a comprehensive prospective
sample collection protocol to develop a
biorepository of biological samples with
highly annotated clinical information.
Unique BRAVE patients (n = 220) were
allocated to this second independent
clinical validation cohort of this locked
classifier (Figure 1). These 220 patients
were recruited from 30 BRAVE clinical sites
between 2013 and 2019. Of these 220
patients, 81 patients were excluded because
of prespecified criteria for proper sample
collection, shipment, and storage, as shown
in Figure 1. Of the remaining 139 patients
who underwent molecular testing with the
classifier, 43 did not have diagnostic
pathology (i.e., they lacked a reference truth
label). Ten did not have slides available for
review, and the remaining 33 were returned
a nondiagnostic label, most commonly
because of insufficient tissue or no
pathologic abnormalities identified.
Comparative descriptions between the
excluded nondiagnostic cohort and this
independent validation cohort are
presented in the Results section. Ninety-six
patients who had diagnostic pathology
(reference truth labels) were included in
this new validation cohort (Figure 1). The
BRAVE study was reviewed and approved
by the Western Institutional Review Board
(20130158, 20141155, and 20130705) and
site-specific institutional review boards.
Written informed consent was obtained
from all BRAVE study participants.
The Envisia Genomic Classifier:
Development and Initial Validation
Training and previous validation of the
classifier (Veracyte) have been previously
described (11–13). Briefly, the Envisia
Classifier is a diagnostic test trained to
accurately identify molecular UIP in
TBBxs, predicting conventional diagnostic
histopathology UIP features interpreted
by experienced pathologists (11–13).
The test provides a binary result of UIP or
non-UIP (11).
Richeldi, Scholand, Lynch, et al.: A Molecular Classifier to Identify UIP
BRAVE Patients (n=447)
Training and Clinical Validation
(n=227)
Allocated to Independent Clinical
Validation (n=220*)
Reasons for Exclusions (n=81)
Missing TBB samples (n=25)
1–2 TBBs (secondary endpoint) (n=34)
Shipping/handling exclusions (n=18)
Other exclusions (n=4)
Patients with Classifier Results
(n=139)
Patients Without Diagnostic Pathology
(n=43)
Missing pathology slides (n=10)
Patients with nondiagnostic pathology
(n=33)
Patients Scored in Independent
Clinical Validation (n=96)
Figure 1. CONSORT (Consolidated Standards of Reporting Trials) diagram illustrating the derivation
of the patients who were allocated to the independent clinical validation (220 patients) of the Envisia
Genomic Classifier from 447 BRAVE (Bronchial Sample Collection for a Novel Genomic Test) patients.
All technical exclusions were prespecified and applied prospectively before scoring in validation.
*These 220 patients are independent of those used in the classifier development and the prior
validation. TBB = transbronchial biopsy.
The test was developed using biopsies patients manifesting ILD. Histopathology
obtained for histopathology as truth labels diagnoses of UIP as a reference truth label
and TBBx for molecular testing (more for this study were fulfilled when pathology
specifically, the RNA sequencing derived from met either UIP or probable UIP criteria
TBBx samples as an input for the classifier) in accordance with the 2018 American
from patients enrolled in the prospective Thoracic Society guidelines for IPF
noninterventional BRAVE study (11). diagnosis (4). The terms “favor UIP”
and “difficult UIP,” as described in the
initial validation of the Envisia Classifier
Pathological and Radiological Review
Pathology slides were prepared from performance (11) and which were used in
lung biopsies by participating study-site this study, were determined by a consensus
personnel and centrally reviewed by a team of the expert review pathologists to be in
of expert pathologists without access to the category of probable UIP. All other
clinical information other than their diagnostic histopathology was categorized
knowledge that the lung biopsies were as non-UIP. These histopathological labels
obtained for histopathology diagnosis in were used to derive UIP or non-UIP
213
 ORIGINAL ARTICLE

Table 1. Clinical Demographics of the Envisia Genomic Classifier Validation Group (Synopsis of Veracyte Design History File
[DHF]: 007-067P: BRAVE Local Radiology
Demographics Clinical Validation (N = 96)
Report Review [CV-3]). These original
HRCT reports were systematically
reviewed by two blinded reviewers (S.M.B.,
Sex, n (%)
F 41 (43) a United States–based board-certified
M 55 (57) pulmonary/critical care physician, and
Age, yr, mean (SD) 62.8 (12.1) L.R.L., a master’s level clinical research
Smoker, n (%) professional) to ensure the radiologic
Yes 48 (50)
No 48 (50)
features described could be expressed within
Study-site type, n (%) the context of the diagnostic HRCT criteria
U.S. academic 41 (43) defined by the Fleischner Society Guidelines
U.S. community 48 (50) (typical UIP, probable UIP, indeterminate
European academic 7 (7) for UIP, and features most consistent with
Biopsy type, n (%)
Surgical 61 (64) non-IPF diagnosis) (8).
TBBx 1 (1) Both reviewers were blinded to the
Cryobiopsy 34 (35) clinical information of the patient, the
UIP frequency in study, n (%) results of the Envisia Classifier, and the
By pathology 58 (60)
By radiology 10/65 (15)
histopathological diagnosis associated with
each HRCT scan report. Each reviewer
Definition of abbreviations: TBBx = transbronchial biopsy; UIP = usual interstitial pneumonia. interpreted the findings in the body of the
local report and the final interpretation
reference standard truth labels for each 1 to 10 scale, were determined. The extent according to the Fleischner Society
subject, as previously described (11). All of lung fibrosis was estimated as .10% of criteria (8). These two interpretations for
laboratory and data analysis personnel lung volume in the absence of consolidation each HRCT report were independently
remained blinded to the identity and or progressive massive fibrosis. Local documented and subsequently compared. If
reference standard results for each patient radiology readings were performed by the there was discordance between the two
until after the scoring of the validation by a local radiologist review according to their reviews, the two reviewers were allowed to
third party was complete. own practice style without prespecified confer until an agreement was reached.
For patients with HRCT scans available study guidelines. If there was persistent disagreement
for review, the HRCT scans were reviewed between the two reviewers regarding the
both by the local radiologist and by central Methodology for Review of Local interpretation of the HRCT report,
radiologists consisting of two expert HRCT Scan Reports a third reviewer would have served
radiologists. Central radiology readings HRCT scan reports by site-specific local as a tie breaker. There were no cases
were reviewed according to Fleischner radiologists were available for review for 85 of that required interpretation by a third
Society criteria for UIP (typical UIP, the 96 patients. To maintain the integrity of reviewer.
probable UIP, and indeterminate for UIP) the interpretation of the initial HRCT report,
(8) and specific non-IPF ILD diagnoses. a prospectively designed protocol for HRCT
Primary, secondary, and tertiary differential review was implemented. A synopsis of this Biological Samples, Processing, and
diagnoses, with associated confidences on a protocol is included in the online supplement Analysis
For molecular testing, total RNA was
Table 2. Histopathological Patterns Represented in the Envisia Genomic Classifier extracted from three to five TBBx per patient
Validation Group and pooled by subject, and a single whole-
transcriptome library was generated and
sequenced as previously described (11). The
Clinical Validation (N = 96) locked Envisia Classifier, whose training
and validation has been previously
UIP, n (%) described, was used to make a molecular
UIP 32 (33) UIP designation. The results are presented
Probable UIP 26 (27)
Other patterns, n (%) as a binary result of UIP or non-UIP, as
RB, SRIF 3 (3) previously described (11).
HP, favor HP 9 (9)
Sarcoidosis 4 (4)
NSIP, cellular NSIP, favor NSIP 9 (9) Statistical Analysis
Bronchiolitis 8 (8) Descriptive statistics are reported for clinical
Organizing pneumonia 1 (1) demographic data of the final validation
Other 4 (4) cohort. The primary study endpoints of test
Definition of abbreviations: HP = hypersensitivity pneumonitis; NSIP = nonspecific interstitial sensitivity and specificity were prespecified in a
pneumonia; RB = respiratory bronchiolitis; SRIF = smoking-related interstitial fibrosis; UIP = usual clinical validation study protocol. Negative and
interstitial pneumonia. positive predictive values are computed at the

214 American Journal of Respiratory and Critical Care Medicine Volume 203 Number 2 | January 15 2021
ORIGINAL ARTICLE

UIP prevalence of the validation set. All
confidence intervals (CIs) are two-sided, 95%
intervals unless otherwise noted. Classifier
accuracy is reported as the area under the
receiver operating characteristic curve. The
significance of difference for subgroup
proportions and performance was tested with
the x2 test. Statistical analyses were performed
in R (version 3.2.3; www.r-project.org).
Results
Study Participants
Clinical demographic data of this Envisia
Classifier validation cohort are shown in
Table 1. Overall, there was an even
representation of smokers and nonsmokers
as well as clinical and academic study sites
(Table 1). In addition to a bronchoscopy to
obtain TBBx for molecular testing, 64% of
the patients underwent SLB, and 35%
underwent cryobiopsy to obtain a final
histopathologic diagnosis. Only one patient
of 60.3% (CI, 46.6–73.0%) and a specificity
of 92.1% (CI, 78.6–98.3%) for histology-
proven UIP pattern (Figure 2). With
the study histopathologic UIP prevalence
of 60.4%, the positive predictive value of
an Envisia Classifier UIP result is 92.1%
(CI, 78.6–98.3%), and the negative
predictive value of an Envisia Classifier
non-UIP result is 60.3% (CI, 46.6–73.0%).
In a subset of patients with SLB biopsies
alone, the classifier performance was
similar (see Table E1 in the online
supplement).
A B
ROC curve
1.0
0.8
Sensitivity
Among the different UIP patterns
reported by the central pathologists, a
definite UIP pattern was detected in 55% of
all UIP cases. Of note, probable UIP
pathology was more common when the
Envisia Classifier result showed non-UIP
(false negatives [FNs]) compared with
definite UIP pathology (15 of 26 FNs among
probable UIP vs. 8 of 32 FNs among definite
UIP; P = 0.02) (Figure 3). When the
classifier missed UIP (FN), there were more
patients with discordant UIP among
different biopsied lobes compared with
1
0.8
NPV
PPV

NPV/PPV
0.6 * 0.6
whose histopathology diagnosis from
transbronchial biopsy was considered 0.4 0.4
diagnostic was not subjected to the need
for obtaining additional lung tissue by 0.2 0.2
cryobiopsy or SLB. The overall UIP
prevalence (definite UIP and probable UIP) 0.0
AUC = 0.86
0
based on histopathology was 60.4% in this
cohort. The Envisia Classifier validation 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
group contains a representative spectrum of 1-Specificity Prevalence
ILD, including different UIP patterns on
histopathology as well as a variety of non- C
UIP ILD diagnoses (Table 2).
Thirty-three patients who underwent Reference label
lung biopsy had a nondiagnostic pathology Non-UIP UIP
label and were excluded from this study. Of
these patients, 16 had insufficient tissue, 14 Envisia Non-UIP 35 23
had no pathologic diagnosis, and 3 had Genomic
UIP 3 35
chronic inflammation that was not specified. Classifier
Several of these patients with insufficient tissue
Sensitivity 60.3% [46.6 – 73.0]
had nonspecific interstitial fibrosis identified.
Patients who had nondiagnostic pathology Specificity 92.1% [78.6 – 98.3]
labels were compared with patients who were
NPV 60.3% [46.6 – 73.0]
included in the Envisia validation cohort. We
found that the nondiagnostic pathology PPV 92.1% [78.6 – 98.3]
cohort had a higher number of women (73%
UIP prevalence 60.4%
vs. 43%; P = 0.003) and were more likely to
undergo either cryobiopsy (79% vs. 35%) or Figure 2. Clinical validation performance of the Envisia Genomic Classifier. (A) Receiver
TBBx (12% vs. 1%) and less likely to undergo operating characteristic curve for the Envisia Genomic Classifier on the 96 patients scored in
SLB (9% vs. 64%; P , 0.001) compared with validation. *Indicates the prospectively defined decision boundary. (B) Prevalence-adjusted
the Envisia validation cohort. negative predictive value (NPV) and positive predictive value (PPV) curves for the Envisia
Genomic Classifier. The NPV and PPV at the study UIP prevalence of 60%, defined by
reference standard histopathology truth, is shown with a dashed line. The curves represent
Validation Performance of the Envisia the NPV and PPV values expected across different patient population prevalences. (C) Envisia
Genomic Classifier Genomic Classifier performance against reference standard histopathology for UIP pattern.
In these 96 patients (Envisia validation AUC = area under the curve; ROC = receiver operating characteristic; UIP = usual interstitial
cohort), the classifier achieved a sensitivity pneumonia.

Richeldi, Scholand, Lynch, et al.: A Molecular Classifier to Identify UIP 215

 ORIGINAL ARTICLE

Classification Score patients with UIP pathology in all biopsied

–2 –1 0 1 2
lung lobes (Table 3). In contrast, there is no
enrichment of classifier errors in cases with
NSIP Sarcoidosis
pathology diagnoses requiring adjudication
RB
Sarcoidosis Other
by a third pathologist compared with
NSIP
Sarcoidosis OP
those with agreement determined by two
RB
Bronchiolitis/Other Sarcoidosis
reviewing pathologists (Table E2). The
Bronchiolitis
Sarcoidosis
Envisia Classifier sensitivity and specificity
NSIP
HP in the validation group is statistically
Other DIP
Other/NSIP Sarcoidosis equivalent between patients who were
HP
Bronchiolitis greater than 65 years old and those who
Probable UIP/NSIP
Bronchiolitis Inflammatory Bronchiolitis were less than 65 years, between smokers
Probable UIP NSIP
Probable UIP and nonsmokers, and between men and
Probable UIP NSIP
HP NSIP women (Table E3).
HP HP
OP Sixty-five of the HRCT scans provided
Other
Probable UIP by study sites were centrally reviewed by
Probable UIP HP
NSIP Inflammatory Bronchiolitis two expert radiologists according to the
Bronchiolitis Other
Sarcoidosis Fleischner Society criteria for UIP (typical
Other Obliterative Bronchiolitis
UIP NSIP UIP, probable UIP, and indeterminate
SRIF HP
UIP HP for UIP) (8) and specific non-IPF ILD
Probable UIP Probable UIP
Sarcoidosis diagnoses (nonspecific interstitial
HP
NSIP pneumonia, hypersensitivity pneumonitis
Probable UIP Probable UIP
Probable UIP [HP], organizing pneumonia, sarcoidosis,
UIP
Probable UIP HP respiratory bronchiolitis, desquamative
HP HP
interstitial pneumonia, emphysema,
Central Pathology Diagnosis

RB

Central Radiology Diagnosis

Other
NSIP
Bronchiolitis Other inflammatory bronchiolitis, obliterative
Bronchiolitis HP
Probable UIP HP bronchiolitis, lymphangioleiomyomatosis,
UIP HP
Probable UIP NSIP Langerhans cell histiocytosis, eosinophilic
UIP/OP HP
NSIP Sarcoidosis pneumonia, and other). These were
HP
Probable UIP/OP included for descriptive analyses in the
Probable UIP/NSIP HP
UIP OP score plot (Figure 3).
Bronchiolitis
UIP/NSIP HP
UIP
Probable UIP Typical UIP Improved UIP Diagnostic Yield with
Probable UIP/HP Typical UIP
HP HP Radiology
Probable UIP
UIP HP Local HRCT scans for 85 of the 96 patients
UIP HP
Probable UIP NSIP were available for review. The mean
Probable UIP NSIP
Probable UIP timeframe between the 85 local HRCT
UIP HP
UIP Indeterminate for UIP scans with reports available and the
UIP HP
UIP HP transbronchial biopsy/diagnostic procedure
UIP HP
UIP HP was 80 days, with a median of 57 days.
Probable UIP
HP
NSIP Among the 53 cases with histopathologically
UIP
UIP
Probable UIP
Other identified UIP by central review, 18 were
UIP
UIP
Probable UIP
HP
identified as definite or probable UIP
UIP
UIP Probable UIP
by local radiology reports, resulting in
UIP
UIP/NSIP
a sensitivity of 34.0% (CI, 21.5–48.3%); for
UIP
Probable UIP
NSIP
NSIP
the remaining 32 histopathologically non-
UIP
UIP
HP
UIP diagnoses, local radiology reports
Probable UIP
UIP
Typical UIP
Probable UIP
identified 31 accordingly as non-UIP,
Other/NSIP
UIP
Sarcoidosis
HP
resulting in a specificity of 96.9% (CI,
UIP
Probable UIP/Other
Indeterminate for UIP
NSIP
83.8–100%) (Table 4). When the Envisia
Probable UIP
UIP
HP
Indeterminate for UIP
Classifier results are used in conjunction
Probable UIP
UIP
Typical UIP with local radiology, 24 additional patients
with biopsy-proven UIP are identified
(sensitivity of 79.2% [CI, 65.9–89.2%];
UIP specificity of 90.6% [CI, 75.0–98.0%])
UIP(with Non-UIP)
Non-UIP
(Table 5). The negative predictive value for
UIP improves from 47% with HRCT alone
Figure 3. The distribution of Envisia Genomic Classifier scores for the 96 patients in the clinical to 72.5% with the combination of HRCT
validation. Patients are sorted from left to right by increasing score on the x-axis, and the reference and the Envisia Classifier, and the positive

216 American Journal of Respiratory and Critical Care Medicine Volume 203 Number 2 | January 15 2021
ORIGINAL ARTICLE

Table 3. Envisia Genomic Classifier Performance Relative to the Complexity of Lung Lobe–Level Histopathology

Lung Lobe–Level Histopathology Discordance FN (n = 23) TP (n = 35) FP (n = 3) TN (n = 35) P Value

UIP UIP in all lung lobes 12 29 — — 0.01

UIP/non-UIP, nondiagnostic 11 6 — —
Non-UIP Non-UIP in all lung lobes — — 2 28 0.59
Non-UIP/nondiagnostic — — 1 7

Definition of abbreviations: FN = false-negative Envisia Genomic Classifier; FP = false-positive Envisia Genomic Classifier; TN = true-negative Envisia
Genomic Classifier; TP = true-positive Envisia Genomic Classifier; UIP = usual interstitial pneumonia.

predictive value remains above 90% when
Envisia Genomic Classifier results are used
in complement with local radiology.
Discussion
In this second independent validation
cohort of patients enrolled in the BRAVE
study, we demonstrated that the molecular
UIP pattern determined from TBBx using
the Envisia Genomic Classifier has high
reproducibility and sustained accuracy for
the detection of histopathology features of
UIP. The results of this second validation
cohort confirm and expand on the initial
validation performance of the classifier by
identifying a UIP pattern with a sensitivity
of 60.3% and a specificity of 92.1% (Figure
E4). The classifier performance was not
impacted by age, sex, or smoking status in
this study. In addition, the Envisia Classifier
enhanced the diagnostic yield and increased
the sensitivity from 34.0% to 79.2% in
detecting UIP when the classifier was used in
complement with current HRCT criteria for
UIP pattern. Overall, these results suggest
that the recognition of a UIP pattern by the
Envisia Genomic Classifier may be useful as
a surrogate histopathology in combination
with HRCT image patterns and clinical
factors in a multidisciplinary team
discussion to assist clinicians in making a
diagnosis of ILD, especially IPF.
An early and accurate diagnosis of ILD
is essential to initiate appropriate initial
therapies as well as to guide further timely
interventions, such as antifibrotic treatment,
lung transplantation, and palliative care.
Importantly, the early diagnosis of ILD may
not only be more effective in decreasing
the progression of disease but may also lead to
the reversal of HP. For example, the early
identification and removal of an environmental
antigen may enable the reversal of interstitial
abnormalities before the development of
fibrotic HP/pulmonary fibrosis, leading to a
significantly improved outcome.
Unfortunately, delays in ILD diagnosis
are common and are particularly common
in IPF, as noted in the recent INTENSITY
(Interstitial Lung Disease Patient Diagnostic
Journey) survey of patients with ILD that
showed the median time to diagnosis was 7
months, with 43% of patients experiencing a
delay in their diagnosis of greater than 1 year
(16). Diagnostic uncertainty of HRCT ILD
patterns may contribute to the delay
of an accurate ILD diagnosis, with the
consequence of inadequate or potentially
harmful therapies, leading to worse
outcomes in this population (17–22).
Our study showed that the Envisia
Genomic Classifier complements radiology
interpretation of the HRCT scan by local
radiologists, significantly increasing the
diagnostic yield of UIP by over 130% in 24
additional patients (18 of 53 to 42 of 53),
and increases the sensitivity and negative
predictive value while maintaining a high
specificity and positive predictive value in
the detection of UIP. These results highlight
the potential utility of the Envisia Genomic
Classifier as a complement to HRCT images
when the image pattern is inconclusive in
making a UIP diagnosis.
Progressive fibrosing ILD (PF-ILD) has
been recently proposed as a clinical ILD
phenotype on the basis of the clinical
behavior of ILD (i.e., progression of clinical
symptoms, radiographic findings, and/or
pulmonary function testing) regardless of
the primary clinical diagnosis (23). Recently,
the INBUILD trial showed a decreased
rate of decline in FVC with nintedanib
compared with placebo only when the
heterogeneous group of PF-ILD was
grouped as a single entity in patients with
PF-ILD, specifically, in patients with a UIP-
like fibrotic pattern (9). Although this
study suggests a therapeutic response to
nintedanib in patients who have PF-ILD,
individuals and individual subgroups of the
heterogeneous group of ILD may not
respond to nintedanib. There were no
prespecified diagnostic criteria for the
specific PF-ILD diagnoses nor were these
adjudicated, and, hence, the results of the
INBUILD trial must be interpreted with
caution, as several questions are raised
(24). Regardless, a subset of grouped
patients with PF-ILD did not respond
or had adverse events that led to the
discontinuation of nintedanib, suggesting
that obtaining an initial ILD diagnosis to
determine early management strategies and
an assessment of the clinical behavior of the
lung disease remains necessary to optimally
manage patients with ILD. In addition,
patients with a non-UIP fibrotic pattern
had a nonsignificant change in their lung
function with nintedanib, suggesting that
distinguishing patients with a UIP pattern
early may allow better selection of patients
to receive antifibrotic medications, given
the potentially debilitating side effects as
well as consideration for early referral for
lung transplantation in appropriately
selected patients. Lastly, because each
individual patient’s trajectory of decline

Figure 3. (Continued ). standard pathology diagnosis and central radiology diagnosis for each patient are listed on the y-axis. Patients with discordant
usual interstitial pneumonia (UIP) (UIP pattern pathology in one lobe and non-UIP pattern pathology in a different lobe) are indicated with a hatched circle.
Patients with scores ,0.87 are reported as non-UIP, and patients with scores of >0.87 are reported as UIP. Scores are not considered indicative of UIP
pattern severity. DIP = desquamative interstitial pneumonia; HP = hypersensitivity pneumonitis; NSIP = nonspecific interstitial pneumonia; OP = organizing
pneumonia; RB = respiratory bronchiolitis; SRIF = smoking-related interstitial fibrosis.

Richeldi, Scholand, Lynch, et al.: A Molecular Classifier to Identify UIP 217


Table 4. UIP Diagnostic Yield from Local Radiology Alone
Local Radiology Result
Definite/probable UIP, n
Indeterminate for UIP/consistent with non-IPF, n
Sensitivity, % (95% CI)
Specificity, % (95% CI)
NPV, % (95% CI)
PPV, % (95% CI)
UIP prevalence, %
Definition of abbreviations: CI = confidence interval; IPF = idiopathic pulmonary fibrosis;
NPV = negative predictive value; PPV = positive predictive value; UIP = usual interstitial pneumonia.
and the overall prognosis is highly
dependent on his or her clinical diagnosis,
lumping all subgroups of progressive ILD
may hinder the ability to identify and
develop targeted therapies for each group
(24). Early management decisions will be
even more relevant given the increasing
number of clinical trials of both antifibrotic
medications and immunosuppressive
therapies targeted to specific ILD diagnoses
(23). Future studies may include the Envisia
Classifier in differentiating patients with a
UIP pattern to aid in the appropriate
management of patients at the onset and
later in the course of PF-ILD.
Current guidelines for the diagnosis of
IPF conditionally recommend SLB for a
pathologic diagnosis when HRCT and
clinical factors are not definitive for IPF (4).
However, SLB is not without risk and has a
reported in-hospital mortality of 1.7% for
elective procedures and 16% for nonelective
procedures (5). In addition, there may be
select patients who are at high risk for
complications related to SLB and who,
Table 5. UIP Diagnostic Yield from Local Radiology in Conjunction with Envisia
Genomic Classifier Testing
Local Radiology 1 Envisia Classifier
Definite/probable UIP or Envisia Classifier UIP, n
Indeterminate for UIP/consistent with non-IPF and
Envisia Classifier non-UIP, n
Sensitivity, % (95% CI)
Specificity, % (95% CI)
NPV, % (95% CI)
PPV, % (95% CI)
UIP prevalence, %
For definition of abbreviations, see Table 4.
218 American Journal of Respiratory and Critical Care Medicine Volume 203 Number 2 | January 15 2021
Pathology Reference Standard
UIP (n = 53) Non-UIP (n = 32)
18 1 patients with a probable UIP pattern on
35 31
34.0 (21.5–48.3)
96.9 (83.8–100)
47.0 (34.6–59.7)
94.7 (74.0–99.9)
62.4
therefore, may be unable or unwilling to
undergo SLB because of these additional
risks. The Envisia Genomic Classifier is
obtained via bronchoscopy with TBBx, an
easier and safer procedure that can be
performed in the outpatient setting by
general pulmonologists worldwide. As a
result, the classifier can be expeditiously
obtained while avoiding the risks and
complications of SLB.
Histopathology obtained from
both SLB and cryobiopsies have
appreciable interobserver variability in the
interpretation of ILD patterns even among
expert ILD pathologists (25–27). A recent
meta-analysis of more than 10,000 patients
with ILD showed that 12% of all patients
with ILD remain unclassifiable despite
many undergoing SLB, highlighting the
heterogeneity and diagnostic uncertainty of
ILD pathology (28). Although the recent
prospective study demonstrated a good
interobserver results in histopathology
diagnostic accuracy with cryobiopsy
compared with SLB (29), cryobiopsy
Pathology Reference Standard
UIP (n = 53) Non-UIP (n = 32)
42 3 United States, with two centers in Europe,
11 29
79.2 (65.9–89.2)
90.6 (75.0–98.0)
72.5 (56.1–85.4)
93.3 (81.7–98.6)
62.4
ORIGINAL ARTICLE
requires the procedure to be done by
experienced experts in specialized centers,
whereas TBBx is a routine procedure by
pulmonologists in the ambulatory setting,
and its safety is well established worldwide.
Interestingly, in our study, there were more
histopathology in the classifier FN biopsies
(i.e., pathology ultimately showed UIP and
the classifier result showed non-UIP). In
addition, among the FN classifier results,
there was an enrichment of discordance in
the lung lobar pathological diagnosis
(i.e., UIP and a non-UIP diagnosis in
different lobes in a patient). These findings
may highlight the inherent challenge in
achieving a clear ILD diagnosis.
It is important to note that although
UIP detected by either molecular,
radiographic, or histopathologic
identification is present in patients with IPF,
it is not synonymous with IPF and may be
present in patients with other types of ILD,
including chronic HP, connective tissue
disease–associated ILD, and ILD associated
with other exposures, such as asbestosis
exposure. The recognition of a UIP pattern
by the Envisia Classifier should be
interpreted in the context of the HRCT
image patterns and clinical features, ideally
in the presence of a multidisciplinary team
discussion, and does not always confer a
diagnosis of IPF. Similarly, making a
diagnosis of chronic HP is often complex
and is based on clinical features, exposure
history, and specific radiographic pattern
with or without histopathology. A UIP
pattern from the Envisia Classifier in
patients with suspected chronic HP can be
challenging and should be interpreted in
the context of these clinical and radiographic
features.
There are a few limitations of this
study. First, the generalizability of the
patient population is limited to patients
without a recognized definitive HRCT UIP
pattern that required lung biopsy with
diagnostic pathology to obtain a UIP
diagnosis. However, given the multicenter
prospective accrual of patients from both
community and academic centers across the
we believe that the patients with ILD in
this validation cohort are heterogeneous
and representative of a broad spectrum
of patients with ILD. In addition, the
accuracy of the Envisia Genomic Classifier
performance is unknown in patients with
ILD who had a nondiagnostic pathology in
ORIGINAL ARTICLE

lung biopsy. This study did not mandate the
type of procedure that was required to
obtain lung biopsies, only that these biopsies
met the criteria set by two central expert ILD
pathologists of having an adequate tissue
specimen to make an ILD diagnosis.
Biopsies that did not meet this level of rigor
were not included in this study. This study
was not designed to assess or compare lung
biopsies obtained by different diagnostic
techniques (29, 30). Lastly, the investigators
acknowledge the limitation related to the
interpretation of local HRCT reports per
the Fleischner criteria by two independent
reviewers. Given the importance of
capturing a real-world experience of local
pulmonologists receiving local radiology
reports, these summary diagnoses of the
radiology features in a consistent manner
and according to guidelines should provide
a representative picture of the patient
radiology in this study cohort.
In conclusion, the results of this
study support the sustained accuracy
and reproducibility of the molecular
diagnosis of UIP by the Envisia Genomic
Classifier in a second independent validation
cohort. The study highlights the potential
utility of increased identification of UIP by
the Envisia Genomic Classifier when HRCT
interpretation according to guidelines
remains inconclusive for a diagnosis of ILD.
The combined validity and utility of the
Envisia Genomic Classifier has the ability to
identify a UIP pattern of lung fibrosis in
patients with ILD of unclear etiology,
potentially facilitating an earlier IPF
diagnosis in the correct clinical context and
an earlier initiation of treatment without the
need for an SLB. n
Author disclosures are available with the text
of this article at www.atsjournals.org.
Acknowledgment: The authors thank the
following members of their team for their
assistance with the coordination of the clinical
and biological sample collection, processing, and
scoring: Grażyna Fedorowicz, Eric Morrie,
Jeremy Burbanks-Ivey, Germaine Ng, Hannah
Neville, Srinivas Jakkula, Jianchang Ning, Xinwu
Yang, Marla Johnson, and the clinical laboratory
scientists of the Veracyte Clinical Laboratory
Improvement Amendments (CLIA) laboratory.
They also thank the patients and their families
who made this study possible.

References
1. American Thoracic Society; European Respiratory Society. American
Thoracic Society/European Respiratory Society international
multidisciplinary consensus classification of the idiopathic interstitial
pneumonias: this joint statement of the American Thoracic Society
(ATS), and the European Respiratory Society (ERS) was adopted by
the ATS board of directors, June 2001 and by the ERS Executive
Committee, June 2001. Am J Respir Crit Care Med 2002;165:
277–304.
2. Lederer DJ, Martinez FJ. Idiopathic pulmonary fibrosis. N Engl J Med
2018;378:1811–1823.
3. Raghu G, Collard HR, Egan JJ, Martinez FJ, Behr J, Brown KK, et al.;
ATS/ERS/JRS/ALAT Committee on Idiopathic Pulmonary Fibrosis. An
official ATS/ERS/JRS/ALAT statement: idiopathic pulmonary fibrosis.
Evidence-based guidelines for diagnosis and management. Am J
Respir Crit Care Med 2011;183:788–824.
4. Raghu G, Remy-Jardin M, Myers JL, Richeldi L, Ryerson CJ, Lederer DJ,
et al.; American Thoracic Society, European Respiratory Society,
Japanese Respiratory Society, and Latin American Thoracic
Society. Diagnosis of idiopathic pulmonary fibrosis: an official
ATS/ERS/JRS/ALAT clinical practice guideline. Am J Respir Crit Care
Med 2018;198:e44–e68.
5. Hutchinson JP, Fogarty AW, McKeever TM, Hubbard RB. In-hospital
mortality after surgical lung biopsy for interstitial lung disease in the
United States: 2000 to 2011. Am J Respir Crit Care Med 2016;193:
1161–1167.
6. Walsh SLF, Wells AU, Desai SR, Poletti V, Piciucchi S, Dubini A, et al.
Multicentre evaluation of multidisciplinary team meeting agreement on
diagnosis in diffuse parenchymal lung disease: a case-cohort study.
Lancet Respir Med 2016;4:557–565.
7. Ryerson CJ, Corte TJ, Lee JS, Richeldi L, Walsh SLF, Myers JL, et al. A
standardized diagnostic ontology for fibrotic interstitial lung disease:
an international working group perspective. Am J Respir Crit Care
Med 2017;196:1249–1254.
8. Lynch DA, Sverzellati N, Travis WD, Brown KK, Colby TV, Galvin
JR, et al. Diagnostic criteria for idiopathic pulmonary fibrosis:
a Fleischner Society white paper. Lancet Respir Med 2018;6:
138–153.
9. Flaherty KR, Wells AU, Cottin V, Devaraj A, Walsh SLF, Inoue Y, et al.;
INBUILD Trial Investigators. Nintedanib in progressive fibrosing
interstitial lung diseases. N Engl J Med 2019;381:1718–1727.
10. Walsh SLF, Lederer DJ, Ryerson CJ, Kolb M, Maher TM, Nusser R,
et al. Diagnostic likelihood thresholds that define a working diagnosis
of idiopathic pulmonary fibrosis. Am J Respir Crit Care Med 2019;
200:1146–1153.
11. Raghu G, Flaherty KR, Lederer DJ, Lynch DA, Colby TV, Myers JL,
et al. Use of a molecular classifier to identify usual interstitial
pneumonia in conventional transbronchial lung biopsy samples:
a prospective validation study. Lancet Respir Med 2019;7:
487–496.
12. Pankratz DG, Choi Y, Imtiaz U, Fedorowicz GM, Anderson JD,
Colby TV, et al. Usual interstitial pneumonia can be detected in
transbronchial biopsies using machine learning. Ann Am Thorac Soc
2017;14:1646–1654.
13. Choi Y, Liu TT, Pankratz DG, Colby TV, Barth NM, Lynch DA, et al.
Identification of usual interstitial pneumonia pattern using RNA-Seq
and machine learning: challenges and solutions. BMC Genomics
2018;19:101.
14. Case A, Lynch D, Chung J, Huang J, Choi Y, Lofaro L, et al.
Performance and clinical utility of the genomic classifier (Envisia)
for usual interstitial pneumonia in conjunction with local radiology
[abstract]. Am J Respir Crit Care Med 2020;201:A2580.
15. Scholand M, Nathan S, Colby T, Groshong S, Lynch D, Lancaster L,
et al. A follow-on prospective clinical validation of the Envisia
Genomic Classifier [abstract]. Am J Respir Crit Care Med 2020;201:
A2557.
16. Cosgrove GP, Bianchi P, Danese S, Lederer DJ. Barriers to timely
diagnosis of interstitial lung disease in the real world: the INTENSITY
survey. BMC Pulm Med 2018;18:9.
17. Flaherty KR, Thwaite EL, Kazerooni EA, Gross BH, Toews GB, Colby
TV, et al. Radiological versus histological diagnosis in UIP and NSIP:
survival implications. Thorax 2003;58:143–148.
18. Silva CIS, Müller NL, Lynch DA, Curran-Everett D, Brown KK,
Lee KS, et al. Chronic hypersensitivity pneumonitis: differentiation
from idiopathic pulmonary fibrosis and nonspecific interstitial
pneumonia by using thin-section CT. Radiology 2008;246:
288–297.
19. Sumikawa H, Johkoh T, Fujimoto K, Ichikado K, Colby TV, Fukuoka
J, et al. Usual interstitial pneumonia and nonspecific interstitial
pneumonia: correlation between CT findings at the site of
biopsy with pathological diagnoses. Eur J Radiol 2012;81:
2919–2924.
20. Raghu G, Anstrom KJ, King TE Jr, Lasky JA, Martinez FJ; Idiopathic
Pulmonary Fibrosis Clinical Research Network. Prednisone,
azathioprine, and N-acetylcysteine for pulmonary fibrosis. N Engl J
Med 2012;366:1968–1977.
21. Raghu G, Rochwerg B, Zhang Y, Garcia CAC, Azuma A, Behr J, et al.;
American Thoracic Society; European Respiratory society; Japanese
Respiratory Society; Latin American Thoracic Association. An official
ATS/ERS/JRS/ALAT clinical practice guideline: treatment of idiopathic
pulmonary fibrosis. An update of the 2011 clinical practice guideline.
Am J Respir Crit Care Med 2015;192:e3–e19.

Richeldi, Scholand, Lynch, et al.: A Molecular Classifier to Identify UIP 219


22. Chung JH, Chawla A, Peljto AL, Cool CD, Groshong SD, Talbert JL,
et al. CT scan findings of probable usual interstitial pneumonitis have
a high predictive value for histologic usual interstitial pneumonitis.
Chest 2015;147:450–459.
23. Wong AW, Ryerson CJ, Guler SA. Progression of fibrosing interstitial
lung disease. Respir Res 2020;21:32.
24. Raghu G. Nintedanib in progressive fibrosing interstitial lung diseases.
N Engl J Med 2020;382:779–780.
25. Flaherty KR, Travis WD, Colby TV, Toews GB, Kazerooni EA, Gross
BH, et al. Histopathologic variability in usual and nonspecific
interstitial pneumonias. Am J Respir Crit Care Med 2001;164:
1722–1727.
26. Nicholson AG, Addis BJ, Bharucha H, Clelland CA, Corrin B, Gibbs AR,
et al. Inter-observer variation between pathologists in diffuse
parenchymal lung disease. Thorax 2004;59:500–505.
220 American Journal of Respiratory and Critical Care Medicine Volume 203 Number 2 | January 15 2021
ORIGINAL ARTICLE
27. Lettieri CJ, Veerappan GR, Parker JM, Franks TJ, Hayden D, Travis WD,
et al. Discordance between general and pulmonary pathologists
in the diagnosis of interstitial lung disease. Respir Med 2005;99:
1425–1430.
28. Guler SA, Ellison K, Algamdi M, Collard HR, Ryerson CJ. Heterogeneity
in unclassifiable interstitial lung disease. A systematic review and
meta-analysis. Ann Am Thorac Soc 2018;15:854–863.
29. Troy LK, Grainge C, Corte TJ, Williamson JP, Vallely MP, Cooper WA,
et al. Diagnostic accuracy of transbronchial lung cryobiopsy for
interstitial lung disease diagnosis (COLDICE): a prospective,
comparative study. Lancet Respir Med 2020;8:171–181.
30. Maldonado F, Danoff SK, Wells AU, Colby TV, Ryu JH, Liberman M,
et al. Transbronchial cryobiopsy for the diagnosis of interstitial lung
diseases: CHEST guideline and expert panel report. Chest 2020;157:
1030–1042.