Assessment of a Consensus Driven Narrow Band Imaging Pattern Classification System in Barrett*s Esophagus

Endoscopie surveillance of Barrett oesophagus (BO) patients is recommended to
detect high-grade intraepithelial
neoplasia (HGIN) or early cancer (EC) at a curable stage. With standard
endoscopy, however, it is difficult to distinguish
areas with HGIN/EC. In the absence of visible lesions, random biopsies are
obtained for histological assessment of
neoplasia, but these random biopsies may miss dysplastic lesions (sampling
error). The endoscopic detection of early
neoplasia may be improved by the use of Narrow Band Imaging (NBI). Narrow Band
Imaging improves image contrast by taking advantage of the absorption and
scattering properties of human tissue. Light penetrates tissue as a function of
its wavelength. Longer wavelengths penetrate more deeply into the tissue than
shorter wavelengths due to scattering phenomena. Therefore, the blue light
centered at 415 nanometers shows detail on the surface of the mucosa revealing
subtle changes. In addition, hemoglobin absorbs light at both 415 nanometers
(blue) and 540 nanometers (green) making vessels appear with greater contrast.
As a result, superficial mucosal vessels appear brownish while the deeper
vessels appear greenish in color. Using NBI, different mucosal (regular, round,
oval, villous, irregular or abnormal) and vascular (normal, regular, irregular
or abnormal) patterns can be recognized within the BE segment. These patterns
have the ability to predict histology. A recent meta-analysis which included 3
studies with a total of 580 biopsies found good sensitivity (77% - 100%),
specificity (79% to 94%), and accuracy (88% to 96%) of NBI in differentiating
gastric mucosa from intestinal metaplasia [15]. A pooled analysis of 5 studies
and 756 biopsies showed sensitivity (95% CI) of 97% (89% - 99%), specificity of
94% (60% - 99%) and an overall accuracy of 96% (72% - 99%) in predicting
dysplastic from non dysplastic BE using NBI. However, all the studies above
have used different scopes and criteria for identification of dysplasia in BE.

There are only a few randomized controlled trials (RCT) evaluating the efficacy
of NBI over other WLE or chromoendoscopy. One RCT by Kara et al compared
chromoendoscopy to NBI in detecting HGD or esophageal cancer (EAC) in 28 BE
patients [16] and showed that indigo carmine chromoendoscopy and NBI were
comparable in the detection of HGD and EAC. However, this study was likely
underpowered to detect such a difference [15]. A randomized, controlled,
multicenter, crossover study by Sharma et al comparing high definition WLE to
NBI in 116 patients found no difference in the detection of intestinal
metaplasia (85% vs. 86%) [17]. However, this study found that the NBI group
required significantly less number of biopsies per patient (3.7 vs. 8). Wolfsen
and colleagues evaluated 65 patients in a prospective, blinded, tandem
endoscopy study comparing NBI to standard WLE [18]. They found that NBI
detected more patients with dysplasia (57% vs. 43%) with significantly less
biopsies per patient (4.7 vs. 8) when compared to WLE. Based on the findings of
above studies, it is likely that NBI will improve the detection of dysplasia
and/or at the least decrease the number of biopsies needed when used in
conjunction with WLE. This decrease in the number of biopsies has the ability
to make BE surveillance efficient and cost-effective.

However, a simple, consensus driven NBI classification system of BE surface
patterns (mucosal and vascular) has not been established and tested. The
development and assessment of such a classification system could lead to
uniform terminology and could then be prospectively evaluated in BE imaging
studies.

This clinical trial will evaluate the inter-observer agreement of a simple,
consensus driven NBI classification system of surface patterns and its ability
to differentiate dysplastic vs. non-dysplastic BE in patients undergoing BE
screening or surveillance. From previous studies, the expert panel has
developed simplified criteria to detect dysplasia in BE. This study attempts to
validate these criteria in expert academic centers NBI images of the BE mucosa
obtained during endoscopy will be classified by 6 academic endoscopists and
initially. The endoscopists will then be asked to predict histopathology based
on the surface mucosal and vascular patterns. The inter-observer agreement of
this NBI classification system will be tested using kappa statistics. Their
performance will be evaluated for accuracy, sensitivity, specificity, PPV and
NPV of each pattern that is visualized on NBI.

This is a multicenter, prospective, double-blinded study of 200 NBI images from
approximately 50 or more patients enrolled in the study. The study will be
conducted at the following expert academic centers: Veterans Affairs Medical
Center (Kansas City, MO USA), The University of Chicago Medical Center
(Chicago), Amsterdam Medical Center (Amsterdam, Netherlands), , University of
Regensburg (Ausburg, Germany), The Jikei University School of Medicine (Tokyo,
Japan), and Hokkaido University Hospital (Hokkaido, Japan). Patients
undergoing BE screening and BE surveillance will be enrolled into the study.
After meeting eligibility criteria, and obtaining an informed consent, patients
will undergo their routine upper endoscopic examination using HD white light
endoscopy. NBI images will be obtained in areas before the corresponding
biopsies are obtained.
During the course of the upper endoscopy, the BE surface patterns will be
carefully examined. In each of these positions, 4-5 high quality images will be
obtained from different surface patterns initially with HD WLE and then using
NBI. Image capture will be standardized. From the 4-5 images recorded from one
area, only the best high quality image that had corresponding targeted biopsy
will be chosen by the primary investigator. No more than 4 images from
different surface and mucosal NBI patterns within a given patient will be
obtained (only 1 image from one area). All images will be captured using a
high-definition, NBI endoscope (HD-180 endoscopes, Olympus Inc) and stored in
the high quality TIFF format. For the purposes of this study the settings on
the scope will be chosen such that no patient information is included either on
the image recording (i.e. images showing the endoscopic picture only without
any patient identifiers). After images have been obtained, target biopsies will
be obtained from each area and submitted for histopathological evaluation in
separate jars.
Surveillance biopsies will be performed on all patients with known BE according
to the recommended guidelines. Four quadrant biopsies will be taken using
maximum capacity forceps beginning at the top of the gastric folds (indicated
as gastroesophageal junction or GEJ) and moving proximally in 1-2 cm increments
to encompass the complete (baseline) length. Prior to biopsies being obtained,
a detailed examination of the BE surface patterns as described above will be
performed.

The chance on the general risks associated with a diagnostic gastroscopy, such
as irritation of the throat by introduction of the endoscope, difficulty
swallowing or retrosternal pain, is not increased by the use of NBI for the
detection of early neoplasia. The added burden for the patient when
participating in this study will be a prolongation of the endoscopy by 15
minutes to a total of 45 minutes.