Introduction
Patients with Barrett’s esophagus (BE) undergo regular endoscopic surveillance to
detect dysplasia in the early stages when it is amenable to endoscopic treatment to
prevent progression to invasive esophageal adenocarcinoma (EAC).
Because dysplasia is often not visible during endoscopy, there is general agreement
that forceps biopsy sampling according to the Seattle protocol is currently the best
method for detection of dysplasia in the absence of visible lesions. This involves
random forceps biopsies (FBs) obtained from four quadrants at 2-cm intervals along
the BE segment, using large-capacity forceps. However, dysplasia and early cancer
may be patchy or focal and, as a result, can be missed by FBs. FBs are therefore associated
with sampling error and missed dysplasia. Furthermore, the Seattle protocol is time-consuming, especially in longer BE segments.
Studies have shown that approximately half of gastroenterologists do not adhere to
the Seattle protocol [1 ].
Wide-area transepithelial sampling (WATS) with computer-assisted 3 D analysis is a
novel, brush biopsy technique used to broaden the area of sampled BE. The WATS brush
has long hard bristles and is abrasive, so it enables deep transepithelial specimens
to be obtained. Use of the brush is relatively easy and not time-consuming. Tissue
obtained by WATS undergoes computerized neural network analysis that helps identify
the areas on the slide that are most likely to represent neoplastic change. These
areas are then presented to a pathologist for evaluation and diagnosis.
Previous studies have shown a 2 %–42 % absolute increase, or 42 %–329 % relative increase,
in the detection of high grade dysplasia (HGD) or EAC when WATS is used as an adjunct
to FB [2 ]
[3 ]
[4 ]
[5 ]
[6 ]. However, these studies had some methodological limitations. For example, patients
with visible lesions were included; WATS was primarily assessed as an adjunct to FB
instead of as a replacement procedure; and/or studies had a primary focus on the detection
of intestinal metaplasia (IM) and not dysplasia.
Therefore, the purpose of this study was to assess the value of WATS as a replacement
for FBs for the detection of HGD/EAC in BE patients without visible lesions. This
was the first prospective randomized study of this nature in a referral population
of BE patients with a history of dysplasia.
Methods
This was a prospective multicenter randomized study that enrolled patients under endoscopic
surveillance for BE with a history of dysplasia or mucosal adenocarcinoma at 17 participating
medical centers (Table 1 s , see online-only Supplementary material). The institutional review boards of all
participating centers approved the study and informed consent was obtained from all
patients. An on-site initiation visit with training on the protocol and the WATS procedure
was planned at all sites prior to the start of the study and the first 10 cases per
endoscopist were not included in the study analysis.
Study population
We included patients scheduled for a regular imaging endoscopy for BE, with a history
of BE-associated neoplasia, either:
BE in the absence of visible lesions with either low grade dysplasia (LGD) or HGD
diagnosed on FBs ([Fig. 1 ]); or
Flat BE after prior endoscopic resection (ER) for a visible lesion with the resection
specimen showing LGD, HGD, or mucosal EAC with good-to-moderate differentiation, without
lymphovascular invasion, and with negative resection margins. The minimum time between
ER and randomization was 6 weeks.
Fig. 1 Patient study flow diagram. FB, forceps biopsy; WATS, wide-area transepithelial sampling.
Exclusion criteria were: age < 18 years; BE length < 2 cm circumferential extent or
> 10 cm maximum extent; prior ablation therapy; and history of esophageal surgery
other than fundoplication.
Randomization and index endoscopy
Eligible patients were randomized by a computer-generated system into two groups:
FB sampling followed by WATS; or WATS followed by FB. WATS and FB were performed during
the same endoscopy.
Random forceps biopsies
Four-quadrant FB specimens were obtained using a standard biopsy forceps according
to the Seattle protocol at 2-cm intervals along the complete length of the BE segment.
WATS brushing
Two WATS brushes were obtained for every 5 cm segment of BE, starting at the gastroesophageal
junction and moving proximally through the entire BE segment. Bristles were placed
against the mucosal surface and the brush was rotated and repeatedly passed back and
forth until pinpoint bleeding was observed ([Video 1 ]). The first brush was smeared on a glass slide and fixed for Papanicolaou (PAP)
staining and its bristles were cut and placed inside the vial. The process was repeated
with the second brush for the same BE area and this brush was placed directly into
the vial without first being smeared on a slide. In patients with BE of > 5 cm in
length, two new brushes were used for the next 5 cm of BE.
Video 1 Video showing the wide-area transepithelial sampling (WATS) procedure, with the WATS
brush inserted through the working channel of the endoscope and brushing performed
until pinpoint bleeding is observed.
Pathology
All specimens, regardless of tissue acquisition technique, were categorized as either
LGD, HGD, or intramucosal EAC according to previously published criteria for FB [7 ] and WATS [13 ] ([Fig. 2 ]).
Fig. 2 Images showing high grade dysplasia diagnosed using the WATS3 D brush on: a cell block; b smear.
Pathologists received the specimen along with only a random study number and were
blinded to the patient’s history, demographics, endoscopic findings, histologic findings,
and outcome of the other technique.
Forceps biopsy
The local expert pathologist in each study center performed the initial assessment.
All local pathologists worked in BE expert centers and were experienced in the diagnosis
of BE and related neoplastic complications. Next, an independent pathologist with
expertise in gastrointestinal pathology located at a central European pathology laboratory
(G.H.) interpreted the specimens.
Where there were discrepancies, a third BE expert pathologist (S.M.) reviewed the
sample to achieve a consensus diagnosis (defined as the diagnosis rendered by two
of the three pathologists).
WATS
The WATS brush sample that was stained with a modified PAP stain was evaluated with
computer-assisted 3 D tissue analysis using neural networks specifically optimized
for the esophageal mucosa. The computer is capable of detecting very small numbers
of atypical cells on the slide. The images of the most atypical cells were then displayed
on a high resolution video monitor in order for a final diagnosis to be made by the
pathologist.
All WATS specimens selected by the computer were analyzed by two central pathologists
from a single central laboratory (CDx Diagnostics, Inc.), both of whom have extensive
experience in WATS evaluation. The pathologists made the diagnosis independently from
each other. Where discrepancies occurred, a third pathologist reviewed the case to
produce a consensus diagnosis.
When WATS is used for regular clinical care, and not as in the current study, a single
pathologist reviews the slides selected by the computer and, when there is dysplasia,
a second pathologist is requested, as per national guidelines.
Study end points
The primary end point of this study was the concordance/discordance between the detection
of BE-associated HGD/EAC using WATS brushing and FB.
Secondary end points were: (i) the adjunctive value of WATS, defined as the absolute
and relative increases in detection when WATS is added to FB; (ii) the effect of the
order of WATS and FB on the outcome; (iii) procedure time; and (iv) related complications.
Statistical analysis
The intention-to-treat (ITT) analysis included all patients who underwent randomization.
The per-protocol (PP) analysis included all patients with an adequate specimen for
histologic analysis.
Normally distributed continuous variables were summarized using mean and SD; skewed
variables were summarized using median and interquartile range (p25–p75). Categorical
variables were summarized using counts and percentages. All confidence intervals were
reported at the 95 % level. The McNemar test was used to evaluate the primary outcome.
Student’s t test was used to compare normally distributed variables.
The relative difference was defined as the extra cases found with WATS, divided by
the total cases found with FB; the absolute difference as the extra cases found with
WATS divided by all study patients. The number needed to test (NNT) was the number
of patients required to undergo WATS as an adjunct to FB in order to detect one additional
case of HGD/EAC, as compared with the use of FB alone.
The order of randomization was evaluated by the difference in absolute detection of
HGD/EAC between the two study arms (i. e. WATS then FB, or FB then WATS).
The primary outcome was evaluated stratified by study site (Table 2 s ) but, because all sites were highly experienced with BE screening and surveillance,
we did not adjust for study center in the analysis.
Sample size
In a recent US multicenter study, the diagnostic yield of WATS was four times greater
than the yield of FB [6 ]. For this reason, we designed this study to test the superiority of WATS compared
with FB in detecting HGD/EAC by a superiority margin. We therefore hypothesized that
statistically significant differences in detection rates between WATS and FB might
not be of interest unless the difference were greater than a threshold (i. e. the
smallest difference in proportions considered by our research team to be a clinically
significant difference). Our sample size was based on an anticipated improvement in
detection rate with WATS from 27 % to 55 %, a superiority margin of 0.15 (equivalent
to observation of a detection rate of at least of 0.42 for WATS), with 80 % power
and a significance level of 0.05. This led to a target sample size of 147 participants.
This sample size calculation was made for independent samples. The sample size for a paired study can be approximated by the independent calculation
in many practical situations. Generally, this assumption produces overestimates of
sample size [8 ].
Results
Between November 2017 and February 2019, 205 patients were initially screened for
eligibility, 172 of whom fulfilled the inclusion criteria and underwent randomization
(the ITT population) ([Fig. 1 ]; [Table 1 ]). Patients had prior diagnoses of LGD in flat BE (n = 68; 40 %), HGD in flat BE
(n = 32; 19 %), and flat BE after prior ER of a visible lesion (n = 72; 42 %). The
latter group of patients had undergone prior ER for removal of LGD (n = 3; 5 %), HGD
(n = 21; 29 %), or mucosal EAC (n = 48; 66 %).
Table 1
Baseline characteristics of the 172 patients with Barrett’s esophagus (BE) who fulfilled
the inclusion criteria and underwent randomization (the intention-to-treat population).
Total
FB then WATS
WATS then FB
Mean age (SD), years
68 (8)
68 (8)
69 (8)
Sex, male, n (%)
144 (84)
68 (84)
76 (84)
Worst neoplasia before inclusion, n (%)
68 (40)
26 (32)
42 (46)
32 (19)
20 (25)
12 (13)
72 (42)
35 (43)
37 (41)
Esophagitis, Los Angeles grade, n (%)
163 (95)
76 (94)
87 (96)
8 (5)
4 (5)
4 (4)
1 (1)
1 (1)
0 (0)
Hiatal hernia grade, n (%)
50 (29)
29 (36)
21 (23)
81 (47)
31 (38)
50 (55)
37 (22)
17 (21)
20 (22)
4 (2)
4 (5)
0 (0)
BE extent, mean (SD), cm
3 (3)
3 (3)
3 (3)
5 (3)
5 (3)
5 (3)
FB, forceps biopsy; WATS, wide-area transepithelial sampling; LGD, low grade dysplasia;
HGD, high grade dysplasia.
In total, 25 patients were excluded from the ITT analysis owing to inadequate specimens
on either WATS (n = 23) or biopsy (n = 2). The WATS excluded cases all suffered from
either air-drying artefact, as a result of either improper application of cells onto
the slide, and/or poor application of PAP staining on the slide. The inadequate FBs
were the result of insufficient tissue for diagnosis.
In 30 out of 172 cases (17 %), there was a discrepancy between the local “on-site”
pathologist and the “central” pathologist regarding the FB diagnosis for the presence
or absence of HGD/EAC. Of these 30 cases, 19 were diagnosed with HGD/EAC by the central
study pathologist, but with nondysplastic BE (NDBE)/LGD by the local onsite pathologist.
In 11 of 30 cases, the central pathologist diagnosed NDBE/LGD, whereas the onsite
pathologist diagnosed HGD/EAC. These 30 cases were then reviewed by the third pathologist,
as per the protocol. The final “consensus” diagnosis of these cases was as follows:
12 HGD/EAC and 18 NDBE/LGD.
For WATS, there was a discrepancy between the two initial readings for the presence
or absence of HGD/EAC in 2 /172 cases (1 %), both of which were scored as no HGD/EAC
by the third reviewer.
Yield of HGD/EAC detection
In the ITT analysis, we found no statistically significant difference in the yield
of HGD/EAC detection with either WATS or FB (P = 0.36) ([Table 2a ]; Table 2 s ).
Table 2
Yield of high grade dysplasia/esophageal adenocarcinoma for forceps biopsy (FB) and
wide-area transepithelial sampling (WATS) on
a intention-to-treat analysis
WATS
Positive
Negative
Total
FB
Positive
21
12
33
Negative
18
121
139
Total
39
133
172
b per-protocol analysis
WATS
Positive
Negative
Total
FB
Positive
21
9
30
Negative
18
99
117
Total
39
108
147
a P = 0.36. b P = 0.12.
The PP analysis also showed no significant difference in the detection of HGD/EAC
between the two modalities (P = 0.12) ([Table 2a ]).
Discrepant cases between WATS and FB
Discrepant cases were evaluated in the PP analysis. The 18 WATS-positive/FB-negative
cases had a mean (p25–p75) BE segment of C3M6 (C1–6; M4–8). The referral diagnoses
were flat LGD (n = 7; 39 %), flat HGD (n = 2; 11 %), or a visible lesion that was
resected prior to the study (n = 9; 50 %). The worst diagnoses with FB for these patients
were LGD (n = 15; 83 %) and IM (n = 3; 17 %) ([Table 3 ]).
Table 3
Detection of intestinal metaplasia (IM), low grade dysplasia (LGD), or high grade
dysplasia (HGD)/esophageal adenocarcinoma (EAC) by forceps biopsy (FB) and wide-area
transepithelial sampling (WATS) on per-protocol analysis.
WATS
HGD/EAC
LGD
IM
Total
FB
HGD/EAC
21
3
6
30
LGD
15
37
9
61
IM
3
19
34
56
Total
39
59
49
147
The nine WATS-negative/FB-positive patients had a mean(p25–p75) BE segment of C2M5
(C0–3; M3–5). The referral diagnoses were flat LGD (n = 3; 33 %), flat HGD (n = 3;
33 %), or a visible neoplastic lesion that was resected prior to the study (n = 3,
33 %). The worst diagnoses on WATS samples were LGD (n = 3; 33 %) and IM (n = 6; 67 %).
Yield of HGD/EAC detection with addition of WATS
Considering WATS as an adjunct to FB, the absolute increase in HGD/EAC detection was
10 % (18/172; 95 %CI 6 % to 16 %; P < 0.001). The relative increase was 55 % (18 /33; 95 %CI 36 % to 72 %; P < 0.001). The detection rate increased from 19 % (33/172; 95 %CI 14 % to 26 %) to
30 % (52/172; 95 %CI 23 % to 38 %). In our study population, the NNT to detect one
additional case of HGD/EAC was 10 (172/18).
Impact of randomization
In the study group “FB-WATS”, absolute detection increased by 11 % (9/81; 95 %CI 6 %
to 21 %). In the study group “WATS-FB”, the absolute increment of HGD/EAC detection
was 10 % (9/91; 95 %CI 5 % to 18 %). Differences between study groups were not statistically
significant (1 %; 95 %CI −8 % to 10 %; P = 0.49).
Time
The mean (95 %CI) procedural times for FB and WATS were 6.6 (5.9 to 7.1) and 4.9 (4.1
to 5.4) minutes, respectively. The mean difference between FB and WATS was 1.6 minutes
(95 %CI 1.01 to 2.25; P < 0.001). The procedure time for WATS and FB combined was 11.2 (10.5 to 14.0) minutes.
When WATS was added to FB, the mean additional time was 4.8 minutes (95 %CI 4.4 to
5.2; P < 0.001).
Complications
No complications related to the procedure occurred.
Discussion
In this prospective multicenter randomized study, we aimed to evaluate whether WATS
could replace FBs in the detection of HGD or cancer in BE. For the primary end point
of our study, we demonstrated that there was no significant increase in the detection
of HGD/EAC for WATS as compared with FB. While WATS detected an additional 18 HGD/EAC
cases that were missed with FB (i. e. WATS positive/FB negative), 12 other cases with
HGD/EAC on the FB were missed with WATS (WATS negative/FB positive). Among the additional
cases that were detected with WATS, the majority (i. e. 83 %) were diagnosed with
LGD on FB.
Considering WATS as an adjunct to FB, our study confirmed the findings of prior studies,
with a significant improvement in the detection of HGD/EAC seen, with a relative increase
of 55 % and a NNT of 10 patients in our enriched study population. The previous studies
comparing WATS with FB for the detection of HGD/EAC, all had the aim of establishing
the value of WATS as an adjunct to FB. In 2011, Anandasabapathy and Johanson reported
outcomes for a smaller brush and an older generation computer system [2 ]
[3 ], and Vennalaganti recently published outcomes using the same generation device as
in the current study [4 ]. The relative yield of dysplasia detection increased significantly in all studies
with 42 % [2 ], 88 % [3 ], and 329 % increases [4 ]. The relative increase of 55 % in the current study lies well within this spectrum
and confirms prior findings that, if WATS is used as an adjunct to FB, the detection
of HGD or EAC increases significantly. However, the addition of WATS to FB would also
increase costs and procedure times, so cost-effectiveness studies would be relevant.
Our primary aim was to evaluate WATS as a replacement for FB. In this regard, as well
as the extra cases found with WATS (WATS positive/FB negative), the cases missed with
WATS (WATS negative/FB positive) are also relevant. WATS detected an additional 18
cases in our study (10 % of the study population), but 12 other cases (7 %) were WATS
negative/FB positive. Overall, the difference was not statistically significant with
a P value of 0.36. It should be noted that, of the extra cases detected with WATS, the
vast majority (83 %) were found to have LGD in the FB, meaning these patients already
had an indication for repeat endoscopy within 6 months or consideration for ablative
therapy based on their FB findings.
Although it was not the primary aim of the prior studies to evaluate WATS as a replacement
for FB, which it was in our analysis, comparisons can be made. Anandasabapathy reported,
in an enriched study population, that 16/151 cases (11 %) were WATS + /FB−; yet more
patients were WATS−/FB + (23/151; 15 %). Johanson, in a screening population, reported
14/1183 (1.2 %) WATS + /FB− cases and 11/1183 (0.9 %) WATS−/FB + cases. The outcomes
reported by Vennalaganti in an enriched US study population appeared to differ: 23/160
(14 %) were WATS + /FB−; yet only 1/160 (0.6 %) were WATS−/FB + . One major limitation
of the latter study that may have contributed to this discrepancy was the inclusion
of patients with a visible lesion that might have been brushed but not sampled with
the FB protocol. Codipilly, in a recent meta-analysis, reported that WATS3D was negative for dysplasia in 62.5 % of cases where FB identified dysplasia [6 ].
WATS has several clinical advantages over FB. WATS is easier to perform and decreases
the time of the procedure, and it leads to one sample rather than multiple samples
each with their own costs. Therefore, one may argue that WATS could replace FB because
of its secondary advantages. In this case, a noninferiority study would require 1952
patients in an enriched study population to prove, with 80 % power, that a 5 % difference
in dysplasia detection was not statistically significant [9 ]. This number would be even higher in an actual BE screening population, where the
baseline risk for HGD or cancer is lower. Although our study was not designed as a
noninferiority trial, the similar rates of dysplasia detection in combination with
the secondary advantages may suggest that there could be a role for WATS to replace
FB in the future because of its secondary advantages.
This study has important strengths. It is the first study that systematically compared
WATS as an alternative to FB for the detection of BE-related dysplasia in a well-defined
study population and in the absence of clearly visible lesions. The study was conducted
in 17 international expert centers by dedicated endoscopists, after on-site training
and after 10 lead-in cases that were not used for the study analysis. All pathologists
in the current study were experienced with BE neoplasia, but all FB were reviewed
by a single central pathologist. Where there was discordance, a third reading was
performed for adjudication, providing a high quality gold standard for histology.
Several limitations in our study may have biased the detection rates of WATS and/or
FB. First, there is no gold standard for HGD/cancer. Pathologists that provided the
WATS diagnosis derived from a single central laboratory (CDx Diagnostics, Inc.). This
may potentially have led to a different, either lower or higher, threshold for the
diagnosis of HGD/EAC than for FB. Therefore, it is uncertain whether a WATS + /FB−
HGD diagnosis represents comparable risk for progression to (advanced) EAC as an FB
diagnosis of HGD. Because most patients in the current study underwent ablation therapy,
there is no follow-up data available. Interestingly, a recent study assessed progression
rates to HGD/EAC of an initial WATS diagnosis of either NDBE, crypt dysplasia, or
LGD in 4545 patients who were followed for a mean of 2 years. Annual progression rates
of 0.1 %, 1.89 %, and 3.47 % were found for baseline NDBE, crypt dysplasia, or LGD,
respectively, which are comparable to progression studies performed with FB [10 ]. A new study with long-term endoscopic follow-up for a WATS + /FB− diagnosis of
HGD has been initiated to evaluate this further.
We attempted to minimize the uncertainty for the histologic diagnosis with the use
of a consensus diagnosis and by standard WHO criteria for dysplasia, regardless of
the type of tissue acquisition technique used. Pathologists still disagreed on the
distinction between HGD/EAC versus NDBE/LGD based on the FB specimen in 17 % of patients,
a finding in line with the prior literature [11 ]
[12 ].
In the FB-WATS arm, endoscopists may have avoided – consciously or unconsciously–
the FB lesions when using the WATS brush, fearing that the brush would stick in the
wound. This may have led to bias towards a lower detection rate for WATS. It may also
be that FBs were concentrated in the region previously biopsied with dysplasia according
to the Seattle Protocol, resulting in a higher detection rate for FBs.
It has been proven that adherence to the Seattle protocol is poor, leading to lower
rates for dysplasia detection [1 ]. All participating endoscopists and pathologists were experts in the field and all
adhered strictly to the Seattle protocol. Furthermore, the expert setting of both
endoscopists and pathologists is highly sensitive for detecting dysplastic lesions.
This may have caused bias towards a higher neoplasia detection rate for FBs. In a
less experienced setting, WATS may potentially detect more dysplasia as it may compensate
for a lower lesion detection rate and sampling. A repetition of the current study
in a community setting would be valuable to show whether the implementation of an
AI algorithm in combination with wide-area sampling could raise community practice
to the expert level.
Patients in the current study had a recent diagnosis of dysplasia and the results
are not generalizable to a NDBE screening population. We initiated this study in an
enriched population for efficiency reasons, of note WATS and/or FB had no direct implications
for these patients because an indication for ablation therapy already existed. The
real clinical relevance therefore is probably lower in a low prevalence surveillance
population.
It has to be noted that, despite prestudy training of all sites and investigators,
a significant number of WATS brushes turned out to be inadequate owing to air drying
problems caused by late fixation for the PAP stain. In a newer WATS generation, the
PAP stain is replaced with ThinPrep solution to eliminate any potential technical
errors, which may prevent air-drying issues in the future. The risk of technical errors
still needs to be taken into consideration when endoscopists want to replace standard
FB with WATS, because an inadequate specimen would lead to another endoscopy being
required.
The hypothesis that WATS would detect two times as much dysplasia as FBs and that
the dysplasia detection rate for FBs would be 27 % was too optimistic. The calculated
sample size was however an overestimation because it was based on independent outcomes.
In conclusion, this multicenter randomized study confirmed that the combination of
WATS as an adjunct to FB increases dysplasia detection; however, WATS as a standalone
technique did not detect significantly more HGD or cancer as compared with FBs in
an enriched BE population with known dysplasia.
