Introduction
Small and diminutive colorectal polyps are the most common findings during colonoscopies
[1]
[2]. Sending such polyps routinely for histopathology evaluation is costly and can likely
be replaced by optical diagnosis [1]
[3]. However, current modalities for optical diagnosis cannot reliably distinguish between
low and high risk dysplasia or provide an accurate prediction of sessile serrated
lesions (SSL) [2]
[4]
[5]
[6]. Thus, patients with advanced colorectal neoplasia might be at risk of inappropriate
management and potential surveillance interval delays when undergoing optical diagnosis.
As the risk of colorectal polyps harboring advanced pathology increases with size,
a prudent implementation of optical diagnosis might ensure the safety of patient management
while confidently introducing optical diagnosis into routine clinical practice. Currently,
optical diagnosis is typically used for diminutive (≤ 5 mm) colorectal polyps [7]
[8]
[9]
[10], although some authors have suggested expanding its application to polyps up to
10 mm [11]. It has even been suggested that pathology cannot be regarded as the reference standard
for diagnosing polyps ≤ 3 mm, especially when a high confidence optical diagnosis
identifies an adenoma [12]. To date, no study has evaluated the impact of different size cutoffs on the appropriate
management of patients undergoing optical diagnosis. We hypothesized that a lower
polyp size cutoff (e. g. 1–3 mm) would be associated with a lower risk of misclassifying
advanced neoplasia or even cancer when using optical diagnosis. The aim of this study
was thus to evaluate how the application of different cutoffs (1–3 mm, 1–5 mm, 1–10 mm)
would affect the safety and efficacy of optical diagnosis.
Methods
Study design and patients
This study was a post hoc analysis of data from three prospective single-center studies
(two centers, 22 staff endoscopists; see Table 1 s in the online-only Supplementary Material), in which all patients underwent optical
diagnosis for all polyps ≤ 10 mm found in the study cohorts [13]
[14]. The study population included patients aged 45–80 years undergoing complete elective
colonoscopies at the Montréal University Hospital Center (CHUM) [13]
[14] and VA Medical Center White River Junction (Vermont, USA). Exclusion criteria were
known inflammatory bowel disease, active colitis, coagulopathy, familial polyposis
syndrome, poor general health (American Society of Anesthesiologists class > 3), and
missing or unclear data on demographic or colonoscopy characteristics. Study outcomes
are reported by following the Standards for Reporting Diagnostic accuracy studies
(STARD) recommendations [15]. Each study was approved by the institutional research board (IRB numbers: NORD
study: 16.367; OPTIVISTA study: 17.135; VA study: 921356) and was registered at ClinicalTrials.gov
(NCT04032912 and NCT03515343, respectively) for CHUM.
Study procedures
Patients underwent standard bowel preparation. Participating endoscopists included
board-certified gastroenterologists and fellows with various levels of expertise in
optical diagnosis. During colonoscopies, endoscopists optically evaluated polyps ≤ 10 mm
using different image-enhanced endoscopy equipment: Optivista (1 and 2 Optivista Enhanced
[OE] settings), I-SCAN (1, 2, and 3 settings) (both Pentax Medical, Tokyo, Japan)
[13]
[14]
[16], and narrow-band imaging (NBI; VA study). Polyps were then classified based on the
validated NBI International Colorectal Endoscopic (NICE) classification system as
hyperplastic or adenoma in both centers. An additional assessment was performed in
polyps classified as hyperplastic to evaluate for the presence of serrated features,
according to the Sano classification [16], and any such polyps were defined as SSLs [16]. For each polyp, endoscopists also recorded whether the optical diagnosis was made
with high or low confidence. A high level of confidence in optical diagnosis was assigned
when a polyp had endoscopic color, surface, and/or vessel features associated with
a specific type of histology in the NICE classification [17]. Common colonoscopy quality metrics such as cecal intubation and quality of bowel
preparation, as well as size, location, and morphological characteristics of each
polyp based on the Paris classification [18], were also documented. For analysis, polyps were stratified into three groups according
to the endoscopic size: 1–3 mm, 1–5 mm, and 1–10 mm.
Definition of advanced polyp histology
All 1–10 mm polyps with tubulovillous or villous histology, traditional serrated adenomas,
any polyp histology with high grade dysplasia, or cancer were considered as having
advanced pathology [5]. As the latest US Multi-Society Task Force on Colorectal Cancer (USMSTF) guideline
recommends a shorter surveillance interval for patients with traditional serrated
adenoma owing to the potential for malignancy, we considered traditional serrated
adenoma as advanced adenoma [5]
[19].
Histopathological assessment was available for all resected polyps. Qualified pathologists
assessed polyp specimens according to current and institutional practice standards.
Polyps were categorized as neoplastic (including adenomatous or SSLs, large hyperplastic
polyps ≥ 10 mm, SSL with dysplasia) or non-neoplastic (including hyperplastic polyps,
inflammatory or mucosal prolapse, etc.) [20].
Surveillance interval calculation
Postcolonoscopy surveillance intervals based on optical diagnosis were determined
for each patient based on a combination of the high confidence optical pathology prediction,
the histopathology results of polyps optically diagnosed with low confidence, and
the histopathology outcomes of all other concomitant polyps. Poor bowel preparation
and positive family history of colorectal cancer were considered in final decisions
on surveillance intervals. The reference standard surveillance interval was based
on histopathological outcomes using the most recent (2020) USMSTF guideline [5]. Therefore, four different possible surveillance intervals were assigned to the
patients: one based on actual histopathology outcomes, and three based on high confidence
optical pathology prediction using cutoffs of 1–3 mm, 1–5 mm, and 1–10 mm coupled
with the histopathology reports of polyps with low confidence optical diagnosis.
Study outcomes
The primary outcome was the proportion of patients in whom a polyp with advanced pathology
undergoing optical diagnosis was misdiagnosed as a nonadvanced or non-neoplastic polyp,
resulting in an inappropriately delayed follow-up of either 2 or 7 years for those
patients. This outcome was determined for each of the polyp size groups (1–3 mm, 1–5 mm,
and 1–10 mm) in an attempt to determine the optimal size threshold for safe implementation
of optical diagnosis. Thus, we calculated 1) the proportion of polyps with advanced
pathology in each size group, and 2) the proportion of patients with advanced polyps
who would have been assigned a delayed follow-up based on the NICE classification
system.
Secondary outcomes included the agreements between surveillance intervals based on
the optical diagnosis of polyps of the three size groups and the pathology-based recommendations.
Other secondary outcomes were the diagnostic properties of optical prediction for
neoplastic rectosigmoid polyps, including accuracy, sensitivity, specificity, positive
predictive value, and negative predictive value (NPV). The proportion of histopathology
examinations that could be avoided and the proportion of patients who could receive
an immediate surveillance recommendation were also calculated for each of the three
size cutoffs.
Statistical analyses
Continuous variables are presented as mean (SD) or median (range), as appropriate.
Categorical variables are presented as proportions with 95 %CIs.
The diagnostic characteristics of optical diagnosis were calculated by sub-stratifying
polyps into hyperplastic polyps and adenomas (excluding SSLs) within each of the three
polyp size groups. The reduction in pathology examinations was calculated for: 1)
the reference standard – the number of polyps sent for histopathology evaluation divided
by the total number of polyps; 2) optical diagnosis – the number of polyps 1–3 mm,
1–5 mm, and 1–10 mm, respectively, optically diagnosed with high confidence divided
by the total number of polyps. The proportion of patients who could have received
immediate surveillance interval recommendations was calculated for: 1) reference standard
– the total number of patients without polyp identification during colonoscopy (normal
colonoscopy) divided by the total number of patients; 2) optical diagnosis – the sum
of all patients without any polyps (normal colonoscopy) and patients with only polyps
1–3 mm, 1–5 mm, and 1–10 mm, respectively, optically diagnosed with high confidence
divided by the total number of patients.
Comparing 1–3 mm polyps with 4–5 mm polyps would introduce bias related to the size
estimation by the endoscopists and histology determination by the pathologists. Therefore,
the polyp size groups partially overlapped, and observations from individuals tended
to be correlated. To compare the proportions of outcomes of interest using different
size cutoffs, we used generalized linear models (i. e. binomial regressions) and a
logit link to analyze all correlated errors and population-averaged estimates. To
allow for within-subject observations that are equally correlated, we used an exchangeable
working correlation matrix with robust standard errors. Hence, the separate regression
models were fitted for our primary outcomes. The complete statistical methods are
described in Table 2 s. The surveillance interval agreements between optical diagnosis for different polyp
size cutoffs and pathology were calculated for both the whole cohort of patients and
the cohort of patients for whom optical diagnosis could have changed the recommended
next colonoscopy (e. g. excluding patients with normal colonoscopy, polyps ≥ 10 mm
in size, and poor bowel preparation). The agreements between the surveillance intervals
were compared between the different size cutoffs using Cohens Kappa–Fleiss adjusted
standard error [21]
[22]. Moreover, the proportions of correct and incorrect (shorter or longer) surveillance
intervals using optical diagnosis were calculated for three size groups.
All point estimates are presented with 95 %CIs, and a P value of < 0.05 was considered to indicate statistical significance. SPSS version
26.0 (IBM Corp., Armonk, New York, USA) and MedCalc Version 19.4 (MedCalc Software,
Ostend, Belgium) were used for analyses.
Results
Patient, procedure, and polyp characteristics
During the study period, 3921 patients underwent colonoscopy, and 3374 met the inclusion
criteria and were included in the final analysis (Fig. 1 s). The median age of patients was 66.0 years, and 75.2 % were male. Nearly a third
(29.6 %) of colonoscopies were performed for screening. Details of patient and colonoscopy
characteristics are presented in [Table 1]. During colonoscopies, 5906 polyps 1–5 mm in size and 1385 polyps 6–10 mm in size
(total 1–10 mm polyps = 7291) were detected. Among polyps sized 1–3 mm, 1–5 mm, and
1–10 mm with optical polyp evaluation, 2588/3212 (79.0 %), 4813/5783 (81.5 %), and
6033/7142 (82.7 %), respectively, were diagnosed with high confidence. Polyp characteristics
are presented in [Table 2].
Table 1
Demographic and clinical characteristics of patients.
|
Total number of patients, n
|
3374 (100)
|
|
Age, median (range), years
|
66.0 (45–80)
|
|
Male sex, n (%)
|
2537 (75.2)
|
|
ASA class, n (%)
|
|
|
792 (23.5)
|
|
|
1871 (55.5)
|
|
|
711 (21.1)
|
|
Antithrombotic medication use (Yes)[1], n (%)
|
909 (26.9)
|
|
Family history of CRC in first-degree relatives (Yes)[2], n (%)
|
397 (11.8)
|
|
Colonoscopy characteristics[3]
|
|
Colonoscopy indications, n (%)
|
|
|
998 (29.6)
|
|
|
144 (4.3)
|
|
|
1288 (38.2)
|
|
|
76 (2.3)
|
|
|
384 (11.4)
|
|
|
86 (2.5)
|
|
|
396 (11.7)
|
|
Cecal intubation during colonoscopy (Yes)[5], n (%)
|
3260 (96.6)
|
|
Boston Bowel Preparation Scale ≥ 6[6], n (%)
|
3104 (92.0)
|
|
Number of patients with polyps, n (%)
|
|
|
822 (24.4)
|
|
|
1684 (49.9)
|
|
|
2283 (67.7)
|
|
|
2477 (73.4)
|
ASA, American Society of Anesthesiologists; FIT, fecal immunochemical test; CRC, colorectal
cancer.
1 Missing = 9 (0.3 %).
2 Missing = 1936 (57.4 %), information on the family history of CRC was only available
for patients from CHUM center.
3 Missing = 2 (0.06 %).
4 Other indications included surveillance due to family history of CRC, pre- and post-graft
or organ donation, change in bowel habits such as constipation, post-polypectomy surveillance,
screening for inflammatory diseases, ruling out diverticulitis, abdominal pain, celiac
disease follow-up.
5 Missing = 2 (0.06 %).
6 Missing = 11 (0.3 %).
Table 2
Characteristics of the detected polyps stratified by size.
|
Clinicopathological characteristics of polyps
|
Polyp size cutoff
|
|
1–3 mm
|
1–5 mm
|
1–10 mm
|
|
Number of polyps, n/N (%)
|
3278/7655 (42.8)
|
5906/7655 (77.2)
|
7291/7655 (95.2)
|
|
Anatomical location, n (%)
|
|
|
2432 (74.2)
|
4448 (75.3)[1]
|
5472 (75.1)[2]
|
|
|
846 (25.8)
|
1458 (24.7)
|
1819 (24.9)
|
|
Polyp size, mean (SD), mm
|
2.4 (0.6)
|
3.4 (1.2)
|
4.2 (2.0)
|
|
Histopathology results, n (%)
|
|
|
738 (22.5)
|
1259 (21.3)
|
1453 (19.9)
|
|
|
1997 (60.9)
|
3718 (63.0)
|
4648 (63.7)
|
|
|
11 (0.3)
|
24 (0.4)
|
64 (0.9)
|
|
|
2 (0.1)
|
5 (0.1)
|
8 (0.1)
|
|
|
3 (0.1)
|
4 (0.1)
|
10 (0.1)
|
|
|
70 (2.1)
|
200 (3.4)
|
343 (4.7)
|
|
|
–
|
1 (0.02)
|
2 (0.03)
|
|
|
457 (13.9)
|
695 (11.8)
|
763 (10.5)
|
|
Polyps with advanced pathology[3], n (%)
|
16 (0.5)
|
34 (0.6)
|
84 (1.2)
|
|
Serrated lesions[4], n (%)
|
73 (2.2)
|
204 (3.5)
|
353 (4.8)
|
1 Missing = 3 (0.1 %).
2 Missing = 5 (0.1 %).
3 Including tubulovillous adenoma and villous adenoma, traditional serrated adenoma,
polyp with high grade dysplasia and cancer.
4 Including sessile serrated adenoma, traditional serrated adenoma.
Proportion of polyps with advanced pathology in the respective groups
Among polyps sized 1–3 mm, 1–5 mm, and 1–10 mm, 0.5 %, 0.6 %, and 1.2 % of polyps,
respectively, were found to have advanced pathology ([Table 2]). Significant differences were noted in advanced histopathology proportions when
comparing the 1–3 mm group vs. 1–10 mm group and 1–5 mm vs. 1–10 mm groups (Table 2 s, Table 3 s).
Primary outcome
When using optical diagnosis for polyps 1–3 mm, 1–5 mm, and 1–10 mm, the number of
patients with advanced adenomas undergoing optical polyp diagnosis (n = 79) resulting
in delayed surveillances of either 2 or 7 years would have been 3 (3.8 %), 12 (15.2 %),
and 20 (25.3 %), respectively ([Table 3]). For both surveillance delay durations, the differences between polyps sized 1–3 mm,
1–5 mm, and 1–10 mm, were statistically significant (Table 2 s).
Table 3
Number of patients with surveillance delays for 79 patients with advanced pathology.
|
Patients with advanced polyps
|
No delay, n (%)[1]
|
2-year delay, n (%)[1]
|
7-year delay, n (%)[1]
|
Total with delay, n (%) [95 %CI]
|
|
1–3 mm (n = 14)[2]
|
11 (13.9)
|
0 (0)
|
3 (3.8)
|
3 (3.8) [0.008–0.1]
|
|
1–5 mm (n = 32)[3]
|
20 (25.3)
|
2 (2.5)
|
10 (12.7)
|
12 (15.2) [0.1–0.2]
|
|
1–10 mm (n = 79)[4]
|
59 (74.6)
|
3 (3.8)
|
17 (21.5)
|
20 (25.3) [0.2–0.4]
|
1 Compared with surveillance intervals based on pathology results.
2 Missing = 2 (2.5 %).
3 Missing = 4 (5.1 %).
4 Missing = 8 (10.1 %).
In patients for whom the optical diagnosis of 1–3 mm polyps resulted in either a 2-
or 7-year delay compared with the surveillance intervals calculated based on the pathology
results (n = 3), 33.3 % (1/3) and 66.7 % (2/3) of delays were due to misdiagnosis
of an adenoma and villous component, respectively.
In the patients for whom the optical diagnosis of 1–5 mm polyps resulted in either
a 2- or 7-year delay compared with the surveillance intervals calculated based on
the pathology results (n = 12), 16.7 % (2/12) and 83.3 % (10/12) of delays were due
to misdiagnosis of an adenoma and villous component, respectively.
In the patients for whom the optical diagnosis of 1–10 mm polyps resulted in either
a 2– or 7-year delay compared with the surveillance intervals calculated based on
the pathology results (n = 20), 10 % (2 /20) and 90 % (18 /20) of delays were due
to misdiagnosis of an adenoma and villous component, respectively.
Surveillance interval agreements
Surveillance interval agreements are presented in [Fig. 1]. In the whole cohort of patients (n = 3374), the agreement between surveillance
intervals based on the high confidence optical diagnosis of polyps 1–3 mm and pathology-based
recommendations was 97.2 % (95 %CI 0.97–0.98). Moreover, the agreements between high
confidence optical diagnosis with polyp size cutoffs of 1–5 mm and 1–10 mm and pathology-based
recommendations were 95.5 % (95 %CI 0.95–0.96) and 94.2 % (95 %CI 0.93–0.95), respectively
(all P < 0.001) ([Fig. 1]).
Fig. 1 The percentage of surveillance interval agreement (correct assignment of surveillance
intervals) between histopathology and optical diagnosis in all patients with valid
determination of surveillance intervals, and in patients for whom optical diagnosis
could have affected their next recommended surveillance intervals, using the Narrow-band
imaging International Colorectal Endoscopic (NICE) classification system and different
cutoff points for size.
In the cohort in which patients with normal colonoscopy, polyps > 10 mm, and poor
bowel preparation were excluded, the agreements between surveillance intervals based
on the high confidence optical diagnosis of polyps 1–3 mm, 1–5 mm, and 1–10 mm and
pathology-based recommendations were 96.2 % (95 %CI 0.95–0.97), 93.6 % (95 %CI 0.92–0.95),
and 92.1 % (95 %CI 0.91–0.93), respectively. The agreements between polyps 1–3 mm
and 1–5 mm, between 1–3 mm and 1–10 mm, and between 1–5 mm and 1–10 mm were different
(P < 0.001).
The details of surveillance interval agreements are presented in Table 4 s and Table 5 s.
NPV for neoplastic rectosigmoid polyps
Overall, 16.4 %, 73.3 %, and 8.2 % of polyps 1–10 mm in size were optically predicted
as hyperplastic (NICE 1), adenoma (NICE 2), and SSLs, respectively (Table 6 s).
The NPV of optical diagnosis for diagnosing rectosigmoid neoplastic polyps was as
follows: 1–3 mm: 81.4 % (95 %CI 78.0–84.4); 1–5 mm: 80.9 % (95 %CI 78.0–83.6); 1–10 mm:
80.6 % (95 %CI 77.7–83.3) ([Table 4]). Moreover, the accuracy of optical diagnosis for distinguishing neoplastic from
hyperplastic polyps (regardless of polyp location) was only moderate for all three
polyp size groups (1–3 mm: 78.3 % [95 %CI 76.7–79.9]; 1–5 mm: 80.3 % [95 %CI 79.2–81.4];
1–10 mm: 81.0 % [95 %CI 80.0–82.0]) ([Table 4]).
Table 4
Diagnostic performance of optical diagnosis[*] for differentiating hyperplastic from adenomatous polyps in patients with at least
one polyp 1–3 mm, 1–5 mm, 1–10 mm in size, respectively.
|
Polyp size cut-off
|
Diagnostic properties (adenoma vs. hyperplastic)
|
|
In rectosigmoid polyps
|
In all polyps
|
|
%
|
95 %CI
|
%
|
95 %CI
|
|
1–3 mm
|
|
|
73.8
|
68.0–79.0
|
88.0
|
86.5–89.4
|
|
|
66.9
|
61.4–70.3
|
52.3
|
48.6–55.9
|
|
|
55.4
|
51.8–59.0
|
83.3
|
82.2–84.3
|
|
|
81.4
|
78.0–84.4
|
61.7
|
58.4–64.9
|
|
|
68.8
|
65.2–72.2
|
78.3
|
76.7–79.9
|
|
1–5 mm
|
|
|
78.9
|
75.1–82.4
|
91.4
|
90.5–92.3
|
|
|
59.6
|
56.0–63.2
|
47.4
|
44.6–50.2
|
|
|
56.5
|
54.1–58.9
|
83.8
|
83.0–84.5
|
|
|
80.9
|
78.0–83.6
|
65.0
|
62.2–67.7
|
|
|
67.3
|
64.6–69.9
|
80.3
|
79.2–81.4
|
|
1–10 mm
|
|
|
83.8
|
80.8–86.4
|
92.6
|
91.8–93.4
|
|
|
54.9
|
51.5–58.3
|
43.3
|
40.7–45.9
|
|
|
60.1
|
58.2–62.0
|
84.1
|
83.5–84.7
|
|
|
80.6
|
77.7–83.3
|
64.4
|
61.7–67.1
|
|
|
67.8
|
65.5–70.1
|
81.0
|
80.0–82.0
|
PPV, positive predictive value; NPV, negative predictive value; NICE, Narrow-band
imaging International Colorectal Endoscopic classification.
* Optical diagnosis using the NICE classification system and image-enhanced endoscopy.
Sessile serrated polyps/adenomas were not considered in the analysis.
Reduction in histopathology examinations and allocation of immediate surveillance
intervals
Use of optical diagnosis would have resulted in a 33.5 % (95 %CI 0.32–0.35), 62.3 %
(95 %CI 0.61–0.63), and 78.2 % (95 %CI 0.77–0.80) reduction in histopathology examinations
for polyps of 1–3 mm, 1–5 mm, and 1–10 mm, respectively ([Fig. 2]). Furthermore, optical diagnosis could have allowed 41.0 % (95 %CI 0.39–0.43), 58.2 %
(95 %CI 0.56–0.60), and 73.3 % (95 %CI 0.72–0.75) of patients, respectively, to be
given immediate, same-day surveillance interval recommendations. These proportions
were greater than the corresponding proportions if the recommendations were followed
based on pathology outcomes (P < 0.001 for all) (Table 2 s).
Fig. 2 Reduction in histopathology examinations and proportion of patients who could have
received immediate surveillance interval recommendations.
Discussion
In this study, which included 3374 patients with 7655 polyps undergoing optical diagnosis,
we found that when limiting optical diagnosis to polyps sized 1–3 mm, the proportion
of delayed follow-up assignments in patients who had polyps with advanced pathology
was exceedingly low. Only a few polyps with serrated or villous pathology were found
in the 1–3 mm group (n = 73; 2.2 %). If the optical diagnosis is limited to 1–3 mm
polyps, the proportion of delayed surveillance intervals for patients with advanced
neoplastic polyps is lower compared with using optical diagnosis for polyps up to
5 mm or up to 10 mm. Using optical diagnosis for 1–3 mm polyps exclusively resulted
in only 0.5 % of advanced neoplastic polyps and only three patients (3.8 %) with a
7-year delay in the next surveillance colonoscopy. In contrast, when 4–10 mm polyps
were included in the optical evaluation, 1.2 % of polyps had advanced pathology, and
3 (3.8 %) and 17 (21.5) patients had a 2-year and 7-year delay in their next surveillance
colonoscopy, respectively. As the proportion of advanced pathology increases with
polyp size (P < 0.001), so does the rate of inappropriately delayed surveillance intervals.
Notably, we considered adenomas with a villous component as adenomas with advanced
pathology. However, some studies have found no association between villous adenomas
and an increased risk of neoplasia [23]
[24]. The latest European Society of Gastrointestinal Endoscopy guidelines [7] do not consider polyps with a villous component as “advanced” polyps. However, the
2020 USMSTF guidelines on which we based our study still consider these polyps as
advanced. When villous polyps are excluded from advanced pathology criteria, the surveillance
delays for 11 patients with advanced pathology were 9.1 %, 18.2 %, and 18.2 % for
3-, 5-, and 10-mm cutoffs, respectively.
It is critically important that colorectal adenocarcinomas are not inappropriately
discarded when using the “resect and discard” strategy. It is often recommended to
use NICE 3 classification for flat-depressed or ulcerated morphology (Paris IIc and
III) to potentially identify adenocarcinomas among small polyps. A recent analysis
by Vleugels et al. evaluated optical diagnosis for polyps up to 10 mm and found that
it would have resulted in five T1 cancers being discarded without histopathology evaluation
and appropriate management [11]. In this study, the prevalence of T1 cancers among polyps 1–10 mm was 0.33 % [11]. All cancers had Ip or Is morphology and were often judged through optical diagnosis
as NICE 2 adenomas [11]. Thus, in this study, as in our current cohort where no cancer was recorded, no
correlation between NICE 3 and Paris IIc/III morphology was found for the diagnosis
of adenocarcinoma. In our cohort, out of 5346 polyps predicted to be adenomas in the
1–10 mm polyps, 763 (14.3 %) were evaluated to be hyperplastic or SSL during histopathology
examination. We did not encounter any adenocarcinoma among the polyps sized 1–10 mm.
Thus, the best approach seems to be using a smaller cutoff to limit the risk of mismanaging
advanced colorectal neoplasia within a “resect and discard” strategy.
Starting optical diagnosis at the low threshold of 1–3 mm might be feasible and ensure
a cost-effective and safe approach to implementing the “resect and discard” strategy
in routine clinical practice. Although the highest reduction in pathology examinations
was naturally found when optical diagnosis was expanded to include 1–10 mm polyps
(78.2 %), limiting optical diagnosis to 1–3 mm polyps significantly reduced the need
for pathology examinations (33.5 %), as well as increasing the safety profile compared
with larger size cutoffs. Furthermore, a significant proportion of patients could
have received an immediate surveillance recommendation, even when limiting optical
diagnosis to 1–3 mm polyps (73.3 % in the 1–10 mm group vs. 41.0 % in the 1–3 mm group).
The results of our study support the use of optical diagnosis for 1–3 mm polyps especially
in light of recent evidence indicating the unreliability of histopathology assessment
for this polyp size group [25]. A recent study comparing optical diagnosis of 1–3 mm polyps with histopathology
outcomes found that about 15 % of polyps recorded as adenoma by optical diagnosis
were reported as normal mucosa by pathology experts [25]. Another study reported a similar discrepancy, with 28.9 % of 1–3 mm polyps having
mismatched optical and pathological diagnoses [12]. These findings suggest that high confidence optical diagnosis is a safe method
for accurate adenoma identification for 1–3 mm polyps, given the potential for pathology
evaluations to erroneously report adenomatous polyps as normal mucosa. Furthermore,
multiple recent studies have identified interrater variability between pathologists,
or that expert high confidence diagnoses of 1–3 mm polyps matched interpretation assisted
by artificial intelligence (AI) but not the pathology results. Polyps previously diagnosed
as hyperplastic might be reclassified as adenoma or sessile serrated adenomas after
slide reassessment by another pathologist [11]
[26]
[27]
[28]
[29].
The appropriate size cutoff for optical diagnosis is also relevant for future developments
in AI-assisted optical diagnosis. AI-assisted optical diagnosis has improved detection
with promising accuracy [30]
[31]. Despite recent research efforts to improve the diagnostic precision of AI models,
similarly to regular optical diagnosis, it cannot distinguish between different adenoma
entities such as high vs. low grade dysplasia, or reliably identify serrated or villous
pathology. Limiting optical diagnosis to 1–3 mm polyps will help decrease the risk
of inappropriate management of advanced adenomas, regardless of the optical diagnosis
modality used.
Some strengths and limitations of this study should be mentioned. To our knowledge,
this is the first study to evaluate polyp size cutoffs for implementation of optical
diagnosis. We included the data from two academic centers with variable endoscopist
experience in optical diagnosis, reflecting real-world practice. Thus, it is possible
to cautiously generalize the results to community practices. Study limitations include
the post hoc nature of the analysis, and the fact that polyp size was based on endoscopists’
measurements. Endoscopists tend to overestimate polyp size compared with the size
measured during pathological examination [32]
[33]. However, the method reflects general clinical practice and will remain a limitation
until better techniques are widely available to improve real-time polyp measurement
during colonoscopy. Additionally, the NICE classification system does not accurately
distinguish SSLs from hyperplastic polyps, resulting in misclassification of some
polyps. Other optical diagnosis classification systems were not used because of the
multiplicity of centers. The NICE classification has not been validated for blue-light
imaging; however, there was no decrease in diagnostic performance when compared with
other optical imaging techniques [34]. One major limitation is the biased calculation of surveillance intervals due to
the lack of data on the family history of colorectal cancer for patients from the
VA Medical Center. Consistent with other studies [11]
[35], our study did not reach the American Society for Gastrointestinal Endoscopy Preservation
and Incorporation of Valuable endoscopic Innovations (PIVI) benchmark of an NPV of
≥ 90 % to support the use of a “diagnose and leave” strategy for rectosigmoid polyps
≤ 10 mm [9]. In a sub-analysis of per-endoscopist NPVs, only six expert endoscopists reached
the recommended PIVI benchmark for implementation of this strategy in each size group
(Table 7 s).
In conclusion, this study showed that limiting optical diagnosis to polyps sized 1–3 mm
resulted in an excellent safety profile with a very low risk of inappropriate management
of advanced adenomas, which makes routine clinical implementation of the “resect and
discard” strategy feasible. Implementing a 3-mm cutoff could be a starting point for
endoscopists to feel comfortable with the “resect and discard” strategy, with the
potential of implementing a 5-mm cutoff, once optical diagnosis becomes more popular,
and endoscopists become more comfortable with its use.
