Introduction
Quantitative measurement of the detection skills of gastroenterology fellows performing
colonoscopy is challenging. Fellows are typically supervised by attending doctors
during colonoscopy; therefore, lesion detection generally reflects the combined detection
efforts of a fellow and an attending [1].
One way to quantitatively assess detection skills is to allow the fellow to perform
a complete examination of the colon with resection of all lesions detected and then
to have the attending doctor perform a complete second examination of the entire colon.
This allows calculation of a miss rate for the fellow. In our experience, using this
tandem colonoscopy format is instructive for fellows. Tandem colonoscopy creates an
increased appreciation for the challenges of detection during colonoscopy.
This approach can also lead to identification of specific weaknesses among individual
fellows or within a group of fellows that could direct additional specific instruction
designed to improve detection. In this report, we describe our experience with fellow
miss rates and types of lesions not identified in 100 consecutive tandem examinations.
The initial colonoscopy was performed by the fellow without interference by the supervising
colonoscopist, and then followed by a second examination by a single expert colonoscopist
with a known high detection rate for both adenomas and sessile serrated lesions (SSLs).
Patients and methods
The participating gastroenterology fellows were in their second or third year of training
at a single major US academic institution. Fellows perform about 200 colonoscopies
in their first year.
Ten different fellows performed the initial examinations. We conducted these examinations
as an educational and quality control exercise. The Indiana University Institutional
Review Board determined that oversight was not required for this data analysis. Patients
were selected randomly for inclusion. Fellows were generally not instructed that their
performance was going to be assessed in an individual patient. However, inclusion
of the patient in the assessment was generally obvious to the fellow from the absence
of detailed supervision by the attending doctor. Only cases in which the fellow examined
and cleared the entire colon without comment or interruption by the attending doctor
were included. In many patients, the attending doctor repeated segmental examinations
by the fellow, but none of those patients were included in this report. The fellow
was allowed to accept input from any technicians or nurses observing the examination,
and all lesions detected by technicians and nurses were counted as detections by the
fellow. After retroflexion in the rectum to complete the fellow’s examination, either
the fellow or the attending doctor reinserted the colonoscope to the cecum. Any lesion
detected on the second insertion was counted as a miss for the fellow’s examination,
as was any lesion detected by the attending doctor on the second withdrawal. The same
colonoscope, as well as any add-on attachment such as Endocuff Vision (Olympus Corp.,
Center Valley, Pennsylvania, United States), were used for both the first and the
second examinations.
For each examination, we separately recorded the time for withdrawal (which included
time for inspection, cleaning, and biopsy or polypectomy), and the size, shape, and
location of each lesion as estimated by the endoscopist. Lesions of the same apparent
histologic group and in the same section of the colon were placed in the same bottle
for histologic examination in the first colonoscopy. This process was repeated in
the second colonoscopy in separate bottles from the first colonoscopy. Lesions in
the serrated class located in the rectosigmoid and ≤ 5 mm were noted by the first
endoscopist and if detected at the second endoscopy, were not counted as missed lesions.
Outcome definitions
We measured the adenoma miss rate (AMR), defined as the number of conventional adenomas
discovered by the attending doctor alone divided by the total number of conventional
adenomas discovered during the procedure. Other measures included adenoma detection
rate (ADR), defined as the proportion of colonoscopies in which at least one adenoma
was found, and adenomas per colonoscopy (APC), defined as total number of adenomas
discovered divided by total number of colonoscopies. Similar definitions were used
for SSL miss rate, SSL detection rate (SSLDR), and SSL per colonoscopy (SSLPC). We
accepted the clinical pathologists’ diagnosis of adenoma, SSL, and hyperplastic polyp.
Statistical analysis
We report the descriptive characteristics of colonoscopy procedures, absolute number
of lesions detected, fellow miss rates, and detection rates with 95 % confidence intervals
(CIs). We used Jeffrey’s binomial procedure to calculate the confidence intervals
for proportions. We calculated confidence intervals for APC/SSLPC using standard error
of the mean. Withdrawal times were compared using related samples Wilcoxon rank test.
The level of significance was set at 0.05. All analyses were conducted using SPSS
27 (IBM, New York, United States).
Results
One hundred patients underwent tandem examinations. The average age of the cohort
was 63.5 years (standard deviation [SD] 11.3) years and 58 were men. Eighty-six were
white and 24 had a family history of colorectal cancer ([Table 1]). [Table 1] shows indications for the procedures along with the use of adjunct devices during
the colonoscopy. The median Boston Bowel Preparation Scale score was 9 (interquartile
range [IQR], 9–9). Median withdrawal time for fellows was 17.01 minutes (IQR, 12.2
to 25.0) compared with the attending, 8.36 minutes (IQR, 6.6 to 11.2) (P < 0.001).
For examinations in which no polyps were removed, median withdrawal time for fellows
was 12 minutes (IQR, 9.02 to 12.8) vs. 7 minutes (IQR, 4.4 to 8.5) for the attending
doctor (P = 0.005). The mean number of first examinations per fellow was 10.
Table 1
Patient demographics, procedure indications and use of adjuncts during the procedure
|
Male gender
|
58
|
|
Age, mean (SD, range)
|
63.5 (11.3, 23–88)
|
|
Race
|
|
|
86
|
|
|
8
|
|
|
6
|
|
Family history of CRC
|
24
|
|
Indication
|
|
|
15
|
|
|
43
|
|
|
25
|
|
|
7
|
|
|
5
|
|
|
4
|
|
|
1
|
|
Endocuff use
|
75
|
|
NBI use
|
13
|
|
Cap use
|
7
|
SD, standard deviation; CRC, colorectal cancer; NBI, narrow band imaging.
Fellows detected at least one conventional adenoma in 58 patients (95 %CI, 48.2 to
67.3). The attending doctor detected at least one conventional adenoma in 41 patients
during the second examination and the overall ADR was 69 (95 %CI, 59.5 to 77.4). Fellows
detected a SSL in three patients (95 %CI, 0.9 to 7.8). The attending doctor detected
at least one SSL in 15 patients and the overall SSLDR was 18 (95 %CI, 11.4 to 26.4).
[Table 2] shows the number of lesions detected in the first colonoscopy and the second colonoscopy
according to histology and size. The overall miss rate for conventional adenomas was
30.5 % (95 %CI, 25.5 to 35.8) and for SSLs, 85.7 % (95 %CI, 69.5 to 95.0) (P < 0.001).
The miss rate among lesions ≥ 10 mm was 26.7 % (95 %CI, 9.7 to 51.7) for conventional
adenomas vs. 90.9 % for SSLs (95 %CI, 64.7 to 99.0) (P = 0.001). Ten of the 14 precancerous
lesions ≥ 10 mm that were missed were SSLs. Fellows also missed one of two hyperplastic
polyps ≥ 10 mm and half of 34 hyperplastic polyps 6 to 9 mm. SSLs were missed by seven
different fellows with no SSL identified in the patients colonoscoped by the other
three fellows. SSLs ≥ 10 mm were missed by six fellows with no SSL ≥ 10 mm identified
in the patients colonoscoped by the remaining four fellows. [Fig. 1] shows three examples of SSLs missed by fellows in the study.
Table 2
Detection by fellow and attending doctor according to size and histology.
|
Pathology
|
Size range
|
Found by fellow
|
Additional lesions found by attending doctor
|
Total
|
|
Splenic flexure or proximal
|
Distal to splenic flexure
|
Splenic flexure or proximal
|
Distal to splenic flexure
|
|
Hyperplastic polyp
|
1–5 mm
|
32
|
48
|
15
|
46
|
141
|
|
6–9 mm
|
2
|
15
|
5
|
12
|
34
|
|
≥ 10 mm
|
0
|
1
|
0
|
1
|
2
|
|
Sessile serrated lesion
|
1–5 mm
|
3
|
0
|
7
|
1
|
11
|
|
6–9 mm
|
0
|
0
|
4
|
2
|
6
|
|
≥ 10 mm
|
1
|
0
|
7
|
3
|
11
|
|
Conventional adenoma
|
1–5 mm
|
118
|
35
|
50
|
26
|
229
|
|
6–9 mm
|
30
|
18
|
6
|
7
|
61
|
|
≥ 10 mm
|
8
|
3
|
3
|
1
|
15
|
Fig. 1 Three of the large (≥ 10 mm) sessile serrated lesions missed by gastroenterology
fellows during colonoscopy. Each lesion is seen in white light (a, c, e) and narrow band imaging (b, d, f).
The fellows’ APC rate was 2.12 (95 %CI, 1.4 to 2.8) vs. 3.05 (95 %CI, 2.2 to 4.0)
for both examinations combined. The SSLPC of the fellows was 0.04 (95 %CI, 0 to 0.1)
which increased to 0.28 (95 %CI, 0.1 to 0.4) for both examinations.
Discussion
In this study we found that second- and third-year gastroenterology fellows missed
a significant percentage of conventional adenomas, and a striking percentage of SSLs.
These data suggest that detection of SSLs may have a longer learning curve than detection
of conventional adenomas. Previous studies have suggested that the degree of missing
SSLs is higher among experienced endoscopists than the missing of conventional adenomas,
particularly in the proximal colon [2]
[3]. In addition, SSLs contribute disproportionately to interval cancers [4]
[5]
[6]
[7]
[8]
[9]. Our data suggest that the problem of missing SSLs is present during fellowship,
and efforts to correct the problem should be undertaken during fellowship.
The overall miss rate for conventional adenomas in the study is not substantially
different from that identified in previous tandem studies of colonoscopy. In two meta-analyses
of tandem studies, an overall miss rate for adenomas of 22 % to 26 % was identified
[10]
[11]. It should be remembered that the absolute miss rate calculated for any detection
method will be affected by the detection capacity of the second method. In this case,
the second colonoscopy was performed in all cases by an endoscopist with a high detection
rate for both adenomas and SSLs [12]. Given this, the miss rate for conventional adenomas by the second- and third-year
fellows in this study is not out of line with previous miss rate studies of practicing
endoscopists [13]
[14]
[15]
[16]. Further, the ADR recorded by the fellows as well as the rate of APC were both relatively
high compared to previous studies, suggesting that the conventional adenoma detection
skills of the fellows were acceptable for practicing endoscopists. Despite that, the
detection of SSLs by gastroenterology fellows was quite poor.
We performed second examinations in this study as an instruction tool for gastroenterology
fellows. Anecdotally, the fellows consistently felt that the exercise was valuable
for impressing upon them the need to perform careful and detailed withdrawal examinations
and to improve their recognition of the subtle lesions. Beyond the value for individual
fellows, the exercise indicated the need to create an atlas of serrated class lesions
for our gastroenterology fellows to study, to generally improve their skill in detecting
these lesions [17].
The strengths of the study include that we performed a second exam of the entire colon
rather than segments. Also, the second examination was performed by an endoscopist
with a high rate of detection. Absolute rates of missed lesions by fellows could vary
if second examinations are performed by multiple attendings with variable detection
skills. Use of a single high-detecting attending to perform second examinations will
elevate the observed miss rates, but does not diminish the central observation that
SSL miss rates by fellows are high compared to AMRs. Limitations include the single-center
design, which might limit generalizability.
It is possible that artificial intelligence (AI), which at this writing has just recently
been approved by the Food and Drug Administration for use in the United States in
clinical practice, will eliminate the need for careful training in lesion detection.
This remains uncertain at this time. Some recent data suggest that current AI programs
may sometimes have difficulty recognizing subtle SSLs [18].
Conclusions
In summary, we found through use of tandem examinations designed to improve fellow
instruction that detection of SSLs by second- and third-year gastroenterology fellows
was poor, and substantially worse than detection of adenomas. Special training in
detection of SSLs is warranted for gastroenterology fellows, as well as testing of
training methods to establish their efficacy.
