Impact of endoscopist withdrawal speed on polyp yield: implications for optimal colonoscopy withdrawal time
Abstract
Summary
Background
In 2002, a U.S. Multi-Society Task Force on Colorectal Cancer recommended that the withdrawal phase for colonoscopy should average at least 6–10 min. This was based on 10 consecutive colonoscopies by two endoscopists with different adenoma miss rates.
Aims
To characterize the relationship between endoscopist withdrawal time and polyp detection at colonoscopy, and to determine the withdrawal time that corresponds to the median polyp detection rate.
Design
Procedural data from out-patient colonoscopies performed at the Mayo Clinic, Rochester during 2003 were reviewed. Endoscopists were characterized by their mean withdrawal time for a negative procedure and individual polyp detection rate.
Results
A total of 10 955 colonoscopies performed by 43 endoscopists were analysed. Median withdrawal time was 6.3 min (range: 4.2–11.9); polyp detection rate was 44.0% (all polyps), 29.8% (≤5 mm), 5.9% (6–9 mm), 6.7% (10–19 mm), 2.1% (≥20 mm). Longer withdrawal time was associated with higher polyp detection rate (r = 0.76; P < 0.0001); this relationship weakened for larger polyps (r = 0.19 for polyps 6–9 mm, r = 0.28 for polyps 10–19 mm, r = 0.02 for polyps ≥20 mm). Overall median polyp detection rate corresponded to a withdrawal time of 6.7 min.
Conclusion
Our findings support a colonoscopy withdrawal time of at least 7 min, which correlates with higher colon polyp detection rates.
Introduction
Colon cancer is the second leading cause of cancer-related death in the United States. Routine screening, beginning at age 50 years reduces mortality from colon cancer.1–5 Published studies have demonstrated colonoscopy to be the most effective available screening tool in detecting colon neoplasia, surpassing flexible sigmoidoscopy and double-contrast barium enema.6, 7 Accordingly, in July 2001, Medicare coverage for screening colonoscopy was introduced. Since that time, there has been a dramatic increase in the performance of screening colonoscopies.8, 9
Despite its utility, colonoscopy has inherent limitations, some of which are operator-dependent. Several published studies have reported colonoscopy adenoma miss rates ranging from 6% to 27%.10–14 There are several factors that can hinder adenoma detection. The skill of the endoscopist affects the thoroughness of the examination. Extensive training and experience are necessary to maximize the accuracy and safety of colonoscopy. An inadequate bowel preparation can limit the visualization of the colonic mucosa. A few patients are unable to tolerate a complete colonoscopy despite usual amounts of sedation and analgesia.
Identification of premalignant polyps depends on the quality of the endoscopist's examination of the colonic mucosa. By convention, this inspection is performed primarily during the withdrawal phase of colonoscopy. In 2002, a U.S. Multi-Society Task Force on Colorectal Cancer recommended that the withdrawal phase for colonoscopy, excluding time required for biopsy and polypectomy, should average at least 6–10 min.15 However, this recommendation was based on limited evidence – observation of 10 consecutive colonoscopies by two endoscopists, one with a known higher adenoma miss rate than the other.16
The aim of this study was to assess endoscopists’ procedural ‘speed’ and use this speed as a predictor of endoscopists’ likelihood of detecting polyps. Are endoscopists who typically perform colonoscopies more quickly less likely to detect polyps? Are those who perform colonoscopies more slowly any more likely to find polyps? Is there an ideal speed that maximizes polyp detection? Our objective was to characterize the relationship between endoscopist withdrawal time and neoplasia detection at colonoscopy. Using this data, the withdrawal time that corresponded to the median polyp detection rate (PDR) was calculated in order to derive evidence-based recommendations for minimum acceptable withdrawal time.
Methods
Study setting
Procedural data from all routine colonoscopies performed in the out-patient endoscopy unit at Mayo Clinic, Rochester between 1 January 2003 and 31 December 2003 were reviewed using the Mayo Endoscopic database. This is an institutional computerized database, which captures endoscopic procedural information for all endoscopic procedure performed at the Mayo Clinic, Rochester. Comprehensive patient information before, during and postprocedure and procedure details, are recorded. A trained endoscopy nurse enters data relayed verbally by the doctor while the procedure is being performed. For example, when the endoscope is first inserted into the rectum, the nurse will press a function key that will stamp ‘time of insertion’ into the database. Similar time stamps mark ‘extent of exam’ and ‘time of withdrawal’. The stamped times are downloaded into a spreadsheet and procedure time and endoscope insertion and withdrawal time can be calculated. Multiple mandatory fields requiring data entry minimize incomplete data collection and enhance the quality control of this database.
To validate the accuracy of the procedure times recorded in the database, the insertion and withdrawal times for 78 colonoscopies were prospectively recorded by the endoscopist while the nurse independently entered endoscope insertion, maximal extent and withdrawal times into the database, per our usual practice. The observed times were obtained by the endoscopist using a clock in the endoscopy room. The observed times were compared with the corresponding times recorded in the database.
At our institution, patients referred from elsewhere for management of known colonic polyps are scheduled on a separate ‘complex endoscopy’ schedule that allots more time per procedure. For the purposes of this study, only procedures performed on the ‘routine endoscopy’ schedule were analysed. Procedures in which the examination was incomplete (caecum not reached), bowel preparation was suboptimal, patient had prior colonic resection, or a trainee was involved were excluded from the analysis. Study approval was obtained from the Institutional Review Board.
Patient demographic information, polyp detection and polyp size were analysed. Endoscopists’ visual estimate was used to measure polyp size. Occasionally, an open biopsy forceps (6 mm) was used to calibrate polyp size although this was at the discretion of the individual endoscopist. For colonoscopies in which multiple polyps were detected, the size of the largest lesion was recorded for analysis. Polyp size was classified as 1–5 mm, 6–9 mm, 10–19 mm, or ≥20 mm. The PDR (number of colonoscopies with polyp detected/total number of colonoscopies performed) was calculated for each endoscopist. Colonoscopy volume (number of colonoscopies performed fulfilling study criteria) and experience level (number of years since initial Gastroenterology Board certification or foreign equivalent) were also recorded for each endoscopist.
Time taken for colonoscope withdrawal (from caecum to anal verge) for normal colonoscopic examinations (i.e. no polyps detected, no therapy performed) was extracted from the database for each procedure. Haemorrhoids and diverticula were not excluded since their detection was considered unlikely to prolong colonoscopy withdrawal time. Analysis was restricted to those endoscopists who performed a minimum of 50 negative colonoscopies during the study. This ensured the standard error for mean procedure withdrawal time was <0.5 for all colonoscopists in the study. Minimizing the variance in withdrawal time provided a more precise estimate of each endoscopist's true withdrawal speed.
One of the major limitations of the endoscopic database is the lack of histological data to fully characterize polyps. To counteract this limitation, polyps were categorized according to size (1–5 mm, 6–9 mm, 10–19 mm, ≥20 mm) and 50 polyps from each of the four size categories were randomly selected with pathology reports reviewed to characterize polyp histology (e.g. hyperplastic, adenomatous, malignant).
Statistical analysis
Mean and median values with ranges were described for the recorded data. Data were analysed using jmp software (SAS Institute Inc., Cary, NC, USA). Pearson's correlation coefficient (r) served to measure the linear relationship between withdrawal time and polyp detection. For the purposes of correlation coefficient calculation, PDRs per endoscopist were related to the endoscopists’ mean withdrawal time so that two data points (a rate and a time) arose for each endoscopist.
A multivariate model was developed to gauge the independent effects of withdrawal time on polyp detection while adjusting for the effects of experience level and procedure volume. Separate regression models were developed for polyps of differing sizes. The outcome variable of the model (PDR per endoscopist) was dichotomized based on greater or less than the median yield of all endoscopists. Odds ratios and their 95% confidence intervals served to describe the influence of withdrawal time on polyp detection.
Results
Patient population
Between 1 January and 31 December 2003, 11 444 colonoscopies that met our study criteria were performed by 49 endoscopists. Of these colonoscopies, 10 995 were performed by 43 endoscopists who each carried out at least 50 negative colonoscopies. Overall, 54.4% of patients were male, mean age was 62.5 years. The demographic features of this patient cohort are shown in Table 1. Polyps were detected in 4837 (44.0%) of these 10 995 procedures; greatest polyp size was ≤5 mm in 3275 (29.8%) procedures, 6–9 mm in 644 (5.9%), 10–19 mm in 499 (6.7%), ≥20 mm in 235 (2.1%) and unspecified size in 184 (1.7%) patients (Figure 1).
| Total colonoscopies | 10 995 |
| Mean age (S.E.) | 62.5 (0.12) |
| Gender (male) | 54.4% |
| Colonoscopies, n (%) | |
| Without polyps | 6158 (56.0) |
| With polyps | 4837 (44.0) |
| Polyp size [mm; n (%)] | |
| ≤5 | 3,275 (29.8) |
| 6–9 | 644 (5.9) |
| 10–19 | 499 (6.7) |
| ≥20 | 235 (2.1) |
| Size unspecified | 184 (1.7) |
Illustration of mean polyp detection rates for all endoscopists according to polyp size.
Endoscopist characteristics
The average endoscopist withdrawal time for normal diagnostic procedures varied from 4.2 to 11.9 min (median: 6.3 min). Median endoscopist experience level was 9 years (range: 1–25) while median annual colonoscopy volume was 242 cases (range: 103–580).
Polyp detection rates
The endoscopist PDR varied from 23.6% to 65.6% of cases (median 42.7%). This median PDR corresponded to a withdrawal time of 6.7 min (Figure 2), with a close correlation between overall PDR and mean endoscopist withdrawal time (r = 0.76, P < 0.0001). The 25th percentile PDR correlated with a withdrawal of 5.1 min; 75th percentile correlated with a withdrawal of 8.8 min; 90th percentile correlated with a withdrawal of 11.9 min (Figure 2).
Relationship of mean withdrawal time per endoscopist and polyp yield. Horizontal axes illustrate 25th, 50th, 75th and 90th percentiles for polyp detection rates and their intersection with withdrawal times.
On multivariate analysis, longer withdrawal time [>6.3 min (median)] was a statistically significant independent predictor of a higher PDR (OR = 11.8, 95% CI: 2.3–78.4, P = 0.005). Greater endoscopist experience was associated with lower PDRs (r = −0.39, P = 0.009). This finding appeared to be driven primarily be detection of smaller polyps, i.e. as polyp size increased, the inverse relationship between detection rate and experience weakened [r = −0.32 (polyps ≤5 mm), r = −0.23 (polyps 6–9 mm), r = −0.15 (polyps 10–19 mm) and r = 0.01 (polyps ≥20 mm)]. There was no association between procedure volume and PDRs (r = 0.12, P = 0.43).
Polyps of varying sizes
The PDR for polyps ≤5 mm varied from 9.1% to 55.2% (median 29.8%). There was a close correlation between the detection rate for polyps ≤5 mm and mean endoscopist withdrawal time (r = 0.67, P < 0.0001, Figure 3). The PDR for polyps 6–9 mm varied from 0.8% to 14.7% (median 5.2%). There was no significant relationship between PDR and mean withdrawal time for this size category (r = 0.19, P = 0.23). The PDR for polyps 10–19 mm varied from 1.4% to 10.1% (median 4.6%). There was no significant relationship between PDR and mean withdrawal time for this size category (r = 0.28, P = 0.07). The PDR for polyps ≥20 mm varied from 0% to 7.9% (median 1.6%). There was no significant relationship between PDR and mean withdrawal time for this size category (r = 0.02, P = 0.88).
Proportion of adenomas among polyps of various sizes.
Validation of procedure times
The withdrawal times (r = 0.88, P < 0.0001) and insertion times (r = 0.97, P < 0.0001) independently observed during colonoscopy (n = 78) correlated closely with those times recorded in the database.
Validation of polyp histology
Among a subset of 200 colonoscopies with polyps removed, adenomas became increasingly prevalent as polyp size increased, as illustrated in Figure 3. Adenomas were detected in 56% (28 of 50) of polyps ≤5 mm, 64% (32 of 50) of polyps 6–9 mm, 76% (38 of 50) of polyps 10–19 mm and 82% (41 of 50) of polyps ≥20 mm. A single cancer was found in the 6–9 mm category; four cancers were found among polyps ≥20 mm.
Discussion
This is the first study from a tertiary referral setting to provide evidence to support guidelines for a minimum acceptable withdrawal time at colonoscopy. As previous studies have demonstrated, a close relationship exists between colonoscopy time and polyp yield,17 with a lower yield observed among endoscopists with shorter withdrawal times. In the context of the dramatic rise in the performance of screening colonoscopy in the United States over the past several years,8 these findings have assumed increasing importance. Based on our findings, a withdrawal time of at least 7 min corresponds to a detection rate above the 50th percentile for polyps of all sizes. If a minimum withdrawal time of 9 min was recommended, although this would correspond to detection rates above the 65th percentile for polyps <20 mm, the additional 2 min spent on withdrawal would not increase the yield of larger polyps (still at 65th percentile) compared with a 7-min withdrawal, as shown in Table 2. In an attempt to balance ideal expectations and practical reality, a minimum withdrawal time of 7 min would appear reasonable.
| Withdrawal time (min) | Polyp yield | ||||
|---|---|---|---|---|---|
| All sizes | ≤5 mm | 6–9 mm | 10–19 mm | ≥20 mm | |
| 5 | 25th | 25th | 50th | 40th | 65th |
| 7 | 55th | 50th | 60th | 50th | 65th |
| 9 | 75th | 75th | 65th | 65th | 65th |
| 11 | 85th | 90th | 70th | 75th | 65th |
To date, the study from Sanchez et al. is the only publication to demonstrate a clear association between longer total procedure time at colonoscopy and increasing polyp yield.17 In that study the investigators used total procedure time for a negative colonoscopy as a surrogate marker for the real measure of interest, withdrawal time. In this study withdrawal time alone was used as the outcome parameter, which provided a more precise measure. The published literature addressing the specific impact of withdrawal time on neoplasia yield is similarly scarce. Rex and colleagues produced an important study in this area, although this was based on limited data.16 The withdrawal technique of two experienced endoscopists known to have differing PDRs was observed for 10 consecutive procedures; the colonoscopist with the lower miss rate had a longer withdrawal time (8 min, 55 s) compared with his colleague with a higher miss rate (6 min, 41 s). This disparity in procedure performance led experts to recommend that a minimum withdrawal time of 6–10 min be adopted for colonoscopy.10
The literature is replete with evidence that miss rates vary among colonoscopists.12–14 Of concern, experienced endoscopists failed to detect adenomas in approximately 25% of patients in the National Polyp Study.14 Two studies in which tandem colonoscopies were used as a mechanism of evaluating neoplasia miss rates have provided further evidence of the limitations of colonoscopy; miss rates for adenomas ≥10 mm varied from 0% to 6% while 12–13% of adenomas 6–9 mm and 15–27% of adenomas ≤5 mm were missed.12, 13 It would appear that inter-endoscopist disparity in polyp detection is greatest for small polyps. In the present study, most of the association of withdrawal time and polyp detection is driven by detection of smaller (≤5 mm) polyps. This most likely reflects that larger polyps are readily visible and harder to miss. In addition, the lower prevalence of large polyps broadens the confidence intervals for their detection rates. These less precise statistical estimates impair the ability to demonstrate statistically significant associations with withdrawal time.
Interestingly, an inverse correlation was observed between endoscopist experience level and PDR, although again this appeared to be driven primarily by detection of smaller polyps; as polyp size increased, the inverse relationship between detection rate and experience weakened. This suggests that more experienced endoscopists may have chosen not to document smaller polyps which they judged to be clinically insignificant. Conversely, less experienced endoscopists may have performed a more meticulous examination and conscientiously removed all diminutive polyps. No association was observed between procedure volume and polyp detection which likely reflects the endoscopy practice at the Mayo Clinic where procedures are assigned to endoscopists based on their availability/clinical responsibilities, etc. Therefore, procedure volume did not serve as a surrogate marker for endoscopist expertise, complexity of procedure, procedure indication (i.e. pretest probability of polyp detection) or procedure duration. Moreover, the inclusion criteria of this study precluded analysis of procedures involving trainees; therefore, procedure volumes would be underestimated for those colonoscopists who spend larger proportions of their time teaching. These factors may explain the absence of association between procedure volumes and PDR.
However, the clinical importance of polyps ≤5 mm should not be dismissed, especially in the population of patients who, despite national guidelines, may never undergo subsequent colonoscopy. The process of adenomatous transformation to cancer is thought to take approximately 10 years.18 Thus, small adenomas that were missed 10 years previously may have progressed to a higher risk lesion (high-grade dysplasia or adenoma ≥10 mm). Indeed, our analysis of polyps ≤5 mm revealed that 56% of these lesions were adenomatous. Reassuringly, our findings are consistent with those of Church who demonstrated a 49% prevalence rate of adenomas among 4381 polyps ≤5 mm.19
Several limitations of this study deserve mention. First, the retrospective nature of database research will never duplicate the methodological rigour of a prospective trial. However, in this setting, the excellent quality control of the institutional database strengthens our findings. As evidence of the integrity of the database contents, the close correlation of recorded procedure times with those taken by an independent observer is reassuring. Secondly, the lack of histological data in the database deprived us of a true gold standard for determining the nature of all polyps. However, this flaw was somewhat counteracted by the random review of a selection of polyps of various sizes which served to characterize the adenomatous content, and indirectly the neoplastic potential, of polyps of various sizes. Reassuringly, the proportion of adenomas among each size category is in keeping with prior published research.20 Of note, the imprecise measurement of polyp size (endoscopist's estimate) represents a further study limitation. Thirdly, the Mayo Clinic is a tertiary referral centre and, as such, the generalizability of any findings will be questioned. The endoscopic practice at this institution is open access, both within the institution and to the doctors in the surrounding area, which somewhat reduces the tertiary referral bias of the patient population. However, repeating this study in the community setting would settle this issue. A community practice-based study recently published in abstract form showed lower PDRs among endoscopist whose colonoscopy withdrawal times were shorter than the median among their colleagues.21 Finally, this endoscopic database does not capture postprocedure events, such as bleeds, perforations, or hospitalizations relating to the procedure which precluded analysis of this data.
In conclusion, this study adds important evidence upon which future recommendations for minimum colonoscopy withdrawal time can be based. A minimum withdrawal of 7 min represents a reasonable balance between optimal polyp detection and practice efficiency. In order to make firm recommendations for a minimum withdrawal time further studies in varied settings will be required to validate our findings. Of course, any recommended time would need to be adapted to the clinical setting. For example, prolonging the withdrawal phase in a patient who is reluctant to return for repeat screening in the future would be advisable. However, shortening this minimum withdrawal time should be discouraged as this would correspond to PDRs less than the median. This study will hopefully prompt further discussion regarding an ideal minimum withdrawal time, leading to evidence-based guidelines.
Acknowledgement
Dr Harewood is funded by an American Society of Gastrointestinal Endoscopy Career Development Award.
