The success of current and proposed strategies to reduce colorectal cancer incidence and mortality is fundamentally based on measurement accuracy. The risk of histologic progression to high-grade dysplasia or cancer for adenomas ≥10 mm is approximately double that of adenomas <10 mm,1–3 and current international postpolypectomy surveillance guidelines use polyp size to determine the risk of colorectal neoplasia after screening colonoscopy. Measurement accuracy is also critical to optical diagnosis, a proposed colonoscopic management paradigm in which adenomas are detected, resected, and then discarded without pathologic confirmation; and rectosigmoid hyperplastic polyps are detected and left without resection.4,5 The safety of optical diagnosis is dependent on restricting its use to diminutive (5 mm or less) polyps, because their risk of cancer is considered negligible.6,7 By underestimating polyp size, endoscopists are at risk of unknowingly resecting and discarding a cancerous polyp, or worse, leaving it behind.
There have been increasing concerns regarding the accuracy of polyp measurements obtained endoscopically,8–11 but the role of technology bias in measurement error remains understated.12Figure 1 illustrates the extent of image distortion from the convex fish-eye lens of modern colonoscopes, where polyps in the center of the display appear magnified, and polyps in the periphery appear small and warped. A polyp image spanning both peripheral and central sections of the display may thus appear very different in terms of size, shape, and orientation than the polyp in reality (see grid lines in Fig. 1). A proof-of-concept measurement grid was recently used to unmask this form of technology bias, which resulted in colonoscopists achieving 99.8% accuracy for high-confidence classification of polyps into clinically relevant size categories.12
Because most studies investigating measurement error were limited by small sample sizes, few endoscopists, and data from single centers,9–11 we found it difficult to gauge the urgency for discussions on improvements in colonoscope design, as well as clinical care and research-based consensus statements that are fundamentally dependent on polyp size. To address these limitations, we conducted a population-wide, observational study within all hospitals of the government-funded health system in Brisbane, Australia. The aim of this study was to evaluate the densities of colorectal polyps individually measured at colonoscopy.
All colonoscopies performed over a 12-month period between December 1, 2014, and November 30, 2015, were included. We intentionally sourced retrospective data to exclude observation bias. Data were extracted manually from all colonoscopy reports via the electronic medical record at all 9 centers (ProVation MD, Wolters Kluwer). All polyps detected and noted in each report were eligible for inclusion. We extracted the size of every polyp that was detected and given an individual measurement. We excluded polyps in which the polyp size was not individually specified in the colonoscopy report.
The manual abstraction process utilized a data collection form and detailed abstraction protocol, which were pilot tested. During pilot testing, we found that polyp measurements recorded in groups were ambiguous. For example, an endoscopist may have reported: “Three polyps measuring 3 to 5 mm were found in the ascending colon...,” and we could not be sure if the endoscopist meant that one 3-mm, one 4-mm, and one 5-mm polyp were measured, or each polyp was estimated at 3 to 5 mm in size. Because our focus was the density distribution of polyps exactly measured during colonoscopy, we therefore only included polyps that were given individual measurements.
Data were abstracted by 8 research assistants, who were internal medicine residents. Each research assistant was individually trained in accessing, interpreting, and coding the electronic colonoscopy records. The training included an assessment, in which the research assistants had to access, interpret, and code 5 colonoscopy reports. The codes were compared against the model answers, and concordance was achieved with all report reviewers.
Data collection was centralized across 8 computer terminals in a single academic endoscopy center. Data were collected over 10 sessions, with each session lasting 6 hours. Data were abstracted hospital by hospital. The research assistants worked in pairs, and, for each hospital, each pair was given a specific 3-month period to review. Data were abstracted twice, and corrections were made in pairs. All questions on data interpretation and coding were directed to S.S. and D.G.H., who reviewed discrepancies for ongoing training and quality control.
The primary outcome was the frequency of individual reported polyp sizes. Data management and analyses were performed in R, version 3.2.0, in which a kernel density function was used to generate the density distribution. A smoothing parameter was selected such that density at every integer point could be uniquely defined.
Approval to conduct the research was obtained by the Metro South Human Research Ethics Committee.
A total of 12,597 electronic colonoscopy reports were individually reviewed, hospital by hospital, and 8591 individual size measurements from 18,276 detected polyps (47%) were obtained.
The density distribution of exact polyp measurements is presented in Figure 2A. There are a number of features which suggest that the shape of this distribution does not represent the shape of the distribution of true polyp size. Abrupt increments at 3, 5, 10, 12, 15, 20, 25, 30, 35, and 40 mm measurements, combined with the abrupt decreases at 2, 4, 6, 9, 11, 13, 14, 16, 19, 21, 24, 26, 29, 31, 34, 36, 39, and 40 mm measurements created a jagged, “saw-toothed” density pattern that did not appear naturally shaped. Figure 2B demonstrates the density distribution across the 0- to 20-mm range to highlight that most polyp measurements were 5 mm in size. There was also a sharp, cliff-shaped reduction in density at 6 mm compared with 5 mm. We found that individual participating hospitals produced similarly shaped density distributions. The density distribution of 2597 polyps from an exemplar academic endoscopy center is shown in Figure 2C.
Our study investigated measurement bias at colonoscopy through systematic analysis of 8591 individual polyp measurements recorded from 12,597 colonoscopies. Results suggest that endoscopists were biased toward specific numbers and biased against other numbers. This suggests that the reported population-wide density distribution of polyp measurements may not reflect reality.
A considerable number of small polyps (6–9 mm) are likely misclassified as diminutive or large (10 mm or greater), depending on their proximity to 5- and 10-mm sizes. We believe that the abrupt spikes of 5-, 10-, 15-, and 20-mm sizes in the density distribution is anomalous, and may be due to a human preference for numbers that are multiples of 5 and 10.8 Furthermore, measurement bias would also explain our findings of comparatively high densities of diminutive polyps and the sharp reduction in density of nondiminutive polyps. Previous experimental data have demonstrated that the majority of small polyps may be incorrectly considered diminutive,12 which would also influence the true population densities of larger polyps (Fig. 1). Our findings add to the complexities of appropriately defining and managing advanced lesions detected during colonoscopy.13 Size thresholds (10 mm, 20 mm) are an important factor for therapeutic decisions, which may be considerably influenced by human and technologic measurement bias.
Although previous studies investigating measurement error during colonoscopy have adopted the polyp measurements from pathologists as the true polyp size,10,11,14,15 pathologists are not immune to human measurement bias and have also been shown to exhibit terminal digit preference and clustering of polyp measurements at 5-mm intervals.8 Commonly, polyps are not stored in individual pots, and specimens are measured after formalin fixation with a simple ruler and without magnification. Furthermore, polyps postresection are distorted physically by different resection techniques because endoscopists invariably traumatize, deform, or fragment polyps by crushing, piecemeal resection, diathermy, disintegration, and suctioning. In contrast, endoscopists do not measure polyps directly but measure distorted images of polyps magnified on a 2-dimensional display monitor.12 Also, individual polyp sizes are usually not provided in routine histopathologic reports that generally report the size of retrieved tissue fragments. Routine clinical decisions are thus predominantly based on endoscopic images.
Our study is limited because the true size of unresected polyps was unknown. We chose not to compare pathologic and histologic sizes because resection specimens sent to pathologists are morphologically different, and are measured differently from the preresection polyp images seen by endoscopists. The natural history of polyp growth is not well understood, and the density distribution of polyp size may also be affected by selection bias during colonoscopy from variations in clinical performance. Also, the same electronic documentation system was used for all 8591 measurements. The specific shape of the density distribution was thus influenced by a common electronic menu option in which only whole numbers could be selected. The population-wide density distribution shown in this study is a product of different technology and human biases. The “saw-toothed” density curve, the abrupt fall in 6-mm compared with 5-mm densities, and the spike in 10-mm compared with 9-mm densities were suggestive of biased measurements favoring 5-mm and 10-mm sizes. Importantly, this study suggests that measurement error is a systemic phenomenon among colonoscopists.
Our findings have implications for colonoscopic practice. They cast doubt over the validity of international postpolypectomy surveillance guidelines, which advise shorter surveillance for 10-mm or greater lesions. Inappropriately long surveillance intervals may contribute to postcolonoscopy interval colorectal cancer. They also question the historical accuracy of polyp size data and risk estimates upon which surveillance guidelines were based. Finally, measurement bias threatens the safety of optical diagnosis strategies for real-time determination of polyp histology. Optical diagnosis promises to transform current colonoscopic practice by producing upfront savings of more than $33 million dollars annually in the United States alone.16 However, it relies heavily on including only diminutive polyps to limit the risk of unrecognized cancer being discarded and curative surgery not offered, leading to poor outcomes that may have been preventable.6,7 Also, up to 81% of patients would potentially litigate and seek financial compensation for a delayed diagnosis of cancer as a result of optical misdiagnosis.17
Measurement bias is an understated, systemic contributor to operator dependence in colonoscopy. Improvements in the accuracy of colonoscopic polyp measurement, and discussion of size-based clinical decisions among surgeon and physician endoscopists alike, as well, are therefore urgently needed. The true size, shape, and orientation of an unresected polyp may be better judged with accurate viewing distance information and without the illusionary effect of a convex fish-eye lens. The concept of 3-dimensional displays in colonoscopy has been recently introduced,18 and the provision of direct stereoscopic information, combined with new processor software to overlay measurement grids12 may further improve polyp measurement, clinical care, and research-based consensus statements.
The authors thank Drs Jason C.K. Tong, Nathan T. Jeffery, Harry Jin, June Yong Shin, Stephanie Giddy, Moein Amin, and Matthew O. Crozier for assistance with data collection. Dr Sakata acknowledges financial assistance from the Australian Postgraduate Award PhD and The Royal Australasian College of Surgeons Foundation for Surgery PhD Scholarships.
1. Martínez ME, Baron JA, Lieberman DA, et alA pooled analysis of advanced colorectal neoplasia diagnoses after colonoscopic polypectomy. Gastroenterology. 2009;136:832–841.
2. Lieberman DA, Rex DK, Winawer SJ, Giardiello FM, Johnson DA, Levin TRUnited States Multi-Society Task Force on Colorectal Cancer. Guidelines for colonoscopy surveillance
after screening and polypectomy: a consensus update by the US Multi-Society Task Force on Colorectal Cancer. Gastroenterology. 2012;143:844–857.
3. Cairns SR, Scholefield JH, Steele RJ, et alBritish Society of Gastroenterology; Association of Coloproctology for Great Britain and Ireland. Guidelines for colorectal cancer screening and surveillance in moderate and high risk groups (update from 2002). Gut. 2010;59:666–689.
4. Abu Dayyeh BK, Thosani N, Konda V, et alASGE Technology Committee systematic review and meta-analysis assessing the ASGE PIVI thresholds for adopting real-time endoscopic assessment of the histology of diminutive colorectal polyps. Gastrointest Endosc. 2015;81:502.e1–e16.
5. Kamiński MF, Hassan C, Bisschops R, et alAdvanced imaging for detection and differentiation of colorectal neoplasia: European Society of Gastrointestinal Endoscopy (ESGE) Guideline. Endoscopy. 2014;46:435–449.
6. Gupta N, Bansal A, Rao D, et alPrevalence of advanced histological features in diminutive and small colon polyps. Gastrointest Endosc. 2012;75:1022–1030.
7. Butterly LF, Chase MP, Pohl H, Fiarman GSPrevalence of clinically important histology in small adenomas. Clin Gastroenterol Hepatol. 2006;4:343–348.
8. Plumb AA, Nickerson C, Wooldrage K, et alTerminal digit preference biases polyp size measurements at endoscopy, computed tomographic colonography, and histopathology. Endoscopy. 2016;48:899–908.
9. Ahmad I, Levine JB, Anderson JCEndoscopic measurement of colorectal polyps: how do we measure up? Gastroenterology. 2016;150:769–771.
10. Anderson BW, Smyrk TC, Anderson KS, et alEndoscopic overestimation of colorectal polyp size. Gastrointest Endosc. 2016;83:201–208.
11. Eichenseer PJ, Dhanekula R, Jakate S, Mobarhan S, Melson JEEndoscopic mis-sizing of polyps changes colorectal cancer surveillance recommendations. Dis Colon Rectum. 2013;56:315–321.
12. Sakata S, McIvor F, Klein K, Stevenson ARL, Hewett DGMeasurement of polyp size at colonoscopy: a proof-of- concept simulation study to address technology bias [published online ahead of print October 3, 2016]. Gut. doi: 10.1136/gutjnl-2016–312915.
13. Moran B, Dattani M“SPECC and SPECULATION”: Is a significant polyp benign or an early colorectal cancer? How do we know and what do we do? Colorectal Dis. 2016;18:745–748.
14. Izzy M, Virk MA, Saund A, Tejada J, Kargoli F, Anand SAccuracy of endoscopists’ estimate of polyp size: a continuous dilemma. World J Gastrointest Endosc. 2015;7:824–829.
15. Levene Y, Hutchinson JM, Tinkler-Hundal E, Quirke P, West NPThe correlation between endoscopic and histopathological measurements in colorectal polyps. Histopathology. 2015;66:485–490.
16. Kessler WR, Imperiale TF, Klein RW, Wielage RC, Rex DKA quantitative assessment of the risks and cost savings of forgoing histologic examination of diminutive polyps. Endoscopy. 2011;43:683–691.
17. Sakata S, Lee AH, Kheir AO, et alPatient acceptance of the optical diagnosis
and misdiagnosis of diminutive colorectal polyps [published online ahead of print December 6, 2016]. Gastroint Endosc. doi: 10.1016/j.gie.2016.11.031
18. Sakata S, Grove PM, Stevenson AR, Hewett DGThe impact of three-dimensional imaging on polyp detection during colonoscopy: a proof of concept study. Gut. 2016;65:730–731.