188
EDITORIAL
nature publishing group
COLON/SMALL BOWEL
see related article on page 180
Trouble in Paris (Classification): Polyp Morphology Is in The Eye of The Beholder Samir Gupta, MD, MSCS1, 2
Abstract: Key challenges to colonoscopy outcomes include polyp detection, appropriate polyp resection, and prediction of recurrent polyps. The Paris classification of gastrointestinal neoplasia has been used to attempt to address these challenges based on the hypothesis that the visual appearance of a polyp (e.g., sessile, flat, depressed) has an impact on these outcomes. Although the Paris classification has been widely used as a measurement tool in research, and reported to predict outcomes such as risk for high-grade dysplasia and invasive carcinoma, interobserver variability associated with this classification scheme has not been studied. In the current issue of the Red Journal, van Doorn et al. studied the interobserver variation of Paris classification in 85 colorectal polyps assessed by seven expert colonoscopists. They found that interobserver variation measured by kappa was only moderate (kappa=0.42; 95% confidence interval: 0.39–0.45). These findings suggest that without methods to improve interobserver variability, the Paris classification cannot routinely be used for research or routine practice. New approaches to characterizing polyp appearance may be required to use morphology as a predictor of clinical outcomes. Am J Gastroenterol 2015; 110:188–191; doi:10.1038/ajg.2014.411
Studies suggest that 5–11% of all newly diagnosed colorectal cancers are interval cancers occurring after colonoscopy (1–3). These findings have lead to intense efforts among endoscopists to improve colonoscopy outcomes. Key challenges to optimiz-
ing colonoscopy outcomes include three steps: polyp detection, appropriate polyp resection, and polyp recurrence prediction. Polyp detection is critical, because variation in adenoma detection rates across colonoscopists has been closely linked to risk for cancer after colonoscopy (4,5). Once polyps are detected, appropriate, complete resection is recommended to prevent cancer development. However, prior work has shown that this is difficult to achieve, particularly for larger size and sessile serrated polyps, and perhaps for lesions with a subtle (slightly raised, flat, or depressed) appearance (6). Further, to optimize outcomes, complete resection should not be attempted for polyps likely to harbor invasive cancer and/or lymph node involvement (7,8). After polyps have been identified, removed, and assessed pathologically, the final step is to stratify the individual for risk of recurrent polyps and cancer in order to make recommendations for surveillance. Recurrence prediction is challenging, because risk-stratification strategies for predicting future polyps and colorectal cancer are imprecise, and lead to both under- and over-surveillance (9). The Paris classification of gastrointestinal neoplasia, including colorectal polyps, has been used to attempt to address the challenges surrounding polyp detection, complete resection, and recurrence based on hypotheses that the visual appearance of a polyp has an impact on these outcomes (see Table 1) (7,8). This classification schema places polyps into visual categories ranging from Ip (pendunculated), Is (sessile), IIa (slightly raised) to III (excavated/ulcerated; see Figure 1). Seminal work demonstrated that assessment of polyp morphology using the Paris classification can predict the likelihood of invasive cancer and even lymph node involvement (7,8). Other work using the Paris classification has shown that polyps appearing to be slightly raised, flat, or depressed (technically referred to as non-polypoid neoplasia) may be more common than previously understood, and that detection rates for these polyps vary substantially across endoscopists (10–12). In addition, some investigators have used the Paris classification to link appearance to risk for high-grade dysplasia and carcinoma (10,11). Awareness of the Paris classifi-
1 Division of Gastroenterology, Department of Internal Medicine, Veterans Affairs San Diego Healthcare System, San Diego, California, USA; 2The Division of Gastroenterology, University of California San Diego, La Jolla, California, USA. Correspondence: Samir Gupta, MD, MSCS, Division Of Gastroenterology, Department of Internal Medicine, Veterans Affairs San Diego Healthcare System, San Diego, California, USA. E-mail: [email protected] Received 3 November 2014; accepted 1 December 2014
The American Journal of GASTROENTEROLOGY
|
VOLUME 110 JANUARY 2015 www.amjgastro.com
Editorial
189
Challenge
Hypothesized clinical and/or research importance
Detection
•
Awareness of visual appearance may improve detection, particularly of subtle lesions.
•
Suboptimal detection leads to imprecise outcome assessment in prospective research (e.g., classification of a missed lesion as metachronous).
• From a quality improvement standpoint, a benchmark for detection of subtle lesions might be a valuable quality metric. Appropriate resection
• Assessment of visual appearance may contribute to the decision to remove a lesion endoscopically. • Visual appearance might be a risk factor for incomplete resection
Recurrence
• Visual appearance may be associated with risk for metachronous polyps
Type Ip (pedunculated)
Type Is (sessile)
Type IIa (non-polypoid, slightly elevated)
Type IIb (non-polypoid, flat)
Type IIc (non-polypoid, slightly depressed)
Type III (non-polypoid, excavated)
Figure 1. Paris Classification of Gastrointestinal Neoplasia (adapted) (8).
cation, particularly with respect to characterizing slightly raised, flat, or depressed lesions, has likely been a contributing factor to increased detection of these subtle polyps (especially in the West) over the last 15 years. Indeed, prior to dissemination of the Paris classification, it is likely that many Western endoscopists did not know what they were looking for when it came to visually subtle polyps (13). Although the Paris classification has been used widely as a measurement tool in research, interobserver variability associated with this classification scheme has not been studied. For subjective measurement tools, this is a critical issue, as high interobserver variability means that individual assessors rate the same outcome differently. Such variability can contribute to bias that may underestimate the strength of associations between predictors (in this case Paris classification categories) and outcomes of interest. In the current issue of the Red Journal, van Doorn et al. (14) report on interobserver variation for adjudication of the Paris classification in a research set of 85 colorectal polyps assessed by seven expert colonoscopists. When initially tested, interobserver agreement for the Paris classification was moderate at best, with a Fleiss kappa statistic of just 0.42 (95% confidence interval (CI): 0.39–0.45). The numerical estimate of kappa is usually interpreted as being less than chance (
EDITORIAL
nature publishing group
COLON/SMALL BOWEL
see related article on page 180
Trouble in Paris (Classification): Polyp Morphology Is in The Eye of The Beholder Samir Gupta, MD, MSCS1, 2
Abstract: Key challenges to colonoscopy outcomes include polyp detection, appropriate polyp resection, and prediction of recurrent polyps. The Paris classification of gastrointestinal neoplasia has been used to attempt to address these challenges based on the hypothesis that the visual appearance of a polyp (e.g., sessile, flat, depressed) has an impact on these outcomes. Although the Paris classification has been widely used as a measurement tool in research, and reported to predict outcomes such as risk for high-grade dysplasia and invasive carcinoma, interobserver variability associated with this classification scheme has not been studied. In the current issue of the Red Journal, van Doorn et al. studied the interobserver variation of Paris classification in 85 colorectal polyps assessed by seven expert colonoscopists. They found that interobserver variation measured by kappa was only moderate (kappa=0.42; 95% confidence interval: 0.39–0.45). These findings suggest that without methods to improve interobserver variability, the Paris classification cannot routinely be used for research or routine practice. New approaches to characterizing polyp appearance may be required to use morphology as a predictor of clinical outcomes. Am J Gastroenterol 2015; 110:188–191; doi:10.1038/ajg.2014.411
Studies suggest that 5–11% of all newly diagnosed colorectal cancers are interval cancers occurring after colonoscopy (1–3). These findings have lead to intense efforts among endoscopists to improve colonoscopy outcomes. Key challenges to optimiz-
ing colonoscopy outcomes include three steps: polyp detection, appropriate polyp resection, and polyp recurrence prediction. Polyp detection is critical, because variation in adenoma detection rates across colonoscopists has been closely linked to risk for cancer after colonoscopy (4,5). Once polyps are detected, appropriate, complete resection is recommended to prevent cancer development. However, prior work has shown that this is difficult to achieve, particularly for larger size and sessile serrated polyps, and perhaps for lesions with a subtle (slightly raised, flat, or depressed) appearance (6). Further, to optimize outcomes, complete resection should not be attempted for polyps likely to harbor invasive cancer and/or lymph node involvement (7,8). After polyps have been identified, removed, and assessed pathologically, the final step is to stratify the individual for risk of recurrent polyps and cancer in order to make recommendations for surveillance. Recurrence prediction is challenging, because risk-stratification strategies for predicting future polyps and colorectal cancer are imprecise, and lead to both under- and over-surveillance (9). The Paris classification of gastrointestinal neoplasia, including colorectal polyps, has been used to attempt to address the challenges surrounding polyp detection, complete resection, and recurrence based on hypotheses that the visual appearance of a polyp has an impact on these outcomes (see Table 1) (7,8). This classification schema places polyps into visual categories ranging from Ip (pendunculated), Is (sessile), IIa (slightly raised) to III (excavated/ulcerated; see Figure 1). Seminal work demonstrated that assessment of polyp morphology using the Paris classification can predict the likelihood of invasive cancer and even lymph node involvement (7,8). Other work using the Paris classification has shown that polyps appearing to be slightly raised, flat, or depressed (technically referred to as non-polypoid neoplasia) may be more common than previously understood, and that detection rates for these polyps vary substantially across endoscopists (10–12). In addition, some investigators have used the Paris classification to link appearance to risk for high-grade dysplasia and carcinoma (10,11). Awareness of the Paris classifi-
1 Division of Gastroenterology, Department of Internal Medicine, Veterans Affairs San Diego Healthcare System, San Diego, California, USA; 2The Division of Gastroenterology, University of California San Diego, La Jolla, California, USA. Correspondence: Samir Gupta, MD, MSCS, Division Of Gastroenterology, Department of Internal Medicine, Veterans Affairs San Diego Healthcare System, San Diego, California, USA. E-mail: [email protected] Received 3 November 2014; accepted 1 December 2014
The American Journal of GASTROENTEROLOGY
|
VOLUME 110 JANUARY 2015 www.amjgastro.com
Editorial
189
Challenge
Hypothesized clinical and/or research importance
Detection
•
Awareness of visual appearance may improve detection, particularly of subtle lesions.
•
Suboptimal detection leads to imprecise outcome assessment in prospective research (e.g., classification of a missed lesion as metachronous).
• From a quality improvement standpoint, a benchmark for detection of subtle lesions might be a valuable quality metric. Appropriate resection
• Assessment of visual appearance may contribute to the decision to remove a lesion endoscopically. • Visual appearance might be a risk factor for incomplete resection
Recurrence
• Visual appearance may be associated with risk for metachronous polyps
Type Ip (pedunculated)
Type Is (sessile)
Type IIa (non-polypoid, slightly elevated)
Type IIb (non-polypoid, flat)
Type IIc (non-polypoid, slightly depressed)
Type III (non-polypoid, excavated)
Figure 1. Paris Classification of Gastrointestinal Neoplasia (adapted) (8).
cation, particularly with respect to characterizing slightly raised, flat, or depressed lesions, has likely been a contributing factor to increased detection of these subtle polyps (especially in the West) over the last 15 years. Indeed, prior to dissemination of the Paris classification, it is likely that many Western endoscopists did not know what they were looking for when it came to visually subtle polyps (13). Although the Paris classification has been used widely as a measurement tool in research, interobserver variability associated with this classification scheme has not been studied. For subjective measurement tools, this is a critical issue, as high interobserver variability means that individual assessors rate the same outcome differently. Such variability can contribute to bias that may underestimate the strength of associations between predictors (in this case Paris classification categories) and outcomes of interest. In the current issue of the Red Journal, van Doorn et al. (14) report on interobserver variation for adjudication of the Paris classification in a research set of 85 colorectal polyps assessed by seven expert colonoscopists. When initially tested, interobserver agreement for the Paris classification was moderate at best, with a Fleiss kappa statistic of just 0.42 (95% confidence interval (CI): 0.39–0.45). The numerical estimate of kappa is usually interpreted as being less than chance (
