Simple Clinical Risk Score Identifies Patients with Serrated Polyps in Routine Practice

As colonoscopy with the removal of precancerous lesions is considered the gold standard modality in the prevention of colorectal cancer (1), concerns were raised by studies showing limited effectiveness of colonoscopy in the proximal colon (2–4). The serrated neoplastic pathway may contribute to the occurrence of some postcolonoscopy cancers, as some precursor lesions, especially sessile serrated adenomas/polyps (SSA/P), are easily overlooked during colonoscopy (5, 6) and are more challenging to remove endoscopically (7). It is generally accepted that large, proximal, or dysplastic (LPD) serrated polyps (SP) purport significant risk for malignant transformation, whereas nondysplastic small distal serrated polyps do not (8–11). Some studies indicate that 31% of hyperplastic polyps and 27% of nonadenomatous polyps are missed during colonoscopy (12, 13), which is consistent with a high variability in serrated polyp detection among endoscopists (i.e., ranging from 8%–32% for all serrated polyps, and from 1%–18% for proximal serrated polyps; refs. 14, 15). Taken together these findings highlight the need for improving detection of these lesions. To this end, Kahi and colleagues (16) recently proposed a minimal detection target of 5% for proximal serrated polyps.

Previous studies identified risk factors associated with an increased risk of having serrated polyps, such as smoking (17–22), alcohol consumption (20), and obesity (17, 19, 20, 22), whereas daily aspirin and nonsteroidal anti-inflammatory drug (NSAID) use seems to be protective (19–23). Nevertheless, data on risk factors for serrated polyps are still controversial, probably due to large variation between endoscopists in serrated polyp detection (14, 15), relatively small number of patients with serrated polyps examined (19, 20), inclusion of patients with distal serrated polyps only (18, 23), or a predominant male population (21).

In an attempt to clarify the risk profile associated with having ≥1 LPD SP, we first trained the endoscopists in the detection of nonpolypoid colorectal lesions (24, 25), with focus on quality indicators (26), and subsequently examined these risk factors in a derivation cohort. We assumed that a priori estimation of the risk for having LPD SPs may heighten the vigilance of trainees and less experienced endoscopists in detecting such lesions, thereby reducing the high variability in serrated polyp detection. We finally developed and validated a simple risk score, that is, the SP risk score, as a potential tool for improving the recognition of these lesions in routine practice.

In an ongoing initiative, aiming to improve the quality of colonoscopic cancer prevention in South Limburg, the Netherlands, we familiarized our endoscopists on the recognition, classification, and management of nonpolypoid colorectal lesions, as described previously (24, 25), with focus on quality indicators (26). Subsequently, we included all patients undergoing elective colonoscopy at our university hospital, from February 2008 to February 2010, in a cross-sectional study (i.e., derivation cohort). We excluded patients aged less than 18 years, those with hereditary colorectal cancer syndromes [i.e., known gene mutations or fulfilling the World Health Organization (WHO) criteria for serrated polyposis syndrome], inflammatory bowel disease, a personal history of colorectal cancer, or prior colonic surgery. We also excluded patients in whom colorectal cancer was detected at the time of colonoscopy, to prevent potential bias from underestimating the serrated polyp rates in this situation. As we reported on the individual detection rates of ≥1 LPD SP among the endoscopists, we excluded data from endoscopists conducting less than 50 colonoscopies during the study period, to attenuate the effects of random error by stratification. In patients who underwent multiple colonoscopies during the study period, data from one colonoscopy only were evaluated: index colonoscopy data were used when a second colonoscopy was conducted because of onset of new symptoms, polypectomy, or postpolypectomy surveillance, while data from the second colonoscopy were used in cases of poor bowel preparation or incomplete first colonoscopy. The study was approved by the local Institutional Review Board (MEC 10-4-020) and registered in the Dutch Trial Register (NTR 1891).

According to routine clinical care, patients received 2 to 4 L polyethylene glycol solution for bowel preparation. Colonoscopies were conducted by 8 gastroenterologists and 6 trainees, using conventional white-light colonoscopy. The endoscopists were familiarized with the detection of nonpolypoid lesions (24, 25), but unaware of the main study hypothesis and clinical risk profiling associated with having serrated polyps. Complete clinical, endoscopy, and pathology data were collected and registered into a standardized database. Proximal serrated polyps were defined as serrated polyps located proximal to the splenic flexure. Size of polyps was visually assessed using a biopsy forceps or mini-snare; large serrated polyps were defined as sized 6 mm or more. We used a cutoff of 6 mm for large serrated polyps for the following reasons: serrated polyps are usually smaller than adenomas, less than 6 mm in size (10, 27), approximately 30% to 50% of all SSA/Ps are sized 6 to 9 mm (5, 10, 28), and experts recently recommended removal of all serrated polyps, excepting distal serrated polyps less than 6 mm in size (11). Histopathologic classification of all polyps was conducted by 2 experienced gastrointestinal pathologists (A. Driessen and R. Riedl) according to the WHO classification (29). Serrated polyps comprised hyperplastic polyps, SSA/Ps without dysplasia, and dysplastic serrated polyps (i.e., SSA/Ps with dysplasia and traditional serrated adenomas; TSA). Advanced adenomas were defined as adenomas sized 10 mm or more, containing high-grade dysplasia or any villous components. The pathologists were unaware of the study hypothesis.

A questionnaire including demographics, smoking, body mass index (BMI), medication (i.e., use of aspirin/NSAIDs), and alcohol consumption was administered to all eligible patients shortly after the colonoscopy. Inconsistent or missing data were verified through medical records and recorded as missing in case this information could not be retrieved. Patients who completed the questionnaire and provided informed consent were included in the final analyses. Smoking status was categorized into current, former, or never and BMI classified in normal (i.e., <25 kg/m²), overweight (i.e., ≥25–<30 kg/m²), and obesity (i.e., ≥30 kg/m²). Medication use was categorized as daily versus nondaily/no use, and alcohol consumption was divided into very low/no alcohol consumption (≤1 glass per day) and alcohol use (>1 glass per day). Indications for aspirin use were primary or secondary prevention of cardiovascular events, for which standard daily doses of 100 mg were used. Personal history of adenomatous or serrated polyps was retrieved from medical records.

The primary outcome measure of this study was the risk profile associated with having ≥1 LPD SP. Although the majority of SSA/Ps are featured by a large size and/or proximal location (11), recent studies suggest that a subset of SSA/Ps, in particular those harboring dysplasia, may be smaller in size and more distally located in the colon (25, 30). Thus, to ensure capturing of all SSA/Ps, we specifically examined the group of LPD SPs (8–11, 25, 30–32). Unadjusted logistic regression analyses were conducted using a total of 12 clinically relevant parameters. As a minimum of 10 events is recommended per independent variable (33), we estimated that 120 or more events will be needed. On the basis of previous data (5), we estimated that the prevalence of ≥1 LPD SP in our cohort will be at least 4%. Assuming a questionnaire response rate of 60% (34) and inclusion of on average 200 patients undergoing colonoscopy each month, we expected 24 months will be needed to accomplish data collection. Adjusted logistic regression analyses were conducted to identify independent risk factors for having ≥1 LPD SP, using variables selected from the unadjusted logistic regression analyses (only variables with P ≤ 0.20 were included). The Hosmer–Lemeshow goodness-of-fit statistic was used to test reliability of the model.

Secondary aim of this study was to develop a SP risk score. Independent risk factors were incorporated into the risk score, and for each risk factor, we assigned weight in the risk score by using the ORs from the adjusted logistic regression analyses and rounded it to the nearest whole number. Total scores were obtained by summing up the individual scores, and the risks for having ≥1 LPD SP were calculated. Patients were subdivided into an average-risk group (i.e., a priori defined as having a lower cumulative prevalence of LPD SPs than the previously proposed minimum detection target of 5%, and corresponding to a total score <5 points; ref. 16) and a high-risk group (i.e., a priori defined as having a higher cumulative prevalence of LPD SPs than the suggested reference standard, and corresponding to a total score ≥5 points; ref. 16). In addition, the sensitivity and specificity of the risk score to identify patients with ≥1 LPD SP were calculated.

For this purpose, a second cohort was prospectively accrued at our institution, comprising all consecutive patients undergoing colonoscopy from February 2010 to February 2012. In assembling this cohort, we used a similar methodology (i.e., inclusion and exclusion criteria) as in the derivation cohort. On the basis of their total score, patients in the validation cohort were subdivided into the average-risk group (i.e. corresponding to a total score <5 points) and high-risk group (i.e., corresponding to a total score ≥5 points). Subsequently, the sensitivity and specificity of our risk score were calculated.

Mean (SD) and numbers (percentages) were used to describe continuous and categorical variables, respectively. Differences in continuous variables were analysed using the independent samples t test and differences in categorical values using the χ² test and Fisher exact test, when appropriate. Two-sided P values ≤0.05 were considered statistically significant and all ORs are presented with 95% confidence intervals (CI). Statistical analyses were conducted using the Statistical Package for the Social Sciences version 19.0.

In the derivation cohort, a total of 5,246 colonoscopies were conducted in 4,753 patients. Of these, 1,100 patients were excluded (<18 years, n = 18; hereditary colorectal cancer syndrome, n = 39; inflammatory bowel disease, n = 356; personal history of colorectal cancer, n = 172; prior colonic surgery, n = 81; unattainable for questionnaire, n = 290; or newly diagnosed colorectal cancer at colonoscopy, n = 144). Figure 1A illustrates the study diagram for the derivation cohort. A total of 3,653 patients received the questionnaire, of whom 994 (27.2%) patients did not respond to the request (i.e., nonresponders) and the remaining 2,659 (72.8%) patients returned the questionnaire (i.e., responders). Responders were older [mean (SD) age 60 (14) years vs. 52 (18) years, P < 0.001) and more often had ≥1 colorectal polyp (37.7% vs. 30.8%, P < 0.001), ≥1 adenoma (29.8% vs. 23.4%, P < .001), and ≥1 LPD SP (5.8% vs. 2.2%, P < 0.001) compared with nonresponders. No significant differences were found between responders and nonresponders about gender (males, 45.2% vs. 42.7%, P = 0.174), personal history of ≥1 adenoma (9.5% vs. 8.8%, P = 0.510), or personal history of ≥1 SP (4.6% vs. 4.4%, P = 0.765). A total of 316 (8.7%) nonparticipants (providing no informed consent, n = 201; or returning a blank questionnaire, n = 115) and patients receiving an examination by an endoscopist conducting less than 50 colonoscopies (n = 99) were also excluded resulting in a total of 2,244 (61.4%) patients finally analyzed.

In the validation cohort, a total of 5,883 colonoscopies were conducted in 5,266 patients. After exclusion, data from 2,402 patients were finally examined. Figure 1B illustrates the study diagram of the validation cohort.

Cecal intubation rates were 90.2% in symptomatic patients and 95.7% in asymptomatic patients, 30.6% had ≥1 adenoma and 15.9% had ≥1 SP. As shown in Supplementary Table S1, in the derivation cohort, individual detection rates of ≥1 LPD SP ranged from 2.9% to 7.8% among gastroenterologists and from 2.3% to 12.1% among trainees.

Data from 2,244 patients in the derivation cohort were finally analyzed. The mean (SD) age of the study population was 60 (14) years and 46.1% were males. In 80.3%, colonoscopy was conducted for symptoms, whereas 10.1% underwent colonoscopy for surveillance (i.e., previous adenoma/colorectal cancer) and 9.6% for screening indications. Table 1 depicts the baseline characteristics of the study population in the derivation and validation cohort subdivided according to presence of ≥1 LPD SP. In the derivation cohort, the prevalence of ≥1 LPD SP was higher in patients referred for surveillance versus symptoms or screening (13.3% vs. 5.3%, P < 0.001; 13.3% vs. 7.4%, P = 0.045, respectively). No significant differences were found in prevalences of ≥1 LPD SP between patients referred for symptoms versus screening (5.3% vs. 7.4%, P = 0.187). Patients with ≥1 LPD SP had significantly more adenomas and advanced adenomas compared with patients with nondysplastic small distal or no serrated polyps. Baseline characteristics about risk factors and the presence of ≥1 LPD SP in the derivation and validation cohort are presented in Table 2.

With regard to LPD SPs, results from the unadjusted and adjusted logistic regression analyses for the presence of ≥1 LPD SP are summarized in Table 3. Variables associated with the presence of ≥1 LPD SP (i.e., P ≤ 0.20) in the unadjusted logistic regression analyses were incorporated into the adjusted logistic regression analyses. Adjusted logistic regression analyses showed that age of more than 50 years (OR 2.2; 95% CI, 1.3–3.8; P = 0.004), a personal history of ≥1 SP (OR 2.6; 95% CI, 1.3–4.9; P = 0.005), current smoking (OR 2.2; 95% CI, 1.4–3.6; P = 0.001), and nondaily or no aspirin use (OR 1.8; 95% CI, 1.1–3.0; P = 0.016) were independent risk factors for the presence of ≥1 LPD SP. Separate logistic regression analysis was conducted using a cut-off value of ≥10 mm for large serrated polyps, which did not significantly change the results.

With regard to adenomas, results from the unadjusted and adjusted logistic regression analysis for the presence of ≥1 adenoma are summarized in Supplementary Table S2. Adjusted logistic regression analyses showed that age of more than 50 years (OR 3.8; 95% CI, 2.8–5.2; P < 0.001), male gender (OR 1.8; 95% CI, 1.5–2.2; P < .001), a personal history of ≥1 adenoma (OR 2.8; 95% CI, 2.0–3.9; P < 0.001), and current smoking (OR 2.0; 95% CI, 1.5–2.7; P < 0.001) were all independent risk factors for the presence of ≥1 adenoma.

Independent risk factors were incorporated into the SP risk score (Table 3). The Hosmer–Lemeshow goodness-of-fit statistic was P = 0.279 for our model. Figure 2A depicts the percentage of patients with ≥1 LPD SP per score category and corresponding 95% CI in the derivation cohort. On the basis of their total score, patients were subdivided into an average-risk group (i.e., total score <5 points, n = 1950) and a high-risk group (i.e., total score ≥5 points, n = 294). Prevalence of ≥1 LPD SP was 5.0% in the average-risk group versus 14.6% in the high-risk group (Table 4A). Patients in the high-risk group had a 3.2-fold increased odds of having ≥1 LPD SP than those in the average risk group. The sensitivity and specificity of our SP risk score to identify patients with ≥1 LPD SP were 30.5% (95% CI, 23.5%–38.5%) and 88.1% (95% CI, 86.6%–89.4%), respectively. Additional logistic regression analyses including different cut-off levels for age (i.e., ≥55 years; ≥60 years; ≥65 years, as well as age in decades) did not improve the sensitivity and specificity of our risk score.

Figure 2B depicts the percentage of patients with ≥1 LPD SP per score category and corresponding 95% CI in the validation cohort. The prevalence of ≥1 LPD SP was 6.5% in the average-risk group and 17.2% in the high-risk group (Table 4B). As compared with the average-risk group, patients in the high-risk group had a 3.0-fold increased odds of having ≥1 LPD SP. The sensitivity and specificity of the SP risk score were 33.2% (95% CI, 27.0%–40.0%) and 93.5% (95% CI, 92.4%–94.5%), respectively.

In this study, we found that 6.3% of all patients undergoing colonoscopy had ≥1 LPD SP. Age of more than 50 years, a personal history of serrated polyps, current smoking, and nondaily/no aspirin use were independent risk factors for having LPD SPs. According to our SP score, patients in the high-risk group had a 3.0-fold increased odds of having ≥1 LPD SP as compared with those in the average-risk group. We suggest that preassessment of the clinical risk for having LPD SPs may heighten the vigilance of endoscopists in recognizing such lesions, and might reduce the observed variability in detection.

Several studies suggest that LPD SPs, in particular SSA/Ps, might contribute to the occurrence of postcolonoscopy cancers through a serrated neoplastic pathway (10, 35). It became clear that serrated polyps often have a subtle endoscopic appearance, thereby explaining the operator-dependent variation in their detection, ranging from 8% to 32% (14, 15). This high variability in detection highlights the need for education and training, for which a systematic approach is needed, that is, by means of video-training, computer-aided programs, and learning from experts (14, 15). In this context, the simple clinical score described in our study, might be a useful tool as we assume that a priori estimation of the risk for having LPD SPs may increase the awareness of the endoscopist to detect such lesions, through careful examination of the proximal colon (i.e., longer withdrawal time, twice examination or retroflexion, use of selective dye-based or digital chromoendoscopy, or perhaps, additional tools, such as caps or endocuffs; ref. 36).

Several studies have shown that increasing age is associated with the presence of both adenomas and synchronous serrated polyps (19, 21, 37, 38), whereas the association between age and serrated polyps only remains controversial (19–21, 37, 38). In the current study, 51.1% of all patients with ≥1 LPD SP had ≥1 synchronous adenoma, at a mean (SD) age of 63 (11) years, suggesting that aging may increase the risk for having both LPD SPs and adenomas (19–21, 23). Some previous studies found an association between female gender and presence of serrated polyps (9, 39), whereas others, including the current study, did not (40, 41). Most of these studies observing gender differences were retrospective in nature and based on histopathology data only. As the current study was prospective and based on standardized data registration, we assume that this setting might provide more accurate estimates of the prevalence of serrated polyps in relation to gender. Our data are also in line with studies showing no association between obesity and presence of serrated polyps (37, 38) but at odds with others (19, 20, 22). These discrepancies might possibly be explained by the wide variations in the composition of populations examined. Moreover, we found that daily aspirin use has a protective effect against LPD SPs, in line with earlier studies (19–21, 23). The strong association between current smoking and presence of ≥1 LPD SP is in agreement with existing literature data (18–21, 37, 38, 42). Of note, (current) smoking has been also associated with proximal colorectal cancers, showing extensive DNA methylation, BRAF mutation, and microsatellite instability (43, 44). These shared molecular features in serrated polyps and colorectal cancers arising through the serrated neoplastic pathway (39, 45–47) may imply that smoking has an important role in the serrated carcinogenesis.

Some studies suggested that the risk factors for having serrated polyps might differ from those associated with adenomas (19, 38). In our study, current smoking and increasing age were associated with both the presence of adenomas and serrated polyps, whereas male gender was an independent risk factor for the presence of ≥1 adenoma only. An interesting finding of this study was that a personal history of ≥1 adenoma was associated with the presence of ≥1 adenoma, whereas a personal history of ≥1 serrated polyp was associated with the presence of ≥1 serrated polyp.

In the present study, we found that patients with ≥1 LPD SP significantly more frequently had adenomas and advanced adenomas compared with patients with nondysplastic small distal or no serrated polyps, which is in line with previous data by us (5) and others (31, 32). Taken together, these data suggest that older age, prior history of serrated polyps, current smoking, and nondaily/no aspirin use, as well as presence of adenomas, may define a risk phenotype which is associated with synchronous LPD SPs. This observation is of relevance as it may highlight a subgroup of patients in whom multiplicity of lesions may require personalized surveillance, with regard to the frequency of examination or technique used (i.e., chromoendoscopy, either dye-based or digital-based techniques). In a previous study by our group, we found that presence of (advanced) adenomas during colonoscopic examination might be considered a “red flag” for synchronous serrated polyps (5). The SP risk score proposed in the current study extends these observations, by identifying an a priori risk profile for having ≥1 LPD SP.

Some methodologic issues of this study need to be further addressed. As a strength, all endoscopists were familiarized with the recognition of nonpolypoid colorectal lesions before commencing this study. This was essential, as nearly half of the serrated polyps have a nonpolypoid appearance (5, 6). We used a standardized endoscopic reporting system, including quality benchmarking, and applied the WHO histologic classification of serrated polyps. As a potential limitation, in our study, questionnaire responders were older than nonresponders, in line with other Dutch surveys (34), and had more often colorectal lesions. As responders did not differ from nonresponders with regard to gender, personal history of adenomas or serrated polyps, we assume the differences found in prevalence of colorectal polyps may rather reflect the effect of aging than differences in response. To mitigate this source of bias, logistic regression analyses were conducted adjusting for age. Second, as the questionnaire was distributed to the patients shortly after the colonoscopy, the possibility of a recall bias cannot be excluded. Nevertheless, other studies, using a similar methodology as our study, found no differences in the reported risk factors among patients receiving the questionnaires before versus shortly after the colonoscopy (48). Third, the SP risk score proposed in our study was developed in a real-life cohort, including symptomatic and asymptomatic patients, and hence generalizability to a screening population needs to be further examined. Fourth, the ability of identifying patients with ≥1 LPD SP, using the clinical SP risk score proposed in our study is only moderate with a sensitivity of 33% and specificity of 94%, albeit comparable with other clinical scores (49, 50). Finally, not unexpected, prevalences of ≥1 LPD SP and SSA/Ps without dysplasia were higher in the validation cohort than the derivation cohort. As the endoscopists and pathologists were unaware of the study hypothesis, we believe that these findings rather reflect improvements in their learning curves through performance.

In conclusion, the present study indicates that age of more than 50 years, a personal history of ≥1 serrated polyp, current smoking, and nondaily/no aspirin use are all independent risk factors for having ≥1 LPD SP. Risk preassessment of patients undergoing colonoscopy, using the simple clinical SP risk score proposed in our study, might enable identification of patients at higher risk of having ≥1 LPD SP. Careful colonoscopic inspection in these patients, especially of the proximal colon, may reduce the currently observed variation among endoscopists in the detection of these lesions and finally improve the quality of examination.

S. Sanduleanu is a consultant/advisory board member of PENTAX BV NL. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.A.M. Masclee, S. Sanduleanu

Development of methodology: M.W.E. Bouwens, A.A.M. Masclee, S. Sanduleanu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.W.E. Bouwens, E.J.A. Rondagh, A.L. Driessen, R.G. Riedl, S. Sanduleanu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.W.E. Bouwens, B. Winkens, A.L. Driessen, S. Sanduleanu

Writing, review, and/or revision of the manuscript: M.W.E. Bouwens, E.J.A. Rondagh, R.G. Riedl, A.A.M. Masclee, S. Sanduleanu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.W.E. Bouwens, E.J.A. Rondagh

Study supervision: A.A.M. Masclee, S. Sanduleanu

Other: Critical revision of the manuscript for important intellectual content and final approval; S. Sanduleanu

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.