respondents was “often” or “almost always” concerned about developing cancer. Six respond-ents (9%) had clinical levels of depression and/or anxiety. The levels of anxiety, depressionor cancer worries were not associated with the screening results. Perceived advantages ofsurveillance outweighed disadvantages according to 88% of respondents. Conclusion Resultsindicate that, although EUS is more invasive than MRI, there is no significant difference in thepercentage of respondents who perceived one of the screening methods as more burdensome.Almost one third of respondents worry frequently about cancer, however levels of anxietyand depression are comparable to those of the general population and did not seem to beassociated with participation in the surveillance program. From a psychosocial point of viewPC surveillance in high-risk individuals seems feasible.

Su1135

Antithrombotic and/or Anticoagulant Use is Not Associated With a HigherFalse Positivity Rate in CRC Screening Using FITMaaike Denters, Marije Deutekom, Anne F. van Rijn, Patrick M. Bossuyt, Paul Fockens,Evelien Dekker

Background: No consensus exists on what to do with antithrombotics and anticoagulantsin the setting of fecal occult blood testing. It has been argued that their use could have anadverse effect on test characteristics yet discontinuation of these drugs is associated with anincreased risk of thromboembolic disease. Little is known on the effect of the use ofantithrombotics and anticoagulants on the outcome of the newer fecal immunohistochemicaltest (FIT). Aim: to compare the positive predictive value of FIT for neoplasia in usersand non-users of antithrombotic/anticoagulant agents, using colonoscopy as the referencestandard. Methods: Data were collected in a Dutch FOBT based colorectal cancer screeningpilot in the Amsterdam region. In 2006 and 2008 asymptomatic persons aged 50 to 74from a predefined target area were invited to participate in FIT screening. A single OC-sensor was used with a cut-off value of 50ngHb/ml. In the absence of contraindications, acolonoscopy was advised in FOBT positives. Antithrombotic and/or anticoagulant use >3times weekly (warfarin, acenocoumarol, carbasalate calcium and/or clopidogrel) was recordedat the outpatient clinic in all FOBT positive persons prior to the colonoscopy. Positivepredictive value (PPV) of FIT for neoplasia was defined as the proportion of persons withat least one adenoma or CRC at colonoscopy. Advanced neoplasia was defined as anyadenoma ≥10mm or an adenoma with a villous component >20% or high-grade dysplasia.Results: Colonoscopy results and data on anticoagulant use were available for 510 FITpositives: 88 persons (16%) were on anticoagulant use (63 carbasalate calcium; 22 warfarinor acenocoumarol; 7 clopidogrel) and 422 (76%) were not. Patients on anticoagulants weremore likely to be male (66% versus 50%), the average age was 61±7 in non-users and 65±7in users. No significant difference in PPV for neoplasia was observed between the two groups:74% in non-anticoagulant users versus 68% in anticoagulant users (p=0.31). Also, nodifference in PPV for advanced neoplasia was detected: 47% in non-anticoagulant usersversus 44% in anticoagulant users (p=0.69). Conclusion: Patients on anticoagulants orantithrombotics at the time of performing a FIT are not more likely to have a false positiveFIT result at colonoscopy. Continuation of anticoagulant or antithrombotic medicationduring screening with FIT does not seem to affect test accuracy.

Su1136

Population-Based CRC Screening Pilot: Results of First Round ScreeningJennifer J. Telford, Denis Petrunia, Laura Sware, Mark Elwood, David A. Owen, ChenZhou, Fareeza Khurshed, Andrew Coldman

Background: CRC screening has been shown to decrease CRC mortality and incidence andto be cost-effective compared to not screening at conventional levels of willingness to pay.Objective: To develop the infrastructure and processes for a population based CRC screeningprogram and test in 2 communities. Methods: Individuals 50-74 years of age were recruitedfrom two rural communities by local promotion, primary care physicians and mailed invita-tions. A call-center confirmed eligibility: 1) No prior history of ulcerative or Crohn's colitis,2) No history of CRC, 3) No lower endoscopy in the last 5 years, and 4) No rectal bleeding.Two fecal immunochemical tests (FIT, OC-Auto Micro 80, Polypmedco Inc, NY) were mailedto eligible individuals. Participants with a positive FIT (either test > 99 ng/mL) or a 1stdegree relative (FDR) with CRC were assessed by a nurse navigator for colonoscopy. Datacollected at colonoscopy included: cecal intubation, bowel preparation quality, withdrawaltime, polyp location, size, method of resection and completeness of resection and retrieval.Pathology was classified by the highest risk lesion found on a per patient basis. High riskadenomas were defined as > 9 mm, villous features, high grade dysplasia, and sessile andtraditional serrated adenomas. Immediate and adverse events within the 30 days followingcolonoscopy were recorded. Serious adverse events (SAE) were defined as resulting inhospitalization or death. Participants completed satisfaction surveys. Results: 3,826 indi-viduals registered from 01/09-04/10 and 91% returned a FIT suitable for analysis. The meanage was 62 years (SD 7) and 55% were female. Of the 3,490 with no FDR with CRC, 277(8.7%) had a positive FIT. 253 (95%) underwent colonoscopy: 7 (2.8%) had cancer, 75(29.7%) high risk adenomas, 9 (3.6%) >2 low risk adenomas, and 55 (21.7%) < 3 low riskadenomas. The mean time from FIT result to colonoscopy was 44 days. Of the 336 (8.8%)with a FDR with CRC, 247 underwent colonoscopy: 13 (5.3%) had high risk adenomas, 7(2.8%) >2 low risk adenomas, and 45 (18.2%) < 3 low risk adenomas. Of the 589 colonoscop-ies performed, cecal intubation was 95.3%, bowel preparation was adequate in 98.8% andthe mean withdrawal time was 7.5 minutes (SD 2.4). There were no deaths and 4 (0.7%)SAEs: one each of perforation, bleeding, diverticulitis, and partial small bowel obstruction.Of 663 participant survey responses (81% response rate), 92% found the FIT easy to useand 96% were satisfied with the program. Conclusion: The Colon Check pilot programestablished effective processes to screen for CRC resulting in a high proportion of FIT kitsreturned and appropriate participants undergoing colonoscopy in a timely and quality-assured manner. The positivity rate and positive predictive value of the FIT used was inkeeping with prior publications.

S-413 AGA Abstracts

Su1137

Development and Validation of a Natural Language Processing ComputerProgram to Evaluate the Quality of Colonoscopy ReportsHendrik Harkema, Faraz Bishehsari, Evan S. Dellon, Melissa I. Saul, Wendy Chapman,Carrie M. Farmer, Ateev Mehrotra, Robert E. Schoen

Introduction: Poor quality colonoscopy may compromise the effectiveness of screening inreducing colorectal cancer mortality and incidence. Quality indicators of colonoscopy havebeen identified, but measuring quality from manual review of colonoscopy and pathologyreports is cumbersome and costly. Natural language processing (NLP) is a field of computerscience that abstracts relevant structured data from free-text language. Measuring quality ofreports using a NLP computer program would allow for automated review. Aim: To developand validate a NLP-based computer program for measuring the quality of colonoscopyreports. Method: A NLP computer program was developed that abstracts the necessary datato measure published colonoscopy quality indicators from major gastroenterology societies,including documentation of cecal landmarks and bowel preparation quality, reporting with-drawal time, andmaking appropriate follow-up recommendations based on polyp histopatho-logy (Table 1). The NLP was then tested on a new validation set of 453 colonoscopy and226 associated pathology reports that were also manually reviewed by 3 physicians. Thereports were randomly sampled from 32,485 2008-9 reports from the 10 hospitals in theUPMC health system. Results: The average agreement between physician annotators wasKappa =0.82. Compared to the manual review, the NLP tool had a mean accuracy of 86.7(SD 10.7) for colonoscopy and 91.5 (SD 7.5) for pathology variables. The difference betweenthe clinician-abstracted and NLP-abstracted quality scores varied from 0.2 to 13.6% (Table1). The NLP program was successfully able to integrate colonoscopy and pathology reportsto calculate an adenoma detection rate. Of the major quality indicators, withdrawal time wasdocumented in <1% of reports, appropriate follow-up recommendations in 14-24%,adenomadetection rate 28-29%, bowel prep quality in 50-58%, and cecal landmarks in 65-69%.(Table 1). Conclusions: A NLP-based computer program was successfully developed andvalidated. There is great potential for an NLP computer program to automatically assess thequality of colonoscopy reports, but to be used widely the tool needs further refinement andvalidation on a national set of colonoscopy reports.Table 1 Performance of the NLP tool vs. clinicians measured by the overall accuracy ondifferent quality indicators.

Su1138

Anal PAP Versus High-Resolution Anoscopy in the Screening for High-GradeAnal Dysplasia Among HIV-Positive Patients: Are We Using the Best ScreeningTool?Alexander O. Mallari, Theresa Schwartz, Emmanuel C. Gorospe, Amneris Luque, StephenRauh, Roberto Corales

BACKGROUND: The increasing incidence of anal cancer in HIV-positive patients demandsan effective screening tool for early dysplastic changes. The diagnostic accuracy of anal Papremains uncertain for high-grade anal dysplasia. AIM: To assess the diagnostic accuracy ofanal Pap compared to high-resolution anoscopy (HRA) in detecting high-grade anal dysplasia.METHODS: We retrospectively reviewed the records of all HIV-positive patients who under-went anal Pap and HRA at an urban HIV clinic from June 2009 to June 2010. Baselinedemographic data and independent risk factors for anal dysplasia were analyzed. Paired analcytology and HRA-derived histopathology results were retrieved. Cytology and histologywere reported using the 2001 Bethesda system. Estimates of sensitivity, specificity, positivepredictive value (PPV) and negative predictive value (NPV) of anal Pap were calculated,using HRA-derived histopathology as the gold standard. Agreement between anal Pap andHRA was estimated using kappa statistics. RESULTS: A total of 187 patients were included,yielding 187 paired anal Pap and HRA specimens. Mode of transmission of HIV was mainlysexual (84.0%). Most patients were white (65.8%), males (81.3%) with mean age 47.2 +/-9.7 years. Mean duration of HIV positivity was 13.8 +/- 7.1 years with absolute CD4 countof 548.8 +/- 299.3 copies/mL. 58.3% had homosexual orientation. 62.6% had engaged inreceptive anal intercourse (RAI) while 96.3% had a history of not using condoms. 32.6%of patients reported a history of previously treated anogenital warts. Using histopathologyas the standard, the frequency of anal intraepithelial neoplasia (AIN) of any grade was 59.4%.In detecting dysplasia of any grade, anal Pap showed a sensitivity of 53.2%, specificity of75.0%, PPV of 75.6% andNPV of 52.3%. Concordance between anal cytology and histopatho-logy showed that in 13 (7.0%) paired specimens, both tests detected high-grade anal dysplasia.In detecting high-grade disease (AIN II and III), anal Pap showed a sensitivity of 18.3%,specificity of 21.4%, PPV of 65.0% and NPV of 3.2%. Kappa agreement between anal PapandHRA-derived histopathology was 0.20 (95%CI 0.12-0.31). (see Table 1). CONCLUSION:Although anal Pap has 62.0% accuracy in detecting any grade of dysplasia, it performed poorlyin detecting high-grade anal dysplasia. Overall agreement with standard histopathology waspoor. HRA if performed for high-risk HIV-positive patients may improve the early diagnosisand treatment of high-grade anal dysplasia.Table 1. Cytology and histopathology category concordance for paired specimens

AG

AA

bst

ract

s

AG

AA

bst

ract

s

Su1139

Can an Individual Risk Profile for CRC Be Used as Triage Test in CRCScreeningInge Stegeman, Thomas R. de Wijkerslooth, Esther M. Stoop, Monique van Leerdam,Evelien Dekker, Ernst J. Kuipers, Paul Fockens, Roderik A. Kraaijenhagen, Patrick M.Bossuyt

Background Colorectal cancer (CRC) is one of the leading causes of cancer related death.Detecting cancer or one of its precursors at an early stage can prevent premature death andmay reduce cancer morbidity. Due to its non-invasive character, triaging for colonoscopywith FOBT is an attractive method. However, FOBT has a less than perfect specificity andsensitivity. Risk stratification, based on established risk factors for advanced neoplasia, couldincrease the sensitivity of FOBT-based screening. This way, the diagnostic yield from screeningcould be increased with a similar number of colonoscopies. We explored the use of a riskprediction model in CRC screening. Methods We collected data in the Colonoscopy orColonography for Scree¬ning (COCOS) study, a multicenter screening trial conducted inAmsterdam and Rotterdam, The Netherlands between June 2009 and October 2010. Forthis study 5,924 randomly selected, asymptomatic men and women between 50 and 75years of age were invited to undergo colonoscopy. After colonoscopy, participants wereidentified with one or more advanced neoplasia (carcinomas and advanced adenomas).Screening participants were asked for one sample FIT (OC-sensor) and to complete arisk questionnaire prior to colonoscopy. Based on the questionnaire data, we developed amultivariable risk model with the following factors: total calcium intake, family history, ageand FIT result. We evaluated goodness-of-fit, calibration and discrimination, and comparedit to a model based on primary screening with FIT. Calibration refers to the agreementbetween calculated risk and observed outcomes. Discrimination expresses how well the riskmodel distinguishes between cases and non-cases. Results 1,236 invitees participated in thestudy, of which 1,022 (83%) completed the questionnaire. FIT results and colonoscopyoutcome were known for 787 (77%)of the participants who completed the questionnaires,of which 67 had advanced neoplasia (8.5%); 5 (7.5%) had CRC and 62 (92.5%) advancedadenoma. The risk based model significantly increased the goodness-of-fit compared to theprimary FIT screening model (p=0.0006). Discrimination improved with the risk-basedmodel (AUC: from 0.630 to 0.745), but the gain was not statistically significant (p=0.06).Calibration was good (Hosmer-Lemershow test; p=0.90). With a FIT strategy and using acut off of 50 ng/ml, 74 of the 787 (9.4%) participants would have been referred forcolonoscopy, including 22 persons with advanced neoplasia (29%). If, hypothetically, the74 participants with the highest calculated risk had been referred for colonoscopy, thisgroup would have included 21 (27%) persons with advanced neoplasia. Conclusion Riskbased stratification can increase the accuracy of FOBT in triage for colonoscopy in CRCscreening, but the actual gains in the yield of screening are probably modest.

Su1140

Predominance of Left-Sided Colorectal Cancer Among AsiansChanda K. Ho, Hal F. Yee, Ma Somsouk

Background: Previous literature has suggested that there are racial/ethnic differences in theanatomical distribution of colorectal cancer (CRC) lesions. Specifically, some studies haveshown that Asian patients may have a higher likelihood of developing distal or left-sidedcancers. These results potentially support a role for flexible sigmoidoscopy as an effectivescreening modality for this patient population that has historically been difficult to screen.Research to date, however, does not indicate whether the relative decrease in the proportionof left-sided lesions among non-Asians is attributable to exposure to CRC screening. Aims:1) To compare the anatomic location of colorectal cancer stratified by race/ethnicity 2) Toexamine the level of exposure to CRC screening tests on the proportion of left-sided lesions.Methods: We undertook a retrospective cohort study of all new colorectal cancer diagnosesfrom 1997-2009 at San Francisco General Hospital (SFGH). Diagnoses of CRC were deter-mined by ICD-9 code in the electronic medical record and confirmed by pathology. Lesionslocated at or distal to the descending colon were categorized as left-sided after review ofthe medical records. We analyzed the proportion of left-sided CRC lesions stratified by raceand duration of utilization of our health care system. For those patients in the system greaterthan 1 year prior to their diagnosis of a left-sided CRC, we measured what proportion hadany previous exposure to CRC screening. Results: From 1997-2009 there were 342 newdiagnoses of CRC at SFGH, of which 58% were men. There were 134 Asians, 69 Caucasians,68 African-Americans, 61 Latino patients, and 10 patients classified as “Other.” There were218 (64 %) left-sided cancers and 115 right-sided cancers. Asians had a statistically higherproportion (73%, p<0.007) of left-sided CRC compared to all other racial groups -Caucasians52%, African-Americans 54%, Latinos 65.5%. Of the 342 patients, 204 (60%) had been inthe SFGH system for at least a year. Of these patients, any previous CRC screening wasevident in 13% of Asians, 21% of African-Americans, and 13% of Latinos. Conclusions:Asians have a higher likelihood of developing left-sided CRC. There are likely biologicalfactors driving the increased prevalence of left-sided lesions among Asians since exposureto CRC screening did not significantly affect the proportion of left-sided lesions. Therefore,among Asians, a CRC screening strategy using flexible sigmoidoscopy may be very cost-effective in detecting and preventing CRC. The low rate of exposure to CRC screening acrossall racial groups highlights an important gap in prevention services underutilized by theurban poor.

S-414AGA Abstracts

Su1141

Efficacy and Safety of Colonoscopy in the UK NHS Bowel Cancer ScreeningProgrammeTom J. Lee, Roger G. Blanks, Matthew D. Rutter, Sue Moss, Andrew F. Goddard, AndrewChilton, Claire Nickerson, Richard J. McNally, Julietta Patnick, Colin Rees

BACKGROUND Colonoscopy is a central investigation in all colorectal cancer (CRC) screen-ing strategies. Success of CRC screening is dependent on the quality of colonoscopy. TheNHS Bowel Cancer Screening Programme (BCSP) offers biennial faecal occult blood testingto adults aged 60 -74 years. Colonoscopy is offered to individuals with a positive faecaloccult blood (FOB) test. All colonoscopists practicing within the screening programme arerequired to meet predefined standards through summative assessment and are subject toongoing quality assurance. In this study we examine the quality of colonoscopy in the NHSBCSP and describe the measures taken by the BCSP to achieve high quality colonoscopy.Comparison of current quality indicators with existing quality standards and evidence fromthe UK pilot study of FOB screening will be undertaken. METHODS The NHS BCSP nationaldatabase prospectively collects detailed data on all screening colonoscopies. Data from thefirst 3 years of the programme (August 2006 to August 2009) were analysed. Colonoscopyquality indicators (adenoma detection rate, polyp detection rate, colonoscopy withdrawaltime, caecal intubation rate, rectal retroversion rate, polyp retrieval rate, mean sedationdoses, patient comfort scores, bowel preparation quality and adverse event incidence) werecalculated for this period. All screening centres were contacted directly to verify adverseevent data. RESULTS In the study period, 2,269,983 individuals returned FOB tests and36,460 colonoscopies were performed. Mean caecal intubation rate was 95.2% and meanwithdrawal time for normal procedures was 9.2 minutes. The mean adenoma detection rate(ADR) per colonoscopist was 46.5%. ADR did not differ between prevalent and incidentrounds (p=0.90). Patient comfort scores were high and adverse event rates low. CONCLU-SIONS The NHS Bowel Cancer Screening Programme provides high quality colonoscopy asdemonstrated by high caecal intubation rate, adenoma detection rate and comfort scoresand low adverse event rate. This quality is achieved by ensuring that BCSP colonoscopistsare trained to a high standard and that these standards are maintained through ongoingquality assurance measures.

Su1142

Soluble Transferrin Receptor Level in Patients With GastrointestinalNeoplasm: Is It Useful for Improving Screening Efficacy?Seyed Amir Mirbagheri, Behtash Saeidi, Alireza Momeni, Behtash G. Nezami, KavehHajifathalian

BACKGROUND AND OBJECTIVES: soluble transferrin receptor (sTfR) is a marker of IronDeficiency (ID), and could be used to detect gastrointestinal cancers through ID, even beforedevelopment of anemia. We evaluated the usefulness of sTfR as a possible diagnostic toolfor detection of gastrointestinal (GI) malignant and pre malignant lesions. METHODS: westudied non-anemic patients referred for gastrointestinal endoscopy. sTfR was evaluated inthose with gastrointestinal cancers or polyps ≥ 10mm (case group, 97 patients). 95 ageand sex matched patients with normal upper and lower endoscopy were enrolled as thecontrol group. Additionally, serum ferritin concentrations and a guaiac-based fecal occultblood test (FOBT) were obtained for each individual. RESULTS: 63 patients had a high sTfRlevel in case group (64.9%) compared to only 6 (6.3%) in controls (P=0.000). In this study,Patients with GI lesions were 27.5 times more likely than controls for having a high sTFRlevel (OR: 27.5, 95% CI: 10.9 - 69.4). ROC curve of sTfR as a diagnostic test for GI cancershad an area under the curve (AUC) of 0.922 (95% CI: 0.885 - 0.959). Using a cut off valueof 8.3 mg/l, sTfR had a sensitivity of 64.9% and specificity of 93.7% for detecting patientswith premalignant or malignant gastrointestinal lesions, with a positive predictive value(PPV) of 91.3% and a negative predictive value (NPV) of 72.3%. FOBT was positive in 10(10.5%) healthy controls and only 28 (28.9%) cases; However this difference remained tobe significant (p = 0.005). Ferritin levels did not show any significant difference betweencases and controls (p = 0.392). CONCLUSIONS: We suggest that measuring serum sTfRlevel could be a useful adjunct to regular fecal occult blood testing in detection of bothupper and lower GI cancers, especially at early stages of disease, by identifying the casesthat remain undetected by FOBT. Moreover, negative sTfR result together with negativeFOBT may help clinicians rule out GI neoplasm with better confidence.

Su1143

Fecal Beta2-Microglobulin and Creatine Kinase B mRNA Assays as a Markerfor Colorectal Cancer ScreeningShigeru Kanaoka, Yasushi Hamaya, Shigeru Kuriyama, Mutsuhiro Ikuma, HiroakiMIyajima

Background and Aims: We reported that fecal COX-2 plus MMP-7 mRNA assays is usefulfor colorectal cancer (CRC) screening. Since approximately 20% CRC subjects with negativeresults of fecal COX-2 plus MMP-7 mRNA assays exist, improving sensitivity is one ofchallenging for this assay. We reported that the number of exfoliated cells is more abundantin CRC subjects than healthy subjects (Br J Cancer 2010). Therefore, fecal beta2-microglobu-lin (B2M) mRNA assay would be a marker for detection of CRC. Additionally, we identifiedcreatine kinase B (CKB) as a candidate marker through microarray analysis comparing geneexpression between fecal RNA of a CRC subject with negative result of fecal COX-2 mRNAassay and fecal RNA of a healthy subject. In this study, we performed a study of whetherfecal multi-markers RNA panel (COX-2, MMP-7, B2M and CKB) is useful or not for detectionof CRC and advanced adenoma. Methods: Stool samples from 111 subjects with CRC, from24 subjects with colorectal advanced adenoma (AA; at least 10mm in diameter), 113 healthysubjects were obtained. Standard histological techniques were used to classify malignancyat 0 to IV stage according to TNM classification, yielding stage 0 (n = 11), I (n = 24), II(n = 37), III or IV (n = 39). We purified RNA from routinely collected stool samples andanalyzed mRNA expressions of COX-2, MMP-7, B2M, and CKB by quantitative real-timeRT-PCR with standard plasmid DNA, compared the results with those of a single fecal