Risk for Advanced Colorectal Neoplasia in Asymptomatic Adults is Effectively Stratified by Phenotypic Features: Presidential Poster

Abstract

Purpose: While colorectal cancer (CRC) screening is effective and cost-effective for reducing CRC incidence and mortality, it is underutilized, inefficient, and costly. The ability to stratify risk for advanced neoplasia (AN) could improve screening efficiency and uptake. Several risk factors for AN are not used in practice because of inability to integrate them to produce a risk estimate. Study aim is to create a riskstratifying index for AN (CRC, advanced adenomas, serrated polyps ≥ 1 cm). Methods: We measured socio-demographic and physical features, medical and family history, and lifestyle factors in 50-80 year olds who underwent 1st-time screening colonoscopy between 12/2004 and 9/2011, linking these factors to endoscopic and histologic findings. Using best-subsets logistic regression, we derived a risk equation on a randomly selected 2/3s of the sample. The model selected was based on optimal statistical metrics, including Mallow's C and frequency with which each variable appeared in the top 20 models. Based on model coefficients, we assigned points to each variable to create a risk score (range, -13 to 13). Scores with comparable risk magnitudes were collapsed into risk categories. The model was then tested on the remaining sample. Results: Among 3,025 subjects in the derivation set (mean age 57.3 ± 6.5 years; 52% women), prevalence of AN was 9.4% (including 26 CRCs). Model variables include age, sex, smoking, ethanol use, diet, marital status, NSAID and aspirin use, physical activity, education level, and metabolic syndrome (P-value for fit = 0.09; c-statistic = 0.78). Respective risks of AN in the low- (scores of -13 to -5), intermediate- (scores of -4 to 2) and high- (scores of 3 to 13) were 1.50% (95%, 0.07-2.8%), 6.94%, and 27.3% (P-value for trend < 0.001), with respective cohort proportions of 23%, 59%, and 18%. Ten low-risk subjects had AN (0 CRCs, 6 distal). Based on finding a distal sentinel polyp, sigmoidoscopy would have detected 7 (70%) ANs. Among the 1,475 subjects in the test set (mean age 57.2 ± 6.5 years; 52% women), AN prevalence was 8.4%. Risk of AN in the low-risk subgroup was 2.73% (CI, 1.25-5.11%) and was 5.57% and 25.4% in the intermediate- and high-risk subgroups, respectively (P<0.001), with cohort proportions of 23%, 59%, and 18%. Nine low-risk subjects had AN (0 CRCs, 5 distal, 6 detectable by sigmoidoscopy). Conclusion: This risk index effectively stratifies the risk for AN among asymptomatic adults, identifying a low-risk subgroup of 23% that may be screened effectively and efficiently with tests other than colonoscopy and a high-risk subgroup of 18% for which colonoscopy may be preferred. If validated in other settings, this index could increase the efficiency and uptake of CRC screening.