An Electronic Medical Record-based Scoring System for Estimating the Risk of Advanced Colorectal Neoplasia in Veterans: 2016 ACG Category Award (CRC Prevention)

Abstract

Introduction: Colorectal cancer (CRC) screening is particularly important for Veterans, who are at highrisk because of lifestyle and comorbid conditions. Each year, nearly 3 million colonoscopies are performed on or through the VA healthcare system. Understanding individual risk for advanced colorectal neoplasia (AN) could allow for tailoring screening based on risk for AN, and would improve screening efficiency Methods: Through remote electronic data extraction, we identified veterans who had a diagnostic or screening colonoscopy between 2002 and 2009 at one of 14 VAMCs. Colonoscopies for inflammatory bowel disease or neoplasia surveillance were excluded. We used natural language processing to identify the most advanced finding and location within the colorectum as proximal (splenic flexure) or distal. Logistic regression identified age and sex-adjusted factors that were independently associated with AN (CRC, adenomas or sessile serrated polyps 1 cm or larger, and adenomas with villous histology or highgrade dysplasia). Points for values of each variable were determined from the beta-coefficient for that variable, were summated and related to risk for AN. Scores were collapsed into 4 risk strata. Results: A total of 90,691 eligible colonoscopies were identified. Mean (SD) patient age was 61.7 (9.4) years; 5.2% (n=4,673) were women. Among 72,527 (80%) patients with known race, 55,180 (76.1%) were Caucasian, 14,155 (19.5%) were Black, 2,498 (3.4%) were Hispanic, and 734 (1.0%) were other. Overall prevalence of AN was 8.9% (n=8,081), of which 14.5% (n=1,171) was CRC. Among the subgroup of 66,725 (74%) with complete data, demographics were comparable to the overall sample; the prevalence of AN was 8.8% (N=5,859), of which 14.1% (n=826) was CRC. Independent risk factors were age, sex, and use of tobacco and COX-1 and COX-2 NSAIDs. Four risk strata (very low-to-high) were created where the risk of AN was 2.9% (score of < 0), 5.3% (score of 0-1), 8.6% (score of 2-3), and 11.4% (score of 4-6) respectively, with a goodness-of-fit (GOF) P value = 0.99 and c-statistic = 0.58. Corresponding CRC risks were 0%, 0.5%, 0.9%, and 2.1%, with a GOF P-value = 1.00 and c-statistic = 0.64. Results among the 22,804 (33%) screening colonoscopies were comparable to the overall findings. Conclusion: This new scoring system uses electronic medical record-based data to estimate and stratify the risk for AN and CRC and may be useful for risk-based tailoring of CRC screening, improving its efficiency among Veterans.