Study data published in Gut describe the derivation of a predictive model for assessing the risk for advanced colorectal neoplasia (AN) in average-risk adults without gastrointestinal symptoms. The model—which incorporated sociodemographic, physical, and lifestyle factors—was found to estimate risk for AN with high discrimination, “identifying a lower risk subgroup that may be screened non-invasively and a higher risk subgroup for which colonoscopy may be preferred,” according to the study authors.

The predictive model was developed using data from patients undergoing first-time screening colonoscopy at the Indiana University Medical Center. To be eligible to participate, participants must have been aged 50 to 80 years. Patients were also required to have no high-risk family history and no personal history of colorectal polyps.

Prior to colonoscopy, participants were asked to complete a 50-item survey that captured sociodemographic characteristics, family history, medical history, lifestyle habits, and medication use. The risk equation was derived using data from two-thirds of the sample (“derivation subset”). From 20 candidate variables, a 13-variable model was constructed to predict AN risk. The risk equation gave a numerical output ranging from -13 to +13, with higher scores indicating greater risk. The scores were separated by 3 risk categories: low-risk (-13 to -5), intermediate-risk (-4 to 2), and high-risk (≥3). The performance of the model was then tested in the remaining third of the cohort (“validation subset”).

Continue Reading

The derivation subset comprised 3025 patients (mean patient age, 57.3 ± 6.5 years), 52% of whom were women. The prevalence of AN in this group was 9.4%. Per the predictive model, 23%, 59%, and 18% of patients in the derivation subset were assigned to the low-risk, intermediate-risk, and high-risk subgroups, respectively.

Variables associated with increased risk for AN included age, male sex, cigarette smoking, significant alcohol use, metabolic syndrome, and red meat consumption. Features associated with lower risk included being married or living with a significant other, advanced education, regular use of aspirin and non-steroidal anti-inflammatory drugs, and regular physical activity.

The projected AN risk in the low-, intermediate-, and high-risk groups of the derivation subset was 1.5% (95% CI, 0.72-2.74%), 7.06% (95% CI, 5.89-8.38), and 27.26% (23.47-31.30%), respectively (P <.001). Model calibration was high, demonstrating good fit with the data (P =.69). Discrimination between patients with and without AN was also deemed good (c-statistic =.78). Ten participants in the low-risk group had AN (1.5%), amounting to the figure forecasted by the model.

The validation subset contained 1475 patients; the AN prevalence was 8.4%. Demographic and clinical characteristics were similar to those in the derivation group. AN risk in the validation cohort was 2.73% (95% CI, 1.25-5.11%) for the low-risk group, 5.57% (95% CI, 4.12-7.34%) for the intermediate-risk group, and 25.79% (95% CI, 20.51-31.66%) for the high-risk group (P for trend <.001). Respective sample proportions in the low-, intermediate-, and high-risk subgroups in the validation cohort were the same as those in the derivation cohort: 23%, 59%, and 18 respectively. Nine patients in the low-risk group had AN.

Based on these data, the investigators surmised that the predictive model developed in this study was able to accurately assess risk for AN in asymptomatic adults. Model calibration and discrimination were high in the derivation cohort; this trend continued in the split-sample validation set.

Primary limitations of the model include its derivation from a largely White cohort, which may limit generalizability. The large number of variables, “some of which may be difficult for users to both understand and respond accurately” to, constituted another study limitation, the investigators added. Further model development is necessary to expand its scope and precision.

“Using integrated risk calculators embedded within an electronic medical record, the current model could be used in real time to facilitate shared decision-making

for [colorectal cancer] screening, a process that requires consideration of risk that is integrated with patient preferences for and projected adherence to screening,” the study authors wrote. If validated in other settings, the model “has potential to advance the uptake and efficiency” of screening in this setting, they concluded.


Imperiale TF, Monahan PO, Stump TE, Ransohoff DF. Derivation and validation of a predictive model for advanced colorectal neoplasia in asymptomatic adults. Gut. Published online September 29, 2020. doi:10.1136/gutjnl-2020-321698