Clinical variables associated with colorectal cancer on colonoscopy: a prediction model.

Abstract

We prospectively evaluated the ability of various indications for colonoscopy to predict colorectal cancer. Indications and findings were prospectively recorded for 1223 consecutive colonoscopies performed during an 18-month period at a University Hospital. Colonoscopies performed on 981 patients for indications which included colorectal cancer in the differential diagnosis were included in the study. A group of 653 patients was randomly selected to derive a model predictive of colorectal cancer at colonoscopy; the remaining 328 patients were used to validate the model. Colorectal cancer was found in 44/981 patients (4.5%). Univariate analysis of the derivation set showed that age > 55, occult bleed, anemia, iron deficiency, weight loss, and abnormal CT scan were associated with finding colorectal cancer at colonoscopy in the derivation set, p < 0.05. History of polyps was negatively associated with finding colorectal cancer. Stepwise selection was used to develop a predictive model by using three independent variables-age > 55, iron deficiency, and weight loss. Assigning a score of 1 to each variable, cancer was present in 0% [95% confidence intervals (CI), 0-0.6%], 5.3% (95% CI, 2.3-8.3%), and 17.9% (95% CI, 4.8-31%) of patients in the validation sample with a score of 0, 1, and > or = 2, respectively (p < 0.001 by chi2 for trend). The model was predictive of finding colorectal cancer at colonoscopy, independent of the location or stage of the cancer. Age, iron deficiency, and weight loss are important independent predictors of colorectal cancer in patients referred for colonoscopy.