Prediction of colorectal cancer by a patient consultation questionnaire and scoring system: a prospective study.

BACKGROUND Current NHS guidelines for referral of patients with colorectal symptoms classify many as high risk but fail to identify a significant number of cancers in the low-risk group. We describe a practical scoring method to predict colorectal cancers. METHODS From October, 1999, 2268 patients with distal colonic symptoms, referred by general practitioners, completed a patient consultation questionnaire linked to a computerised record. Referrals were prioritised with a malignancy risk score by a senior colorectal surgeon separately from the general practitioner's letter and from the questionnaire. A weighted numerical score was derived from weighting of primary symptoms and symptom complexes and was calculated automatically when the questionnaire data were entered into the computer program. Analysis by receiver-operating characteristics assessed the scoring systems. Sensitivities and specificities of scoring systems were compared with McNemar's test. FINDINGS Of the 2268 patients, 95 had colorectal cancer. The average weighted numerical score was significantly higher for patients with cancer than for non-cancer patients (mean 76.5 [95% CI 72.2-80.9] vs 44.5 [43.6-45.4]; p<0.0001). At similar cancer detection rates, the malignancy risk score derived from the patient consultation questionnaire and the weighted numerical score graded lower proportions of referrals as urgent than did the current NHS guidelines (43.1% and 39.8% vs 49.8%; p<0.0001). INTERPRETATION The patient consultation questionnaire depends on history alone and is easily reproducible. In conjunction with the weighted numerical score, which removes operator bias, it can be used as an accurate system for prediction of symptomatic colorectal cancer.