Poor predictive value of lower gastrointestinal alarm features in the diagnosis of colorectal cancer in 1981 patients in secondary care

BACKGROUND: Clinicians are advised to refer patients with lower gastrointestinal (GI) alarm features for urgent colonoscopy to exclude colorectal cancer (CRC). However, the utility of alarm features is debated. AIM: To assess whether performance of alarm features is improved by using a symptom frequency threshold to trigger referral, or by combining them into composite variables, including minimum age thresholds, as recommended by the National Institute for Health and Care Excellence (NICE). METHODS: We collected data prospectively from 1981 consecutive adults with lower GI symptoms. Assessors were blinded to symptom status. The reference standard to define CRC was histopathological confirmation of adenocarcinoma in biopsy specimens from a malignant-looking colorectal lesion. Controls were patients without CRC. Sensitivity, specificity, positive predictive values (PPVs) and negative predictive values were calculated for individual alarm features, as well as combinations of these. RESULTS: In identifying 47 (2.4%) patients with CRC, individual alarm features had sensitivities ranging from 11.1% (family history of CRC) to 66.0% (loose stools), and specificities from 30.5% (loose stools) to 75.6% (family history of CRC). Using higher symptom frequency thresholds improved specificity, but to the detriment of sensitivity. NICE referral criteria also had higher specificities and lower sensitivity, with PPVs above 4.8%. More than 80% of those with CRC met at least one of the NICE referral criteria. CONCLUSIONS: Using higher symptom frequency thresholds for alarm features improved specificity, but sensitivity was low. NICE referral criteria had PPVs above 4.8%, but sensitivities ranged from 2.2% to 32.6%, meaning many cancers would be missed.