Identifying patients with suspected colorectal cancer in primary care: derivation and validation of an algorithm.

backgroundEarlier diagnosis of colorectal cancer could help improve survival so better tools are needed to help this.

aimTo derive and validate an algorithm to quantify the absolute risk of colorectal cancer in patients in primary care with and without symptoms. DESIGN AND

settingCohort study using data from 375 UK QResearch® general practices for development and 189 for validation.

methodIncluded patients were aged 30-84 years, free at baseline from a diagnosis of colorectal cancer and without rectal bleeding, abdominal pain, appetite loss, or weight loss in the previous 12 months. The primary outcome was incident diagnosis of colorectal cancer recorded in the next 2 years. Risk factors examined were age, body mass index, smoking status, alcohol status, deprivation, diabetes, inflammatory bowel disease, family history of gastrointestinal cancer, gastrointestinal polyp, history of another cancer, rectal bleeding, abdominal pain, abdominal distension, appetite loss, weight loss, diarrhoea, constipation, change of bowel habit, tiredness, and anaemia. Cox proportional hazards models were used to develop separate risk equations in males and females. Measures of calibration and discrimination assessed performance in the validation cohort.

resultsThere were 4798 incident cases of colorectal cancer from 4.1 million person-years in the derivation cohort. Independent predictors in males and females included family history of gastrointestinal cancer, anaemia, rectal bleeding, abdominal pain, appetite loss, and weight loss. Alcohol consumption and recent change in bowel habit were also predictors in males. On validation, the algorithms explained 65% of the variation in females and 67% in males. The receiver operating curve statistics were 0.89 (females) and 0.91 (males). The D statistic was 2.8 (females) and 2.9 (males). The 10% of patients with the highest predicted risks contained 71% of all colorectal cancers diagnosed over the next 2 years.

conclusionThe algorithm has good discrimination and calibration and could potentially be used to help identify those at highest risk of current colorectal cancer, to facilitate early referral and investigation.