Twenty-three studies were included in the review (81,464 participants, range 112 to 43,791). The main potential source of bias in the included studies was potential for partial (48% of studies) and differential (26% of studies) verification bias.
Only three studies reported sufficient data to calculate 2x2 tables for the accuracy of individual symptoms to suggest a diagnosis of colon cancer. The symptom with the greatest utility in ruling in a diagnosis was rectal bleeding (summary positive likelihood ratio 5.31, 95% CI 1.65 to 17.07, I2=98.7%; three studies). Summary positive likelihood ratios for other single symptoms (abdominal pain, weight loss, diarrhoea, constipation, anaemia, change in bowel habit, bloating) ranged from 0.88 (95% CI 0.63 to 1.15; one study) for bloating to 3.98 (95% CI 2.81 to 5.64) and 4.62 (95% CI 3.03 to 7.06) for the two studies that assessed anaemia. Negative likelihood ratios were all close to one (range 0.75 to 0.86 for symptom evaluated in at least thee studies).
Six studies reported data on the accuracy of rectal bleeding combined with a second symptom. The only pair of symptoms to suggest a greater potential for ruling in colon cancer than rectal bleeding alone was rectal bleeding combined with anaemia (LR+ 7.88, 95% CI 2.65 to 23.4) but this was only assessed in a single study.
Data on predictive values were reported in the other studies (results available in the paper).