We agree with Professor Rubin and Dr Hamilton that the choice of population studied can introduce spectrum bias that will affect the sensitivity and specificity of alarm features for the diagnosis of colorectal cancer. This may limit the generalisability of the findings of the current study to primary care, a fact we acknowledged in our discussion. The vast majority of alarm symptoms that are recommended currently to prioritise access to urgent investigation in order to exclude colorectal cancer performed poorly in secondary care studies.1 2 It is likely that their accuracy in primary care will be even lower due to the high prevalence of symptoms, but low incidence of colorectal cancer. Our paper did not make specific recommendations for alterations to guidelines for referral with suspected colorectal cancer, but did draw attention to the fact that more data are required from well-designed prospective studies in order to support the validity of these recommendations.

In terms of the number of primary care-based studies …