Statistics from Altmetric.com
The recent systematic review by Ford et al (Gut 2008;57:2545–53) of alarm features of colorectal cancer is a very thorough piece of work. We agree with the conclusion that there is little evidence to support NICE guidance on rapid referral: indeed we have previously drawn attention to this.1 2 However, the study selection employed for this review means that its findings will not be helpful in improving that guidance.
An understanding of the significance of features of possible colorectal cancer must be based on their sensitivity and specificity in the appropriate population. Selection of patients for investigation occurs in primary care; therefore their significance has to be studied using primary care populations. Three studies in the review were labelled as originating (in part at least) in primary care. Only one of these studies included unselected patients presenting in primary care, but these were merged with secondary care patients for analysis.3 The other two studies dealt with patients who had already been selected for further investigation. It is, of course, extremely difficult to undertake prospective studies of colorectal symptoms presenting in primary care which utilise colonoscopy or barium enema investigation, for the reason that the authors give: that symptoms of colorectal cancer are also very prevalent in the “healthy” primary care population. Even so, it is not clear why the one good primary care prospective study which did investigate all patients was omitted.4
The difficulty with using secondary care data to drive primary care decisions is that positive predictive values are much higher in secondary care cohorts than primary care cohorts.5 This is because a significant filtering process occurs in primary care, with only a minority of patients being referred for further investigation.6 Failure to appreciate this has led in the past to inappropriate conclusions, one example being 12% of apparently healthy subjects aged 50–80 years who were considered to qualify for an urgent referral for possible colorectal cancer.7 We show this graphically in fig 1, which demonstrates the changing positive and negative predictive values for rectal bleeding, which is assumed to have a sensitivity of 40% for colorectal cancer, with a specificity of 98%. We have added two vertical lines, illustrating the background annual incidence in two different populations: the general population aged over 40 years, which is very similar to the primary care consulting population (0.25%), and the 2 week wait clinic (7%).8 9
Furthermore, we disagree with the suggestion that high specificity should be used to drive cancer diagnostics. Of course, the yield of cancers will be high in this group. However, specificity cannot be viewed without considering frequency. Abdominal masses (which the authors feel should be targeted for rapid investigation) are rare. Indeed over half of patients with symptomatic colorectal cancer in primary care have no high risk feature at all, only having constipation, diarrhoea, abdominal pain or mild degrees of anaemia. Are they to be ignored because their diagnosis is inefficient? The usual view in cancer diagnostics is to maximise sensitivity, until the false-positive rate becomes unacceptable. This seems much more reasonable.
The findings of this review will be helpful for specialists seeking to prioritise their referred patients, but should be viewed with extreme caution in any future review of referral guidelines for primary care.
Competing interests: None.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.