Pitfalls in the construction of cancer guidelines demonstrated by the analyses of colorectal referrals.

INTRODUCTION

The aim of this study was to develop a system to compare and validate cancer referral guidelines, identifying the pitfalls in their development and provide a mechanism to evaluate their efficacy.

PATIENTS AND METHODS

3302 patients referred from primary care with colorectal symptoms over a 3-year period were assessed. All participants had a comprehensive history obtained via a questionnaire that incorporated all colorectal symptoms. The questionnaires were completed prior to assessment at the hospital. All patients were then assessed at the Colorectal One Stop Clinic (CROSC), underwent investigation and diagnosis achieved. All data were entered into a databank. Current prioritisation guidelines and tools that are used to assess colorectal referrals were applied to this colorectal databank to test their efficacy for cancer detection and referral prioritisation. Sensitivity and specificity for cancer detection and referral rates were assessed.

RESULTS

Cancer was detected in 156 patients (4.7%). All prioritisation models (NHS guidelines, Weighted Numerical Score [WNS], Netherlands, Harvard, Mersey, and Somerset) differentiated cancer from non-cancer patients. The use of a few symptoms as risk predictors (e.g. NHS guidelines) causes a decrease in specificity in contrast to a comprehensive risk tool, for example, the WNS at a score of 50 (NHS 54.1%, WNS 62.9%). This results in a significantly higher referral rate (NHS 47.6%, WNS 39.4%) and identifies fewer cancers (NHS 80.1%, WNS 85.9%). Non-evidence based modifications of the NHS guidelines (Somerset and Mersey) caused a further deterioration in specificity, which was reflected in an increased referral rate. Using the WNS, which is objective and a continuous scale, allows adaptation of the referral threshold, balancing sensitivity and specificity to the resources available within a hospital. For example, the WNS of > or = 40 has a sensitivity of 96.8% for cancer detection.

CONCLUSIONS

Accurate prospective data collection into a data bank allows testing of referral guidelines as well as providing an adjunct to guideline construction.