A123
The development and validation of a clinical prediction rule to aid general practitioners in identifying symptomatic patients at high risk of breast cancer
Colin McCowan, Peter T Donnan, John A Dewar, Alastair M Thompson, Tom Fahey
1University of Dundee, UK, 2Ninewells Hospital, NHS Tayside, Dundee, UK, 3Royal College of Surgeons, Ireland, Dublin, Ireland
Background
Breast cancer is the most common cancer in women and is a major concern for the National Health Service. General Practitioners are the first point of contact for women reporting breast problems. This work reports on the development and validation of a clinical prediction rule (CPR) for diagnosis of cancer for women presenting with breast problems.
Method
A prospective cohort study of 802 patients presenting at the symptomatic breast clinic (derivation cohort) and 97 patients within 11 general practices (validation cohort) in Tayside, Scotland was performed between January and July 2007. A standardised data collection form was completed for all patients including findings from clinical examination and the gold standard outcome of histological diagnosis of breast cancer was defined from regional cancer audit records.
Results
Logistic regression analysis identified five factors independently predictive of cancer diagnosis: increasing age (adjusted odds ratio 1.1, 95%25 CI 1.07,1.13), presence of a discrete lump (OR 13.44, 95%25 CI 3.9,46.5), thickening (OR 8.2, 95%25 CI 2.3,29.2) or lymphadenopathy (OR 3.5, 95%25 CI 1.2,10.0), and a lump greater than two centimetres in size (OR 6.7, 95%25 CI 2.5,17.9). All 8 patients in the derivation cohort with a lump tethered to the skin or chest wall were also diagnosed with cancer. Discriminatory performance was good (Area Under Receiver Operating Characteristic curve = 0.94) and calibration was also good (Hosmer Lemeshow Goodness of Fit statistic = 5.21, p=0.634). Validation of the CPR on the general practice cohort was performed with good discriminatory performance (AUROC = 0.92). The CPR also had good calibration with the 5 patients subsequently diagnosed with cancer identified in the two highest deciles of risk (HLGoF statistic = 7.02, p=0.742).
Conclusion
The derived clinical prediction rule allows GPs to assess which patients are at high risk of cancer and require referral to a symptomatic breast clinic. It may perform more accurately than current guidelines but requires further external validation in other populations and settings due to the small numbers in the validation cohort.
