BACKGROUND: Stratification of patients for treatment of ductal carcinoma in situ (DCIS) is suboptimal, with high systemic overtreatment rates. METHODS: A training set of 95 tumours from women with pure DCIS were immunostained for proteins involved in cell survival, hypoxia, growth factor and hormone signalling. A generalised linear regression with regularisation and variable selection was applied to a multiple covariate Cox survival analysis with recurrence-free survival 10-fold cross-validation and leave-one-out iterative approach were used to build and test the model that was validated using an independent cohort of 58 patients with pure DCIS. The clinical role of a COX-2-targeting agent was then tested in a proof-of-concept neoadjuvant randomised trial in ER-positive DCIS treated with exemestane 25 mg day(-1)± celecoxib 800 mg day(-1). RESULTS: The COX-2 expression was an independent prognostic factor for early relapse in the training (HR 37.47 (95% CI: 5.56-252.74) P=0.0001) and independent validation cohort (HR 3.9 (95% CI: 1.8-8.3) P=0.002). There was no significant interaction with other clinicopathological variables. A statistically significant reduction of Ki-67 expression after treatment with exemestane ± celecoxib was observed (P