238 Background: Numerous novel therapies for castration relapsed prostate cancer (CRPC) have led to a rapidly evolving approach to its management. Rationalisation of treatment combinations and sequencing of therapy requires identification of men who are more likely to benefit from a particular treatment. An unmet clinical need exists in this domain. We present the findings of a review of patients treated with abiraterone at our centre, and describe a novel predictive algorithm in this setting employing previously undefined pre-determinants of response. Methods: Patients with CRCP treated with abiraterone post-docetaxel at Queen Elizabeth Hospital between August 2010 and February 2012 were identified. Electronic patient records were utilised to extract variables including patient demographics, tumor characteristics, treatment history and other potential predictors of response such as prostate-specific antigen (PSA), hemoglobin (Hb), and alkaline phosphatase (ALP). Outcome measures included overall survival (OS), adverse events and PSA response rate. OS was estimated using the Kaplan-Meier method, and univariate and multivariate analyses were performed on potential prognosticators. Multivariate Beta coefficients were used to generate a predictive algorithm to identify two distinct risk groups. Results: Sixty one patients met the inclusion criteria. From starting abiraterone, the median OS was 12.6m, and median duration of follow-up was 11.5m. In univariate analysis seven factors impacted OS: age, response duration to androgen deprivation therapy (ADT), Hb, time from diagnosis to starting abiraterone, and ALP. Subsequent multivariate analysis identified three independent predictors of OS: duration of response to ADT (HR: 0.95, p=0.006), performance status (HR: 0.71, p=0.013), and baseline Hb (HR: 0.47, p=<0.001). A predictive algorithm dividing the cohort into high- and low-risk groups derived a diverging Kaplan-Meier curve without overlapping 95% CI’s. The low risk group did not reach median survival. Conclusions: This retrospective review has identified a predictive scoring algorithm for response to abiraterone in CRPC. We suggest that further analysis in the form of external validation is needed in order to justify its use in individualisation of patient management and stratification of patients for prospective clinical trials.