OP0005 Performance of predict tool in predicting breast cancer survival in an Asian setting

Abstract

<h3>Background</h3> Predict Tool is a free online prognostic calculator that aids patient counselling and clinical decision making in the management of women with early breast cancer. We validated its prognostic performance in an Asian setting. <h3>Methods</h3> Within the University Malaya Hospital-Based Breast Cancer Registry, we identified 1676 patients who were diagnosed with non-metastatic disease from 1998 to 2006. Calibration of Predict Tool was done by comparing the predicted 5-year and 10-year overall survival (OS) with patients' actual 5-year and 10-year survival. Discrimination of the model was tested by receiver operating characteristic (ROC) analysis. <h3>Findings</h3> In the whole cohort, Predict Tool predicted 5-year OS accurately: 79.5% compared with observed survival of 79.9% (difference of –0.4%, 95% confidence interval [CI] –2.4 to 1.6). However, it did not perform well in women younger than 40years: 5-year OS was overestimated by 10.2% (95% CI 4.1–16.3). Predict Tool was also overoptimistic in predicting long-term survival; predicted 10-year OS was 70.6% compared with observed 10-year OS of 65.6% (difference of 5.0%, 95% CI 0.9–9.1). In terms of discrimination, Predict Tool performed fairly with an area under ROC curve of 0.77 (95% CI 0.75–0.80) for 5-year OS and 0.78 (95% CI 0.73–0.83) for 10-year OS. <h3>Interpretation</h3> Although Predict Tool performs well in predicting 5-year OS in this cohort of multiethnic Asian women, it was overoptimistic for predicting long-term survival. 10-year OS might be affected by systematic differences in background mortality risk between the population in which the model was developed (UK) and the Malaysian population. Predict Tool will be useful in aiding decision making related to systemic therapy in resource-limited Asian settings where microarray-based prognostication is economically unfeasible.