Abstract P6-07-04: Evaluation of Adjuvant! Online for primary operable grade 2 breast cancers with 10-year follow-up

Abstract

Abstract Background: In primary operable breast cancers, Adjuvant!Online (AOL) uses demographics and clinic-pathological markers to predict the 10-year breast cancer specific survival (BCSS) with and without adjuvant systemic therapy. Grade 2 BC is a difficult group for adjuvant treatment decisions. We compared the observed with the AOL-estimated BCSS rate in our centre in consecutive women primary operated for a grade 2 BC. Patients and Methods: All patients with grade 2 primary operable breast cancer from UZL diagnosed between 1/1/2000 and 31/3/2002. After 10 years of follow-up, patients were evaluated for BCSS and dead from other cause. For each patient, the adjuvant! Online algorithm (version 8) was applied to predict BCSS taking the patients' adjuvant therapy into account. Adjuvant! Online was validated by assessing the calibration and discrimination. Discrimination was assessed by the area under the ROC curve. The expected BCSS rate was obtained by averaging the predicted probabilities of BCSS. Next, logistic calibration was performed by predicting observed BCSS within 10 years of follow-up with the odds of the predicted probability of BCSS (odds BCSS). In this analysis, the intercept should be 0 and the coefficient for odds BCSS 1 for perfect calibration of the predicted probabilities. A likelihood ratio test is used to test the null hypothesis of perfect calibration. Results: Six patients were lost to follow-up. The median age at diagnosis of the remaining 445 patients was 56 years (IQR 48 to 67 years, range 26 to 89 years). In 433/445 cases (97%), the prediction from AOL was available. The median AOL predicted probability for BCSS was 89.1% (IQR 83.2 to 94.6%, range 20.3 to 98.9%). The observed BCSS rate was 93.3% whereas the AOL-expected rate was 86.7%. The likelihood ratio test for perfect calibration was statistically significant (p < 0.0001), with an intercept of −1.55 (95% CI −2.33 to −0.84) and a coefficient for odds BCSS of 0.55 (95% CI 0.19 to 0.92). AOL strongly underestimated the probability of BCSS. The AUC was 0.65 (95% CI 0.53 to 0.76). For patients that did not die from BC within 10 years, the median estimated BCSS probability by AOL was 89.6% (IQR 83.5 to 94.6 %), whereas for patients that died from BC the median estimated probability was 83.9 % (IQR 77.9 to 94.5%). Conclusion: Although Adjuvant! Online achieved moderate discrimination between those that did and those that did not die from BC within 10 years FU, it clearly underestimated the probability of BCSS in our sample. Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P6-07-04.