PO-065 Assessment of prognostic models for stage II/III colon cancer: a hospital-based prospective cohort study

Abstract

IntroductionMultiple predictors of colon cancer prognosis to aid treatment decisions have been proposed. However, it is unclear whether calculators trained on one population yield accurate predictions for another population with different demographics and healthcare. We evaluated five colon cancer prognosis calculators trained on data from American patients in a community-based cohort of 1401 Australian patients with stage II-III colon cancer.Material and methodsData from prospectively recruited (Australian patients were submitted) to online predictors from the Memorial Sloan Kettering Cancer Centre (MSKCC), MD Anderson Cancer Centre (MDA), Mayo Clinic (MC) and Adjuvant! Online. Predicted outcomes were compared to observed outcomes to assess calculator calibration and discrimination.Results and discussionsOverall, we observed pessimism in the predictions of survival across the prognosis calculators, with the exception of the MC calculator. However, this calculator is limited to stage III patients receiving chemotherapy. Calculator discrimination tended to be similar to that observed from AJCC7 staging for relapse-free survival, but was superior for cancer-specific and overall survival.Comparison of the calculators’ performance against each other did not identify a consistently superior predictor, although the MDA cancer-specific survival (CSS) predictors exhibited worse calibration than CSS predictions from Adjuvant! Online.ConclusionDifferences in predicted versus observed overall, recurrence-free and cancer-specific survival were observed in our cohort of Australian colon cancer patients. Our findings suggest that colon cancer calculators are not readily transferable but require recalibration for different populations.