Accuracy of the online prognostication tools PREDICT and Adjuvant! for early-stage breast cancer patients younger than 50 years

Importance Online prognostication tools such as PREDICT and Adjuvant! are increasingly used in clinical practice by oncologists to inform patients and guide treatment decisions about adjuvant systemic therapy. However, their validity for young breast cancer patients is debated. Objective To assess first, the prognostic accuracy of PREDICT's and Adjuvant! 10-year all-cause mortality, and second, its breast cancer–specific mortality estimates, in a large cohort of breast cancer patients diagnosed <50 years. Design Hospital-based cohort. Setting General and cancer hospitals. Participants A consecutive series of 2710 patients without a prior history of cancer, diagnosed between 1990 and 2000 with unilateral stage I–III breast cancer aged <50 years. Main outcome measures Calibration and discriminatory accuracy, measured with C-statistics, of estimated 10-year all-cause and breast cancer–specific mortality. Results Overall, PREDICT's calibration for all-cause mortality was good (predicted versus observed) mean_difference: −1.1% (95%CI: −3.2%–0.9%; P = 0.28). PREDICT tended to underestimate all-cause mortality in good prognosis subgroups (range mean_difference: −2.9% to −4.8%), overestimated all-cause mortality in poor prognosis subgroups (range mean_difference: 2.6%–9.4%) and underestimated survival in patients < 35 by −6.6%. Overall, PREDICT overestimated breast cancer–specific mortality by 3.2% (95%CI: 0.8%–5.6%; P = 0.007); and also overestimated it seemingly indiscriminately in numerous subgroups (range mean_difference: 3.2%–14.1%). Calibration was poor in the cohort of patients with the lowest and those with the highest mortality probabilities. Discriminatory accuracy was moderate-to-good for all-cause mortality in PREDICT (0.71 [95%CI: 0.68 to 0.73]), and the results were similar for breast cancer–specific mortality. Adjuvant!'s calibration and discriminatory accuracy for both all-cause and breast cancer–specific mortality were in line with PREDICT's findings. Conclusions Although imprecise at the extremes, PREDICT's estimates of 10-year all-cause mortality seem reasonably sound for breast cancer patients <50 years; Adjuvant! findings were similar. Prognostication tools should be used with caution due to the intrinsic variability of their estimates, and because the threshold to discuss adjuvant systemic treatment is low. Thus, seemingly insignificant mortality overestimations or underestimations of a few percentages can significantly impact treatment decision-making.