Adjuvant chemotherapy (CT) for breast cancer patients improves survival, but estimating true benefit on the individual level remains a challenge. Cox models provide the possibility to estimate survival probaility for a particular individual with a given set of covariates. In this study, we used Cox modelling to develop a tool to predict benefit of adjuvant CT.
We divided the study population into 2 groups: patients who did not receive adjuvant CT and patients who received adjuvant CT to perform multivariate distant-disease free survival analyses and predictions according to the Cox model (proportional hazard model) for both populations. The covariates used in the models were age at diagnosis, tumour size, number of positive axillary lymph nodes, multifocality, tumour grade, lymph vascular space invasion, status for ER, PR, HER2 and Ki67 rate. By calculating individual probabilities of distant-disease free survival of patients who received chemotherapy with both models (with and without CT), we estimated benefit of adjuvant chemotherapy from the difference between predictions.
Both models in the training sets had excellent discrimination (concordance index > 0.7) and calibration. Tumor burden factors (tumor size, nodal status) had relatively more weight in the no CT model and histological factors (grade, KI67) in the CT model. The mean absolute benefit of adjuvant chemotherapy was only 2% as the majority of patients had an excellent prognosis. There was a strong correlation between prognosis and benefit of chemotherpay (r = 0.96). We report mean benefit according to the risk of distant recurrence in this table.
|Prognosis without CT (% at 5 years)||0-10||11-20||21-30||31-40||41-50||51-60||61-70||71-80||81-90||91-100|
|Absolute benefit of adjuvant CT (%)||32||36||35||30||25||19||13||8||2||0|
Only patients with estimated distant disease free survival less than 85% at 5-years had benefit from adjuvant CT. We demonstate the strong correlation between prognosis and prediction of benefit of adjuvant CT with simple Cox models, suggesting that all prognostic tests are also predictive.
Clinical trial identification
All authors have declared no conflicts of interest.