Conditional survival prediction in elderly patients with hormone receptor positive locally advanced breast cancer...

Conditional survival (CS) analysis reveals dynamic changes in long-term survival, offering cancer survivors more accurate prognosis. We aimed to develop a prognostic model for elderly patients with hormone receptor-positive locally advanced breast cancer (HR+ LABC) based on CS.

Data for patients aged > 65 years with HR+ LABC were obtained from the Surveillance, Epidemiology, and End Results database (2010-2021, n = 8,450) and randomly divided into training (n = 5,915, 70%) and validation (n = 2,535, 30%) sets. CS analysis assessed changes in 10-year survival over time. Multivariable Cox regression identified prognostic factors for breast cancer-specific mortality (BCSM, defined as death due to breast cancer) and overall mortality (OM, defined as death from any cause) for model development.

For the entire cohort, 10-year OM and 10-year BCSM were 58.7% (95% CI 56.8%-60.4%) and 28.0% (95% CI 26.6%-29.4%), respectively. Both OM and BCSM improved over time according to CS analysis. Competing risks regression identified seven independent factors associated with BCSM: age, T stage, N stage, histological grade, HR status, chemotherapy, and radiotherapy. Ten prognostic factors were identified for OM, with additional factors including marital status, race, and surgery (all p < 0.05). The CS models demonstrated strong concordance between predicted and observed outcomes in both training and validation sets. The median time-dependent area under the curve (AUC) for the 10-year period remained above 0.73 for both models.

This study investigated the CS patterns in elderly patients with HR-positive LABC and develop personalized CS prediction models. These models may assist in patient counseling, guide clinical decision-making, and support healthcare resource allocation during follow-up. However, several limitations remain before clinical application, including the absence of important clinical variables, limited applicability to non-surgical patients, and the lack of prospective validation. Future implementation will require external validation and further model refinement.