Development and internal validation of prognostic nomograms for overall and breast cancer-specific survival in de novo stage IV HER2-positive breast cancer.
Abstract
The widespread adoption of contemporary human epidermal growth factor receptor 2 (HER2)-targeted therapies has altered survival in HER2-positive metastatic breast cancer, potentially limiting the applicability of older prognostic models. We aimed to develop and internally validate nomograms for overall survival (OS) and breast cancer-specific survival (BCSS) in patients with de novo stage IV HER2-positive breast cancer diagnosed during 2018-2021. Using the Surveillance, Epidemiology, and End Results (SEER) database, 1,516 patients were randomly assigned to training (n = 1,061) and internal hold-out validation (n = 455) cohorts. Random Survival Forest analysis with a 98% out-of-bag concordance index criterion selected variables for multivariable Cox regression models predicting 1-, 3-, and 5-year survival. Discrimination, calibration, clinical utility, and risk stratification were assessed. Nine variables were retained for OS and seven for BCSS. Concordance indices in the training and validation cohorts were 0.764 and 0.741 for OS and 0.747 and 0.736 for BCSS, respectively. Areas under the time-dependent receiver operating characteristic curves (AUCs) ranged from 0.73 to 0.82. Calibration was generally acceptable, and decision curve analysis suggested potential net clinical benefit. High-risk patients had markedly poorer OS (hazard ratios, 5.05 in training and 5.07 in validation) and BCSS (4.32 and 4.09, respectively; all p < 0.001). Sensitivity analyses addressing the time-varying chemotherapy effect and competing risks broadly supported the primary findings. The nomograms showed good internal discrimination and may support individualized prognostic assessment once initial treatment information is available. However, SEER does not record specific HER2-targeted regimens; therefore, this cohort represents a contemporary calendar period rather than a treatment-confirmed dual-HER2-blockade population. External validation in treatment-annotated cohorts is required, and 5-year estimates should be interpreted cautiously given the median follow-up of 27 months.