Abstract PS1-09-13: Development of Staging Guidelines for Metastatic Breast Cancer: Which method is best?

Abstract Background: Breast cancer staging is a foundational component for determining prognosis, and multiple methods have been used to create staging guidelines. However, it remains unclear which statistical approach is best. As such, we sought to compare 3 methods for developing staging guidelines, using de novo metastatic breast cancer (dnMBC) as an example. Methods: Patients with dnMBC in the National Cancer Database (2010-2022) were identified (N=48,832). Variables used for staging included clinical T-category, grade, ER, PR, HER2, histology, bone-only metastases, brain-only metastases, visceral metastases, and number of metastatic sites. Stage groups were assigned based on 3-year overall survival (OS) rates: stage group IVA >70%, IVB 50-70%, IVC 25-<50%, and IVD <25%. Statistical methods: (1) We conducted an analysis of all possible variable combinations (AC), yielding 643 unique characteristic profiles; OS was estimated via Kaplan-Meier, which was used to assign a subgroup (IVA-D). (2) Recursive partitioning analysis (RPA) grouped patients with similar OS based on these same disease characteristics; 3-year OS rates were estimated for each terminal node; subgroups (IVA-D) were assigned. (3) Bootstrapping was used; the RPA was repeated 1000 times (B-RPA); subgroups (IVA-D) were determined based on the subgroup each profile fell into most often. Overall, patients were categorized as stage IVA-D based on these 3 methods (AC, RPA, B-RPA), resulting in 3 stage group assignments. Differences in stage group assignments were classified as minor (1 group) or major (>1 group). Performance metrics (Akaike Information Criterion, AIC) were compared. Results: Application of the 3 methods yielded similar findings (Table), although the AC analysis had the best performance (lowest AIC), followed by the B-RPA and RPA. Patient-level Comparison (N=48,832). The same stage group (IVA-D) was assigned to 87.8% of patients by all 3 methods. Comparing B-RPA to AC, 10.6% of patients had a minor stage discrepancy and 0.4% had a major discrepancy. The RPA compared similarly to AC (10.9% minor, 0.4% major). Profile-level Comparison (N=643). Comparing B-RPA to AC, there was 60.7% agreement in stage assignments, 31.4% minor discrepancies, and 6.8% major discrepancies. Comparing RPA to AC was similar (60.8% agreement, 30.8% minor, 7.3% major). Of the 160 most common profiles, there was 85% agreement for both B-RPA and RPA. Of the 160 least common characteristic profiles, there was only 36.3% and 36.9% agreement for the B-RPA and RPA, respectively. Conclusions: Although the AC analysis had the best performance, there was a high level of agreement in stage assignment at the patient level. Potentially due to smaller patient samples in some profile subgroups, there was a moderate degree of discrepancies at the profile level, although most were minor. These findings will help inform future staging guideline development. Citation Format: J. K. Plichta, V. Rey, M. Blasingame, S. Luo, S. M. Thomas. Development of Staging Guidelines for Metastatic Breast Cancer: Which method is best? [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS1-09-13.