Comparing Random Survival Forests and Cox Regression for Nonresponders to Neoadjuvant Chemotherapy Among Patients With Breast Cancer: Multicenter Retrospective Cohort Study (Preprint)

• DOI: 10.2196/preprints.69864
• Publication Date: 2025-04-08T16:17:29Z
• Authors: Yudi Jin; Min Zhao; Tong Su; Yanjia Fan; Zubin Ouyang; Fajin Lv
• Keywords: Preprint

<sec> <title>BACKGROUND</title> Breast cancer is one of the most common malignancies among women worldwide. Patients who do not achieve a pathological complete response (pCR) or clinical (cCR) post–neoadjuvant chemotherapy (NAC) typically have worse prognosis compared to those these responses. </sec> <title>OBJECTIVE</title> This study aimed develop and validate random survival forest (RSF) model predict risk in patients with breast pCR cCR post-NAC. <title>METHODS</title> We analyzed no pCR/cCR post-NAC treated at First Affiliated Hospital Chongqing Medical University from January 2019 2023, external validation Duke Surveillance, Epidemiology, End Results (SEER) cohorts. RSF Cox regression models were using time-dependent area under curve (AUC), concordance index (C-index), stratification. <title>RESULTS</title> The cohort included 306 cancer, aged 40-60 years (204/306, 66.7%). majority had invasive ductal carcinoma (290/306, 94.8%), estrogen receptor (ER)+ (182/306, 59.5%), progesterone (PR)– (179/306, 58.5%), human epidermal growth factor 2 (HER2)+ (94/306, 30.7%) profiles. Most presented T2 (185/306, 60.5%), N1 (142/306, 46.4%), M0 (295/306, 96.4%) staging (TNM meaning “tumor, node, metastasis”), 17.6% (54/306) experiencing disease progression during median follow-up 25.9 months (IQR 17.2-36.3). External (N=94) SEER (N=2760) cohorts confirmed consistent patterns age (40-60 years: 59/94, 63%, vs 1480/2760, 53.6%), HER2+ rates (26/94, 28%, 935/2760, 33.9%), prevalence (89/94, 95%, 2506/2760, 90.8%). In internal cohort, achieved significantly higher AUCs 1-year (0.811 0.763), 3-year (0.834 0.783), 5-year (0.810 0.771) intervals (overall C-index: 0.803, 95% CI 0.747-0.859, 0.736, 0.673-0.799). robust generalizability: showed 1-, 3-, 0.912, 0.776, respectively, while maintained performance 0.771, 0.729, 0.702, respectively. Risk stratification identified 25.8% (79/306) high-risk reduced time (&lt;i&gt;P&lt;/i&gt;&amp;lt;.001). Notably, improved net benefits across decision thresholds analysis (DCA); similar results observed studies. also promising different molecular subtypes all datasets. Based on predicted scores, stratified into high- low-risk groups, notably poorer outcomes group group. <title>CONCLUSIONS</title> model, based solely clinicopathological variables, provides tool for identifying approach may facilitate personalized treatment strategies improve patient management practice.