Dynamic evolution, prediction and patient stratification of chemotherapy-induced neutropenia in a predominantly breast cancer cohort: A decision-support study for building a bundle care strategy

Background: Chemotherapy-induced neutropenia (CIN) is a common dose-limiting toxicity in patients with solid tumors, often leading to infections, treatment delays, or dose reductions. However, studies on the dynamic patterns of CIN across multiple chemotherapy cycles and their prediction remain limited. Objectives: To longitudinally observe CIN evolution across two consecutive cycles, develop a predictive model for severe CIN in cycle 2 (T2) based on cycle 1 (T1) characteristics, and classify patients by early recovery capacity to form a personalized bundle care strategy. Methods: This single-center retrospective cohort study using electronic medical record (EMR) data enrolled 138 patients with solid tumors (89.1% breast cancer) who received ?2 chemotherapy cycles between January 1, 2015, and January 31, 2025, at Zhuhai Maternal and Child Health Care Hospital, China. Neutrophil kinetic parameters were collected for T1 and T2. A multivariable logistic regression model predicted grade 3–4 CIN in T2 using T1 variables, with internal bootstrap validation. K-means clustering based on T1 recovery patterns identified patient subgroups. Results: Severe CIN incidence decreased from 88.0% in T1 to 42.7% in T2 (p<0.001); neutrophil nadir rebounded from 0.40 × 109/L (IQR: 0.10–0.70) in cycle?1 to 1.05 × 109/L (IQR: 0.50–1.80) in cycle?2 (p<0.001). The prediction model (age, T1 nadir, T1 days to nadir, T1 recovery duration) achieved a test AUC of 0.677 and an accuracy of 78.6%. Out of the 138 total patients, 120 with complete recovery data were utilized for clustering analysis, identifying three groups: rapid (n=21), similar (n=79) and slow (n=20). The slow-recovery group exhibited significantly higher rates of febrile neutropenia (25.0% vs. 10.1% and 4.8%, p=0.043) and chemotherapy delays (35.0% vs. 15.2% and 9.5%, p=0.021). Conclusion: CIN severity generally improves from first to second cycle, but with substantial inter-individual heterogeneity. A T1-based prediction model combined with recovery stratification can identify high-risk patients for personalized CIN management. Further validation in diverse populations is warranted.