Development and clinical validation of a nursing risk prediction model for chemotherapy-induced febrile neutropenia in patients with cancer

Background Febrile neutropenia (FN) is one of the most serious yet potentially preventable complications of systemic chemotherapy. However, practical tools that support nursing-led early risk stratification and workflow-ready preventive actions remain limited. Methods We performed a real-world cohort study including 2,125 patients with cancer receiving systemic chemotherapy at a single institution. Patients were randomly split (7:3) into a derivation cohort (70%) and an internal validation cohort (30%). Candidate predictors were prespecified to ensure bedside nursing assessability and routine clinical availability, incorporating both conventional clinical/laboratory factors and nursing-relevant indicators (e.g., nutritional risk, mucositis, and self-monitoring adherence). Predictors were selected using penalized regression, followed by multivariable logistic modeling to estimate FN risk in the first evaluable chemotherapy cycle. Model performance was assessed by discrimination and calibration, and clinical utility was examined using decision curve analysis. A nomogram and risk-stratified nursing pathways were developed to translate predicted risk into actionable surveillance and preventive care. Results The final nursing-oriented model showed good discrimination and satisfactory calibration in both the derivation and validation cohorts. Decision curve analysis indicated net benefit across a clinically relevant range of threshold probabilities. Risk stratification based on predicted probabilities was associated with graded increases in FN incidence and adverse clinical outcomes. Conclusion This nursing-oriented FN prediction model provides individualized early-cycle FN risk estimation and operational risk stratification to support targeted surveillance and preventive nursing interventions. External validation across diverse institutions and nursing documentation systems is warranted.