Clinical prognostic risk model optimization based on vascular density in neuroblastoma
Qiaoyin Liu, Jialin Cui, Dong Ren, Lei Yang, Xinyu Wang, Jiatong Xu, Yifan Liu, Cong Zhang, Haoyu Xu, Lejian He, Huanmin Wang, Haiyan Cheng, Guoshuang Feng, Xin Ni, Shengcai Wang, Yaqiong JinABSTRACT
Importance
Elevated microvessel density (MVD) and lymphatic vessel density (LVD) have been linked to poor outcomes in several malignancies. However, their prognostic significance in neuroblastoma (NB) remains unclear.
Objective
To evaluate the prognostic value of MVD and LVD in NB and construct an optimized prognostic model.
Methods
Sixty patients with NB were included. MVD and LVD were quantified using immunohistochemical staining for CD31 and D2‐40, respectively. Cox regression‐based prognostic models combining these markers with the International Neuroblastoma Risk Group (INRG) classification were compared with the standard INRG‐based model using receiver operating characteristic curves, the concordance index (C‐index), decision curve analysis (DCA), net reclassification improvement, and integrated discrimination improvement. Model robustness was assessed using bootstrap with 1000 resamples.
Results
The median MVD and LVD were 24.0/mm 2 and 7.16/mm 2 , respectively. High MVD (area under the curve [AUC] = 0.788) and high LVD (AUC = 0.693) were significantly associated with lower 5‐year event‐free survival (22.2% vs. 78.2% for MVD; 44.8% vs. 81.5% for LVD; both P < 0.05). In the multivariable Cox regression analysis, both LVD (hazard ratio [HR] = 1.088, P = 0.010) and INRG classification (HR = 5.594, P = 0.008) showed statistically significant associations with prognosis in the model incorporating LVD into the INRG classification. The model also enhanced the C‐index (0.747 vs. 0.696 for the original model), AUC, and DCA, with consistent results in internal validation.
Interpretation
Both MVD and LVD were associated with adverse outcomes in NB, while incorporation of LVD provided additional prognostic information beyond the INRG classification. Integrating LVD may improve the predictive performance of INRG‐based models and help identify biologically aggressive tumors among clinically non‐high‐risk patients, providing a foundation for more precise treatment planning.