A novel nomogram for survival prediction in renal cell carcinoma patients with brain metastases: an analysis of the SEER database

Abstract

Background: Existing research on the development of prognostic models for renal cell carcinoma (RCC) patients with brain metastases (BM) remains limited. This study aimed to develop a prognostic prediction model for RCC patients with BM and to identify critical factors influencing clinical outcomes.

Methods: Patients diagnosed with BM between 2010 and 2019 were identified and extracted from the Surveillance, Epidemiology, and End Results (SEER) database. Potential risk factors were initially screened applying the eXtreme Gradient Boosting (XGBoost) and Random Forest (RF) machine learning algorithms. Subsequently, multivariate COX regression analysis was performed to identify independent risk factors for constructing the predictive nomogram. Nomogram performance was comprehensively evaluated based on Harrell's concordance index (C-index), receiver operating characteristic (ROC) curve analysis, calibration plots, and decision curve analysis (DCA). The SHapley Additive exPlanations (SHAP) method was employed to demonstrate the ranking of feature importance affecting patient prognosis at different time points. Moreover, we conducted propensity score matching (PSM) and Kaplan-Meier (K-M) survival analysis to compare clinical outcomes between surgical and non-surgical treatment subgroups.

Results: In total, 982 patients were assigned to the training cohort and 420 to the validation cohort. The constructed nomogram included four clinical variables: histologic type, T stage, N stage, surgery and chemotherapy. The AUC, C-index, calibration curves, and DCA curves showed excellent performance of the nomogram. In addition, the SHAP values indicated that surgical treatment was the most important prognostic risk factor for OS at 6-months, 1-year, 2-years, and 3-years. After further balancing the baseline characteristics between the surgical and non-surgical groups using PSM, we observed that patients with BM who underwent surgical intervention showed significantly better survival outcomes across all subgroups compared to non-surgical patients, though unmeasured confounders may contribute to this association.

Conclusion: We developed a novel nomogram for predicting prognostic factors in RCC patients with BM, offering a valuable tool to support accurate clinical decision-making. Our research also confirmed that surgical intervention was significantly associated with improved survival outcomes for patients with BM.

Keywords: SEER; brain metastases; nomogram; renal cell carcinoma; surgery.

Figure 1

Missing data for research variables, including T stage, N stage, Time (time from diagnosis to therapy), Tumor size, Marital status, Lung meta (lung metastases), Liver meta (liver metastases) and Bone meta (bone metastases).

Figure 2

The results of XGBoost (A) and RF (B) machine learning algorithms filter the top 10 important variables. The results are expressed by coefficient value. (C) Venn analysis of the results of the above two machine algorithms.

Figure 3

Forest plot with hazard ratios (HR) for the optimal prognostic variables of the multivariate Cox regression in the training cohort.

Figure 4

Nomogram for predicting 6-moths, 1-year, 2-year, and 3-year OS for RCC patients with BM in the training cohort.

Figure 5

Nomogram ROC curves to predict 6-months 1-year, 2-year and 3-year OS in the training cohort (A) and in the validation cohort (B). Nomogram calibration curves to predict 6-months 1-year, 2-year and 3-year OS in the training cohort (C) and in the validation cohort (D).

Figure 6

(A–D) DCA analysis predicting 6-months 1-year, 2-year and 3-year OS in the training cohort. (E–H) DCA analysis predicting 6-months 1-year, 2-year and 3-year OS in the validation cohort.

Figure 7

Kaplan–Meier curves for predicting OS of RCC patients with BM in low-risk, medium-risk, and high-risk groups. (A) For all cohort; (B) For training cohort; (C) For validation cohort.

Figure 8

Model interpretation using SHAP (SHapley Additive exPlanations). The importance ranking of clinical characteristics in the XGBoost prognostic model is shown for different timeframes: (A) 6-month, (B) 1-year, (C) 2-year and (D) 3-year models. XGBoost: extreme Gradient Boosting.

Figure 9

PSM-adjusted OS and CSS in brain metastatic RCC patients undergoing surgical treatment. Kaplan–Meier (K–M) survival analysis: (A) OS of all RCC patients with BM; (B) CSS of all RCC patients with BM; (C) OS of RCC patients with BM in the ccRCC subgroup; (D) CSS of RCC patients with BM in the ccRCC subgroup; (E) OS of RCC patients with BM in the non-ccRCC subgroup; (F) CSS of RCC patients with BM in the non-ccRCC subgroup; (G) OS of RCC patients with BM not receiving chemotherapy; (H) CSS of RCC patients with BM not receiving chemotherapy; (I) OS of RCC patients with BM receiving chemotherapy; (J) CSS of RCC patients with BM receiving chemotherapy.