Lymphocyte to C-reactive protein ratio and improved prognostication of renal cell carcinoma: machine learning-based analysis

Purpose To investigate utility of a novel ratio, lymphocyte to C-reactive protein ratio (LCR), for improved prognostic stratification in RCC. Methods We conducted a retrospective machine learning–based analysis in a multicenter RCC cohort who underwent surgery (Radical or Partial Nephrectomy). Cut-off for “low” or “high” LCR was delineated by receiver operating characteristics/area under curve (ROC/AUC). Multivariable Analysis was conducted for predictors of all-cause mortality (ACM) and cancer-specific mortality (CSM). LCR, along with UCLA Integrated Staging System (UISS) factors, were included into MVA model to evaluate CSM. Employing Classification and Regression Tree (CART), we developed a decision tree based on hazard ratios (HR). The predictive performance was calculated using ROC/AUC. Results We analyzed 2071 patients [low LCR n=598/high LCR n=1473; cut-off=237.87 (AUC = 0.80)], median follow-up 60 months). Low LCR is an independent risk factor for ACM (HR = 2.75, p<0.001) and CSM (HR = 2.70, p<0.001). Including LCR (HR = 2.60, p<0.001), tumor stage (Stage II: HR = 6.7, Stage III: HR = 6.5, Stage IV: HR = 26.8, p<0.001) nuclear grade (HR = 3.0, p<0.001) and ECOG (HR = 1.6, p=0.002) in the CART, patients were re-stratified [(HR<1.0 (low-risk), HR = 1.0-2.0 (intermediate-risk), HR>2.0 (high-risk)]. AUC of 0.91 was revealed for the modified UISS model (AUC = 0.85 in the classic UISS model). Our findings are limited by retrospective design. Conclusion This study provides evidence supporting the prognostic significance of LCR in RCC. While low LCR was associated with poorer survival outcomes, incorporating LCR into the UISS model enhanced prognostic stratification in RCC patients. Further investigation is requisite to validate our findings.