A Prognostic Model Using Post-Steroid Neutrophil-Lymphocyte Ratio Predicts Overall Survival in Primary Central Nervous System Lymphoma.

Background: Ratios of differential blood counts (hematological indices, HIs) had been identified as prognostic variables in various cancers. In primary central nervous system lymphomas (PCNSLs), higher baseline neutrophil-lymphocyte ratio (NLR) in particular was found to portend a worse overall survival. However, it was often observed that differential counts shift drastically following steroid administration. Moreover, steroids are an important part of the arsenal against PCNSL due to its potent lymphotoxic effects. We showed that the effect of steroids on differential blood cell counts and HIs could be an early biomarker for subsequent progression-free (PFS) and overall survival (OS). Methods: This study retrospectively identified all adult patients who received a brain biopsy from 2008 to 2019 and had histologically confirmed PCNSL, and included only those who received chemoimmunotherapy, with documented use of corticosteroids prior to treatment induction. Different blood cell counts and HIs were calculated at three time-points: baseline (pre steroid), pre chemoimmunotherapy (post steroid) and post chemoimmunotherapy. Tumor progression and survival data were collected and analyzed through Kaplan−Meier estimates and Cox regression. We then utilized selected variables found to be significant on Kaplan−Meier analysis to generate a decision-tree prognostic model, the NNI-NCCS score. Results: A total of 75 patients who received chemoimmunotherapy were included in the final analysis. For NLR, OS was longer with higher pre-chemoimmunotherapy (post-steroid) NLR (dichotomized at NLR ≥ 4.0, HR 0.42, 95% CI: 0.21−0.83, p = 0.01) only. For platelet-lymphocyte ratio (PLR) and lymphocyte-monocyte ratio (LMR), OS was better for lower post-chemoimmunotherapy PLR (dichotomized at PLR ≥ 241, HR 2.27, 95% CI: 1.00 to 5.18, p = 0.05) and lower pre-chemoimmunotherapy (post-steroid) LMR (dichotomized at LMR ≥25.7, HR 2.17, 95% CI: 1.10 to 4.31, p = 0.03), respectively, only. The decision-tree model using age ≤70, post-steroid NLR >4.0, and pre-steroid (baseline) NLR <2.5 and the division of patients into three risk profiles—low, medium, and high—achieved good accuracy (area-under-curve of 0.78), with good calibration (Brier score: 0.16) for predicting 2-year overall survival. Conclusion: We found that post-steroid NLR, when considered together with baseline NLR, has prognostic value, and incorporation into a prognostic model allowed for accurate and well-calibrated stratification into three risk groups.