Neutrophil to lymphocyte ratio as a predictor for treatment of radiation-induced brain necrosis with bevacizumab in nasopharyngeal carcinoma patients

Radiation-induced brain necrosis (RN) may occur in 3%–24% of patients who receive radiotherapy for head and neck tumors.1 Corticosteroids have long been viewed as the first-line treatment for RN.2 Bevacizumab has been proved to be superior to the classic corticosteroids treatment by many studies and is being used increasingly, but with potential toxicity.3-5 Neutrophil/lymphocyte ratio (NLR), platelet/lymphocyte ratio (PLR), monocyte/lymphocyte ratio (MLR) and mean platelet volume (MPV) are widely used systemic inflammation indicators and have been well studied in their efficacy in predicting prognosis in multiple researches.6 This study aimed to explore the changes in these four biomarkers during bevacizumab or corticosteroids treatment and their potential predictive value of treatment response. Our study included 110 patients that were diagnosed with RN after radiation therapy for nasopharyngeal cancer and treated with bevacizumab, and additional 169 RN patients received corticosteroids treatment. Table 1 and Table S1 present the basic patient information. All detailed methods were provided in the Supporting Materials. We first looked into NLR, MLR, PLR and MPV changes during bevacizumab treatment (Figure 1). NLR level decreased during bevacizumab treatment in all patients overall (Figure 1A). In the effective group, a similar trend was observed while statistics difference was only seen between baseline and treatment 3 (Figure 1B). On the other hand, NLR level decreased between each treatment with baseline in the ineffective group (Figure 1C). Focusing on the individual effect of bevacizumab treatment, we found that NLR decreased after three courses of treatment compared with that at baseline in all patients, the effective and ineffective groups (Figure 1D-F). Interestingly, MLR decreased after three courses of bevacizumab treatment in all patients and the ineffective groups, but not in the effective group (Figure 1G-L). In addition, no significant change was observed in PLR or MPV after treatment 3 in all patients, the effective or ineffective group (Figure S1). We further investigated the efficacy of baseline NLR and MLR in stratifying patients by treatment response (Figure 2). In the effective group, both baseline NLR and MLR were lower than those in the ineffective group separately (Figure 2A,B). Moreover, the area under the curve (AUC) analysis showed favourable discrimination of baseline NLR (AUC .699, 95% CI .574–.825, Figure 2C) and baseline MLR (AUC .708, 95% CI .597–.820, Figure 2D). Furthermore, we constructed a prediction model for treatment response to bevacizumab using multivariable logistic regression. As a result, baseline NLR, the interval between diagnosis of brain necrosis and treatment with bevacizumab (IBT), and the interval between radiotherapy (IRB) and diagnosis of brain necrosis were identified as independent predictors (Table 2 and Table S2). The calculation formula for the response score was shown below: 5.520–.535 × baseline NLR – .025 × IRB – .078 × IBT. The predicted probability of effective treatment response was calculated using 1/(1 + exp [−response score]). Radiation approach (Conventional radiotherapy vs. IMRT) Subsequently, we evaluated the performance of the prediction model. The optimal cut-off values of the response score, baseline NLR, IRB and IBT were determined as .940, 3.571, 49.4 and 27.0. The model showed satisfactory discrimination (AUC .855, 95% CI .756–.955, Figure 2E). In addition, the calibration curve demonstrated good calibration of the model (Figure 2F). The Hosmer–Lemeshow test indicated no deviation from the perfect match with a p value of .317. Moreover, the decision curve analysis demonstrated that the model is clinically useful (Figure 2G). This model could inform a clinician how big the possibility is that a certain patient would respond to bevacizumab treatment and avoid adverse effects brought by bevacizumab on patients that would not respond well. In the meantime, we also studied how these biomarkers changed after corticosteroids treatment. We found that NLR decreased and MLR increased after treatment, while PLR and MPV did not change significantly in all patients overall; subgroup analysis showed that MLR increased and PLR decreased in the ineffective group (Figure S2). Moreover, baseline NLR, MLR, PLR or MPV displayed no significant difference between the effective and ineffective groups (Figure S3) and were not predictors for corticosteroids treatment response (Table S3). We hypothesized that the association between NLR and bevacizumab treatment may be relevant to the pathogenesis of RN. Although the pathogenesis of RN largely remains unknown, endothelial cell dysfunction has been proposed as the primary cause.7, 8 In brief, radiation-induced endothelial damage leads to blood-brain barrier (BBB) destruction, and hypoxia and generation of hypoxia-inducible factor-1α in local tissue, which strongly mediates the upregulation of vascular endothelial growth factor (VEGF).7, 8 Additionally, the breakdown of BBB would lead to peripheral immune cells infiltrating into the brain and leak out of antigen from central nervous system to the peripheral blood, both could activate the immune system. Lymphocytes have been found to infiltrate into the central nervous system and are partly responsible for neuronal damage and clinical symptoms.9, 10 Bevacizumab treatment would lead to BBB normalization which may decrease the leak out of antigen and infiltration of immune cells including both neutrophils and lymphocytes.8 In this study, patients with lower baseline NLR tend to have positive responses to bevacizumab treatment which can reduce VEGF. This may indicate a larger role of VEGF elevation and BBB breakdown in the pathophysiology in these patients. Apart from NLR, other independent predictors (i.e., IRB and IBT) are negatively correlated with bevacizumab treatment response, which is interpretable since the shorter duration after pathological changes would naturally indicate less irreversible damage of the brain thus leading to better treatment outcomes. In conclusion, we confirmed the predictive role of NLR in treatment response to bevacizumab in RN patients and constructed a reliable prediction model by integrating NLR, IRB and IBT. Additional external and prospective studies are needed to validate the prediction model as well as to investigate the role of immunity and BBB in the pathogenies of RN, which could provide potential intervention targets in the future. This work was supported by the National Natural Science Foundation of China (grant numbers: 81925031 and 81820108026), and the Science and Technology Program of Guangzhou (grant number: 202007030001) to Yamei Tang; the National Natural Science Foundation of China (grant number: 81872549) to Yi Li; the National Natural Science Foundation of China (grant number: 82003389) to Honghong Li; the Youth Program of National Natural Science Foundation of China (grant number: 81801229) to Yongteng Xu. The authors declare no potential conflict of interest. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.