Neutrophil to Lymphocyte Ratio (NLR) has demonstrated its promising potential as a prognostic biomarker in multiple cancer types. It is considered NLR is related to immunologic status. However, induction chemotherapy and the immune regulation drugs commonly alter NLR before radiotherapy, which might compromise the conclusion. The goal of the present study was to verify the predictive value of NLR in the absence of chemotherapy and immune regulation era NPC patients. In this retrospective study in the endemic region, we included 552 NPC patients from 1996 to 2002 with available pretreatment peripheral CBC data. Selected patients were treated with 2D conventional radiotherapy only. Optimal NLR cutoff was determined by receiver operating curve analysis, and the whole cohort was divided into high NLR and low NLR group based on the cutoff value. Kaplan Meier method and log-rank test were used to plot and compare survival, univariate and multivariate analysis were performed using Cox regression hazards model. 96% of the included patients were locally advanced or metastasized. The median value of NLR was 2.82(IQR, 1.94-4.56). With OS as the endpoint, NLR cutoff value was 2.29. It was discovered that NLR cutoff value was associated with T stage (p=0.036), and borderline associated with clinical stage (p=0.060). In multivariate analysis, compared with lower NLR group (≤2.29), the patients in higher NLR group (>2.29) was highly associated with worse overall survival (HR, 1.46; 95%CI, 1.14-1.85; p=0.002). For the first time in the exclusion of the hematological influence, this study verified the survival outcome was extremely related to the original status of immune before radiotherapy in NPC patients. Immunotherapy might play an essential role in the treatment of NPC in the future.

This study aims at employing computed tomography (CT)–based radiomics analysis within the primary tumor among NPC patients for the selection of candidate features, which can be correlated to EBV status. Data for biopsy-proven NPC patients dispositioned to definitive (chemo)radiation therapy at a single Chinese institution between 2005-2012 were scanned (n=202). Pretreatment contrast-enhanced CT (CECT) images and contours of the gross primary tumor were extracted in DICOM-RT format for patients with known EBV status. Serum EBV antibody levels were determined by titration using an Immunoenzymatic assay with 400 set as a cutoff value for the test (less than 400 denotes a negative test and vice versa). A total of 60 radiomics features were selected from the categories intensity direct (n = 11), neighborhood intensity difference (NID; n = 5), gray-level co-occurrence matrix (GLCM; n = 22), gray-level run length (GLRL; n =9), and shape (n = 13). The features were tested for correlation with volume, mean image intensity, and image intensity standard deviation, and they were removed if the absolute value of the Spearman correlation was greater than 0.8. After features with the highest average redundancy were excluded, 26 radiomics features remained. Potential candidate features for a multivariate logistic regression model were determined using the least absolute shrinkage and selection operator (LASSO); the optimal value of the regularization parameter lambda was determined to be 0.03. To measure the performance of the model and correct for optimism, Hosmer-Lemeshow test and bootstraping method (n=1000) were applied. False discovery rate (FDR) was evaluated using permutation test. The final area under the curve (AUC) was determined to assess the model performance. The final cohort included a total of 202 NPC patients with known EBV status. Median age was 47 (IQR: 39-52) and 72.8% were males. The disease was staged as II, III, or IV in 5.5%, 43.8%, and 50.7% of the patients, respectively. Based on EBV serological testing, patients were categorized into EBV-positive (56.9%) and EBV-negative (43.1%). After calibration, the LASSO fit selected four features. The most discriminative feature was Sphericity (0.00585). Other candidate features were the neighborhood gray tone difference based feature Texture Strength (0.01384), GLCM feature Correlation (0.01390), and statistical feature Kurtosis (0.04222). The apparent AUC and confidence intervals (CI) for the prediction of EBV status of the primary tumor was 0.73 [95% C.I.: 0.66-0.80]. This study represents the first attempt to use high-throughput imaging analytics to correlate imaging features to NPC biology in terms of EBV status. Our data shows that CECT-based radiomics features, specifically shape and intensity features, can discriminate between EBV-positive and EBV-negative tumors.