Red blood cell distribution width as a predictor of survival in nasal-type, extranodal natural killer/T-cell lymphoma

We retrospectively enrolled 191 nasal-type, extranodal natural killer/T-cell lymphoma (ENKTL) patients newly diagnosed from 2008 to 2016 at the Sichuan Cancer Hospital, in order to evaluate the relationship between disease outcomes, demographic and clinical factors, and red blood cell distribution width (RDW). C-index, fisher's exact test, univariate analysis, and cox regression analysis were applied. The median age of patients was 44 years and 134 (70%) were men. The cutoff of RDW was 46.2 fL determined by Cutoff Finder. Patients with RDW≤46.2 fL had significantly better progression-free survival (PFS) (3-year PFS, 80.4% vs. 63.1%; P=0.01) and overall survival (OS) (3-year OS, 83.2% vs. 65.5%; P=0.004) than those with RDW>46.2 fL. Multivariate analysis demonstrated that elevated RDW is an independent adverse predictor of OS (P=0.021, HR=2.04). RDW is an independent predictor of survival outcomes in ENKTL, which we found to be superior to both the prognostic index of natural killer lymphoma (PINK) and the Korean Prognostic Index (KPI) in discriminating patients with different outcomes in low-risk and high-risk groups (all P < 0.05). The new models combining RDW with the International Prognostic Index (IPI), KPI, and PINK showed more powerful prognostic value than corresponding original models. RDW represents an easily available and inexpensive marker for risk stratification in patients with ENKTL treated with radiotherapy-based treatment. Further prospective studies are warranted to confirm the prognostic value of RDW in ENKTL.


INTRODUCTION
Nasal-type, extranodal natural killer/T-cell lymphoma (ENKTL) is a highly aggressive and heterogeneous non-Hodgkin lymphoma (NHL). It is most common in East Asia and Latin America (upwards of 10% of all NHLs), but rare in North America and Europe (less than 1% of all NHL) [1]. ENKTL accounts for 5-10% of Research Paper all malignant lymphomas in China [2]. Depending on the primary site of the cancer, ENKTL is divided into upper aerodigestive tract NK/T-cell lymphoma (UNKTL) and extra-upper aerodigestive tract NK/T-cell lymphoma (EUNKTL). UNKTL typically causes the vast majority of cases of "lethal midline granuloma", a syndrome which involves the paranasal sinuses, orbits, and lymph nodes, and even develops outside of the nasopharynx [1].
The red blood cell distribution width (RDW), which is traditionally known as anisocytosis, reflects the size heterogeneity of erythrocyte volume. RDW has been used to discriminate between different types of anemia [3]; however, emerging evidence suggests that anisocytosis is involved in a variety of human disorders such as cardiovascular disease, venous thromboembolism, diabetes, community-acquired pneumonia, as well as other acute or chronic conditions [3]. The role of RDW in patients with various cancers was investigated by several studies [4], and RDW has proven to be an independent prognosis factor in breast cancer [5], esophageal squamous cell carcinoma [6], lung cancer [7], and diffuse large B-cell lymphoma (DLBCL) [8]. However to the best of our knowledge the prognostic value of RDW in T-cell NHL has never been investigated. Therefore, we performed this study to evaluate the prognostic significance of RDW in patients with ENKL.

Patient characteristics
The main baseline characteristics of the 191 patients studied are listed in Table 1. The median age was 44 years (range = 15-86 years). Most of enrolled patients were male (134, 70.2%). There was an Ann-Arbor (AA) staging I/II predominance. In all, 184 patients (96.3%) were classified into AA I/II and 7 cases (3.7 %) into stage III/IV. The mean RDW was 45.7 (range = 36.3-65.0) (data not shown).
The optimal cut-off value of RDW for OS and PFS was determined by Cutoff Finder via the method of ROC curve (Euclidean distance). These two outcomes have the same cut-off value of 46.2 fL. For OS and PFS, the areas under the curve (AUC) were 0.60 and 0.56, sensitivities were 61.1 and 56.9, and specificities were 62.8 and 61.7, respectively ( Figure 1). In all, 107 (56.0%) cases had low RDW (≤ 46.2 fL) and were classified into the low-risk group, whereas 84 (44%) patients had high RDW (>46.2 fL) and were classified into the high-risk group. The highrisk group tended to have a higher death ratio than that of the low-risk group (38.8% and 19.4%, respectively). Regional lymphadenopathies were more frequent in patients with low RDW than in other patients. Patients in the high-risk group tended to have ECOG PS ≥ 2, and IPI scores 2-5. Other characteristics showed no significant differences across high-or low-risk groups with different RDW levels (Table 1).
Significant differences in chemotherapy regimens of patients was based on RDW (P=0.011); however, we found no significant associations between RDW and treatment response, chemotherapy cycles, and treatment modalities ( Table 2).

Survival and prognostic factors
Of the 191 patients studied, 54 patients died after a median follow-up of 30 months (range = 2-97). The 3-year OS and PFS in the whole group were 75.4% and 72.8%, respectively. The results of RDW distinguished two groups with significantly different survival outcomes ( Figure 2). Patients with RDW≤46.2 fL had significantly better PFS than those with RDW>46.2 fL (3-year PFS, 80.4% vs. 63.1%; P=0.01, Figure 2A) and OS (3-year OS, 83.2% vs. 65.5%; P=0.004, Figure 2B). Due to the existence of significant differences in chemotherapy regimens of patients based on RDW, we conducted subgroup analysis. We discovered that RDW is a significant predictor of overall survival for patients treated with P-Gemox, LVD, or VDLP regimen, but not CHOP or CHOP-like regimen (P=0.032, P=0.201, respectively, Figure 3

Combining RDW to IPI, KPI, and PINK score improves survival prediction and risk stratification
We conducted a model for combining RDW to the IPI, KPI, and PINK. Briefly, patients with an elevated RDW level (RDW>46.2 fL) were allocated a score of 1, while patients without an elevated RDW level received a score of 0. We then added the RDW score to the IPI, KPI, and PINK scores to generate this new prognostic model. We performed C-index analysis to evaluate the discriminatory impact of RDW on OS. IPI and PINK scores were found to be significant with C-index analysis in OS (0.607 and 0.537, respectively) (Figure 4A, 4C; Table 4). No significance was found between KPI and OS with a C-index of 0.588 ( Figure 4B; Table 4). After combining RDW with the IPI, KPI, and PINK scores, the new prognostic model showed significant association with OS, and survival prediction and risk stratification were improved as indicated by C-index (0.640, 0.639, and 0.603, respectively) ( Figure 4D, 4F, 4E; Table 4).
Furthermore, subgroup analysis was conducted to further investigate the effect of RDW to IPI, KPI, and PINK score. The RDW failed to significant stratify patients with IPI scores 0-1 (P>0.05, data and figure not shown). However, the RDW distinguish patients with good outcomes from those with inferior outcomes among patients with IPI scores≥2 ( Figure 5). For the KPI and PINK model, the RDW can stratify patients into a highrisk group or low-risk group ( Figure 6; Figure 7).

DISCUSSION
This retrospective study included 191 patients newly diagnosed with ENKTL to clarify the prognostic value of baseline RDW in pretreatment patients with ENKTL. Our results demonstrate that RDW is an independent prognostic factor for OS and PFS in patients with ENKTL given radiotherapy-based treatment. Moreover, combining RDW to IPI, KPI, and PINK scores can improve survival prediction and risk stratification. To our knowledge, this is the first report on the prognostic value of RDW in patients with ENKTL.
The RDW was used for the differential diagnosis of anemia for decades. However, in the recent years, RDW has gained notable renaissance. A variety of studies have found RDW to be a simple, robust, and convenient parameter associated with various human disorders such as cardiovascular disease [9], acute pulmonary embolism [10], diabetes [11], kidney disease [12,13], liver disease [14], and chronic obstructive pulmonary disease (COPD) [15], but even more importantly, with overall mortality in the general population [3,16,17]. Recently, studies have demonstrated that RDW is prognostic factor for a variety of cancers [4]. Accumulating evidence has shown that RDW has important clinical significance. However, is RDW a risk factor or a simple epiphenomenon of an underlying biological or metabolic imbalance? Lippi et al. demonstrated that an elevated RDW involved both impaired erythropoiesis and abnormal erythrocyte metabolism and survival that mirrored a variety of abnormalities, such as shortening of telomeres length, oxidative stress, inflammation, erythrocyte fragmentation, poor nutritional status, hypertension, dyslipidemia and abnormality of erythropoietin function [3]. Although the IPI, KPI and PINK are important prognostic models in patients with ENKTL with specific treatments, they have not been previously evaluated in ENKTL with radiotherapy-based treatment. In this study, they show significant association with OS and PFS in univariate analysis, whereas they also have some disadvantages. A shared disadvantage is that more than 70% of the cases were classified into the low-risk group, which is consistent with previous studies [18,19]. The RDW can effectively divide the low-risk patients into two groups, yet it failed to significantly stratify patients with IPI scores 0-1. Additional analysis found that the IPI, KPI and PINK were unable to clearly categorize high-and intermediate-risk groups. The high-risk group determined by IPI, KPI and PINK were further classified into two groups with significantly different survival outcomes with RDW. In order to investigate the impact of IPI, KPI, and PINK by RDW, these models were utilized to evaluate the RDW low-risk patients. All three models failed to significantly show significant prognostic value in patients with RDW≤46.2 fL. Finally, we constructed a new model by combining RDW with the IPI, KPI, and PINK scores. The new model showed more powerful prognostic value than the corresponding original models. Thus, we have determined RDW as a novel and powerful prognostic factor, which may be more useful than previous prognostic models. This study merits further prospective and multicenter studies to validate the prognostic value of RDW in ENKTL.
sedimentation rate (ESR) [27]. Additionally, in the healthy population and in patients with DLBCL, RDW was correlated with CRP [8,28]. RDW is also associated with nutritional indicators, including albumin, iron, folate, and vitamin B12 in patients with cancer [8,29]. Though the mechanism of RDW association with cancer patient survival is unclear, a possible explanation is that high levels of RDW mirror underlying inflammatory and malnutrition conditions which weaken erythrocyte maturation and causes inadequate production of the hormone erythropoietin, malnutrition, or oxidative damage. As such, the decreased survival seen in patients  with elevated RDW may be a symptom of the underlying inflammation and malnutrition. Additional research is needed to explain the chemical and molecular mechanism relating RDW and cancer patient mortality. According to National Comprehensive Cancer Network (NCCN) guidelines for Non-Hodgkin's lymphoma version 2.2015, the risk factors for ENKTL include age>60 years, ECOG PS score≥2, B symptoms, increased LDH, regional lymphadenopathies, local tumor invasion, histological expression of Ki-67, Epstein-Barr virus (EBV) DNA > 6.1E+7/ml, and IPI score >2. With the exception of EBV DNA, due to insufficient data, the rest of the risk factors were evaluated in our study. While half of these risk factors show significant association with survival in the univariate analysis, four factors (age>60 years, B symptoms, regional lymphadenopathies and Ki-67>50%) were not significantly prognostic for OS and PFS. Though age>60 years as a risk factor is investigated in many studies, some studies found that it was not predictive of survival [30][31][32]. Several previous studies also did not find significant associations between B symptoms and survival outcomes [33][34][35]. B symptoms was constructed in the pre-asparaginase era, and might lose its value in the era of asparaginase therapy. Regional lymphadenopathies was recognized as an independent risk indicator for worse OS in KPI [30]. Nevertheless, in our study, regional lymphadenopathies was not a significant predictor for OS, which is consistent with results presented by Kim et al. [36]. This difference might be related to better local control in these two studies, because regional lymph nodes might be included in the radiation field. The relationship between Ki-67 expression and outcome with lymphoma showed strong heterogeneity among various studies [37]. In Au et al.'s study overall survival did not differ significantly between patients with and without Ki-67>50% in univariate analysis [19]. We also found that AA III/IV shows no association with OS and PFS of patients with ENKTL, which is inconsistent with the results of other studies [30,36]. This may result from the limited number of AA III/IV (7, 3.7%) in our study and differences in primary treatment. However, a second study has reported results consistent with ours [19]. Notably, due to some common limitations such as retrospective research, sample size imbalance between Ann-Arbor staging I/II and III/IV, uniform treatment (radiotherapy-based treatment), clinical heterogeneity for ENKTL, and the fact that some risk factors show negative results in our study, further well designed prospective studies are necessary.
In conclusion, this study demonstrates a dependent association between high RDW and poorer clinical outcomes in ENKTL patients treated with radiotherapy-

Statistical analysis
All statistical analyses were conducted with IBM SPSS 20.0 (SPSS Inc., Chicago, IL), R software version 3.2.3, and Cutoff Finder. A web-based R software (Cutoff Finder, http://molpath.charite.de/cutoff) was used to determine the optimal cut-off value for RDW and HB [42]. Overall survival (OS) was defined as the period from clear diagnosis to death, lost follow-up or deadline. Progression free survival (PFS) was defined as the period from clear diagnosis of the tumor to first tumor progression, death, lost follow-up or deadline. Fisher's exact test was used to identify associations between categorical variables. The Kaplan-Meier method was used to investigate OS and PFS, and the log-rank test was used to evaluate the difference. When the P value of log-rank was close to 0.05, breslow test and tarone-ware test were utilized. When the P value was lower than 0.05, the corresponding factor was added into the multivariate analysis. Multivariate analysis was conducted by Cox proportional hazard model. Discrimination for survival data was investigated utilizing the C statistic with concordance index (C-index) [43,44]. The C-index can evaluate the model's ability to classify individual patients into risk groups with different prognoses by estimating the probability of concordance between predicted and observed outcomes. C-index was calculated using Hmisc R package in R software version 3.2.3 [45]. A two tailed P value <0.05 was considered statistically significant.

ACKNOWLEDGMENTS AND FUNDING
This study was supported by grants from Health and Family Planning Commission of Sichuan Province universal application project (17PJ516). We thank Michelle Wei from the Duke University for its linguistic assistance during the preparation of this manuscript and the reviewers for their intellectual support.