Prognostic models to predict survival in patients with advanced non-small cell lung cancer treated with first-line chemo- or targeted therapy

Background We aimed to assess the prognostic role of neutrophilia, lymphocytopenia and the neutrophil-to-lymphocyte ratio (NLR), and to design models to define the prognosis of patients receiving first-line chemo- or targeted therapy for advanced non-small cell lung cancer (NSCLC). Materials and Methods We retrospectively analysed 401 consecutive patients with advanced NSCLC treated with first line chemo- or targeted therapy. Patients were stratified into two groups with pre-treatment NLR ≥ 3.7 (Group A) vs. < 3.7 (Group B). The best NLR cut-off was identified by ROC curve analysis. Results At baseline 264 patients had NLR≥3.7 (Group A), whilst 137 had lower NLR (Group B). Median OS was 10.8 months and 19.4 months in the two groups (p < 0.001), while median PFS was 3.6 months and 5.6 months, respectively (p = 0.012). At multivariate analysis, ECOG-PS≥2, stage IV cancer, non-adenocarcinoma histology, EGFR wild-type status and NLR were predictors of worse OS. Stage IV cancer, wild type EGFR status and NLR≥3.7 were independent prognostic factors for worse PFS. Patients were stratified according to the presence of 0-1 prognostic factors (8%), 2-3 factors (73%) and 4-5 factors (19%) and median OS in these groups was 33.7 months, 14.6 months and 6.6 months, respectively (p < 0.001). Similarly, patients were stratified for PFS based on the presence of 0-1 prognostic factor (15%), 2 factors (41%) and 3 factors (44%). The median PFS was 8.3 months, 4.6 months and 3.3 months respectively (p < 0.001). Conclusion Pre-treatment NLR is an independent prognostic factor for patients with advanced NSCLC treated with first-line therapies.


INTRODUCTION
Lung cancer is the leading cause of cancer related death in both men and women [1]. Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancers. NSCLC is often insidious, producing no symptoms until the disease is advanced. Over the past few years a variety of prognostic and predictive factors have been investigated in patients with NSCLC, and several prognostic models proposed to predict the outcome of patients with NSCLC after surgery [2−4] or chemotherapy and/or radiotherapy [5−6]. Despite this the prognosis for patient with advanced NSCLC remains dismal, and novel approaches are required in order to optimize patient outcomes and guide treatment decisions.
Inflammation contributes to the pathogenesis www.impactjournals.com/oncotarget and progression of lung cancer. For example, chronic inflammatory lung diseases such as sarcoidosis and chronic obstructive pulmonary disease (COPD) are associated with a higher risk of lung cancer [7], whilst the chronic use of anti-inflammatory drugs seems to reduce the risk [8]. In vitro studies suggest that direct cell-cell interactions between neutrophils and NSCLC cells can induce the release of inflammatory mediators, which may promote tumor cell proliferation [9]. Indeed, NSCLC cells might secret immunoreactive IL-8 and stimulate polymorphonuclear neutrophils (PMNs) to release Arginase 1. Both molecules inhibit T-cell proliferation and favour tumor cell progression [10]. An elevated neutrophil count has been associated with poor prognosis in patients with NSCLC treated with chemotherapy, with a difference in overall survival (OS) of approximately 9 months compared to those with normal neutrophil count (19.3 vs. 10.2 months) [11].
Markers of inflammation, such as the neutrophilto-lymphocyte ratio (NLR), and their clinical significance in NSCLC patients are still under evaluation. NLR is an easily measurable parameter of systemic inflammation. Increased pre-treatment NLR has been demonstrated to be associated with poor outcome for various types of cancers including gastric cancer [12], advanced pancreatic cancer [13], hepatocellular carcinoma [14], colorectal  [15], bladder cancer [16], malignant mesothelioma [17], ovarian cancer [18] and renal cell carcinoma [19−22].
The aim of this study was to assess the prognostic role of pre-treatment neutrophilia, lymphocytopenia and NLR and to design a model to define the prognosis of patients receiving first-line chemo-or targeted therapy for advanced NSCLC.

Patient characteristics
Five hundreds and twenty-one patients were treated with first-line therapies. Of these, 401 patients (275 males and 126 female) were included in the NLR analysis, while 120 patients were excluded for the lack of data on pretreatment NLR.
The best NLR cut-off was ≥ 3.7 vs. < 3.7, as identified by ROC curve analysis ( Figure 1). Patients were further divided into two groups according to NLR. Two hundred and sixty-four patients (66%) had NLR ≥ 3.7 at baseline (Group A), while 137 (34%) had lower NLR (Group B).

Overall survival (OS)
The median OS from first-line therapy was 14.4 months (95% CI 12.4 to 16.9) in the total population. Two hundred and forty eight patients died during follow-up.
Furthermore, univariate analysis showed that gender (male), ECOG-PS ≥ 2, current or former smokers, tumor  stage IV, wild-type EGFR status, neutrophilia and NLR ≥ 3.7 were significantly associated with worst PFS (Table 3). Multivariate Cox regression analysis revealed that tumor stage IV, wild-type EGFR status and NLR ≥ 3.7 were independent prognostic factors for worse PFS ( Table 3).
As for response to first-line therapy, we compared by chi-squared test the rate of disease progressions in patients with NLR ≥ 3.7 vs. < 3.7. In the subgroup with higher NLR, the progression rate was 81% vs 71% in patients with NLR < 3.7, showing a significant difference between these two populations (p = 0.03).

DISCUSSION
A better understanding of the complexity of tumor-immune interactions in patients with NSCLC has facilitated an increased interest and development of immune therapeutic strategies. However the effects of chemotherapy and targeted agents on the immune system and its role in tumour response to therapies remains less clear.
Previous groups have examined pre-treatment NLR in NSCLC. In early-stage disease, NLR was positively associated with the prognosis for patients treated with stereotactic radiation [23] and surgery [24−26]. Zhang et al. showed that preoperative lymphocytopenia correlated with lymphatic invasion and a shorter disease-free survival (DFS, 318 vs. 669 days) in 142 patients with NSCLC who underwent lobectomy and lymph node dissection and adjuvant chemotherapy. Neutrophil count however was not associated with DFS [27]. Elevated pre-treatment NLR was significantly associated with worse OS in 81 advanced EGFR-mutated NSCLC patients treated with first-line EGFR TKIs [28], as well as in 199 never smokers with advanced NSCLC receiving gefitinib or standard chemotherapy [29]. In addition, Yao et al. found that high NLR was associated with shorter OS and PFS in patients with advanced NSCLC treated with first-line platinumbased chemotherapy [30]. More recently, three different meta-analyses have confirmed the prognostic role of NLR in patients with lung cancer [31−33].
To the best of our knowledge, this is the largest study investigating for the role of neutrophilia, lymphocytopenia and NLR in patients with NSCLC treated with first-line therapies. In our study, ECOG-PS ≥ 2, IV tumor stage, non-adenocarcinoma histology, EGFR wild-type status and NLR were predictors of worse OS, whilst tumor stage IV, wild-type EGFR status and NLR ≥ 3.7 were independent prognostic factors for worse PFS. Based on these data, we designed two prognostic models for OS and PFS. Median OS differed according to the presence of 0-1, 2-3 or 4-5 prognostic factors, thus identifying patients with good, intermediate and poor prognosis (33.7 months vs. 14.6 months vs. 6.6 months, p < 0.001). Similarly, when patients were stratified according to the presence of 0-1, 2 or 3 prognostic factors for PFS, different duration of response to first-line therapies became evident (8.3 months vs. 4.6 months vs. 3.3 months, p < 0.001).
There are some limitations to this study. First, this is a retrospective study, which is susceptible to bias in data selection and analysis. Other inflammatory markers such as C-reactive protein (CRP) or procalcitonin, are not routinely measured in our institutions and so were not included. Furthermore, NLR differs among individuals and can present fluctuations due to concurrent infections and other medications, factors that cannot be fully accounted for in this study. Moreover, some patients (13/41) with EGFR mutated tumors were not treated with EGFR-TKIs  Oncotarget 26922 www.impactjournals.com/oncotarget as first-line therapy due to unavailable data on EGFR mutational status at time of the beginning of treatment.
Despite these limitations, our study suggests that pre-treatment NLR is associated with PFS and OS in patients treated with first-line therapies for advanced NSCLC. Prospective studies are needed to evaluate and validate the prognostic models described here and adequately assess the potential role of NLR in guiding treatment decisions, patient selection, and clinical trials design.

Study population and data collection
The study population was adults with a histological or cytological diagnosis of locally advanced or metastatic NSCLC treated with first-line chemotherapy or targeted therapy according to EGFR mutational status at two institutions (Università Politecnica Marche, Italy and Chelsea & Westminster Hospital, UK) between 1 st May 2009 and 31th October 2014. Tumor stages were assessed according to the tumor-node-metastasis (TNM) criteria and included patients with stage IIIB and IV, as well as patients in stage IIIA not suitable for surgery, as defined in AJCC version 7 [34].
Patients were ineligible if they had received surgery or radiotherapy within one month from the start of firstline therapy and if they presented factors that could influence NLR (see below). Data was retrospectively collected from patients' medical records.
Treatment with first-line chemotherapy or TKIs was continued until evidence of disease progression on scans, unacceptable adverse events, or death. Followup generally consisted of regular physical examination and laboratory assessment (hematologic and serum biochemical measurements), and imaging studies by computed tomography (CT) or magnetic resonance imaging (MRI) scans was carried out according to local procedures every 8-12 weeks.
The OS was defined as the time from the beginning of first-line treatment until death from any cause. Progression free survival (PFS) was defined as the time from beginning of treatment to disease progression or death from any cause. Patients without tumour progression or death at the time of the data cut-off for the analysis or at the time of receiving an additional anticancer therapy were censored at their last date of adequate tumour evaluation.
Peripheral blood samples were obtained 1 to 7 days before the start of first-line therapy. Patients without available data on pre-treatment NLR and those with baseline comorbidities that might influence NLR, such as chronic lymphocytic leukaemia (CLL), chronic inflammatory diseases, and recent signs of infection or therapy with steroids or granulocyte colony stimulating factor (G-CSF), were excluded.

Statistical analysis
PFS and OS were estimated using Kaplan-Meier method with Rothman's 95% confidence intervals (CI) and compared across the groups using the log-rank test. Patients with a stable disease (SD), partial remission, and a complete remission were considered as responders.
Pre-treatment NLR was calculated by dividing the absolute neutrophil count by the absolute lymphocyte count and potential factors associated with outcome were evaluated, including patients' age (≥ 70y vs. < 70y), gender, tumor stage, histology, EGFR mutational status, Eastern Cooperative Oncology Group-Performance Status (ECOG-PS), smoking history, neutrophil count, lymphocyte count, and NLR. We determined by ROC analysis the value that best discriminated between good and poor survival.
Cox proportional hazards models were applied to explore patients' characteristics predictors of survival in univariate-and multivariable analysis. Variables not fitting at univariate analysis were excluded from the multivariate model. No-multicollinearity of the grouped co-variates was checked. Significance level in the univariate model for inclusion in the multivariate final model was more liberally set at a 0.2 level, according to Hosmer et al. [35,36]. The likelihood ratio test was conducted to evaluate the improvement in prediction performance gained by backward elimination of variables from the prognostic model [37]. All other significance levels were set at a 0.05 value and all P values were two-sided. Statistical analyses were performed using MedCalc version 11.4.4.0 (MedCalc Software, Broekstraat 52, 9030 Mariakerke, Belgium).