Neutrophil-To-Lymphocyte and Platelet-To-Lymphocyte Ratios as Prognostic Markers of Survival in Patients with Head and Neck Tumours—Results of a Retrospective Multicentric Study

Background: The neutrophil-to-lymphocyte ratio (NLR) and the platelet-to-lymphocyte ratio (PLR) may be useful for drawing conclusions about the survival of head and neck squamous cell carcinoma (HNSCC) patients. Methods: Clinical data of 156 patients managed for HNSCC at two head and neck surgery centres were analyzed retrospectively. We studied the relationships between survival and PLR as well as NLR. Results: With regards to 5-year survival, the difference between the two groups with PLR values lower or higher than the threshold was statistically significant (p = 0.004), and we found the same for disease-free survival (p = 0.05), and tumour-specific mortality (p = 0.009). Concerning NLR, the difference in tumour-specific survival was statistically significant (p = 0.006). According to the multivariate analysis, NLR values higher than the threshold indicated an enhanced risk for overall as well as for tumour-specific mortality. Conclusion: In HNSCC patients, a high NLR may be considered as an independent risk factor for 5-year overall survival.


Introduction
Head and neck squamous cell carcinoma (HNSCC) of mucosal origin is the sixth most common malignancy, with an estimated 900,000 new cases and about 350,000 fatalities a year worldwide [1]. The availability of an easy-to-determine prognostic factor could improve the effectiveness of individualized therapy.
Certain formed elements of the blood, as well as their ratios, can be used to establish the diagnosis and to predict prognosis.

Materials and Methods
The study was approved by the Scientific and Research Ethics Committee of the Hungarian Medical Research Council on 14 December 2015 (Nr. 8951-3/2015/EKU (0444/15).
We conducted a retrospective analysis on the clinical data of patients managed for primary HNSCC at two head and neck surgical centres (Uzsoki Street Hospital, Budapest and Medical Centre of the Hungarian Defence Forces, Budapest), between 2000 and 2014. The criteria for inclusion comprised the diagnosis of squamous cell carcinoma confirmed by histology, and the availability of platelet, neutrophil, and lymphocyte counts measured at the earliest one month before treatment. The exclusion criteria were as follows: the lack of an R0 resection, the presence of a synchronous tumour, inflammatory conditions (fistula of the neck, pneumonia, wound infection, abscess, cholecystitis, cannula sepsis, endocarditis, urinary tract infection, Crohn's disease, ulcerative colitis, etc.), thromboembolic events (deep vein thrombosis, pulmonary embolism, myocardial infarction), corticosteroid therapy, and treatment with platelet aggregation inhibitors during the month before blood sampling.
The data available on the factors predisposing to malignancy (smoking, alcohol consumption, infection by human papillomavirus or by the Epstein-Barr virus, etc.) were incomplete and hence, we did not take these into account in our study. Overall survival was defined as the period from the first day of treatment to the day of death from any cause-or to the last date of observation. Disease-free survival was calculated as the time from the start date of treatment and to the date of the detection of recurrence. Missing mortality data were supplemented with information obtained from public health databases maintained by the government.
Based on the above criteria, 393 out of the 549 patients were excluded owing to incomplete laboratory data and hence, 156 patients were included in the study. They were staged according to the 7th edition of the TNM classification of the American Joint Committee on Cancer. Due to the occasional uncertainty of the clinical information available, we had to work with diverse survival data. In particular, overall, disease-specific, and tumour-specific survivals were analysed using data available from 156, 150, and 149 patients.

Statistical Analysis
The relationships among categorical variables were analysed with the cross-table method and the χ 2 test. The non-parametric Mann-Whitney, U-, and Kruskal-Wallis tests were used to study the associations between the continuous variables PLR, NLR, and the subgroups of patients. The survival of patients with PLR and NLR values lower or higher than the threshold was analysed with the log-rank test and plotted as a Kaplan-Meier survival curve. The correlations between PLR, NLR, and survival were controlled for age, gender, and the characteristics of the malignancy, and then subjected to multivariate Cox regression analysis. The joint effect of discrete, multiple categorical variables was appraised with the Wald test. The validity of Cox models was verified by testing the proportional hazard assumption with the Grambsch and Ternau score test [19], the goodness of fit on 4 quantiles of risk with the May and Hosmer test [20], and the global significance of the models with the likelihood ratio (LR) test. All calculations were performed with version 14.2 of the Stata statistical software [21].
The optimal threshold values of PLR and NLR were determined as follows: the effect of individual PLR and NLR thresholds was studied as independent variables in logistic regression models designed to assess 5-year survival. Rounding PLR and NLR values to the accuracy of 1 and of 0.1, we calculated the area under the ROC curve (AUROC) for every threshold level. The optimal threshold values most effective for distinguishing among the patients by 5-year mortality, as determined based on AUROC peaks were 107 for PLR ( Figure 1A

Statistical Analysis
The relationships among categorical variables were analysed with the cross-table method and the χ² test. The non-parametric Mann-Whitney, U-, and Kruskal-Wallis tests were used to study the associations between the continuous variables PLR, NLR, and the subgroups of patients. The survival of patients with PLR and NLR values lower or higher than the threshold was analysed with the logrank test and plotted as a Kaplan-Meier survival curve. The correlations between PLR, NLR, and survival were controlled for age, gender, and the characteristics of the malignancy, and then subjected to multivariate Cox regression analysis. The joint effect of discrete, multiple categorical variables was appraised with the Wald test. The validity of Cox models was verified by testing the proportional hazard assumption with the Grambsch and Ternau score test [19], the goodness of fit on 4 quantiles of risk with the May and Hosmer test [20], and the global significance of the models with the likelihood ratio (LR) test. All calculations were performed with version 14.2 of the Stata statistical software [21].
The optimal threshold values of PLR and NLR were determined as follows: the effect of individual PLR and NLR thresholds was studied as independent variables in logistic regression models designed to assess 5-year survival. Rounding PLR and NLR values to the accuracy of 1 and of 0.1, we calculated the area under the ROC curve (AUROC) for every threshold level. The optimal threshold values most effective for distinguishing among the patients by 5-year mortality, as determined based on AUROC peaks were 107 for PLR ( Figure 1A) and 3.9 for NLR. ( Figure 1B)

Results
The mean age of the subjects was 58.1 years (± SD: 8.7); and the proportion of females was 19.9%. The median duration of follow-up was 36 months, whereas the shortest duration was 4.9 and the longest was 161 months. Sixty-three patients died during the study, and 43 of them from malignant disease. Recurrence occurred in 72 patients. Median survival was 124.4 months (95% CI 71.6-∞), whereas median disease-free survival was 78.5 months (95% CI 33.5-∞).
The distributions of both PLR and NLR were skewed to the right ( Figure 2). Median PLR was 113 (IQR: 84.5 to 160), and its value reached or exceeded the threshold of 107 in 53.9% of the patients. Median NLR was 2.8 (IQR: 2.1 to 3.8), and 20.5% of the patients had an NLR value corresponding to the threshold of 3.9 or higher-the latter was observed only in combination with PLR values higher than the threshold. PLR was elevated in 33.3% of the patients, and neither PLR nor NLR was elevated in 46.2%. Considering the demographical parameters and the characteristics of the malignancy, a significant relationship of PLR and NLR values was detected only with tumour stage (Kruskal-Wallis test p < 0.001, and p = 0.004), and with the primary extension of the tumour (Figure 3). Red blood cell

Results
The mean age of the subjects was 58.1 years (± SD: 8.7); and the proportion of females was 19.9%. The median duration of follow-up was 36 months, whereas the shortest duration was 4.9 and the longest was 161 months. Sixty-three patients died during the study, and 43 of them from malignant disease. Recurrence occurred in 72 patients. Median survival was 124.4 months (95% CI 71.6-∞), whereas median disease-free survival was 78.5 months (95% CI 33.5-∞).
The distributions of both PLR and NLR were skewed to the right ( Figure 2). Median PLR was 113 (IQR: 84.5 to 160), and its value reached or exceeded the threshold of 107 in 53.9% of the patients. Median NLR was 2.8 (IQR: 2.1 to 3.8), and 20.5% of the patients had an NLR value corresponding to the threshold of 3.9 or higher-the latter was observed only in combination with PLR values higher than the threshold. PLR was elevated in 33.3% of the patients, and neither PLR nor NLR was elevated in 46.2%. Considering the demographical parameters and the characteristics of the malignancy, a significant relationship of PLR and NLR values was detected only with tumour stage (Kruskal-Wallis test p < 0.001, and p = 0.004), and with the primary extension of the tumour ( Figure 3). Red blood cell count (median: 4.57; IQR: 4.31 to 4.93) was not correlated with PLR (r = −0.019, p = 0.811) or with NLR (r = −0.023, p = 0.770). Platelet count (median: 270 G/L; IQR: 220.5 to 333) exhibited moderate correlation with PLR (r = 0.680, p < 0.001), and weak correlation with NLR values (r = 0.227, p = 0.004). The study population was divided into three groups by primary treatment modality-that is, into groups formed by patients treated with surgery, by radio-/chemotherapy, or radiotherapy. Considering primary and secondary therapies together, 136 patients underwent surgery, 65 were treated with chemotherapy, and 49 with radiotherapy. The study population was divided into three groups by primary treatment modality-that is, into groups formed by patients treated with surgery, by radio-/chemotherapy, or radiotherapy. Considering primary and secondary therapies together, 136 patients underwent surgery, 65 were treated with chemotherapy, and 49 with radiotherapy.  The subjects were also studied after grouping by the site of origin of the tumour ( Table 1). The age distribution, tumour stage, the proportion of patients with a PLR value higher than the threshold, and of those surviving for 5 years differed significantly by tumour location. The proportion of  2 4 6 8 10 12 14 16 Neutrophil-to-lymphocyte ratio B In the box plots, the thick horizontal band represents the median value, the box corresponds to the interquartile range, and the two whiskers indicate 1.5× interquartile range (or minimum/maximum values when located within that range). The circles show outlier values. The study population was divided into three groups by primary treatment modality-that is, into groups formed by patients treated with surgery, by radio-/chemotherapy, or radiotherapy. Considering primary and secondary therapies together, 136 patients underwent surgery, 65 were treated with chemotherapy, and 49 with radiotherapy.  The subjects were also studied after grouping by the site of origin of the tumour ( Table 1). The age distribution, tumour stage, the proportion of patients with a PLR value higher than the threshold, and of those surviving for 5 years differed significantly by tumour location. The proportion of The subjects were also studied after grouping by the site of origin of the tumour ( Table 1). The age distribution, tumour stage, the proportion of patients with a PLR value higher than the threshold, and of those surviving for 5 years differed significantly by tumour location. The proportion of patients with PLR values higher than the threshold was the largest (71.7%) among those with tumours of the hypopharynx, and the smallest (25.0%) among those with oropharyngeal malignancies. Over a period of five years, 24.6% of the patients with a PLR value under the threshold of 107 died, whereas the mortality rate was 46.4% among those with a PLR ≥107-the difference between these two groups was statistically significant (log-rank test χ 2 (df = 1) = 8.14, p = 0.004). Over five years, tumour recurrence occurred in 37.1% of patients with a PLR value lower than the threshold, and in 52.2% of those with a PLR value higher than the threshold. The difference between the two groups was significant also with regard to disease-free survival (log-rank test χ 2 (df = 1) = 4.03, p = 0.045), as well as tumour-specific mortality (log-rank test χ 2 (df = 1) = 6.89, p = 0.009). The mortality rate of patients with an NLR value under the threshold of 3.9 was 32.3%, whereas 56.5% of patients with NLR ≥ 3.9 died. The five-year recurrence rate was 43.0% in the former and 55.2% in the latter group; this difference was not statistically significant (log-rank test χ 2 (df = 1) = 2.21, p = 0.138). By contrast, a statistically significant difference in tumour-specific survival was noted between the two groups (log-rank test χ 2 (df = 1) = 7.43, p = 0.006). We also studied the correlation between the combination of PLR and NLR values and five-year mortality. The subjects could be categorised into the following three groups: (1) Both NLR and PLR were higher than the threshold, (2) only PLR was higher than the threshold, and (3) both NLR and PLR were lower than the threshold. We found significant differences among the three groups in overall survival (log-rank test χ 2 (df = 2) = 12.78, p = 0.002), and in tumour-specific survival (log-rank test χ 2 (df = 2) = 9.92, p = 0.007). However, disease-free survival was not significantly different among these three groups (log-rank test χ 2 (df = 2) = 4.45, p = 0.108). The chance for five-year survival was the highest in patients who had both NLR and PLR values under the threshold, and it was the lowest in those in whom both ratios were elevated (Figure 4).  We analysed the five-year mortality, tumour-specific mortality, and disease-free survival with multivariate Cox-regression, using the impact of PLR and of NLR, as well as the joint effect of these two markers as independent variables. Age, gender, tumour stage, the grade of differentiation, initial region, red blood cell count, and thrombocytosis were used as control variables. In our study population, NLR values higher than the threshold were observed only in combination with elevated PLR values and therefore, studying the effect of isolated NLR elevation was not possible. The assumption for the proportional hazard was not fulfilled in the models designed to study tumourspecific mortality and accordingly, their coefficients are of limited validity. Among the control variables, male gender and advanced tumour stage were associated with an enhanced risk with respect to survival. Further, the likelihood of tumour recurrence was significantly correlated with male gender and with the grade of tumour differentiation. There was an inverse relationship between red blood cell count and the risk of mortality and of recurrence-that is, the elevation of red blood cell count was associated with a lower risk of death or tumour recurrence (Table 2). In models controlled for the extension of T stage-analysed instead of or in combination with TNM stage-the  We analysed the five-year mortality, tumour-specific mortality, and disease-free survival with multivariate Cox-regression, using the impact of PLR and of NLR, as well as the joint effect of these two markers as independent variables. Age, gender, tumour stage, the grade of differentiation, initial region, red blood cell count, and thrombocytosis were used as control variables. In our study population, NLR values higher than the threshold were observed only in combination with elevated PLR values and therefore, studying the effect of isolated NLR elevation was not possible. The assumption for the proportional hazard was not fulfilled in the models designed to study tumour-specific mortality and accordingly, their coefficients are of limited validity. Among the control variables, male gender and advanced tumour stage were associated with an enhanced risk with respect to survival. Further, the likelihood of tumour recurrence was significantly correlated with male gender and with the grade 7 of 12 of tumour differentiation. There was an inverse relationship between red blood cell count and the risk of mortality and of recurrence-that is, the elevation of red blood cell count was associated with a lower risk of death or tumour recurrence (Table 2). In models controlled for the extension of T stage-analysed instead of or in combination with TNM stage-the conclusions regarding overall survival remained unchanged. The models containing platelet count as a continuous control variable similarly did not influence the negative prognostic role of NLR for overall survival. However, the coefficients of the model of tumour-specific mortality did not prove stable upon adjusting the control variables, and the NLR value lost significance following the addition of T stage as a control variable.

Discussion
Patients with a high platelet count are at an increased risk of cancer, as well as certain malignancies which exhibit a closer relationship with thrombocytosis [22,23]. Previous studies confirmed the prognostic significance of platelet count in a variety of solid (renal-cell, hepatocellular, lung, colorectal, gastric, and gynaecological) tumours [24][25][26][27][28][29].
The association between elevated PLR or NLR and worse prognosis was observed also in HNSCC [61,[63][64][65][66], although these studies were limited to tumours of a specific location or stage, or to patients receiving a certain treatment modality. According to our findings, elevated PLR was associated with a significantly inferior five-year survival and disease-free survival, as well as with greater tumour-specific mortality. In case of elevated NLR, by contrast, only the increase of tumour-specific mortality was significant. Primary treatment modality had no significant influence on NLR/PLR values.
RASSOULI et al. found that when considered together as a prognostic marker for survival, the combination of NLR and PLR proves as reliable as does TNM staging [62]. Our study yielded similar findings, because the chance for five-year survival was the highest if neither NLR nor PLR exceeded the threshold level. Conversely, survival rate was the lowest when both NLR and PLR were elevated.
In another study, the prognostic value of NLR was also assessed by the primary location of naso-/oro-/hypopharyngeal or laryngeal tumours; however, only patients undergoing radiotherapy were included. A significant relationship was demonstrated for nasal, hypopharyngeal, and laryngealbut not for oropharyngeal-tumours [60]. In our study, we did not detect any significant differences between NLR and the different primary locations of the tumours. By contrast, the proportion of patients with PLR values higher than the threshold differed significantly by primary site: it was the highest with hypopharyngeal and the lowest with oropharyngeal tumours. HPV status was not available for our analysis, but the frequent occurence of HPV-positive tumours in the oropharynx may give an explanation for the above-mentioned lower incidence of high PLR value at this tumour site. Some studies suggest that HPV infection might alter inflammatory response by altering white blood cell distribution and thus affect the PLR and NLR values [67]. Rosculet et al. obtained similar results in their retrospective study. They found that NLR is an indicator of both disease-free and overall survival, but it does not have independent prognostic significance when HPV status is incorporated into multivariate analysis [68].
When used as prognostic factors, the advantage of NLR and PLR is that both can be determined by routine laboratory methods. However, both are influenced by inflammation or anaemia, and this may undermine their reliability. This is why we defined active inflammation among the exclusion criteria, as well as used red blood cell count as a control variable during the multivariate analysis. According to the multivariate Cox regression analysis controlled in this fashion, PLR had no significant influence on mortality risk in any of the models. By contrast, NLR values higher than the threshold represented an enhanced risk both for overall and for tumour-specific mortality. Neither PLR nor NLR had a significant impact on the risk of tumour recurrence.

Conclusions
NLR values over the threshold of 3.9 (determined using the data of our patients) may be regarded as an independent risk factor for five-year survival in malignancies of the head and neck region. NLR was closely associated, both with the size of the primary tumour and with the stage of the malignant disease. Furthermore, NLR values higher than the threshold represented an enhanced risk regarding overall survival, as well as in other models controlled for the extension of the primary tumour, for the stage of the malignant disease, and for other demographical parameters, or disease characteristics. In view of the foregoing, therefore, we recommend determining NLR in routine clinical practice as a risk factor for five-year overall survival. On the other hand, we could not demonstrate a significant relationship between tumour-free survival and NLR, and this relationship was not consistent with regard to tumour-specific survival.
Nevertheless, it would be desirable to undertake a similar study in a larger population of patients, and with other malignancies (of the thyroid or salivary glands, for example) of the head and neck region.