Prognostic Significance of Preoperative Neutrophil-to-Lymphocyte Ratio in Surgically Resectable Pancreatic Neuroendocrine Tumors.

BACKGROUND The aim of this study was to evaluate the predictive and prognostic value of the preoperative neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR) in pancreatic neuroendocrine tumor (PNET) patients undergoing potentially curative resection. MATERIAL AND METHODS A retrospective review of 172 patients with PNETs was conducted. Kaplan-Meier curves and multivariate Cox proportional models were used to calculate overall survival (OS) and disease-free survival (DFS). The predictive performance of the NLR was compared with other inflammation-based scores and conventional stratification systems using receiver operating characteristic (ROC) curve analysis. RESULTS Elevated NLR and PLR were both associated with advanced AJCC stage and high grade. In the univariate analysis, elevated NLR and PLR were both significantly associated with decreased OS and DFS. In the multivariate analysis, the preoperative NLR, but not the PLR, was an independent risk factor for OS (HR=4.471, 95% CI 1.531-13.054, p=0.006) and DFS (HR=2.531, 95% CI 1.202-5.329, p=0.015). The discriminatory capability of the NLR was superior to that of other inflammation-based scores in OS prediction. Furthermore, the predictive range was expanded by incorporating the NLR into the conventional stratification systems, including the AJCC stage and WHO classification systems. CONCLUSIONS As an independent prognostic factor, an elevated preoperative NLR is superior to the PLR with respect to predicting clinical outcomes in PNET patients undergoing potentially curative resection. The incorporation of the NLR into the existing conventional stratification systems improved the predictive accuracy.

Background

Pancreatic neuroendocrine tumors (PNETs) are a heterogeneous group of neoplasms with multiple clinical characteristics. PNETs are considered more indolent tumors and are associated with better long-term survival rates than with tumors of the exocrine pancreas [1-3]. PNETs account for approximately 1–2% of all pancreatic neoplasms and 7.0% of all neuroendocrine tumors [4]. The annual incidence of PNETs in the United States is estimated to range between 2 and 5 cases per one million individuals but appears to be rising [5]. PNETs can be classified as either functional or nonfunctional. The majority (60%–90%) of PNETs are nonfunctional. Complete surgical resection of a PNET has been suggested to be the only potentially curative treatment for the disease, similar to pancreatic adenocarcinoma [6]. PNETs have a better prognosis than pancreatic ductal adenocarcinoma (PDAC). The 5-year survival rate after radical resection is approximately 86.4% [7]. Several studies have found that many host-related factors affect survival in PNETs. Intrinsic tumor characteristics, such as tumor sizes, stages, and grades; Ki-67 indices; and lymph node involvement, have long been shown to be associated with clinical outcomes [8-10]. Other clinical and pathologic factors used to predict disease progression include patient age at diagnosis, visceral pleural invasion status, and margin status [11]. Information regarding these factors is generally useful, but most of these factors are determined only after surgery. Therefore, it is necessary to search for potential prognostic indicators that are available before surgery.

Increasing evidence suggests that inflammatory cells are an essential component of the tumor microenvironment and play a role in tumor progression [12-15]. Inflammatory chemokines, including neutrophil-attracting cysteine-X-cysteine chemokines, can either be triggered by the tumor itself or be part of the host innate response to the cancer. Markers of systemic inflammation, such as the neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), are generally appealing to clinicians, as these laboratory data are routinely collected prior to surgery and are therefore readily available. The NLR and PLR have been identified as prognostic factors. Elevated NLR and PLR have been shown to be correlated with advanced stages and poor prognoses in a variety of human tumors, including colorectal cancer [16], hepatocellular carcinoma [17], breast cancer [18], gastric neuroendocrine tumors [19], and pancreatic cancer [20]. However, the roles of NLR and PLR and the clinical significance of these parameters in PNETs remain under evaluation. The goal of this study was to assess the prognostic value of the NLR and PLR in patients with PNETs following potentially curative resection. Further, we compared the discriminative ability of the NLR with other inflammation-based scores to determine whether the NLR is a useful marker for predicting patient outcomes. Additionally, we refined the existing stratification systems by incorporating the NLR into the existing TNM staging system or WHO classification.

Material and Methods

Study population

Patients who underwent surgical resection for PNETs from November 2003 to August 2016 at the First Affiliated Hospital, Zhejiang University School of Medicine, were retrospectively reviewed. The diagnosis of PNET was made based on standard histologic criteria. The TNM stage of each PNET was determined based on the American Joint Committee on Cancer TNM Classification, while the grade of each PNET was determined according to the 2010 WHO classification of NETs of the GEP system. Patients were excluded if they: (1) showed clinical evidence of infection or evidence of hyperpyrexia at the time of diagnosis; (2) were treated for recurrent disease; (3) received preoperative radiochemotherapy prior to surgery; (4) had a history of cancer of any type; or (5) did not consent to the use of their medical records for research purposes. We included only those patients who had survived for at least 60 days after surgery to exclude perioperative mortality-related bias. All patients underwent potentially curative resection. Finally, 172 patients with PNETs were included (Figure 1). We also enrolled 172 healthy volunteers with similar age and sex distributions from the physical examination center in our hospital. The NLR was calculated by dividing the absolute neutrophil count by the absolute lymphocyte count. The PLR was calculated by dividing the absolute platelet count by the absolute lymphocyte count, while the lymphocyte-to-monocyte ratio (LMR) was calculated by dividing the absolute lymphocyte count by the absolute monocyte count on preoperative routine blood tests. The study was approved by the Ethics Committee of the First Affiliated Hospital of Zhejiang University School of Medicine.

Figure 1

Flowchart of patient inclusion and exclusion.

Follow-up

Patient follow-up was performed by reviewing hospital records or contacting patient family members. Overall survival (OS) was defined as the time span extending from the date of initial diagnosis until the date of death from any cause or the date of last known contact. Disease-free survival (DFS) was defined as the time extending from the date of surgery to the date of PNET recurrence. Patients who did not have evidence of local recurrence or metastasis at the last follow-up and patients who had died of diseases unrelated to PNETs were censored in the analysis of DFS. Our department follows up with patients every 6 months for the first 5 years after surgery and then yearly thereafter. The following postoperative follow-up data were collected for each patient: clinical symptoms and signs, laboratory test results, and radiological examination results. Once recurrence was confirmed, patients were treated by repeat tumor resection, radiofrequency ablation (RFA), transarterial chemoembolization (TACE), systemic chemotherapy, and somatostatin analogue therapy, according to the sizes, numbers, and locations of their recurrent tumors.

Statistical analysis

All statistical analyses were performed using SPSS 16.0 software (SPSS, Chicago, IL, USA) for Windows. Differences in the NLR and PLR between patients and healthy subjects were evaluated by t tests in the case of normally distributed variables or by the Mann-Whitney U test in the case of abnormally distributed variables. Area under the curve (AUC) values obtained from the receiver operating characteristic (ROC) curve analysis were used to compare the predictive efficacies of the NLR and other inflammation-based scores. The associations between NLR and PLR and other prognostic factors were analyzed using chi-square and Fisher exact tests. The Kaplan-Meier method and the log-rank test were used to calculate OS and DFS. Prognostic analysis was performed using univariate and multivariate Cox regressions models. A p value <0.05 was considered statistically significant.

Results

Patient characteristics

A total of 172 patients with histologically confirmed PNETs were included in the present analysis. These patients were diagnosed at a mean age of 52.92±12.55 years and were evaluated over a mean follow-up period of 48.04±35.2 months. Seventy-three (42.4%) of these patients had grade 1 disease, 76 (44.2%) had grade 2 disease, and the remaining 23 (13.4%) had grade 3 disease. The majority of patients (150/172, 87.2%) had stage I or II disease. A total of 166 patients underwent curative resections (R0 resection, 96.5%), while palliative surgery (R1 resection, 3.5%) was performed for only 6 patients. The operative procedures included the distal pancreatectomy (n=89), pancreaticoduodenectomy (n=53), enucleation (n=23), middle pancreatectomy (n=4), and total pancreatectomy (n=3). The pathology showed lymph node metastasis in 33 (19.2%) patients. At the time of the last follow-up visit, 46 patients had relapsed and 28 patients had died. The 1-, 3- and 5-y OS rates for the entire cohort were 98%, 90% and 78%, respectively, and the 1-, 3- and 5-y DFS rates for the entire cohort were 84%, 72%, and 71%, respectively.

Blood NLRs and PLRs were elevated in patients with PNETs

As shown in Table 1, platelet counts and lymphocyte counts were significantly lower in the blood of patients with PNETs than in the blood of normal volunteers (NVs) (both p<0.05). The PLR, NLR, and neutrophil counts were significantly higher in the patients with PNETs than in NVs (all p<0.05).

Table 1

Comparison of blood cell counts between PNET patients and NVs.

Variables	PNET (N=172)	NVs (N=172)	p Values
Neutrophil counts (109/l)	3.60±1.58	3.08±0.80	0.003
Platelet counts (109/l)	200.69±64.93	216.75±51.32	0.039
Lymphocyte counts (109/l)	1.64±0.56	2.08±0.58	<0.001
NLR	2.48±1.57	1.55±0.45	<0.001
PLR	133.39±58.71	110.48±35.68	<0.001

PNETs – pancreatic neuroendocrine tumours; NVs – normal volunteers; NLR – neutrophil to lymphocyte ratio; PLR – platelet to lymphocyte ratio.

Correlations between the NLR and PLR and other PNET clinical parameters

Baseline patient demographic and clinicopathologic characteristics, which were stratified by NLR and PLR, are summarized in Table 2. Preoperatively, the NLR was >2.31 in 67 (39.0%) patients, while the PLR was >153.4 in 50 (29.1%) patients. Univariate analysis revealed that an increased NLR and PLR were both associated with advanced AJCC stage, high grade, high alkaline phosphatase (AKT) level, and R1 resection (all p<0.05). Additionally, elevated NLR was associated with the presence of vascular lymph node invasion (LVSI) (p=0.027) and perineural invasion (p=0.038). In contrast, an elevated PLR was associated with tumor location (p=0.048).

Table 2

Baseline characteristics of the PNET patients stratified by the NLR and PLR.

Variables	NLR			PLR
	≤2.31 (N=105)	>2.31 (N=67)	p Values	≤151.4 (N=122)	>151.4 (N=50)	p Values
Age (years)			0.576			0.947
≤56	61	36		69	28
>56	44	31		53	22
Albumin (g/l)			0.374			0.097
≤45	81	45		85	41
>45	24	22		37	9
Grade			0.003			0.045
G1	54	19		58	15
G2	40	36		49	27
G3	11	12		15	8
AJCC stage			0.011			0.005
I–II	97	53		112	38
III–IV	8	14		10	12
LVSI			0.027			0.108
No	86	45		97	34
Yes	19	22		25	16
Tumour size			0.071			0.481
≤3.5 cm	67	33		73	27
>3.5 cm	38	34		49	23
Sex			0.068			0.802
Female	62	30		66	26
Male	43	37		56	24
Tumour location			0.139			0.048
Head/uncinate	31	28		35	24
Neck	12	6		14	4
Body/tail	62	33		73	22
Symptomatic diagnosis			0.13			0.085
No	40	18		46	12
Yes	65	49		76	38
AKT (U/l)			0.011			<0.001
≤124	97	53		113	37
>124	8	14		9	13
Radical resection			0.002			0.039
R0	105	61		120	46
R1	0	6		2	4
Perineural invasion			0.038			0.070
No	96	54		110	40
Yes	9	13		12	10
Function			0.724			0.450
No	79	52		91	40
Yes	26	15		31	10

PNETs – pancreatic neuroendocrine tumours; NLR – neutrophil to lymphocyte ratio; PLR – platelet to lymphocyte ratio; LVSI – vascular lymph node invasion; AKT – alkaline phosphatase.

Prognostic significance of variables and cut-off value determination

The results of the univariate survival analysis for each of the clinicopathologic variables are shown in Tables 3 and 4. Both a high NLR and a high PLR were significantly associated with a poor prognosis. Patients with a high NLR or PLR had shorter OS (HR=4.907, 95% CI 2.048–11.756, p<0.001 and HR=3.307, 95% CI 1.499–7.297, p=0.003, respectively) and DFS (HR=4.143, 95% CI 2.229–7.701, p<0.001 and HR=2.617, 95% CI 1.465–4.675, p=0.001, respectively) than patients with a low NLR or PLR (Figure 2). Furthermore, high tumor grade, presence of LVSI and perineural invasion, large tumor, high stage, increased AKT level, high LMR, nonfunctional tumor, and symptomatic tumor were prognostic factors for poor OS (p<0.05 for all). High grade, presence of LVSI and perineural invasion, large tumor, high stage, increased AKT level, high LMR, nonfunctional tumor, and male sex were associated with poor DFS (p<0.05 for all).

Table 3

Variables associated with OS according to the Cox proportional hazards regression model.

Variables	Univariable analysis			Multivariable analysis
	HR	95%CI	p Values	HR	95%CI	p Values
Age (years)			0.076
≤56	Reference
>56	2.085	0.926–4.696
NLR			<0.001			0.006
≤2.31	Reference			Reference
>2.31	4.907	2.048–11.756		4.471	1.531–13.054
PLR			0.003			0.205
≤151.4	Reference			Reference
>151.4	3.307	1.499–7.297		NA	NA
LMR			<0.001			0.228
≤3.22				Reference
>3.22	0.168	0.067–0.425		NA	NA
Albumin (g/l)			0.249
≤45	Reference
>45	1.640	0.707–3.804
Grade			<0.001			0.001
G1	Reference			Reference
G2	6.666	1.491–29.805		0.851	0.114–6.367
G3	45.599	9.676–214.977		5.360	0.616–46.666
AJCC stage			<0.001			0.004
I–II	Reference			Reference
III–IV	4.034	2.609–6.239		3.946	1.568–9.929
LVSI			<0.001			0.531
No	Reference			Reference
Yes	9.777	4.152–23.021		NA	NA
Tumour size			0.023			0.782
≤3.5 cm	Reference			Reference
>3.5 cm	2.557	1.135–5.760		NA	NA
Sex			0.068
Female	Reference
Male	2.161	0.945–4.941
Tumour location			0.084
Head/uncinate	Reference
Neck	1.553	0.201–11.996
Body/tail	0.398	0.177–0.896
AKT (U/L)			<0.001			0.519
≤124	Reference			Reference
>124	5.442	2.37–12.495		NA	NA
Radical resection			0.006			0.025
R0	Reference			Reference
R1	5.566	1.655–18.724		5.059	1.231–20.798
Perineural invasion			<0.001			0.036
No	Reference			Reference
Yes	6.943	3.027–15.925		2.683	1.065–6.763
Function			0.03			0.245
No	Reference			Reference
Yes	0.109	0.015–0.808		NA	NA
Symptomatic diagnosis			0.044			0.069
No	Reference			Reference
Yes	3.475	1.056–11.658		NA	NA

PNETs– pancreatic neuroendocrine tumours; NLR – neutrophil to lymphocyte ratio; PLR – platelet to lymphocyte ratio; LMR – lymphocyte to monocyte ratio; OS – overall survival; LVSI – vascular lymph node invasion; AKT – alkaline phosphatase; NA– not available.

Table 4

Variables associated with DFS according to the Cox proportional hazards regression model.

Variables	Univariable analysis			Multivariable analysis
	HR	95%CI	p Values	HR	95%CI	p Values
Age (years)			0.087
≤56	Reference
>56	1.663	0.929–2.976
NLR			<0.001			0.015
≤2.31	Reference			Reference
>2.31	4.143	2.229–7.701		2.531	1.202–5.329
PLR			0.001			0.275
≤151.4	Reference			Reference
>151.4	2.617	1.465–4.675		NA	NA
LMR			0.001			0.477
≤3.22	Reference			Reference
>3.22	0.365	0.203–0.654		NA	NA
Albumin (g/l)			0.426
≤45	Reference
>45	1.299	0.682–2.470
Grade			<0.001			0.001
G1	Reference			Reference
G2	10.426	3.16–34.398		1.659	0.413–6.663
G3	42.931	12.345–149.297		5.632	1.303–24.349
AJCC stage			<0.001			0.003
I–II	Reference				Reference
III–IV	4.899	3.406–7.045		2.771	1.423–5.397
LVSI			<0.001			0.163
No	Reference			Reference
Yes	5.363	2.981–9.648		NA	NA
Tumour size			0.001			0.611
≤3.5 cm	Reference			Reference
>3.5 cm	2.812	1.547–5.112		NA	NA
Sex			0.001			0.105
Female	Reference			Reference
Male	2.856	1.54–5.297		NA	NA
Tumour location			0.173
Head/uncinate	Reference
Neck	0.424	0.126–1.429
Body/tail	0.615	0.338–1.121
AKT (U/l)			<0.001			0.107
≤124	Reference			Reference
>124	3.7	1.932–7.087		NA	NA
Perineural invasion			<0.001			0.036
No	Reference			Reference
Yes	5.423	2.884–10.197		1.989	1.001–3.95
Function			0.003			0.43
No	Reference			Reference
Yes	0.114	0.028–0.472		NA	NA
Symptomatic diagnosis			0.109
No	Reference
Yes	1.741	0.883–3.430

PNETs – pancreatic neuroendocrine tumours; NLR – neutrophil to lymphocyte ratio; PLR – platelet to lymphocyte ratio; LMR – lymphocyte to monocyte ratio; DFS – disease-free survival; LVSI – vascular lymph node invasion; AKT – alkaline phosphatase; NA – not available.

Figure 2

Kaplan-Meier survival curves showing OS (A) and DFS (B) stratified by NLR in patients who underwent surgery for PNETs and OS (C) and DFS (D) stratified by PLR.

In multivariate analysis, NLR remained significantly associated with OS (HR=4.471, 95% CI 1.531–13.054, p=0.006) and DFS (HR=2.531, 95% CI 1.202–5.329, p=0.015) (Tables 3, 4). Furthermore, WHO grade, AJCC stage, perineural invasion, and radical resection were independent predictive factors for OS (p<0.05 for all measurements, Table 3), whereas WHO grade, AJCC stage, and perineural invasion were independent predictive factors for DFS (p<0.05 for all, Table 4).

ROC curve analysis showed that the AUCs of the NLR, PLR, LMR, and tumor size were 0.785, 0.67, 0.747, and 0.709, respectively, and that the best cut-off values for the above parameters were 2.31, 151.4, 3.22, and 3.5, respectively, as these values were both the most sensitive and the most specific in predicting survival (Figure 3).

Figure 3

ROC curve for the NLR, PLR, LMR, and tumor size in resectable PNETs.

Subgroup analyses of the parameters associated with the NLR

We investigated the prognostic value of the NLR relative to AJCC stage, grade, age, sex, LVSI, tumor size, and perineural invasion (Table 5). We noted a strong association between NLR and OS, which was independent of AJCC stage (p=0.002 for stage I/II; p=0.041 for stage III/IV), tumor size (tumor size ≤3.5 cm, p=0.016; tumor size >3.5 cm, p=0.006), and sex (male, p=0.021; female, p=0.007). We also noted a strong association between NLR and DFS, which was independent of age (age ≤56, p=0.017; age >56, p<0.001), tumor size (tumor size ≤3.5 cm, p=0.002; tumor size >3.5 cm, p=0.007), and sex (male, p=0.005; female, p=0.003). Furthermore, a high NLR was significantly associated with shorter OS in older (>56 y) (Figure 4A) patients with low-grade tumors (1 and 2) (Figure 4B), patients with tumors without perineural invasion (Figure 4C), and patients with tumors without LVSI (Figure 4D). Moreover, a high NLR was associated with poor DFS in patients with stage I/II disease (Figure 5A), patients with low-grade tumors (Figure 5B), patients with tumors without perineural invasion (Figure 5C), and patients with tumors without LVSI (Figure 5D) (p<0.01 for all measurements).

Table 5

Subgroup analysis for OS and DFS according to NLR.

Variables	NLR	N (%)	OS			DFS
			HR	95%CI	p Values	HR	95%CI	p Values
Age (years)
≤56	≤2.31	61 (62.9)	Reference			Reference
	>2.31	36 (37.1)	2.506	0.795–7.9	0.117	2.822	1.203–6.62	0.017
>56	≤2.31	44 (58.7)	Reference
	>2.31	31 (41.3)	24.399	3.162–188.259	0.002	6.105	2.389–15.597	<0.001
Grade
G1/G2	≤2.31	94 (63.1)	Reference			Reference
	>2.31	55 (36.9)	11.078	2.454–50.012	0.002	5.856	2.588–13.252	<0.001
G3	≤2.31	11 (47.8)	Reference
	>2.31	12 (52.2)	2.224	0.587–8.426	0.239	1.512	0.548–4.175	0.425
AJCC stage
Stage I/II	≤2.31	97 (64.7)	Reference			Reference
	>2.31	53 (35.3)	8.094	2.219–29.521	0.002	8.271	2.272–30.106	0.001
Stage III/IV	≤2.31	8 (36.4)	Reference			Reference
	>2.31	14 (63.6)	2.902	1.047–8.042	0.041	1.783	0.87–3.655	0.114
LVSI
No	≤2.31	86 (65.6)	Reference			Reference
	>2.31	45 (34.4)	12.69	1.527–105.478	0.019	5.581	2.159–14.426	<0.001
Yes	≤2.31	19 (46.3)	Reference			Reference
	>2.31	22 (53.7)	2.692	0.954–7.598	0.061	2.16	0.945–4.938	0.068
Tumour size
≤3.5 cm	≤2.31	67 (67)	Reference			Reference
	>2.31	33 (33)	7.129	1.434–35.454	0.016	5.519	1.912–15.930	0.002
>3.5 cm	≤2.31	38 (52.8)	Reference			Reference
	>2.31	34 (47.2)	5.751	1.638–20.191	0.006	2.843	1.323–6.108	0.007
Sex
Female	≤2.31	62 (67.4)	Reference			Reference
	>2.31	30 (32.6)	18.263	2.243–148.732	0.007	5.238	1.772–15.479	0.003
Male	≤2.31	43 (53.8)	Reference			Reference
	>2.31	37 (46.2)	3.806	1.226–11.812	0.021	2.979	1.4–6.339	0.005
Perineural invasion
No	≤2.31	96 (64)	Reference			Reference
	>2.31	54 (36)	12.837	2.871–57.394	0.001	5.685	2.611–12.38	<0.001
Yes	≤2.31	9 (40.9)	Reference			Reference
	>2.31	13 (59.1)	1.34	0.345–5.206	0.673	0.887	0.315–2.497	0.821

NLR – neutrophil to lymphocyte ratio; OS – overall survival; DFS – disease-free survival; LVSI – vascular lymph node invasion.

Figure 4

Kaplan-Meier survival curves for the different PNET subgroups. An NLR >2.31 was significantly correlated with shorter OS in older (>56 years) (A) patients with low-grade tumors (grade 1/2) (B), patients with tumors without perineural invasion (C), and patients with tumors without LVSI (D).

Figure 5

Kaplan-Meier survival curves for the different PNET subgroups. An NLR >2.31 was significantly correlated with poor DFS in patients with stage I/II disease (A), patients with low-grade tumors (B), patients with tumors without perineural invasion (C), and patients with tumors without LVSI (D).

Comparative performance of the NLR and other predictive models

To further evaluate the prognostic values of the NLR, other inflammation-based scores and the conventional stratification systems, an ROC analysis was performed and the AUC values were compared. The NLR had a higher AUC value (0.736; p<0.001) than the PLR and LMR (Figure 6A). However, the conventional staging systems were superior to the inflammation-based scores in the OS prediction for PNETs (Table 6). In addition, the predictive ability of the AJCC staging system was superior to that of the WHO classification in our cohort (AUC value: 0.846 vs. 0.784).

Figure 6

ROC curve for the NLR and other predictive models in resectable PNETs. The NLR had a higher AUC value than the PLR and LMR (A), while the prognostic models incorporating the NLR into the TNM staging system or WHO classification provided improved predictive accuracy compared with the prognostic models of the stratification systems alone (B).

Table 6

Areas under the ROC curves of the conventional staging systems and inflammation-based prognostic scores for predicting OS in PNET patients undergoing potentially curative resection.

Variables	Area under the ROC curve (95% CI)	p Values
Combined predictive models
AJCC stage + NLR	0.916 (0.869–0.962)	<0.001
WHO classification + NLR	0.857 (0.782–0.932)	<0.001
Staging systems
AJCC stage	0.846 (0.768–0.924)	<0.001
WHO classification	0.784 (0.694–0.874)	<0.001
Inflammation-based scores
NLR (≤2.31/>2.31)	0.736 (0.636–0.835)	<0.001
PLR (≤151.4/>151.4)	0.646 (0.529–0.764)	0.015
LMR (≤3.22/>3.22)	0.686 (0.574–0.797)	0.002

ROC – receiver operating characteristic; OS – overall survival; PNETs – pancreatic neuroendocrine tumours; NLR – neutrophil to lymphocyte ratio; PLR – platelet to lymphocyte ratio; LMR – lymphocyte to monocyte ratio.

The model integrating the NLR and the AJCC stage for OS prediction had higher AUC values than that of the AJCC stage alone (0.916 vs. 0.846), while the predictive abilities of the model integrating the NLR and the WHO classification was superior to that of the WHO classification alone (0.857 vs. 0.784) (Figure 6B).

Discussion

The results of the present study showed that the preoperative NLR, but not the PLR, was an independent risk factor for OS (HR=4.471, 95% CI 1.531–13.054, p=0.006) and DFS (HR=2.531, 95% CI 1.202–5.329, p=0.015) in PNET patients undergoing potentially curative resection. Furthermore, we observed that elevated preoperative NLR and PLR were both associated with advanced tumor stages and higher tumor grades. Finally, we showed that the NLR outperformed other inflammation-based scores in terms of its discriminatory capacity. The predictive models incorporating the NLR and conventional stratification systems, including the WHO classification and AJCC stage, showed improved predictive power relative to those of the stratification systems alone.

Increasing evidence has confirmed that systemic inflammation is associated with poorer cancer-specific survival in patients with different types of cancer [21-24]. Among several prognostic scores, the NLR, the PLR, the Glasgow prognostic score (GPS) based on serum C-reactive protein (CRP) and albumin, the prognostic index (PI) based on CRP and WBC counts, and the prognostic nutritional index (PNI) based on albumin and lymphocyte count are recognized as useful for predicting outcomes after surgery in specific host subgroups [25,26]. More recently, the NLR, which can comprehensively reflect inflammatory and immune status in patients with cancer, has been shown to be a reliable marker for predicting the survival of patients with different types of cancer, such as lung cancer [27], colorectal cancer [21], liver cancer [17], breast cancer [18], PDAC [20], and renal cell carcinoma [22]. In addition, many studies have confirmed that the PLR is a marker of patient immune status and long-term survival. PLR is also a prognostic marker in several different tumor types [28,29]. However, Shirai et al. reported that an increased preoperative PLR was not associated with OS in patients undergoing elective pancreatic resection [30]. Another study reported that the PLR has no prognostic value in oesophageal cancer [31]. The heterogeneity across these studies may be attributed to several factors, such as differences in PLR cut-off values, differences in inclusion and exclusion criteria, differences in statistical analysis methodologies (univariate vs. multivariate analyses, differences in the covariates investigated in the multivariate analysis), and differences in treatment schedules. Regarding NETs, the report by Salman et al. revealed that elevated NLR and PLR were associated with a high tumor grade and advanced tumor stage. The study also verified that the NLR and PLR are simple laboratory parameters that can be used to identify NETs with worse outcomes [32]. Yucel et al. investigated 52 patients with NETs and demonstrated the prognostic importance of the NLR in their study [33]. Regarding gastric NET, blood NLRs can also be an independent prognostic factor for RFS and OS [19]. In the present study, we found that the NLR and PLR were significantly higher in patients with PNETs than in matched NVs. Furthermore, the NLR, but not the PLR, is an independent prognostic factor associated with both OS and DFS in patients with PNET. In addition, we found that a preoperative NLR >2.31 was predictive of significantly worse survival in the subgroup of patients with stage I/II or grade 1/2 tumors. Thus, the preoperative NLR may be able to predict a poor prognosis in patients with stage I/II or grade 1/2 tumors. However, the prognostic value of the PLR was limited in our study. Patients with a low PLR at diagnosis showed significantly prolonged OS and DFS compared to patients with a high PLR in the univariate analysis; however, PLR did not show prognostic significance in the multivariate analysis. Nevertheless, given the large number of patients enrolled in this study and the effect of the NLR on prognosis in PNET demonstrated herein, our study is an important addition to the relevant literature on this topic.

The cut-off values for the NLR and PLR were important in our analysis. The cut-off value for NLR that is commonly used many studies is 5.0 [30,31,33,34], while PLR cut-off values varying from 150 to 300 have been used in other studies [29-32]. In most of the above studies, the cut-off values for the NLR and PLR were set empirically. In contrast to those studies [31,33], our study used cut-off values for NLR and PLR that were calculated with an ROC curve based on survival predictions. Our results demonstrated that the best cut-off values for the NLR and PLR were 2.31 and 151.4, with AUCs of 0.785 and 0.67, respectively, indicating that the NLR was superior to the PLR as a predictive factor in patients with PNET undergoing potentially curative resection.

Although the biology underlying the abovementioned increases in the NLR and PLR remains unclear, it is widely accepted that tumor development is associated with inflammation and immunity. Inflammation plays an important role in tumor growth [35-37]. Inflammatory mediators and cytokines, such as epidermal growth factor (EGF), transforming growth factor-β (TGF-β), tumor necrosis factor-α (TNF-α), fibroblast growth factors (FGFs), and interleukins (IL-4, IL-8, IL-10 and IL-13), which are produced by the tumor or as part of the host innate immune response, can promote angiogenesis, cause matrix degradation and cancer progression, and facilitate immunosuppression [38,39]. All of the above pathways converge to activate transcription factors, such as NF-kappaB and STAT3, leading to downstream recruitment of inflammatory mediators and leukocytes in the tumor environment [40]. This microenvironment also potentiates and enhances the neoplastic risk and ultimately promotes metastatic spread. Research has confirmed the existence of a relationship between the inflammatory microenvironments of tumors and the systemic responses induced by tumors. A variety of prognostic markers associated with the presence of a systemic inflammatory response have been described in a previous study [41].

In recent years, WHO classification, TNM stage, distant metastases, surgical margin status, tumor sizes, and Ki-67 indices have been suggested as valuable prognostic factors in patients with PNETs [8-11]. To the best of our knowledge, the existing stratification systems and predictive models for PNET, including the abovementioned TNM staging system, WHO classification and 2 nomograms [42,43], lack indicators of systemic inflammation, which could offer additional information for prognostic evaluation. Herein, we incorporated the NLR into the AJCC stage and WHO classification and showed that the predictive ability of models integrating the NLR and the stratification systems for OS was superior to that of the stratification systems alone. The results support the integration of the NLR into the conventional stratification systems for an improved discriminative ability.

Our study had several limitations that must be considered. First, given its retrospective design, the current study was subject to possible selection bias and diagnostic bias. Second, the NLR, a marker of systemic inflammation, may be affected by many conditions, including chemotherapy toxicity, chronic inflammatory diseases, granulocyte colony-stimulating factor administration, pathogenic inflammation, and other diseases. Therefore, these conditions must be accounted for in clinical practice. Moreover, the present study was conducted at a single institution. The performance of multicentre studies of the markers used herein would strengthen our conclusions. Finally, only patients who underwent surgery were included in the study, and it does not cover most of the advanced cases; thus, our results may not apply to patients without indications for surgical resection due to advanced stage.

Conclusions

As an easily accessible inflammation-based biomarker, the preoperative NLR, but not the PLR, was an independent predictor of OS and DFS in patients who underwent potentially curative resection for PNETs. Furthermore, we confirmed that prognostic models incorporating the NLR into the TNM staging system or WHO classification provided improved predictive accuracy compared with those incorporating the stratification systems alone. Therefore, we recommend that surgeons develop a treatment plan that considers not only the TNM stage but also these prognosis-related serum biomarkers in order to improve personalized therapy for patients with PNETs.