High platelet-to-lymphocyte ratio is associated with poor prognosis in patients with unresectable intrahepatic cholangiocarcinoma receiving gemcitabine plus cisplatin

Background Several systemic inflammatory response (SIR) markers, including platelet-to-lymphocyte ratio (PLR), neutrophil-to-lymphocyte ratio (NLR), lymphocyte-to-monocyte ratio (LMR), and albumin-to-globulin ratio (AGR), have emerged as prognostic markers in various cancers. The aim of this study was to explore the impact of SIR markers on the survival outcomes of unresectable intrahepatic cholangiocarcinoma (IHC) patients. Methods Patients with histologically confirmed, unresectable IHC treated with gemcitabine plus cisplatin (GP) chemotherapy in a single tertiary hospital from 2012 to 2016 were retrospectively reviewed. Progression-free survival (PFS) and overall survival (OS) were determined using unadjusted Kaplan-Meier and adjusted Cox-proportional-hazards analysis. Time-dependent receiver operating characteristic (ROC) analysis was performed to compare the performance of the SIR markers in predicting OS. Results A total of 137 patients received a median of six cycles (interquartile range [IQR], 3–11) of GP chemotherapy with a median observation time of 9.9 months (range, 1.8–54.7 months). The median PFS and OS of all patients were 7.8 months and 9.9 months, respectively. Among the SIR markers, high PLR (> 148) and high NLR (> 5) were associated with a short PFS (Hazard ratio [HR] 1.828, P = 0.006; HR 1.738, P = 0.030, respectively) and short OS (HR 2.332, P < 0.001; HR 2.273, P < 0.001, respectively). Low LMR (< 3.5) and low AGR (< 1.2) were associated with a short OS (HR 2.423, P < 0.001; HR 1.768, P = 0.002, respectively). In multivariable cox-regression analysis, high PLR (HR 1.766, P = 0.009) and distant lymph node (LN) metastasis (HR 2.085, P = 0.001) were associated with a short PFS. High PLR (HR 1.856, P = 0.002) was an independent predictor of a short OS, along with distant LN metastasis (HR 1.929; P < 0.001), low LMR (HR 1.691; P = 0.041), and low level of serum albumin (< 3.5 g/dL) (HR 1.632; P = 0.043). Time-dependent ROC analysis revealed that the area under the curve of PLR for predicting overall survival was greater than that of NLR, LMR, and AGR at most time points. Conclusions High PLR was an independent prognostic factor of a short PFS and OS in patients with unresectable IHC receiving GP chemotherapy.


Background
Intrahepatic cholangiocarcinoma (IHC) is cancer that originates from epithelial cells of the intrahepatic bile duct, which accounts for approximately 10% of all cholangiocarcinomas [1,2]. IHC is a rare disease, though its prevalence varies enormously according to geographic regions. The incidence rates seem to be increasing globally. Surgical resection is the only curative treatment; however, only a minority of patients present with resectable disease. The prognosis for IHC is poor, and it remains a challenge to identify prognostic biomarkers to stratify patients and determine the optimal therapy.
Inflammation is a hallmark of cancer, and the inflammatory response plays an important role in cancer development and progression [3]. Several systemic inflammatory response (SIR) markers, such as platelet-to-lymphocyte ratio (PLR), neutrophil-to-lymphocyte ratio (NLR), lymphocyte-to-monocyte ratio (LMR), and albumin-toglobulin ratio (AGR), have been studied and recognized as prognostic factors in various cancers. Chronic inflammation is a key predisposing factor in the development of biliary tract cancer (BTC) [4], and there are several studies to evaluate the prognostic impact of SIR markers in BTC. However, most studies focused on the preoperative values of these markers in patients with resectable disease. To the best of our knowledge, there have not been any studies that have evaluated the prognostic impact of PLR, NLR, LMR, and AGR together in unresectable IHC patients receiving systemic chemotherapy.
The aim of the present study was to explore the prognostic value of these SIR markers in patients with unresectable IHC receiving first-line chemotherapy.

Study subjects
The medical records of patients diagnosed with unresectable IHC at Seoul National University Hospital from January 1st, 2012 through December 31st, 2016 were retrospectively reviewed. All adults aged 20 years or older with histologically diagnosed IHC, who received gemcitabine plus cisplatin (GP) chemotherapy, were included in the study. Patients who received GP chemotherapy for only one cycle or received best supportive care or locoregional therapy (e.g., transarterial chemoembolization or radiation therapy) were excluded. Patients with a history of concomitant malignancy or rheumatic disease within 5 years or missing laboratory values were also excluded from the study.

Data collection and definitions
Demographic and clinical data were collected, including age, sex, body mass index, Eastern Cooperative Oncology Group performance status (ECOG PS), history of chronic hepatitis B or C or hepatolithiasis, presence of cirrhosis, Charlson comorbidity index, presence of concomitant malignancies or rheumatic disease, history of previous biliary drainage, presence of regional or distant lymph node (LN) metastasis, and site of metastasis. Regional LNs were defined according to the 8th edition of the AJCC staging manual [5]. Baseline laboratory values in venous blood, including complete blood cell count, liver function test, and CA 19-9 levels, were obtained within a week before the initiation of GP chemotherapy in patients without evidence of active infection.
PLR and NLR were calculated as the absolute platelet and absolute neutrophil count divided by the absolute lymphocyte count, respectively. LMR was calculated as the absolute lymphocyte count divided by the absolute monocyte count. AGR was the level of serum albumin divided by the serum globulin level. Because the optimal cut-off values for PLR, NLR, LMR, and AGR have not been established, maximally selected log-rank statistic by Hothorn and Lausen was used to determine the optimal cut-off values that represent the maximum difference in overall survival between groups [6]. R package 'maxstat' (https://CRAN.R-project.org/package=maxstat) was used for this analysis and adjusted p-values were calculated by the approximation based on an improved Bonferroni inequality with an alpha error of 0.05. Patients were stratified by the optimal values, and the clinical characteristics were compared across groups.
Data on survival outcomes were collected. Progressionfree survival (PFS) was defined as the time from chemotherapy initiation until either disease progression or death due to any cause. Overall survival (OS) was the time from chemotherapy initiation until death or the last follow-up. Tumor response was assessed every 6 to 9 weeks according to the Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1 [7]. The response rate was calculated as the proportion of patients with complete response or partial response. The disease control rate was defined as the proportion of patients with complete response, partial response, or stable disease. A minimum time interval of 6 weeks was required for patients to be considered evaluable for disease control.

Statistical analysis
Data were presented as whole numbers with percentages for categorical variables and median with interquartile range (IQR) or mean ± standard deviation for continuous variables. The Chi-square and Mann-Whitney U tests were used to compare the categorical and continuous variables, respectively. Correlation among SIR markers were determined by the Spearman rank correlation coefficients and Pearson correlation coefficient. Estimates of the median follow-up period was calculated by the reverse Kaplan-Meier method [8]. Estimates of PFS and OS were calculated using the Kaplan-Meir method. The log-rank test was used to compare survival outcomes between the groups. All variables with a univariate P < 0.100 were subjected to multivariable analysis using the Cox proportional hazards model with backward elimination. The variance inflation factor was estimated to assess the multicollinearity of the final model. Considering the collinearity between SIR markers, we conducted further multivariable analyses to build separate models. This time, each SIR marker was assessed separately in different multivariable models. All reported P-values are two-sided, and P < 0.05 was considered significant. Time-dependent receiver operating characteristic (ROC) curves were created to assess the performance of PLR, NLR, LMR, and AGR in predicting OS. Time-dependent ROC curves were estimated using the 'timeROC' R package. All statistical analyses were performed using the RStudio version 3.44 statistical software package.

Ethical standards
This study was approved by the Institutional Review Board of Seoul National University Hospital (IRB No. H-1705-120-855) and conducted in conformity with the Declaration of Helsinki.

Clinical characteristics of the study population
A total of 251 patients were diagnosed with unresectable IHC at Seoul National University Hospital from January 1st, 2012 through December 31st, 2016. Of these 251 patients, 31 patients who received best supportive care only, 17 patients who received locoregional therapy, and 16 patients who received chemotherapeutic agents other than GP were excluded from the study. Among the remaining 187 patients who received GP chemotherapy, 22 patients received only one cycle of GP chemotherapy, seven patients had a history of concomitant malignancy or rheumatic disease, and 21 patients had missing laboratory data. These 50 patients were also excluded from the study. The remaining 137 patients were included in the study (Fig. 1).

Comparisons between groups according to optimal cutoff values
The optimal cut-off values for the SIR markers were 148 for PLR (P < 0.001), 5.0 for NLR (P = 0.020), 3.5 for LMR (P < 0.001), and 1.2 for AGR (P = 0.025) (Supplementary Figure 2). Patients were stratified into groups according to these values. The high PLR group (N = 63) had significantly lower albumin levels than the low PLR group (N = 74) (P = 0.030) ( Table 2). There were no other significant differences in the clinical characteristics between the high and the low PLR groups. The clinical characteristics of patients, according to NLR, LMR, and AGR, are summarized in Supplementary Table 1. Patients with high NLR (N = 28), low LMR (N = 110), and low AGR (N = 72) had lower albumin levels compared to those with low NLR (N = 109), high LMR (N = 27), and high AGR (N = 65) (P = 0.011, P = 0.012, and P < 0.001, respectively). The high NLR and low AGR groups had a higher proportion of poor performance status (ECOG PS ≥ 1) than the low NLR and high AGR groups (P = 0.018 and P = 0.035, respectively). The high LMR and low AGR groups had higher number of metastatic sites than the low LMR and high AGR groups (P = 0.001 and P = 0.031, respectively).The low AGR group had a higher proportion of patients with distant LN metastasis compared to the high AGR group (P < 0.001).

Association of SIR markers and survival outcomes
The median PFS was significantly different between the high and low PLR groups (6.0 vs. 8.6 months; P = 0.006) (Fig. 3a). A significant difference was also observed in the median OS for these groups (8.0 vs. 13.4 months; P < 0.001) (Fig. 3b). Similarly, the high and low NLR groups had significantly different median PFS (4.0 vs. 8.6 months; P = 0.030) and OS (6.7 vs. 11.7 months; P < 0.001). The low LMR group had a significantly shorter OS compared to the high LMR group (9.1 vs. 19.1 months; P < 0.001); however, the difference between the median PFS of these two groups was not statistically significant (6.8 vs. 9.9 months; P = 0.087). Similarly, the low AGR group had a significantly shorter OS compared to the high AGR group (8.7 vs. 13.2 months; P = 0.002), but the difference in the median PFS between these groups was not statistically significant (6.5 vs. 9.6 months; P = 0.368). The time-dependent ROC analysis revealed that the area under the curve of PLR for predicting overall survival was greater than that of NLR, LMR, and AGR at most time points (Fig. 4).

Predictive factors for PFS
The results of the univariate and multivariable analyses for PFS are shown in Table 3 Table 2).

Predictive factors for OS
The results of the univariate and multivariate analyses for OS are shown in   Table 3).

Discussion
This study is the first to evaluate the association of PLR, NLR, LMR, and AGR together with PFS and OS in patients with unresectable IHC. We found that high PLR (> 148) was an independent prognostic factor of a short PFS and OS in patients with unresectable IHC receiving GP chemotherapy. Time-dependent ROC analysis revealed that the area under the curve of PLR for predicting overall survival was greater than that of the other inflammatory markers (i.e., NLR, LMR, or AGR) at most time points. SIR markers, including PLR, NLR, LMR, and AGR, have been investigated for an association with the prognosis of BTC. Studies on the association of PLR and survival outcomes in BTC patients are relatively scarce and mostly only included patients undergoing surgical resection [9] [10]. Cho et al. [11] reported that high NLR  Number of metastatic sites (≥3.8) and high PLR (≥121) were independent predictors of a short OS for patients with advanced BTC, including IHC, extrahepatic cholangiocarcinoma, gallbladder cancer, and ampulla of Vater cancer. In our study, high PLR (> 148), but not high NLR (> 5), was an independent predictor of a short PFS and OS in patients with advanced IHC. The discrepancies might be due to the different study populations of the two studies. In the previous study, the distribution of PLR varied according to tumour origin (P = 0.003). There has been accumulating evidence indicating extreme molecular and biological heterogeneity in BTC, according to the anatomical location of the tumour [12,13]. Therefore, it is reasonable to assume that the SIR markers might have different prognostic impacts depending on the cancer type. Another study that included only patients with advanced IHC reported that high NLR (> 2.8) and high PLR (> 128. 3) were associated with a short PFS or OS by univariate analysis, but not multivariable analysis [14]. The reason for this difference is not clear; however, the different cut-off values might be one explanation. Because optimal cut-off values for PLR and NLR have not been established, the previous study chose median values as cut-off values, while we selected cut-off points that showed maximal differences in OS. In addition, the previous study did not investigate the impact of LMR but explored the association of the modified Glasgow Prognostic Score, which is based on the level of albumin and C-reactive protein. There have not been any studies that have evaluated LMR in advanced IHC. However, in advanced gallbladder cancer, our group recently reported that a high monocyte to lymphocyte  ratio (> 0.24) and high PLR (> 108) were associated with poor survival [15]. The exact mechanisms by which PLR predicts survival outcomes in cancer patients are not clear. However, there are plausible mechanisms to explain how thrombocytosis and lymphopenia are associated with poor prognosis in cancer patients. The association between platelets and cancer has been studied since 1865 when Trousseau first described thrombosis in gastric cancer [16]. Tumour cells are known to activate platelets and stimulate platelet aggregation. Activated platelets can mediate cancer cell growth and angiogenesis [17]. They can also directly protect circulating tumour cells from natural killer cellmediated lysis, which promotes metastatic dissemination [18]. In experimental mouse models, the induction of thrombocytopenia reduced the rate of metastasis, while reconstitution with human platelets increased the number of metastases in vivo [19,20]. Clinical studies have shown that thrombocytosis is associated with poor prognosis in several cancers, including breast, lung, colon, gastric, and ovarian cancer [21]. The lymphocyte is a key mediator of immunosurveillance and antitumor immunity [22]. A low number of lymphocytes could be responsible for a weak immune response against cancer cells.
Data on the association of PLR and OS according to the stage of BTC are insufficient; however, studies conducted in other solid cancers suggest that the significance of PLR is greater for metastatic disease than for early-stage disease [23,24]. Furthermore, PLR values were greater in metastatic disease than in early-stage disease. In our study, PLR was not associated with the level of CA 19-9 or organ/lymph node metastasis and the Not surprisingly, there were strong correlations between SIR markers. PLR was positively associated with NLR and negatively associated with LMR. This SIR marker was also negatively associated with albumin, but there was no significant association with AGR. Albumin has been classically known to reflect nutritional status; however, recent studies have shown that the level of albumin is decreased during inflammation, regardless of nutritional status [25]. Globulin plays an important role in immunity and inflammation. Because levels of both albumin and globulin are easily influenced by factors, such as dehydration or oedema, AGR has been suggested and explored as a prognostic factor in several cancers [26]. In the present study, the serum albumin levels and AGR were significantly associated with distant LN metastasis, which was the strongest independent predictive factor of PFS and OS. Low AGR did not remain as an independent prognostic factor of OS after multivariable analysis, which might be due to its association with the presence of distant LN metastasis and ECOG PS.
There are several limitations to the current study. First, it was a single-centre, retrospective study, and the number of patients comprising the study population was small. Due to the rarity of IHC, the number of IHC patients is limited in a single-centre design. However, multicentre studies have a pitfall that stems from different blood processing techniques between laboratories. Second, the current study included only patients who received at least two cycles of GP. We believed that this approach could minimize the potential confounding effect that might result from different anticancer therapies. Thus, the results may not be generalizable to other populations, including patients receiving non-GP chemotherapy, concurrent chemoradiation therapy, locoregional therapy, or best supportive care only. Third, the dichotomous cut-off value of the SIR markers might be arbitrary. Currently, optimal cut-off values for the SIR markers have not been established, and they vary between studies. We chose to use dichotomized cut-offs because previous studies utilizing such cut-offs have shown a better association between PLR and overall survival than those with three risk categories [23]. The optimal cut-off values in the present study, which were determined using a statistical method, were similar to those of the previous studies [26][27][28][29]. Nevertheless, they warrant further validation in large, prospective studies since outcome-oriented approach to select cut-off value might result in overfitting.