Pancreatic ductal adenocarcinoma (PDAC) is the fourth most common cause of cancer-related mortality in the Western world. Only 5%-20% of patients are candidates for curative pancreatectomy and most cases of PDAC are diagnosed at advanced stages, leading to dismal prognosis [1]. Numerous chemotherapy drugs have been investigated alone and in combination for the treatment of advanced PDAC but have shown little or no impact on disease progression or survival. In the two recent clinical trials, FOLFIRINOX (5-fluorouracil, leucovorin, irinotecan, and oxaliplatin) and gemcitabine plus nanoparticle albumin-bound paclitaxel showed superior outcomes compared with gemcitabine alone [1]. However, these new regimens are associated with considerable toxicities and the long-term survival remains unsatisfactory.

Identification of prognostic factors is crucial in guiding personalized medicine. Such factors help patients to avoid unnecessary toxicity and complications from aggressive treatment and can further facilitate stratification for clinical trials. Many studies have examined various preoperative/postoperative factors predictive of treatment outcomes [2]. However, their significance is still questionable, as most have been identified based on small sample sizes with inadequate statistical power. Widely suggested parameters are derived from functional and nutritional status, distant metastasis, and laboratory findings including C-reactive protein (CRP), carcinoembryonic antigen (CEA), and carbohydrate antigen 19-9 (CA 19-9) [3]. However, the role of these parameters is not affirmed, and the risk of bias subsists when the choice of cutoff for the continuous variables is random, leading to the use of different cutoff points across multiple studies, which impedes direct comparison. In addition, regarding advanced disease which is unresectable or metastatic, few risk scoring indices were available, and the majority focused on prediction after chemotherapy—particularly gemcitabine [4]. However, it is more important in the clinical field to discriminate patients who might benefit from chemotherapy from those who cannot. Therefore, the aim of this study is to determine prognostic values of clinic-laboratory parameters routinely evaluated at the time of diagnosis, leading to implementation of a scoring index for use in effective identification of different risk patient groups.

The median follow-up period was 7.9 months (range, 0.1 to 70.5 months) from diagnosis. Baseline patient characteristics are shown in Table 1. The median age was 66 years, and 198 patients (49.1%) were men. Sixty-two patients (15.4%) underwent curative resection, 217 patients (53.8%) had an ECOG PS of 0-1, and 178 (44.2%) and 79 (19.6%) patients had liver metastasis and peritoneal carcinomatosis at the time of diagnosis, respectively. The median value of NLR was 2.92, and the median value of CEA and CA 19-9 was 4.4 ng/mL and 224.5 U/mL, respectively.

Pancreatic cancer is the only human malignancy for which patients' survival has rarely improved during the past 30 years. Chemotherapy remains the mainstay of treatment in metastatic pancreatic cancer. However, current systemic treatments offer only a modest benefit in tumor-related symptoms and survival and these benefits, if any, are usually limited to patients—even with a novel regimen of FOLFIRINOX—with good PS and adequate nutritional status, and abrupt changes in clinical status may occur in debilitated patients with advanced disease despite systemic treatment. Therefore, primary point was (1) how to predict survival of patients at the time of initial diagnosis and (2) how to differentiate patients who might benefit from chemotherapy from those cannot.

In general, pancreatic cancer patients are not clinically separated into prognostic groups, with the exception of surgical status, before treatment. Therefore, limited studies have focused on metastatic disease [3,4]. A previous Korean study reported liver metastasis, peritoneal seeding, high CRP, and low serum albumin as poor indicators of OS. A western study reported age, weight loss, and metastasis as poor prognostic factors. Both studies focused on gemcitabine-treated patients, and are small retrospective analyses with questionable statistical powers. However, pancreatic cancer is a heterogeneous disease and recent introduction of the three-drug combination regimen or the new taxane represent key advances for better control of the disease over gemcitabine, which requires a new prognostic model.

Another debate was cutoff points of routinely measured tumor markers—CRP, CEA, and CA 19-9 [9]. Using a chemiluminescence immunoassay, the standard diagnostic cutoff points for CA 19-9 and CEA were 37 IU/mL and 5 ng/mL, respectively [10]. However, the cut-off points for these variables have not been established in metastatic disease. Some researchers advocate that CA 19-9 serum level of < 100 U/mL implies likely resectable disease, while levels > 100 U/mL may suggest unresectablity or metastatic disease [11]. However, most studies used median values or normal ranges for dichotomization. For dichotomizing continuous variables, splits of some percentile or average are arbitrary and may not be useful in assessing the true prognostic value of variables [12]. In our study, instead, cutoff point was derived using the Contal and O’Quigley method, which selects the cut-off point that maximizes the model likelihood only after all possible cut-offs have been evaluated [6]. In addition, K-adaptive partitioning for the survival model was used to determine minimum p-value.

The purpose of implementing a scoring system is for clinical decision-making and patient risk stratification, using routinely measurable clinical parameters on admission. In this study PS, hemoglobin, WBC, NLR, and CEA were identified as independent prognostic markers. For patients with a score of ≤ 6, we may hypothesize that their survival can be extended up to 12.5 months by chemotherapy, which provides us with a rationale for recommendation of relatively aggressive treatment such as FOLFIRINOX. Conversely, high-risk group (score > 25) patients had an extremely poor prognosis, with a median survival duration of only 1.3 months and the 1-year survival rate was less than 1%. Chemotherapy has no statistical benefit in these patients, but it is impetuous for now to declare that chemotherapy is contraindicated for the high risk group, in which only two patients who received chemotherapy were included for analysis. Intermediate-risk patients showed comparable outcomes with gemcitabine-based treatment. Thus, we can recommend conventional gemcitabine-based therapy to these patients. Ultimately, these findings suggest that this scoring system can be helpful in selection of the "right" patients for the "right" treatment.

Several previous studies reported reduced survival in patients with elevated CA 19-9 levels [13]. However, CA 19-9 was not included in our system. We believe increased false positive results in the presence of obstructive jaundice and inflammation—20% of our patients accompanied obstructive jaundice and 12% with systemic leucocytosis—could severely limit the universal applicability of serum CA 19-9 levels in pancreatic cancer management. Instead, our study showed that NLR, leukocytosis, and CEA were independent. Hypothesis of the role of inflammation in development and prognosis of cancer has been examined and several studies have suggested that NLR and WBC count are indicative of survival [3,14]. Whether the systemic inflammation is malignancy-associated or is the result of any co-morbid conditions that cancer patients may suffer is still unclear, but here we assumed that systemic inflammation might also occur in commonly prevalent conditions. One plausible explanation is that cellular response of blood components might be mediated through endothelial dysfunction, leading to depletion of vasodilator, antithrombotic, and anti-atherogenic properties of the endothelium and may result in capillary leukocytosis and subsequent increased vascular resistance [15]. In addition, high neutrophil counts reflect inflammation, whereas low lymphocyte counts reflect malnutrition [16]. Cytokine-mediated systemic inflammation has been shown to promote angiogenesis, DNA damage response, and tumor invasion [17]. NLR has recently been reported as a prognostic marker in colorectal and lung cancers [14,15], and its prognostic value in pancreatic cancer has also been reported [18]. Although an NLR cut-off value of 4.0 or 5.0 has been shown to have significant prognostic value in resectable cancers, a cutoff point of 3.0 may be reasonable for survival predication, which is consistent with a previous report [18-20].

The significance of CEA in pancreatic cancer has been rarely investigated [3]. CEA functions as an adhesion molecule during metastasis, and Hostetter et al. [21] reported that CEA increases the percentage of colorectal liver metastases from 2% to 48% in vivo. In a small study, CEA was identified as prognostic in advanced pancreatic cancer; superior survival outcome has been reported in patients with low CEA levels (< 2.5 ng/mL) compared to those with high CEA levels [22]. Although the cut-off value for CEA in our study was 7 ng/mL, further evaluation is warranted to elucidate the role of CEA in pancreatic cancer. Another factor, hemoglobin level, has been reported as a poor prognostic factor and is reflective of nutrition or general condition [23].

However, a prognostic model using clinical parameters only had a limitation in that tumor biology could not be reflected. In pancreatic cancer, K-ras mutations occur in 90% of cases, and overexpression of growth factors such as epidermal growth factor, transforming growth factor α is common [24]. In addition, expression of these growth factors and their receptor has been found to show association with decreased survival, thus, a combined approach using both clinical parameters and biologic markers is warranted for implementation of a more robust prognostic model. However, our system showed a Harrell’s c-index of 0.731, indicating sufficient discrimination power for patients according to the risk, and it warrants prospective clinical utility [25]. Another limitation of this study is the retrospective and non-randomized design. Patients underwent heterogeneous treatments including gemicitabine monotherapy or doublet, fluoropyrimidines, oral chemotherapy or even concurrent chemoradiotherapy which might have affected differential influence on survival.

In conclusion, we found that NLR and CEA were independent prognostic factors and suggest 3.0 for NLR and 7 ng/mL for CEA as statistically determinant cut-off values for prognostic dichotomization. In addition, our prognostic scoring system provides rationale for risk stratification of advanced PDAC patients, leading to individualized treatment. Further studies are needed to evaluate the accuracy of this scoring system.