Biliary tract cancer (BTC) comprises a heterogeneous group of tumors arising from the biliary tree-lining cells. These tumors are classified further as gallbladder cancer, ampullary cancer, or cholangiocarcinoma. The latter type is divided phenotypically into intrahepatic and ductal cholangiocarcinomas (hilar and non-hilar) to emphasize the distinctions between these clinically distinct cancers [1,2]. BTC is a relatively rare malignancy in Western populations; approximately 2,500 cases of cholangiocarcinoma and 5,000 cases of gallbladder cancer occur annually in the USA, yielding an average annual incidence of one case per 100,000 people. In contrast, multiple independent studies have documented a steady increase in the worldwide incidence of cholangiocarcinoma [3,4], and BTC is relatively common in Asian countries, including Korea, Japan, China, and India [5], with an approximate incidence of six cases per 100,000 people [5,6]. In 2013, 5,283 patients were newly diagnosed with BTC and 3,783 related deaths were reported in Korea, where BTC is the sixth-leading cause of cancer mortality.

Few treatment options are available for advanced BTC. Although a complete surgical resection remains the only curative treatment option, few patients are candidates for a potentially curative resection at the time of presentation. BTC is relatively chemo-/radio-resistant disease, and many patients are elderly and not healthy enough to tolerate aggressive treatment. Accordingly, these patients have dismal outcomes with a median survival of < 1 year [7]. Currently, combination chemotherapy with cisplatin and gemcitabine (Gem/Cis) is considered the standard first-line palliative treatment, regardless of the primary tumor location, which is based on the results from the ABC 02 trial [8]. On the other hand, the response rate to Gem/Cis is only 26%, with a median overall survival (OS) of ≤ 11.7 months. In other words, this regimen does not constitute an optimal treatment [8]. Therefore, it is essential to estimate the prognosis of a patient with metastatic BTC at the initial presentation, thus enabling patient individualization according to the risk and facilitating treatment optimization.

To date, several reports have addressed the prognostic parameters associated with BTC. For example, lymph node metastasis is a typical poor prognosticator for resectable BTC, and laboratory analyses indicated that a high C-reactive protein level and neutrophil-to-lymphocyte ratio (NLR) are correlated with survival in patients with resectable cancer or those receiving systemic chemotherapy. Despite this, few prognostic studies have addressed metastatic disease at diagnosis [9-12]. Of these, the majority targeted predictions after chemotherapy (particularly Gem/Cis). Although Eastern Cooperative Oncology Group performance status (ECOG PS), NLR or absolute neutrophil counts, hemoglobin, white blood cell (WBC) counts, primary tumor location, and the number of metastatic diseases have been proposed, the significance of these factors is unclear. Most of these parameters were identified using small sample sizes with inadequate statistical power, and their roles have rarely been confirmed independently. Another problem is the risk of bias when the points of cutoff for the continuous variables are random, leading to the use of different cutoff points across studies, which imped a direct comparison between studies. Ultimately, in a real-world clinical setting, the question of "which patients to treat" should be answered prior to "which regimen to use," and that the answer to the former question requires the proposal of prognosticators that could discriminate patients who might and might not benefit from treatment.

Therefore, this paper presents the results of multivariate analysis of routinely evaluated clinico-laboratory parameters "at the time of initial diagnosis," to implement a scoring model that could effectively identify the risk groups, and validate the model in an independent dataset.

To the best of the authors’ knowledge, this is the first study to implement a prognostic scoring model for patients with metastatic BTC at the time of the initial disease presentation. This model might better indicate which patients would benefit from treatment because given the relative rarity of BTC, most previous studies enrolled subjects irrespective of the primary location, disease extent (local or distant metastasis), and disease status (recurrent or unresectable). Moreover, many patients with BTC are quite malnourished and have a poor general state due to the dual effects of malignancy and obstructive jaundice. Because of this population heterogeneity, identification of the prognostic factors is crucial for clinical trial design and stratification factor selection.

Most studies included patients who received palliative chemotherapy, whereas the proportions of patients receiving active chemotherapy was not high, leading to a small number of study samples and low statistical power. This study showed that one clinical parameter (ECOG PS) and four laboratory parameters (serum albumin, CEA, CA 19-9, and NLR) affected the patient survival. ECOG-PS is a well-established, robust prognosticator in a variety of cancers, including BTC [9,11]. Although hypoalbuminemia is not a widely accepted BTC parameter, it is a key component of the modified Glasgow Prognostic Score, which can predict the cancer survival independently of the tumor site [12,18]. The pretreatment serum albumin levels provide useful prognostic information [19] because albumin synthesis reflects malnutrition and tumor-related inflammation through the production of cytokines, such as interleukin 6 [19]. A third of these patients were identified as nutritionally at-risk upon admission, according to the Malnutrition Universal Screening Tool (MUST), leading to the initiation of nutritional care plans [20].

These results are comparable to those of McNamara et al. [9], who reported the associations of a poor ECOG PS, low albumin level, and high NLR with reduced survival in patients with BTC. On the other hand, that study included a heterogeneous dataset that incorporated all the disease stages (I-IV) and a wide range of treatments (curative surgery to palliative therapy), and found that the primary location, stage, lack of surgery, and anemia were also associated with survival. In contrast, McNamara et al. [9] identified ECOG PS as the most significant prognostic factor; the NLR had the biggest impact on survival in this study. NLR, a marker of host inflammation, is a prognostic factor in several cancers. The prognostic significance of NLR might be related to reduced immunocompetence or neutrophilia, which would contribute to a tumor-promoting microenvironment characterized by suppressed host immune surveillance. Furthermore, its significance might be attributable to either a predisposition to chronic inflammation, a known risk factor for cancer development, or cancer-associated inflammation, which was also reported to be a key determinant of disease progression in many cancers [9].

Previous clinical and population-based studies examined the effects of various inflammatory factors on BTC development [21-24]. In metastatic BTC, the large cancer cell burden may lead to paraneoplastic production of myeloid growth factors (e.g., granulocyte colony-stimulating factor), and consequently, leukocytosis. Lymphocytes play a critical role in the tumor defense by inducing immune-mediated cytotoxic cell death; accordingly, lymphopenia in a metastatic setting reflects the reduced immunocompetence and the effects of lympholytic cytokines produced by cancer or host cells. Recent findings suggest that in cancer patients, lymphopenia might arise from tumor-mediated lymphocyte destruction and/or altered lymphocyte homeostasis [25]. Cancer-associated neutrophils also produce vascular endothelial growth factor, matrix metallopeptidase-9, reactive oxygen species, and arginase 1, which encourage tumor development and inhibit the functions of cytotoxic lymphocytes [26,27]. Interestingly, a high PLR was not found to be significant in this multivariate analysis, possibly because of the significant link between NLR and PLR (correlation co-efficient value, 0.446; p < 0.001).

The markers, CEA and CA 19-9, are non-specific because their expression may increase in response to extrahepatic obstruction from any cause, and their cutoff values are more obscure in the context of metastatic BTC. Previous studies usually adopted the median values or normal ranges to dichotomize these continuous variables; however, some of the percentiles and averages were arbitrary and might not reveal the true prognostic value of these variables [28]. Accordingly, the Contal and O’Quigley method, which selects a cutoff point that maximizes the model likelihood only after evaluating all possible cutoffs, was implemented and useful cutoff values of 9 ng/mL for CEA and 120 U/mL for CA 19-9 were estimated in patients with metastatic BTC. On the other hand, this study focused only on the dichotomization of a continuous covariate, assuming that such dichotomization would be possible from a biological point of view; however, more than one cutoff may exist in reality. Furthermore, an ideal cutoff search would work within the framework of a multiple regression model to eliminate the potential influences of other prognostic factors. Nevertheless, this study provides insights into some of the data and outcome-oriented cutoff determination methods.

The retrospective study design introduced some limitations with regard to interpreting the effects of the different treatment regimens used in this study. Gem/Cis has been available and accepted widely in Korea since 2011. Prior to that, various monotherapies and doublet regimens that included 5-FU, gemcitabine, and platinum agents were used. In addition, > 30% of patients failed to start first-line chemotherapy. In the low-risk group, 20% of patients refused chemotherapy because of old age, poor nutritional status, and combined inflammation; in contrast, > 40% of patients in the high-risk group attributed their decision in favor of chemotherapy to their young age. This scoring system is expected to allow more sophisticated treatment decisions. For example, patients with a score of ≤ 2 could expect their survival to be extended by up to 16 months with active chemotherapy (Gem/Cis or other regimens), and these patients would be candidates for more aggressive regimens or clinical trials. On the other hand, high-risk patients (score ≥ 6) are unlikely to benefit from chemotherapy. Nevertheless, it would be imprudent to conclude that chemotherapy is contraindicated for this group because the survival curves are still separated between chemotherapy and best supportive group, even though statistical significance was not reached. This may be related to the lack of power, in which only small number of patients was included for analysis. Alternatively, monotherapy could be a rational option, even though a prospective evaluation will be needed.

Regarding the intermediate group, patients treated with chemotherapy had a median survival of 9.9 months, suggesting that the general survival benefit of chemotherapy has not been established clearly; still, chemotherapy yielded a significant difference in survival (9.9 months vs. without 4.2 months; p=0.006, log-rank test). Therefore, chemotherapy is recommended for patients with a good clinical condition. On the other hand, patients with general deterioration (ECOG 2) or insufficient biliary decompression will receive only small benefits from chemotherapy, and its indications should be considered carefully. Overall, these findings suggest that the proposed scoring system could facilitate the selection of the "right" patients for the "right" treatment plan.

In this study, multiple internal validation analyses were performed. The Harrell’s c-index (0.682), iAUC (0.654), and bootstrapping showed that the proposed model possesses meaningful risk-based patient discrimination power [29]. Given the rarity of BTC, validation studies involving independent datasets are uncommon; internal validation methodologies were considered to be insufficiently credible and it is believed that the inclusion of an independent test set validation is the strength of the present study. Validation using an independent set confirmed that this scoring system clearly divided patients according to risk. On the other hand, there was another limitation. Although the investigation and validation datasets comprised independent populations, all patients were treated at a single center. Therefore, the proposed scoring model needs to be assessed in an external validation dataset to improve the robustness.

In conclusion, a set of prognostic criteria were proposed to determine the survival outcomes of patients with metastatic BTC, which were validated independently. Specifically, this system suggests cutoff values for the ECOG PS, albumin, CEA, CA 19-9, and NLR, which were identified as independent prognostic factors. In addition, the prognostic scoring system provides a rationale for the risk stratification of patients with metastatic BTC. Clinicians should find that this system facilitates treatment-related decision making.