Overall survival in patients with metastatic pancreatic cancer after surgically primary tumor resected: a SEER-based nomogram analysis

Background This study aimed to establish and validate a nomogram to predict overall survival in patients with metastatic pancreatic cancer (mPC) after surgically primary tumor resected. Methods All mPC patients who underwent primary tumor resection at SEER database between 2004 and 2016 were identified. We randomly assigned two-thirds of the patients to the training group and one third to the validation group. In the training group, the Kaplan–Meier survival analysis was used to analyze survival outcomes. A univariate and multivariate cox regression analysis was used to identify significant prognostic factors for establishing a nomogram. The predictive accuracy and discriminative ability were measured by the concordance index (C-index) and risk group stratification. Results A total of 742 patients were included for analysis. Four significant prognostic factors were obtained and included in the nomogram. The nomogram showed an acceptable discrimination ability (C- index:0.711) and good calibration and was further validated in the validation cohort (C- index: 0.727). The nomogram total points (NTP) had the potential to stratify patients into 2-risk groups with a median OS of 11 and 4.5 months (P < 0.001), respectively. The can provide considerable accuracy individual in it can guide clinicians make decisions therapies.


Introduction
Pancreatic cancer (PC) is a fatal disease with poor prognosis and high mortality [1,2]. It is predicted that pancreatic cancer is to be the second leading cause of cancer-associated mortality in the United States by 2030 [3]. Although great advances have been made in surgery, radiotherapy and chemotherapy, the five-year survival rate for pancreatic cancer remains below 8 percent in the United States [4]. Surgical resection remains the mainstay of treatment for PC patients. However, mostly on account of the absence of the specific clinical feature, patients are diagnosed with an advantaged stage and lost the best chance of surgery [5]. The liver is the most common site of metastasis in PC patients. Meanwhile, It was revealed that those patients with liver metastasis had poorest prognoses than other site metastases [6]. Fortunately, some studies have shown that surgical resection of primary tumors can prolong the survival time of patients with metastatic pancreatic cancer (mPC) [7]. However, there is currently a lack of an effective prognostic tool to predict the survival of mPC patients with surgical resection of the primary tumor.
Currently, a nomogram has been widely used in clinical prognosis prediction of cancer patients because of its' simple and exact [8,9]. In addition, nomograms are useful for clinicians to deal with difficult conditions without uniform clinical guidelines. Compared to the traditional AJCC TNM stage, nomogram can predict the survival of patients more accurately [10,11]. Besides, TNM stage can't provide a prognosis prediction for mPC patients with primary tumor resection, it is necessary to build a more personalized prognostic tool that can provide a more precise prognosis prediction for mPC patients with surgically resected primary tumor. There was already research about the nomogram for predicting survival in mPC patients [12]. However, nomograms for predicting survival time of mPC patients with surgically resected primary tumor are scarce.
Therefore, we intended to build a nomogram with the data from SEER database to better predict individualized prognosis in mPC patients with surgically resected primary tumor.

Patients selection
The data of this study was obtained from the Surveillance, Epidemiology, and End Results (SEER) 18-Registry (1973

Data Collection
The following variables were extracted in the analysis: age at diagnosis, gender, race, marital status, tumor size, extension, regional nodes positive, site of metastases, surgery to the primary and metastases, scope regional LN surgery, TNM stage, chemotherapy recode, radiation recode, survival months and vital status. The primary endpoint was overall survival (OS), which was defined as the time from diagnosis to death due to any cause. The survival time of patients who were still alive at the end of the follow-up period was used as a censored. Ethical consent was not needed because no special personal information was recorded in the SEER database.

Statistical analysis
The statistical analyses were performed using R software version 3.6.2 (http://www.rproject.org) for windows and SPSS software (Inc., Chicago, USA, version 23). The significant level was set at 0.05 and all tests were two-sided. Categorical data were presented as frequency and percentage and tested with the Chi-square test.
For nomogram construction and validation, two-thirds of patients were assigned into the training group and one-third of patients were assigned into the validation group. The baseline characteristics of both groups were compared by the chi-square test. In the validation group, significant prognostic variables were obtained by univariate and multivariate Cox regression analyses. According to the result of multivariate analysis, independent prognostic factors were further selected to develop a nomogram to predict the probability of survival at 6-months,12-months, and 24-months. Discrimination was evaluated by using a concordance index (C-index). For reducing bias, Bootstraps with 1000 resamples in the training group and validation group were used for these activities.
Besides, we use the calibration plot to compare the difference both predicted survival probability and the actual survival.

Risk Group Stratification Based On The Nomogram Total Points
To further discriminate OS, we categorized the patients into low-risk and high-risk groups based on nomogram total points (NTP). The optimal value of the cut-off for NTP was decided by the receiver operating characteristic (ROC) curve. The Kaplan-Meier survival curves were used to reveal the stratification of OS in both risk groups. Also, the cut-off value was then applied to the validation group and the Kaplan-Meier survival curves were also performed.

Result
Clinicopathological characteristics A total of 742 eligible mPC patients with primary tumor surgery between 2004 and 2016 were included in the study ( Figure.1). Among them, there were 494 cases in the training group and 248 cases in the validation group. The baseline characteristics of patients did not differ significantly between these 2 groups (Table 1). In the whole group, there were 383 (51.6%) males and 359 (48.4%) females. Most of them were white (n = 597, 80.5%) and married (n = 494, 66.6%). The most common tumor sites were the pancreatic head (n = 401, 54.0%), followed by the pancreatic tail (n = 176, 23.7%). The main grade of them were moderate differentiation (n = 340, 45.8%) and poor differentiation (n = 322, 43.4%).
As for treatment, besides primary tumor resection, most of patients received chemotherapy (n = 443,59.7%), while surgery of metastases (n = 230, 31.0%) and radiotherapy (n = 93, 12.5%) were relatively less. By the end of follow-up, 644 (86.8%) patients had died and 98 (13.2%) patients were still alive (as a censored). The median survival time was 7 months. Kaplan-Meier survival curves turned out the OS differences about stratification by these factors were all statistically significant (Fig. 2).

Nomogram And Risk Classification
Based on the result of cox multivariate analysis, grade (p 0.001), T stage, resected number of regional LN and chemotherapy were integrated into establishing a nomogram (Fig. 3). This model showed that grade and chemotherapy had the greatest effect on prognosis, followed by the T stage and resected number of regional LN. Each subtype within these variables had a corresponding score on the point scale. By adding the scores of each selected variable, we can easily calculate the likelihood of survival at 6-,12-,24months. In the training group and validation group, the C-index for OS was 0.711 and 0.727, respectively, which indicated that the nomogram had an accepted discernible ability. The calibration curves for predicting 6-months, 12-months and 24-months survival probability in the training group and validation group were shown in Fig. 4.
We divided the training group into two subgroups according to the optimal NTP cut-off value, which was identified by the ROC curve. The NTP cut-off value was 115.5. The patients were divided into low-risk subgroups (NTP 115.5,n = 244) and high-risk subgroups (NTP ≥ 115.5 n = 250). The median OS of both groups was 12 months and 4.5 months, respective. In the validation group, the same cut-off value was used to grouping.
The median OS of low-risk was also 13 months and 4 months. The Kaplan-Meier survival curves were used to reveal the stratification of OS in the training and validation groups ( Fig. 5-A and B).

Discussion
In clinical practice, such a dilemma often happens: patients are found to have a distant metastatic site but the primary tumor can be surgically resected. In such a case, surgery is sometimes performed based on the surgeon's experience and personal desire, but whether it has an impact on survival is not clearly defined. Several studies had demonstrated that surgical resection of a primary tumor can prolong the survival time of mPC [6,13,14]. Besides, a study based on the seer database, through propensity score matching, concluded that patients with metastatic pancreatic cancer could benefit from primary tumor resection [15]. In this study, tumor grade, T grade, resected number of regional LN and chemotherapy were confirmed as independent prognostic factors of mPC with primary tumor resection.
The grade is one of the strongest independent prognostic factors. Tumor grading reflects the biological behavior of the tumor and plays an important role in prognosis. It has been demonstrated that tumor differentiation is an independent influencing factor for predicting OS in other similar studies [16,17]. In our study, the HR of OS increased with the degree of tumor differentiation decreases, and patients who were well-differentiation suffered better survival than poor-differentiation patients. Several studies also advised that tumor grade should be incorporated into the current TNM classification, because novel TNMG staging system enables more accurate prognosis prediction within particular clinical stages [18,19]. Chemotherapy is an important treatment method for advanced pancreatic cancer. Studies have shown that concurrent chemoradiotherapy [20][21] after the induction of effective chemotherapy [22][23][24] can alleviate symptoms and improve the survival of patients for advanced pancreatic cancer with a good systemic condition.
According to the National Comprehensive Center Network (NCCN) guidelines (Version Compared with chemotherapy alone, chemotherapy combined with surgery can even increase the resection rate of partial metastatic pancreatic cancer [26]. Similar to the above result, it was found that chemotherapy therapy improved survival time in mPC patients with primary tumor resection (HR = 2.453 for no receiving chemotherapy, 95% CI: 1.981-3.037). Lymph node metastasis is considered to be an important prognostic factor for pancreatic cancer and other gastrointestinal tumors. Pancreatic cancer often metastasizes to distant lymph nodes through complex pathways and develops as a systemic disease [27]. In the early stage of the resectable pancreas, appropriate regional lymph node dissection can reduce the probability of postoperative recurrence and benefit the survival of patients, but the specific number of lymph nodes removed has not been agreed [28][29]. At present, there are few studies on whether primary resection of metastatic pancreatic cancer combined with regional lymph node dissection is beneficial for survival. In our study, Cox multivariate analysis showed that N staging was not an independent prognostic factor in mPC patients with surgically primary resection. However, combined regional lymph node removal can benefit the survival of mPC patients, and the number of lymph nodes removed is positively correlated with survival outcome(HR = 0.766 for 1-3 regional LN resected, 95% CI: 0.484-1.213; HR = 0.612 for ≥ 4 regional LN resected, 95% CI: 0.444-0.842). This result revealed that the resected number of regional LN was identified as an independent prognostic factor.
Nomograms are graphical representations of statistical prediction models that provide the probability of survival for a given outcome [30]. Thus, the variables included in a nomogram should be easily available and measurable. In our study, the nomogram is simple to use and provides a quantitative prognosis for individual patients. Besides, our study uses four variables that are easily accessible in clinical work, making the use of nomogram more convenient. In some studies [31],the preoperative pain was included in the nomogram for pancreatic cancer. Although pain is the common symptom of patients with metastasis pancreatic cancer it is so subjective and the clinical significance remains unclear which makes the result of prognosis prediction inaccurate. Yi-Nan Shen et al [32] study showed that radiological parameters played important parts in the diagnosis and treatment of pancreatic cancers with venous invasion. However, it is difficult to measure and obtain, it is very inconvenient for clinical use.

Availability of data and materials
The datasets analyzed during the current study are available in the SEER repository (https://seer.cancer.gov/).

Ethics approval and consent to participate
Not applicable

Consent for publication
Not applicable.