Development and validation of a nomogram for predicting survival in patients with pancreatic ductal adenocarcinoma after radical pancreatoduodenectomy

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive cancer with a 5 year overall survival (OS) rate of only 9%, remaining the fourth main cause of cancer-related death all around the world [1, 2]. Only about 20% of patients with PDAC have access to radical surgery because of its asymptomatic onset and high metastasizing potential at an early stage [3]. Pancreatoduodenectomy-oriented comprehensive treatment is the basic therapeutic strategy for patients with PDAC located at the head/neck of the pancreas. However, the 5 year OS of these patients still does not exceed 20% in recent years [4]. Of note, patients at same pathological staging and managed with the same remedy present heterogenous survival outcomes, suggesting that the TNM staging system is not effective in prognostic risk stratification for this subset of patients. Consequently, developing a more effective tool for survival prediction is crucial.

PDAC is notorious for eliciting intense local/systemic inflammation, facilitating its progression and resistance to treatment. This process involves the activation of multiple components such as the immune system and the coagulation cascade, which means that the occurrence and progression of tumor is a coordinated manner [5]. Hypercoagulation and malnutrition, being the characteristic pathophysiological alterations in patients with PDAC, are caused by this inflammatory response at the systemic level, greatly affecting patients’ survival. The incidence of cancer-associated venous thromboembolism (VTE) is up to 36% in PDAC patients [6, 7]. In addition, almost 80% of patients with PDAC exhibit “cancer-cachexia syndrome,” which is characterized by anorexia, weight loss, asthenia, and poor survival [8]. Therefore, it is reasonable to consider that biomarkers indicating this network effect may have a high prognostic value.

Fibrinogen (FBG) and prognostic nutritious index (PNI) are some of the biomarkers that have significant value in prognosis prediction of PDAC patients underwent radical resection [9], [10], [11]. However, only a small section of studies were performed to determined the synergistic value of these indicators in this area according to our investigation. Therefore, we aimed at integrating indicators to build a nomogram for predicting OS of these patients and assessed its predictive performance accordingly.

Methods

Results

Nomogram construction and evaluation

A nomogram (Figure 1) was established based on regression formula. The estimated survival probability of individual patients at each time point (1, 2, and 3 years) could be acquired by calculating the total score. Only slight deviations from the ideal 45° line were seen in the 1-, 2-, and 3-year calibration curves (bootstrap method) (Figure 2A–C), which indicated that this model performed well in the present dataset. Time-dependent ROCs (Figure 3A–C) were drawn using the function “survival ROC” of R language; AUCs were 0.713, 0.777, and 0.845 at 1, 2, and 3 years, respectively, indicating the satisfactory accuracy and discrimination ability of the nomogram.

Figure 1:

Nomogram for predicting overall survival. The nomogram was established for the OS prediction of individual patients who underwent open radical pancreaticoduodenectomy at 1, 2, and 3 years by incorporating seven independent predictors (high-grade tumor, VCE, N-stage, anemia, fibrinogen, PLR, and DRR). Abbreviations: VCE, vessel carcinoma embolus; PLR, platelet-to-lymphocyte ratio; DRR, De Ritis ratio.

Figure 2:

Calibration curves. (A–C). Calibration curves for predicting OS at 1, 2, and 3 years. Validation of the calibration curve shows that the predicted probability fits well with the ideal probability.

Figure 3:

Receiver operating characteristic analyses. (A–C). Receiver operating characteristic (ROC) analyses. Time-dependent ROCs were drawn by the function “survival ROC,” areas under curves (AUCs) were 0.713, 0.777, and 0.845 at 1, 2, and 3 years, respectively, indicating that this nomogram had satisfactory accuracy and discrimination ability.

Clinical usefulness of the nomogram

The DCA presented in Figure 4 was performed to assess the clinical usefulness of the nomogram. In DCA, the y-axis represents the net benefit, calculated as the sum of benefits (true positives) and harms (false positives). The horizontal “None” line assumes no patients would die, and the “ALL” line assumes all patients would die. DCA determines the clinical practicability of the nomogram by quantifying the net benefits under different threshold probabilities in the validation data set. Based on the obtained threshold probabilities, our results show that nomogram has greater net utility than existing TNM layer systems. And the DCA curve showed that the nomogram contributed net benefit in threshold probabilities from the range 0.17–0.5, which suggested that the model can provide great clinical benefit for predicting OS.

Figure 4:

Decision curve analysis (DCA) based on the nomogram and TNM stage. DCA determines the clinical practicability of the nomogram by quantifying the net benefits under different threshold probabilities in the validation data set. The horizontal “None” line assumes no patients would die, and the “ALL” line assumes all patients would die. Based on the obtained threshold probabilities, our results show that nomogram has greater net utility than existing TNM layer systems.

Discussion

The survival prediction in patients with PDAC after radical surgery can help guide postoperative treatment and follow-up, which has led many scholars to determine methods for predicting survival in this subset of patients. A prospective study [17] in 355 patients with PDAC who underwent PD at Tongji Hospital showed that age, preoperative CA19-9, adjuvant treatment, TNM staging system, and tumor differentiation were the prognostic factors for survival, based on which a predicated nomogram was developed. Xie et al. [18] analyzed 300 radiological characteristics (extracted from the computer tomography images of 220 patients with PDAC) and selected features by using LASSO to create a radiological score that was related to patients’ prognosis. In addition to the clinical and imageology indicators, genomics plays an important role in survival prediction. A research [19] based on mRNA expression information from the gene expression omnibus database identified genes that are differentially expressed in OS by performing univariate analysis and LASSO Cox regression and established a gene-based prognostic signature. The statistic revealed that gene markers are independent prognostic prediction factors for PDAC. A nomogram combining genetic markers and clinical prognostic factors outperformed AJCC staging in predicting OS. Based on these studies, we comprehensively considered tumor characteristics (e.g., pathological stage) and cancer-induced pathophysiological changes in the hosts (e.g., hypercoagulation and cachexia) and built an integrated nomogram model based on multivariate regression analysis. It could predict individualized survival probability at a certain time point (1, 2, and 3 years) according to the clinicopathological profiles of the patients.

Compared with previous prognostic evaluation methods such as TNM staging system, the nomogram model has considerable predictive performance, as evaluated by the internally verified bootstrap method. Although the current clinical prognosis assessment mainly relies on the TNM staging system, it is not an independent predictor for OS in univariate analysis in this study. Previous studies have shown that the prognosis of patients with the same stage varies greatly, reflecting the insufficient prognostic information of the current TNM stage. DCA curve is an evaluation method first developed by Dr. Andrew Vickers of MSKCC (Memorial Kettering Cancer Research Institute in Si Long) and published in Med Decis Making in 2006. DCA determines the clinical practicability of the nomogram by quantifying the net benefits under different threshold probabilities in the validation data set. In DCA curve, the two curved oblique lines in the figure represent two different clinical prognosis models, the nomogram and TNM staging system. Besides, there are two lines, which represent two extreme situations. The horizontal “None” line assumes no patients would die, and the “ALL” line assumes all patients would die. Based on the obtained threshold probabilities, our results show that nomogram has greater net utility than the single use of existing TNM layer systems. And the DCA curve also showed that the nomogram contributed net benefit in threshold probabilities from the range 0.17–0.5, which suggested that the model can provide great clinical benefit for predicting OS.

In PDAC, lymphatic invasion is one of the hallmarks of tumor aggressiveness and has a poor prognostic significance [20]. The quantities of regional lymph node metastasis is the major part of latest TNM classification (8th edition) [21]. Tumor grade or tumor cell differentiation is another long-established prognostic indicator after the resection of PDAC [22], [23], [24]. Wasif et al. [16] demonstrated that combining tumor differentiation (well-differentiated and moderately differentiated tumors are classified as low-grade tumors and poorly differentiated and undifferentiated tumors are classified as high-grade tumors) and AJCC staging can significantly improve the prognostic stratification of patients, which was further confirmed by Rochefort et al. [25]. The same tumor grading criteria were used in the present study, and high-grade tumors were identified as independent hazard factors for bad prognosis along with AJCC stage 8 N2 and VCE. Intravascular cancer embolus refers to the embolus formed by malignant tumor cell mass found in the blood or lymphatic vessels connected with the tumor mass in the tissue specimen removed after surgery, that is, cancer embolus. This means that the malignancy has invaded blood or lymphatic vessels and is likely to have developed distant metastases, even if no distant metastases of the tumor are visible elsewhere. It is another well-acknowledged signature of cancer aggressiveness.

Platelet and FBG are the characteristic markers of a PDAC-induced hypercoagulable state, which can accelerate cancer progression and is related to bad prognosis. The long duration of thrombocytosis and high plasma FBG levels are major risk factors for thrombosis [26, 27]. VTE is a common complication of advanced PDAC with an estimated incidence of up to 42%, and symptomatic VTE is a major hazard factor for poor prognosis of patients [28]. Moreover, increasing evidence has shown that the combination of platelets and FBG facilitates survival of circulating tumor cells (CTCs) and their anchoring to the vascular wall, thus promoting hematogenous metastasis. Tight adherence to platelets can guard CTCs against shear stresses in the bloodstream [29], thus forming a thrombus easily and promoting their arrest at the endothelium. FBG acts as a “molecular bridge” by enhancing the sustained adhesion and survival of single tumor cell emboli in the target organ vasculature [30], [31], [32]. Adhesion of tumor cells results in platelet activation and the release of several soluble factors, including transforming growth factor-β, which can impair the cytotoxicity and proliferation of NK cells and T cells [33], [34], [35], thus further protecting CTCs from immune surveillance or elimination. Furthermore, FBG participates in the formation of the tumor extracellular matrix and angiogenesis [36], promotes epithelial–mesenchymal transition [32], and affects the tumor infiltration of immune cells [37]. The present study showed high PLR and FBG levels are the independent hazard factors for bad survival, providing evidence to their synergistic effects on cancer progression and prognosis.

Cachexia caused by cancer-induced metabolic perturbation significantly deteriorates patients’ quality of life, increases the risk of anemia, inflammation and anastomotic leakage, and aggravates immunosuppression and chemotherapy resistance, which are closely associated with death of approximately 33% of PDAC patients [38]. The primary indicator of cachexia is inadvertent weight loss >5% of normal body weight or >2% weight loss in patients with a BMI <20 kg/m² in 6 months [13]. In addition to the mechanical compression of tumors and the loss of pancreatic exocrine functions, the tumor-induced systemic inflammatory response and related cytokines, such as interleukin (IL)-1, IL-6, and tumor necrosis factor-α, also lead to cachexia by inducing excessive catabolism of fats and muscles [39]. These inflammatory factors also lead to increased neutrophils and decreased lymphocytes in peripheral blood, which can impair response of B lymphocytes, NK cells, and activated T cells et al., resulting in a decline in lymphocyte-mediated antitumor immune responses and inhibition of timely detection or elimination of tumor cells, ultimately leading to tumor progression [40, 41]. PNI, an emerging prognostic biomarker for PDAC, generally reflects the nutrition statuses of patients and is closely related to cachexia [40], [41], [42]. Cachexia occurrence and a low PNI (<40.5) were significantly correlated to poor OS in univariate analysis; however, intriguingly, there were no statistical difference in multivariate analysis, whereas anemia was proved to be an independent hazard factor. This finding suggests that compared with weight loss and hypoalbuminemia, anemia is the downstream factor for malnutrition, affecting the survival of patients with PDAC. However, this result requires further validation.

AST/ALT was identified by De Ritis [15] and has been known as De Ritis ratio (DRR) since 1957. The normal value of DRR is about 1.15 in healthy people. ALT is located in the cytoplasm of hepatocytes, and AST is located in the cytoplasm and mitochondria of hepatocytes [15]. When the hepatocytes had mild pathological changes, mitochondria remain intact and only the transaminase in the cytoplasm of the hepatocytes was released into the blood. The increase of ALT was more obvious than that of AST, which lead to a decrease of DRR. When hepatocyte damage is severe, the transaminase in hepatocyte plasma and mitochondria is all released into the blood, so the increase of serum AST is higher than that of ALT, with an increase of DRR [15, 43], [44], [45]. Several hypotheses on the association of DRR with cancer are available. AST plays a central role in glycolysis, which is the main energy supply pathway for most cancer cells [46]. Cancer-induced oxidative stress and inflammation can damage hepatocytes and intracellular mitochondria, resulting in a higher level of AST release relative to ALT. Therefore, a higher DRR may generally reflect higher tumor proliferative activity and malignant behavior. Ridel et al. [47] found that in patients with PDAC treated with first-line nab-paclitaxel and gemcitabine, the doubling of DRR was associated with 0.5-fold lower odds of objective response and a 1.4-fold higher relative risk of progression or death (HR=1.38; 95% CI: 1.06–1.80; p=0.017). However, there has been no research of the prognostic effect of DRR for resectable PDAC. Our study showed that DRR>1.1 was an independent risk factor for poor OS in patients underwent PD (HR=2.69; 95% CI: 1.56–4.27; p<0.001). This groundbreaking discovery deserves further demonstration in large clinical cohort and laboratory trials.

There were still some limitations in this study. First of all, this is a retrospective study conducted at a single center with a relatively small sample size, warranting multi-institution cohort studies to verify the results. Second, some promising biomarkers, such as cytokines and D-dimer, were not incorporated because clinical examinations or observations were unavailable, which limited the nomogram performance prediction.

Conclusions

We preliminarily established a prognostic nomogram for predicting OS of patients with PDAC underwent PD(R0) by integrating tumor- and host-related pathophysiological factors. The predictive model presented reasonably good discrimination and calibration abilities. It can improve the current clinical practice and assist clinicians in prognostic risk stratification and decision-making for patients with PDAC.