Patients with PDAC hospitalized at the Sixth People’s Hospital Affiliated with Shanghai Jiao Tong University School of Medicine, from January 2014 to December 2022, were retrospectively screened. The inclusion criteria indicated below: (1) pathological diagnosis of pancreatic ductal adenocarcinoma; (2) no loss to follow-up; (3) relatively complete clinical data. The exclusion criteria were as follows: (1) pathological diagnosis of pancreatic neuroendocrine carcinoma or pancreatic acinar cell carcinoma; (2) patients with second or multiple primary cancers; (3) lack of necessary data (detailed clinicopathological data or follow-up data). A total of 500 patients who met our inclusion criteria, with clinicopathological data collected from medical records, were included in our study.

clinical variables of patients, including age, sex, follow-up time and status, clinical stage, pathological grade, and history of therapy (radical surgery or palliative chemotherapy), activated partial thromboplastin time (APTT), prothrombin time (PT), international normalized ratio (INR), thrombin time (TT), fibrinogen (Fg), fibrin degradation products (FDP), d-dimer (D-D), alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TBiL), direct bilirubin (DBiL), albumin (ALB), globulin (GLB), creatinine (Cr), blood urea nitrogen (BUN), alpha-fetoprotein (AFP), CEA, quamous cell carcinoma antigen (SCC), cancer antigen 125 (CA125), cancer antigen 724 (CA724), cancer antigen 199 (CA199), cancer antigen 242 (CA242), white blood cell count (WBC), neutrophil (NEUT), lymphocyte (LYMPH), monocyte (MONO), red blood cell (RBC), hemoglobin (Hb), mean corpuscular volume (MCV), red cell distribution width (RDW), platelet count (PLT), were collected. Laboratory data were obtained from routine blood tests performed at the time of initial diagnosis. NLR, LMR, PLR, NMR and SII were calculated through the following formula: NLR = neutrophil/lymphocyte; LMR = lymphocyte/monocyte; PLR = platelet/lymphocyte; NMR = neutrophil/monocyte; SII = platelet × neutrophil/lymphocyte, respectively. Telephone follow-ups were conducted for the included patients to obtain their OS data. When P < 0.05, the variable is considered statistically significant for prognosis.

Excel was utilized for data management, while R software was employed for analyzing the data. Categorical variables were summarized using absolute frequencies and percentages, whereas continuous variables were presented as medians with ranges. A risk regression model was applied to evaluate the relationship between blood indicators and prognosis. To circumvent the limitations inherent in median-based dichotomization, time-dependent receiver operating characteristic (ROC) analysis was performed, and the optimal cutoff value (the highest Youden index) was determined. Univariate and multivariate Cox regression analyses were conducted to identify variables that have a significant impact on prognosis, while hazard ratio (HR) and 95% confidence interval (CI) were calculated. The optimal prognostic model was developed by employing LASSO regression with three-fold cross-validation, using the Concordance Index (C-index) as the evaluation metric. A risk score plot was subsequently constructed based on this prognostic model. Kaplan-Meier curve and prognostic calibration curve were plotted to validate the efficacy of the model.

Ethical permission was approved by Ethics Committee of Shanghai Sixth People’s Hospital. The study protocol was designed in accordance with Good Clinical Practice and conducted in compliance with the Declaration of Helsinki. All participants provided written informed consent prior to their inclusion in the study.

This study included a total of 500 PDAC patients, with a median follow-up time of 12 months. The shortest OS was 1 month, and the longest OS was 39 months. By the last follow-up, 215 patients (43%) had died. Among these patients, there were 308 females and 192 males. A total of 438 patients underwent radical resection for pancreatic cancer, while 61 patients, who were not candidates for surgery, received palliative chemotherapy (Figure 1).

To establish a prognostic prediction model for pancreatic cancer, we analyzed the median values and ranges of collected blood indicators (Table 1). Using the median values as cutoff points, we performed univariate Cox regression model analysis to assess their association with prognosis. The following factors, which were showed in Table 2, were found to have a statistically significant association with prognosis: age, TNM stage, treatment, FDP, D-dimer, CEA, CA125, CA199, CA242, NEUT and LYMP (P < 0.05).

To avoid the potential bias associated with using the median value as the cutoff, we additionally plot time-dependent ROC curves for the highest Youden index and identify the optimal cutoff values, which maximizes the difference between sensitivity and specificity to identify the most effective threshold for prognostic prediction (Table 3). Univariate Cox proportional hazards regression models were used to assess the association between patient characteristics and OS, and the results were visualized using a hazard ratio forest plot (Figure 2). The results showed that among the clinical characteristics, TNM stage and treatment had a significant impact on the prognosis of PDAC patients. Among the blood indicators, neutrophil, lymphocyte, CA199, CA125, CEA, NLR, LMR, NMR, and SII were statistically significant in relation to prognosis.

Considering the overlap and potential interference among neutrophil, lymphocyte, NLR, LMR, NMR and SII, we conducted multivariate analysis incorporating TNM stage, treatment, neutrophil, lymphocyte, CA199, CA125, and CEA. Furthermore, NLR, LMR, NMR, and SII were separately analyzed in combination with TNM stage and treatment in multivariate analyses (Figure 2). Multivariate analysis revealed that, in addition to TNM stage (I vs. III: HR = 3.052, 95% confidence interval (CI) = 1.900-4.905, P < 0.001; I vs. IV: HR = 4.815, 95% CI = 2.504–9.258, P < 0.001), which are closely related to the prognosis of PDAC patients, neutrophils (HR = 1.416, 95% CI = 1.037-1.932, P = 0.028), lymphocytes (HR = 0.625, 95% CI = 0.462-0.846, P = 0.002), CEA (HR = 1.820, 95% CI = 1.315-2.518, P < 0.001), CA125 (HR = 1.392, 95% CI = 1.001-1.936, P = 0.049), NLR (HR = 1.748, 95% CI = 1.210–2.525, P = 0.003), LMR (HR = 0.597, 95% CI = 0.430–0.829, P = 0.002), NMR (HR = 2.065, 95% CI = 1.331–3.206, P = 0.001), and SII (HR = 1.751, 95% CI = 1.244–2.466, P = 0.001) are also significantly associated with the prognosis of PDAC patients.

Notably, after multivariate analysis, the association between CA199 levels and prognosis in PDAC patients became less significant compared to that of CEA. The lymphocyte counts showed a negative correlation with prognosis in both univariate and multivariate analyses.

To establish a prognostic model for PDAC patients, we attempted to combine the clinical characteristics—such as gender, age, TNM stage, histologic grade, and treatment—with previously validated independent prognostic factors, including neutrophil, lymphocyte, CEA, CA125, NLR, LMR, NMR, and SII in different combinations. Given the established importance of CA199 in pancreatic cancer, it was also included and subjected to multiple combinations. By calculating the LASSO coefficients, we aimed to obtain an optimal prognostic prediction model. When the value of Log(λ) is small (the left part of the graph), the C-index is relatively high and stable, remaining 0.73, indicating that the model has good predictive accuracy (Figure 3). Based on this, we established a risk scoring model for prognostic prediction in PDAC patients.

Risk score = NLR*0.088 + CA125*0.001 + CEA*0.013 + TNM*0.486 + Grade*0.202 - OPER*0.074 + Age*0.026 - Sex*0.076

It is worth noting that radical surgery was represented by 1 and palliative chemotherapy by 0, with a correlation coefficient of -0.074 for prognosis, indicating that radical surgery can reduce the risk of adverse outcomes associated with prognosis. Meanwhile, male was represented by 1 and female by 2, with a correlation coefficient of -0.076 for prognosis, indicating that male patients have higher risk of poor prognosis than female.

We validated the effectiveness of this prognostic model by plotting a risk factor diagram. We stratified the risk scores into high-risk score and low-risk score groups based on the highest Youden index 3.41. Survival analysis demonstrated that the low-risk score group had a significantly better prognosis compared to the high-risk score group (HR = 3.94, 95% CI = 2.89–5.37, P < 0.001). The prognostic calibration curve also demonstrates that the model accurately predicts the 1-year and 2-year survival rates of PDAC patients (Figure 3).