Preoperative risk stratification of early recurrence in resected pancreatic ductal adenocarcinoma: Novel equilibrium-phase-computed tomography biomarker of extracellular volume

Abstract

BACKGROUND

Predicting early recurrence (ER), (≤ 12 months) after pancreatic ductal adenocarcinoma (PDAC) resection remains challenging. Preoperative biomarkers such as carbohydrate antigen 19-9 (CA19-9) and computed tomography (CT) lack optimal specificity and reproducibility. Extracellular volume (ECV), measured on equilibrium-phase CT to quantify stromal fibrosis, correlates with PDAC progression but its utility for ER prediction is unknown. This study evaluates preoperative CT-ECV as a novel biomarker to predict ER following curative-intent PDAC resection.

AIM

To investigate the utility of CT-ECV for preoperative prediction of ER in PDAC patients after R0 resection.

METHODS

This retrospective study included 93 PDAC patients undergoing R0 resection and preoperative pancreatic CT from January 2020 to November 2023. Clinical and CT features were analyzed. ECV was calculated using unenhanced and equilibrium-phase CT. Univariable and multivariable Cox regression identified ER predictors, followed by receiver operating characteristic analysis. Recurrence-free survival (RFS) was assessed by the Kaplan-Meier method.

RESULTS

Multivariable analysis identified elevated CT-ECV [hazard ratio (HR) = 1.05; 95% confidence interval (CI): 1.02-1.09; P = 0.003], high preoperative CA19-9 (HR = 1.00; 95%CI: 1.00-1.00; P = 0.002), and poor tumor grade (HR = 2.51; 95%CI: 1.20-5.22; P = 0.014) as independent ER predictors. CT-ECV demonstrated comparable predictive accuracy to tumor grade [areas under the curve (AUC): 0.736 vs 0.650; P = 0.202]. Combining CT-ECV and CA19-9 achieved a higher AUC than tumor grade alone (0.759 vs 0.650; P < 0.05). Kaplan-Meier analysis revealed significantly shorter RFS in patients with low CT-ECV (≤ 35.37%), elevated CA19-9 (> 55 U/mL), or poorly differentiated tumors compared to those with high CT-ECV (> 35.37%), low CA19-9 (≤ 55 U/mL), or moderately/well-differentiated tumors.

CONCLUSION

CT-derived ECV is a promising non-invasive biomarker for preoperative ER prediction in PDAC. Combined with CA19-9, it outperforms tumor grade in stratifying recurrence risk, offering a clinically actionable tool for optimizing postoperative management.

Key Words: Pancreatic ductal adenocarcinoma; Equilibrium-phase computed tomography; Extracellular volume; Early recurrence; Carbohydrate antigen 19-9

Core Tip: This study demonstrates that preoperative computed tomography (CT)-derived extracellular volume (CT-ECV) is an independent predictor of early recurrence in pancreatic ductal adenocarcinoma after R0 resection. Elevated CT-ECV, high carbohydrate antigen 19-9 (CA19-9), and poor tumor grade were significantly associated with shorter recurrence-free survival. Combining CT-ECV and CA19-9 improved predictive accuracy over tumor grade alone, offering a non-invasive tool for preoperative risk stratification and personalized postoperative management.

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is the most fatal pancreatic malignancy and the seventh leading cause of global cancer-related deaths, with an estimated 660000 deaths in 2024[1]. The 5-year survival rate has been stagnant at 7%-12% for a long time despite advancements in diagnostics and multimodal therapies[2]. In 2024, the mortality/incidence ratio is close to 0.99 (i.e., nearly one death for every new case) and the annual incidence has continually increased at a rate of 1%[1]. While radical resection remains the sole curative option[3], only 15%-20% of patients qualify for surgery, and over half experience early recurrence (ER) within a median recurrence-free survival (RFS) of 12 months post-resection[4-6]. This aggressive recurrence pattern underscores the critical need for preoperative risk stratification to optimize treatment strategies. Although neoadjuvant therapy shows promise in mitigating micrometastases and improving surgical margins[7,8], concerns regarding tumor progression during treatment and heterogeneous patient responses highlight the urgency for reliable biomarkers to guide therapeutic decisions[9].

Current ER predictors, including tumor grade, tumor node metastasis (TNM) stage, and lymphovascular invasion, are exclusively postoperative, limiting preoperative risk assessment[10]. Carbohydrate antigen 19-9 (CA19-9), the most widely used serum biomarker, correlates with recurrence risk but suffers from suboptimal specificity[11,12]. Computed tomography (CT), pivotal for PDAC diagnosis and staging, provides prognostic insights through features such as perivascular infiltration, peripancreatic tumor infiltration, and peripheral enhancement[7,12,13]; however, its reliance on subjective interpretation restricts reproducibility.

PDAC is pathologically characterized by extracellular matrix expansion driven by collagen deposition, quantified as extracellular volume (ECV)[14,15]. The iodinated contrast agent is uniformly spread across both intravascular and extravascular spaces in the CT equilibrium-phase. CT-derived ECV, calculated by comparing tissue and blood pool (aortic) enhancement corrected for hematocrit, provides an objective measure of fibrotic burden. Previous studies have established CT-ECV as an imaging biomarker for cardiac and hepatic fibrosis. In pancreatic imaging, CT-ECV has also emerged as a noninvasive quantitative biomarker for evaluating pancreatic fibrosis and pancreas-related clinical events. Specifically, Fukukura et al[15] and Wang et al[16] demonstrated that CT-ECV predicts survival outcomes and chemotherapy responses in advanced PDAC patients. Bai et al[17] reported dual-energy CT-ECV as a potential predictor for acute pancreatitis following pancreatoduodenectomy. Furthermore, Kameda et al[18] found a significant negative correlation between pancreatic ECV and glycated hemoglobin levels in cirrhotic patients, suggesting ECV may help clarify the pathophysiology of impaired glucose tolerance in this population[18]. Despite these associations, the role of ECV in predicting ER after curative resection of PDAC remains unexplored.

This study investigates the utility of preoperative CT-ECV as a novel imaging biomarker for predicting ER in PDAC patients undergoing R0 resection, aiming to bridge the gap in preoperative risk stratification and inform personalized therapeutic strategies.

MATERIALS AND METHODS

This retrospective study was approved by the ethics committee of the First Affiliated Hospital of Chongqing Medical University (No. K2023-356), which waived the requirement for informed consent.

Patients

We consecutively reviewed resectable PDAC patients who underwent preoperative pancreatic CT from January 2020 to November 2023. Inclusion criteria included the following: (1) Histologically confirmed PDAC; (2) Patients who received radical (R0) resection; and (3) Preoperative CT examination performed within 15 days pre-surgery. Exclusion criteria included the following: (1) Neoadjuvant chemotherapy before surgery; (2) Intraoperative metastasis (excluding lymph node metastases); (3) Coexisting other malignancies within 5 years other than PDAC; (4) Severe image artifacts or absence of equilibrium-phase CT images; and (5) Inadequate clinical data or follow-up information. Resectable disease was defined as tumors with no major arterial involvement and no contact or ≤ 180° contact without contour irregularity with the portal vein or superior mesenteric vein, according to the National Comprehensive Cancer Network criteria[19]. Patients with PDAC underwent initial eligibility screening based on imaging and medical records, followed by retrospective confirmation of CT scans reviewed by an experienced radiologist (Li YM, 20 years’ expertise in pancreatic imaging). Finally, 93 patients were enrolled in this study (Figure 1). The sample size is determined by the number of patients who meet the criteria and have complete data during the study period.

Figure 1 Flow chart. PDAC: Pancreatic ductal adenocarcinoma; CT: Computed tomography; ER: Early recurrence.

CT examination

Multiphasic contrast-enhanced CT was performed for all patients using two 128-channel multidetector row CT scanners (SOMATOM Definition Flash and SOMATOM Force, Siemens Healthineers). After the non-enhanced scan was performed, a standard bolus dosage of 1.2 mL/kg of nonionic contrast agent (Iopamiro 370 from Bracco Healthcare; or Ultravist 370 from Bayer Healthcare) was injected at a rate of 3.0-5.0 mL/second. The scans for the arterial and parenchymal phases were postponed by 8 seconds and 35 seconds, respectively, after the descending aorta reached 100 Hounsfield units (HU). Portal venous and equilibrium phases were performed with a fixed scan delay of 72 seconds and 3-5 minutes.

The detailed scan parameters were as follows: Tube voltages of 120 kVp, and 200-440 mA with use of automatic exposure control (Care dose 4D; Siemens), collimation of 128 mm × 0.6 mm, gantry rotation time of 0.5 seconds, and spiral pitch of 1.0 (for SOMATOM Definition Flash) or 0.7 (for SOMATOM Force). For image reconstruction, iterative reconstruction algorithms were employed (SAFIRE for SOMATOM Definition Flash; ADMIRE for SOMATOM Force). Images were reconstructed with a thickness of 1 mm and increment of 0.8 mm.

Clinical-pathological data collection

The following clinicopathologic and biochemical data were collected from the electronic medical records and pathological reports: Age, sex, body mass index, bilirubin, albumin, lymphocytes, hematocrit (Hct), alkaline phosphatase and serum CA 19-9 Levels, surgery type, tumor location, tumor diameter, tumor grade, and tumor stage of PDAC according to the American Joint Committee on Cancer (AJCC) 8^th edition. All biochemical variables were routinely measured within 15 days before surgery. According to the international study group for pancreatic surgery definition[20], the type of pancreatic surgery was categorized as standard pancreaticoduodenectomy/distal pancreatectomy or extended pancreaticoduodenectomy/distal pancreatectomy involving concomitant veins or additional organ resection. Radical (R0) resection was defined as the complete removal of all tumors with a histologically negative margin.

Image analysis

Two radiologists with 4 and 6 years of experience in abdominal imaging who were blinded to the clinical data and patient outcomes reviewed CT images at a picture archive and communication system (PACS) workstation (Communication System Carestream Vue PACS, version 12.1). Regions of interest (ROIs) were used to measure the CT values (HU) of the primary PDAC, portal vein and abdominal aorta on the same plane on unenhanced and equilibrium-phase images. For the tumor ROIs, areas of visible vessels, calcifications, and necrotic collections were avoided. For the aorta and portal vein, the ROIs were as large as possible but carefully avoided the aortic wall and calcification. ROIs placements were kept identical and visually co-localized on the CT images during unenhanced and equilibrium phases. The CT-ECV of the tumor were calculated using the following formula:

ECV = [100-Hct (%)] × ΔHU_tumor/(ΔHU_aorta)

Where ΔHU represents the change in HU between unenhanced and equilibrium-phase CT images. To confirm that the equilibrium phase was adequately obtained, ΔHU_aorta and ΔHU_{portal vein} for each patient were compared. Any significant differences between ΔHU_aorta and ΔHU_{portal vein} exceeding ± 10 HU were deemed outliers and removed. All measurements were taken twice by two radiologists and then averaged for analysis. The delineation of ROI is shown in Figure 2.

Figure 2 Computed tomography-derived extracellular volume calculation workflow. The diagram illustrates the extracellular volume (ECV) calculation process in a 61-year-old male with pancreatic ductal adenocarcinoma who underwent radical (R0) resection, with a known hematocrit value of 40.8%. Region of interests were placed on the tumor, portal vein, and abdominal aorta. A and B: In preoperative baseline unenhanced; C and D: Equilibrium phase computed tomography (CT) images. The absolute difference between the ΔHounsfield unit (HU)_aorta and ΔHU_portal measured < 10 HU, confirming measurement validity. Two independent readers calculated CT-ECV values using the formula shown in the figure, with the final patient ECV (15.53%) derived as the mean of both measurements. ECV: Extracellular volume; Hct: Hematocrit; HU: Hounsfield unit.

The same two observers also assessed the following CT variables based on previous imaging studies[7,21]: (1) Margins (well circumscribed vs ill defined); (2) Main pancreatic duct dilation (present vs absent); (3) Parenchymal atrophy (present vs absent, based on the ratio of the main pancreatic duct’s diameter to the pancreatic parenchyma’s width at the same level >0.5); (4) Vein abutment (present vs absent, defined as ≤ 180° of circumferential contact of the tumor with the portal vein, superior mesenteric vein and splenic vein); and (5) Peripancreatic tumor infiltration (tumor with expansive growth beyond the pancreas). Any ambiguous CT findings were addressed by consultation with the senior radiologist, who has 28 years of experience in abdominal imaging.

Follow-up protocol

Every patient in our institution underwent a contrast-enhanced CT scan every three months during the first year after surgery. Magnetic resonance imaging, positron emission tomography/CT or biopsy was performed if necessary to clarify tumor recurrence. By comparing these results with preoperative data, ER was assessed in each patient after R0 resection. RFS was defined as the time from surgery to tumor recurrence, death, or last follow-up examination (censored). Each patient was followed up for at least 1 year and the ER status was reviewed up to November 2024. Recurrence within 1 year after R0 resection was defined as ER.

Statistical analysis

Continuous variables are presented as mean ± SD or medians and ranges conforming to a normal distribution, and compared using the t-test or Mann-Whitney U test. Categorical variables are expressed as numbers and percentages, and compared using the χ² test or Fisher’s exact test. Patients were divided into ER and non-ER (NER) groups according to whether they had ER.

Univariable and multivariable Cox proportional-hazards models were used to identify independent predictors of ER. Variables with a P value of < 0.10 in the univariable analysis were then included for multivariable analysis. The results are presented as odds ratio with 95% confidence interval (CI). To evaluate the predictive performance of the predictor for ER, receiving operating characteristic curve analysis was used, and areas under the curve (AUC) were calculated. The optimal cutoff value was established by the Youden index, defined as the value maximizing the mean of sensitivity and specificity. Comparisons of AUC values between parameters were performed using the DeLong method. Kaplan-Meier survival curves for RFS were generated and analyzed using the log-rank test.

Intraclass correlation coefficient (ICC) was calculated to evaluate inter-observer agreement for attenuations and ECV measurements (ICC: 0.00-0.20, poor correlation; ICC: 0.21-0.40, fair correlation; ICC: 0.41-0.60, moderate correlation; ICC: 0.61-0.80, good correlation; and ICC: 0.81-1.00, excellent correlation). All statistical analyses were performed by SPSS software (version 26.0; IBM Corporation, 2013) and MedCalc Statistical Software (version 20.0.14), with significance at P < 0.05.

RESULTS

Patient characteristics

A total of 156 PDAC patients underwent preoperative pancreatic CT and R0 resection during the study period. After excluding 25 patients who either underwent neoadjuvant chemotherapy before surgery (n = 13), had intraoperative metastasis (n = 7) and coexisting malignancies within 5 years (n = 5), we also excluded 38 patients due to severe image artifacts or absence of equilibrium-phase CT images (n = 23) and inadequate clinical data or follow-up (n = 15). Finally, 93 patients (61.0 ± 9.1 years; 58 males) were included.

Tumors were found in the head (n = 68), body or tail (n = 25) of the pancreas, with an average primary tumor size of 2.8 cm. The AJCC classification identified 62 patients as stage IA or IB, 29 as stage IIA or IIB, and 2 as stage III. Out of the total, 79 patients received standard pancreatic resection, while 14 had an extended version of the surgery. Postoperative adjuvant therapy was administered to 53 patients.

Early tumor recurrence

Patients were categorized into ER (n = 46) and NER (n = 47) groups according to whether the tumor recurred within one year. The median RFS of ER patients was 5 months (interquartile range: 3.25-8.45).

Differences in demographic and clinicopathological characteristics between the ER and NER groups

Our results revealed no significant differences in demographic data, surgery type, biochemical results except CA19-9, tumor location, AJCC tumor stage, postoperative adjuvant therapy and CT findings of the pancreas between the two groups (P > 0.05), whereas significant differences were noted in the CA19-9, maximum tumor diameter, tumor grade, ΔHU_tumor, and CT-derived ECV (P < 0.05). The main characteristics of the patients are summarized in Table 1.

Table 1 Baseline characteristics of patients with pancreatic ductal adenocarcinoma, mean ± SD/n (%).

Characteristic	Non-ER (n = 47)	ER (n = 46)	P value
Age (years)	60.76 ± 1.50	61.27 ± 1.33	0.802
Male sex	29 (61.7)	29 (63.0)	0.867
Female sex	18 (38.3)	17 (37.0)	0.894
BMI (kg/m2), IQR	21.85 (20.50-23.93)	21.78 ± 4.70	0.895
Type of pancreatic surgery			0.165
Standard pancreatic surgery	42 (89.4)	37 (80.4)
Extended pancreatic surgery	5 (10.6)	9 (19.6)
Laboratory results
CA19-9 (U/mL), IQR	54.37 (15.48-268.88)	328.00 (90.45-500.00)	0.006
Haematocrit level (%), IQR	38.80 ± 7.86	38.00 (34.55-41.45)	0.339
Bilirubin (μmol/L), IQR	47.05 (12.98-159.25)	64.00 (10.20-201.55)	0.809
Albumin (g/L), IQR	40.50 (37.00-44.75)	38.97 ± 1.14	0.161
Lymphocytes (109/L), IQR	24.41 ± 1.38	19.50 (14.75-30.75)	0.207
Alkaline phosphatase (μkat/L), IQR	188.0 (88.0-428.0)	154.00 (87.00-461.00)	0.913
Tumor location			0.077
Head	31 (66.0)	37 (80.4)
Body/tail	16 (34.0)	9 (19.6)
Tumor size (cm), IQR	2.0 (1.50-3.50)	2.8 (2.25-3.90)	0.024
AJCC tumor stage			0.867
I	31 (66.0)	31 (67.4)
II/III	16 (34.0)	15 (32.6)
Tumor grade			0.006
G1/2	32 (68.1)	19 (41.3)
G3	15 (31.9)	27 (58.7)
Postoperative adjuvant therapy	26 (49.1)	27 (50.9)	0.430
ΔHUtumor, IQR	53.05 ± 3.43	37.00 (32.66-45.00)	0.003
CT-derived ECV (%)	39.78 ± 1.95	30.01 ± 1.90	0.001
CT findings of the pancreas
Ill-defined margins	13 (27.7)	13 (28.3)	0.948
Main pancreatic duct dilation	21 (44.7)	20 (43.5)	0.979
Parenchymal atrophy	7 (14.9)	11 (23.9)	0.271
Vein abutment	9 (19.1)	17 (37.0)	0.056
Peripancreatic tumor infiltration	11 (23.4)	18 (39.1)	0.102

ER: Early recurrence; IQR: Interquartile range; BMI: Body mass index; CA19-9: Carbohydrate antigen 19-9; ECV: Extracellular volume; HU: Hounsfield unit; ΔHU_tumor: Absolute enhancement of the tumor attenuation between the unenhanced and equilibrium-phase computed tomography images; CT: Computed tomography; AJCC: American Joint Committee on Cancer; G1: Well differentiated; G2: Moderately differentiated; G3: Poorly differentiated.

Interobserver agreement of CT-ECV values

The interobserver reproducibility of CT-derived ECV was excellent (ICC = 0.926, 95%CI: 0.882-0.953). The Bland-Altman plots showed good interobserver agreement, with the mean bias of agreement in CT-ECV values between reader 1 and reader 2 of 0.87% and the upper and lower limits were -8.95% and 7.12%. Neither fixed bias nor proportional bias was observed. Therefore, the average value of CT-ECV measured by the two radiologists was used for further analysis. The Bland-Altman plot is shown in Figure 3.

Figure 3 Bland-Altman plots of the computed tomography-derived extracellular volume differences measured by radiologist 1 and radiologist 2. These Bland-Altman plots showed good agreement. ECV: Extracellular volume.

Predictive factors for ER

The results of univariate and multivariate analyses for ER are shown in Table 2. Univariate analysis revealed that CA 19-9 [hazard ratio (HR) = 1.001; 95%CI: 1.000-1.001; P = 0.050], T stage (HR = 3.256; 95%CI: 1.499-7.071; P = 0.003), poorly differentiated PDAC (HR = 2.542; 95%CI: 1.352-4.780; P = 0.004), vein abutment (HR = 1.778; 95%CI: 0.954-3.313; P = 0.070), ΔHU_tumor (HR = 0.974; 95%CI: 0.958-1.090; P = 0.002) and CT-derived ECV (HR = 1.057; 95%CI: 1.025-1.090; P = 0.001) were significantly associated with ER. In the multivariate analysis, CA 19-9 (HR = 1.000; 95%CI: 1.000-1.001; P = 0.002), poorly differentiated PDAC (HR = 2.506; 95%CI: 1.204-5.215; P = 0.014) and CT-derived ECV (HR = 1.050; 95%CI: 1.017-1.085; P = 0.003) were risk predictors of ER.

Table 2 Univariate and multivariate analyses for early recurrence in patients with pancreatic ductal adenocarcinoma.

Variable	Univariate analysis		Multivariate analysis
	Hazard ratio (95%CI)	P value	Hazard ratio (95%CI)	P value
Age (years)	1.008 (0.974-1.042)	0.660
Male sex	0.842 (0.429-1.650)	0.616
Postoperative adjuvant therapy	0.734 (0.423-1.274)	0.272
CA19-9 (U/mL)	1.001 (1.000-1.001)	0.050	1.000 (1.000-1.001)	0.002a
Tumor location		0.131
Head	1 (Reference)
Body to tail	0.404 (0.125-1.310)
Tumor size (cm)	1.196 (0.959-1.492)	0.112
T stage		0.003		0.319
T1	1 (Reference)		1 (Reference)
T2 + T3	3.256 (1.499-7.071)		1.556 (0.652-3.712)
N, yes	1.211 (0.578-2.539)	0.612
Tumor grade, G3	2.542 (1.352-4.780)	0.004	2.506 (1.204-5.215)	0.014a
AJCC tumor stage		0.942
I	1 (Reference)
II/III	0.975 (0.498-1.912)
CT findings of the pancreas
Ill-defined margins	0.987 (0.511-1.905)	0.968
Main pancreatic duct dilation	0.895 (0.485-1.652)	0.723
Parenchymal atrophy	1.306 (0.654-2.609)	0.449
Vein abutment	1.778 (0.954-3.313)	0.070	2.099 (0.977-4.510)	0.057
Peripancreatic tumor infiltration	0.762 (0.441-1.315)	0.329	0.879 (0.530-1.457)	0.616
ΔHUtumor	0.974 (0.958-0.990)	0.002	0.988 (0.965-1.013)	0.340
CT-derived ECV (%)	1.057 (1.025-1.090)	0.001	1.050 (1.017-1.085)	0.003a

^aP < 0.05.

CA19-9: Carbohydrate antigen 19-9; ECV: Extracellular volume; HU: Hounsfield unit; CT: Computed tomography; AJCC: American Joint Committee on Cancer; ΔHU_tumor: Absolute enhancement of the tumor attenuation between the unenhanced and equilibrium-phase computed tomography images; N: Lymph nodes; CI: Confidence interval; G3: Poorly differentiated.

Diagnostic performance of independent predictors for predicting ER

The AUC for predicting ER using the three identified independent predictors for CT-derived ECV was 0.736 (95%CI: 0.628-0.827), for CA19-9 was 0.686 (95%CI: 0.575-0.784), and for tumor grade was 0.650 (95%CI: 0.538-0.752). The higher AUC of CT-ECV and CA19-9 compared with tumor grade was not statistically significant (P = 0.202 and P = 0.653, respectively). However, the combination of ECV and CA19-9 yielded a significantly higher AUC [0.759 (95%CI: 0.653-0.846)] compared with tumor grade (P = 0.044) (Figure 4).

Figure 4 Receiver operating characteristic curve analysis of the computed tomography-derived extracellular volume, carbohydrate antigen 19-9 and tumor grade for predicting early recurrence. ECV: Extracellular volume; CA19-9: Carbohydrate antigen 19-9; AUC: Area under the curve.

As shown in Table 3, using the Youden index, CT-derived ECV with the cutoff value of ≤ 35.37% was associated with a sensitivity of 80.49% and specificity of 59.42%, CA19-9 with the cutoff value of > 55 was associated with a sensitivity of 90.24% and specificity of 45.24%, tumor grade had a sensitivity of 60.98% and specificity of 69.05%. The combination of ECV and CA19-9 (threshold predicted probability of 0.45) achieved a sensitivity of 87.80% and specificity of 69.52%.

Table 3 The efficacy of each parameter in predicting early recurrence.

Characteristic	Combined ECV and CA19-9	CT-derived ECV	CA19-9	Tumor grade
Cutoff value	0.45	35.37	55.00	Ref.
AUC (95%CI)	0759 (0.653-0.846)	0.736 (0.628-0.827)	0.686 (0.575-0.784)	0.650 (0.538-0.752)
Sensitivity	87.80	80.49	90.24	60.98
Specificity	69.52	59.42	45.24	69.05
P value	< 0.001	< 0.001	0.002	0.005

CA19-9: Carbohydrate antigen 19-9; ECV: Extracellular volume; CT: Computed tomography; AUC: Area under the curve; CI: Confidence interval.

Stratified survival analysis using the risk predictors of ER

When survival analysis was stratified according to the optimal threshold, the Kaplan-Meier curves showed a higher ER rate in patients with low CT-ECV than in patients with high CT-ECV (66.0% vs 24.24%, log-rank test: P = 0.003). The results also showed that patients with high CA19-9 Levels had a higher ER rate than those with low CA19-9 Levels (61.67% vs 17.39%, log-rank test: P = 0.002). Tumor grade was stratified by high/moderate differentiation and poor differentiation, and patients with poorly differentiated PDAC had a higher ER rate than those with highly/moderately differentiated PDAC (65.79% vs 35.56%, log-rank test: P = 0.036) (Figure 5). Representative cases are shown in Figure 6.

Figure 5 Kaplan-Meier survival curves for early recurrence in patients with pancreatic ductal adenocarcinoma. A: Stratified by computed tomography-derived extracellular volume; B: Carbohydrate antigen 19-9; C: Tumor grade. CT: Computed tomography; ER: Early recurrence; ECV: Extracellular volume; CA19-9: Carbohydrate antigen 19-9; G1: Well differentiated; G2: Moderately differentiated; G3: Poorly differentiated.

Figure 6 Computed tomography images. A-D: A 57-year-old female patient with grade G2 (moderately differentiated) pancreatic ductal adenocarcinoma (PDAC) without early recurrence (ER). Axial unenhanced (A) and equilibrium-phase (B) images show regions of interest (ROIs) (in yellow) placed on tumor in the pancreatic head and the aorta. Coronal reconstructed image in portal venous phase shows the dilation of the main pancreatic duct (arrow), and the fat plane between the tumor (T) and superior mesenteric vein (SMV) is clear (C). The CT-derived extracellular volume (ECV) was 37.17%. No tumor recurrence was found in the surgical bed (arrow) after radical resection on follow-up CT (D); E-H: A 70-year-old male patient with grade G3 (poorly differentiated) PDAC with ER. The same ROIs placement demonstration (E and F). Coronal reconstructed image in portal venous phase (G) shows the tumor (T) invading the duodenum (arrow), and interfacing with the SMV without luminal narrowing (arrow). The CT-derived ECV was 28.12%. Tumor local recurrence (black arrow) and multiple hepatic metastases (arrow) occurred 8.3 months after radical resection on postoperative follow-up CT (H); I-L: A 61-year-old male patient with grade G3 (poorly differentiated) PDAC with ER. The same ROIs placement demonstration (I and J). Coronal reconstructed image in portal venous phase shows the pancreatic parenchymal atrophy with main pancreatic duct dilation (arrow), and tumor (T) invade the confluence of portal-SMV (arrow). The CT-derived ECV fraction was 15.53% (K). Hepatic metastasis (arrow) occurred 10.9 months after radical resection on follow-up CT (L). T: Tumor; SMV: Superior mesenteric vein.

DISCUSSION

Our study establishes CT-derived ECV as a novel preoperative predictor of ER in resectable PDAC, demonstrating comparable efficacy to tumor grade (AUC: 0.736 vs 0.650) and showing synergistic value when combined with CA19-9 (AUC: 0.759). These findings address a critical gap in preoperative risk stratification, as traditional predictors such as TNM stage or vascular invasion require postoperative histopathology. The prognostic superiority of CT-ECV over subjective CT features (e.g., vein abutment) and its technical advantages of rapid acquisition (< 3 minutes), minimal radiation, and reproducibility, position it as a practical biomarker for clinical integration.

Notably, low CT-ECV (≤ 35.37%) was correlated with a twofold increased ER risk, aligning with its pathophysiological basis: ECV quantifies fibrotic extracellular matrix expansion, a hallmark of PDAC aggression[15]. Our results are consistent with those of previous studies showing that CT-ECV was useful in estimating the risk of pancreas-related events, such as pancreas fibrosis, postoperative acute pancreatitis and chemotherapy response[15-17]. While fluorodeoxyglucose positron emission tomography/CT (AUC: 0.748) and radiomics (AUC: 0.701) show similar predictive potential[4,22], CT-ECV avoids their limitations of high cost, complex feature extraction, or protocol variability. Although magnetic resonance imaging offers superior soft-tissue contrast for ECV calculation, its longer scan time and contraindications limit practicality compared to equilibrium-phase CT[23].

Several well-known clinical and pathological variables, including T stage, lymph node status, post-surgical adjuvant chemotherapy and tumor differentiation influence PDAC outcomes[12,24]. Poor tumor differentiation, in particular, has been linked to greater heterogeneity, invasiveness, and a higher likelihood of ER. Consistent with previous reports[24,25], our results showed that poorly differentiated PDAC was associated with a twofold increased risk of ER compared to moderately or well-differentiated tumors. Notably, tumor grade exhibited marginally lower predictive accuracy AUC compared to CT-ECV, underscoring the utility of CT-ECV in preoperative evaluation. We found that T stage and lymph node metastasis were not identified as independent predictors of ER, which may be due to the relatively small sample size or the superior discriminatory power of tumor grade in reflecting tumor heterogeneity.

Serum CA19-9 is a well-established prognostic factor in PDAC, with elevated concentrations correlating with higher tumor burden and poorer survival[10,26]. Multiple studies have demonstrated the association between preoperative CA19-9 Levels and ER following resection of PDAC, with threshold values ranging from 100 to 529 U/mL[12,27,28]. In our study, patients experiencing ER exhibited significantly elevated CA19-9 Levels (optimal threshold: 55 U/mL; AUC: 0.686, sensitivity: 90.24%, specificity: 45.24%). While our threshold yielded lower specificity than previous studies, the high sensitivity (90.24%) supports its utility in high-risk screening, particularly given potential confounding effects from biliary obstruction in some patients.

Pancreatic CT remains the preferred imaging modality of choice for pancreatic tumor evaluation. Beyond its diagnostic and staging roles, tumor enhancement features on CT may hold prognostic value[7]. Recent studies have linked morphological and enhancement characteristics, such as tumor diameter, peripancreatic infiltration, and suspicious lymph node metastasis, to tumor recurrence and poor survival after R0 resection[7,12]. In our study, vein abutment and tumor enhancement were significant predictors of ER in univariate analysis but did not emerge as independent predictors in multivariate analysis. This may be attributed to the strong discriminatory power of CT-ECV, CA19-9, and tumor grade, which diminished the contribution of these CT characteristics. The prognostic role of vascular invasion in PDAC has been explored, with some studies indicating that portal or splenic vein invasion is associated with worse outcomes[13,29]. We speculate that tumor is more likely to be systemic when it has invaded the portal vein or its branches. Indeed, recent findings have confirmed the prognostic significance of portal vein circulation tumor cells in PDAC patients[30,31], supporting the hypothesis that vascular invasion reflects systemic disease.

This study has some limitations. First, single-center relatively small R0-resected cohorts may limit generalizability; thus, validation in a larger population of PDAC individuals is warranted. Second, this study used retrospective observational data and selective bias is unavoidable. Prospective studies will be needed to further validate the findings. Third, the small sample size leads to insufficient adjustment efficacy for key confounding variables, such as age and surgery types. Future studies should expand the sample size through multicenter cohorts to more comprehensively control for confounding factors. Fourth, ECV was calculated using a single CT approach, and the prognostic value of ECV from dual-energy CT in post-surgery PDAC needs more investigation. Finally, 3-minute equilibrium-phase timing, although validated for hepatic fibrosis[32], requires optimization for pancreatic applications.

CONCLUSION

Combined CT-derived ECV and CA19-9 provide superior ER prediction compared to tumor grade alone in PDAC patients. As a noninvasive, widely accessible imaging biomarker, CT-ECV offers critical insights for risk stratification and personalized therapeutic strategies in PDAC management.

ACKNOWLEDGEMENTS

All authors would like to thank all the subjects who participated in this study.