We retrospectively and consecutively identified 2,379 patients who underwent pancreatic surgeries at Severance Hospital from January 1, 2007, to August 13, 2020. Inclusion criteria were as follows: patients who (1) underwent pancreatectomy for PDAC; (2) received NAT with or without radiotherapy. One-hundred-and-fifty-four patients were identified. Nine patients were excluded: initially metastatic PDAC (n=4); palliative resection to control hemorrhage (n=1); presence of other malignancy (n=2); and baseline CT examination unavailable (n=2). Finally, 145 patients were included (Fig. 1).

Demographic, laboratory, treatment history, and histopathologic variables were collected from the electronic records by a study coordinator (G.E.) under the guidance of a radiologist (H.K.Y.). Laboratory variables, including serum CA19-9, carcinoembryonic antigen (CEA), and total bilirubin, were routinely measured at baseline and before the decision to perform surgery, and the interval change of each laboratory value was calculated. Lewis antigen status was collected for the patients whose baseline CA19-9 was not elevated [11]. Treatment history included the type and duration of NAT, type of pancreatic surgery, and postoperative adjuvant therapy.

Histopathologic data included resection margin status, tumor differentiation grade, and pathologic TNM staging. A positive resection margin (R1 resection) was defined as when tumor cells were present at the cut-resection margins or within 1 mm from the circumferential resection margins using the British Royal College of Pathology guidelines [12]. Pathologic TNM staging was based on the American Joint Commission on Cancer (AJCC) Cancer Staging Manual, 8th edition.

Our primary outcome was OS, defined as the time from baseline CT to death, with final evaluation on September 22, 2020. Our secondary outcome was RFS, the time from surgery to recurrence. Recurrence was diagnosed by clinicoradiologic evidence or histologic confirmation. When development of recurrence was suspected but not certain on a single imaging study, it was finally diagnosed when interval growth was noted or at least another imaging modality, i.e., magnetic resonance imaging or positron emission tomography–computed tomography, added evidence supportive of recurrence. The time of recurrence was defined as the date of the imaging study where the suspicious lesion was first noticeable in retrospect.

Pancreatic CT was performed as recommended by the National Comprehensive Cancer Network (NCCN) guidelines [13]. CT images were obtained using multidetector row CT scanners. The scanning protocol included unenhanced and contrast-enhanced biphasic imaging in the pancreatic and portal venous phases after intravenous administration of iopamidol (Iopamiro 370, Bracco, Milan, Italy) at a rate of 1.6-4.0 mL/sec depending on the patient’s body weight. Pancreatic and portal venous phase images were obtained by adding 23 seconds to the time of peak abdominal aortic enhancement calculated at the hepatic hilum, and 25 seconds to the end of the pancreatic phase, respectively. Reconstructed slice thickness and interval were 3 mm and 3 mm for both transverse and coronal images.

Two gastrointestinal radiologists (M-S.P. and H.K.Y., with 21 and 5 years of specialized experience in gastrointestinal imaging, respectively) retrospectively and independently reviewed the images. They were blinded to clinical, laboratory, histopathologic, and follow-up results but knew that the patients were treated for PDAC. CT findings were selected from the parameters in the PDAC radiology reporting template proposed by the Society of Abdominal Radiology and the American Pancreatic Association [14]: tumor size, individual tumor-vessel contact (no contact, abutment [≤ 180°], or encasement [> 180°]) and venous stenosis, resectability status as per NCCN Guidelines, presence of suspicious lymph nodes, and invasion to the stomach or duodenum. The tumor diameter was measured on the transverse plane. Each finding was assessed on CTs both at baseline and preoperatively, and then the interval change of the finding was evaluated. We regarded that there was decrease in tumor size and/or tumor-vessel contact when “any degree of radiologic improvement in tumor size and/or tumor-vessel contact” was present [15,16]. To maintain objectivity, radiologic improvement in tumor size was regarded to be present when there was approximately ≥ 3 mm decrease in tumor diameter since baseline. After independent image review, interobserver agreement was evaluated. In the survival analyses, the consensus results of the two radiologists regarding the qualitative features were used.

Univariable Cox proportional hazard analyses were performed to determine potentially significant factors for OS and RFS. Multivariable Cox proportional hazard analyses were performed using the variables that showed p < 0.05 in each univariable analysis. While individual tumor-vessel contact was assessed, collective perivascular assessment results (i.e., changes in tumor-any artery contact and tumorany vein contact or venous stenosis, NCCN resectability at baseline and preoperative CT, and change in the resectability) were used in the multivariable analysis to enhance applicability in practice. Risk stratification criteria, i.e., postNAT response criteria, were proposed using the independent preoperative predictors from the multivariable analyses for OS. Kaplan-Meier survival analysis and log-rank test were performed for the criteria. The discrimination performances of the post-NAT response criteria and post-NAT pathological (yp) TNM staging were evaluated using Harrell’s C-index [17]. The difference in C-indices and 95% confidence interval (CI) of the calibration curve were estimated using bootstrap with 1,000 resampling. The calibration performance was evaluated using a calibration plot of the predicted and observed 3-year OS.

Interobserver agreement for the qualitative CT findings was evaluated with percentage of agreement and Cohen’s kappa coefficient. The kappa value (the level of agreement) was defined as follows: 0.00-0.20, poor; 0.21-0.40, fair; 0.41-0.60, moderate; 0.61-0.80, good; and 0.81-1.00, excellent. As the magnitude of kappa is affected by the prevalence of the attribute [18], some findings of low prevalence were better assessed with percentage of agreement. Interobserver agreement for the quantitative CT findings was evaluated with intraclass correlation coefficient. p < 0.05 was considered to be statistically significant. All statistical analyses were performed by MedCalc for Windows, ver. 20.014 (MedCalc Software, Ostend, Belgium) and R software (ver. 4.1.1., http://www.R-project.org/, Vienna, Austria).

We divided the study sample into the development and test sets to perform internal validation (S1 Fig.). Details are in Supplementary Material.

One-hundred-and-forty-five patients (mean age±standard deviation, 60.8±8.8 years; 80 men and 65 women) with PDAC who underwent curative-intent pancreatic surgery after NAT with (n=98, 67.6%) or without (n=47, 32.4%) radiotherapy were analyzed. 8.3% (12/145) of the patients with locally advanced PDAC received conversion therapy. All the 145 patients completed chemo-(FOLFIRINOX, n=36; modified FOLFIRINOX, n=5; gemcitabine and cisplatin, n=1; gemcitabine and erlotinib, n=2; gemcitabine and nab-paclitaxel, n=3) or chemoradiation therapy (n=98). The interval from the baseline CT to surgery was 6.6±5.8 months. In 34.5% (50/145) of the patients, the baseline serum CA19-9 levels were not elevated, i.e., < 37 U/mL. Forty-eight of the 50 patients in whom the baseline CA19-9 levels were not elevated were Lewis-negative phenotype; Lewis a/b (–/–) in 31, (–/+) in 16, and (+/–) in one patient(s); and the status was unknown for the remaining two patients. Initially elevated CA19-9 normalized in 47.4% (45/95) and remained elevated in 52.6% (50/95). R0 resection was performed in 67.6% (98/145). The median OS and RFS were 36.2 (95% CI, 25.8 to 55.5) and 12.0 (95% CI, 10.7 to 19.0) months, respectively. The other data are listed in Table 1.

The mean diameter of PDACs was 29.8±11.9 mm on the baseline CT and 20.8±10.9 mm on the preoperative CT. A decrease in tumor size was noted in 68.3% (99/145) of the patients. At baseline, resectable, borderline resectable, and locally advanced cases were 49.7% (72/145), 42.1% (61/145), and 8.3% (12/145) of the entire population, respectively; at preoperative CT, they were 53.1% (77/145), 42.1% (61/145), and 4.8% (7/145). A suspicious node was present in 37.2% (54/145) of the patients at baseline and 27.6% (40/145) at preoperative CT. Tumor invasion to the stomach or duodenum was noted in 24.8% (36/145) of the patients at baseline and 21.4% (31/145) at preoperative CT.

Clinical, surgical, and pathological characteristics, and oncologic outcomes of the development and test sets are compared in S5 Table. Significantly higher proportions of patients showed radiologic improvement after NAT regarding various CT findings in the test set compared to the development set (S6 Table). Cox survival analysis of preand perioperative predictors of OS using the development set showed elevated preoperative CA19-9 and stable or increased tumor size on CT were significant independent poor predictors (S7 Table). Harrell’s C-indices of the postNAT response criteria and ypTNM staging did not significantly differ in the development and test sets. There were significant differences in OS and RFS between the groups stratified by post-NAT response criteria. The detailed results are in Supplementary Material.