The stakes could hardly be higher. Even when surgeons achieve what is known as an R0 resection, meaning the tumor is completely removed with microscopically clear margins, more than 80 percent of patients eventually experience local recurrence or distant metastasis. Between 40 and 50 percent of these relapses occur within just 12 months of surgery, a phenomenon clinicians call early recurrence, and it portends a grim long-term outlook. Current guidelines from the Chinese Society of Clinical Oncology and the National Comprehensive Cancer Network recommend that patients with resectable tumors and high-risk features receive neoadjuvant therapy, meaning chemotherapy before the operation, but those guidelines stop short of clearly defining who qualifies as high risk. The new model aims to fill precisely that gap.

The retrospective study was conducted at the Cancer Hospital of the Chinese Academy of Medical Sciences, where researchers reviewed the records of 239 consecutive patients who underwent curative-intent pancreatic resection for histologically confirmed pancreatic ductal adenocarcinoma between January 2017 and June 2021. After applying strict exclusion criteria—ruling out patients who had already received neoadjuvant chemotherapy, those who died within 30 days of surgery, individuals with prior malignancies, incomplete records, or images of unusable quality—131 patients formed the final study population. This group was randomly divided at a 7:3 ratio into a training cohort of 91 patients used to build the model, and an independent validation cohort of 40 patients used to test it.

Every enrolled patient had undergone a contrast-enhanced MRI of the pancreas within four weeks before surgery, performed on 3.0-Tesla scanners following a standardized protocol. The examination captured a battery of sequences, including T1- and T2-weighted imaging and diffusion-weighted imaging, before and after intravenous injection of a gadolinium-based contrast agent. Multiphase contrast-enhanced scanning tracked the agent’s passage through arterial, pancreatic parenchymal, portal venous, and delayed phases. Two radiologists, blinded to all clinical data and outcomes, independently scored a checklist of features defined by China’s evidence-based guideline for pancreatic solid tumor imaging reports: tumor size, location, degree of enhancement relative to normal pancreas, peripancreatic fat infiltration, invasion of adjacent organs, imaging signs of lymph node spread, dilation of the pancreatic and bile ducts, and atrophy of the downstream pancreas.

Among all the imaging patterns examined, one stood out with striking statistical force: peripheral rim enhancement, an appearance in which the tumor’s outer rim lights up with contrast while its core remains dark. In the multivariate analysis, rim enhancement carried an odds ratio of 18.93, meaning patients whose tumors displayed this pattern had nearly nineteen-fold higher odds of early recurrence than those without it. The pattern has a biological logic—prior work, including a study by Lee and colleagues in the journal Radiology, has linked rim enhancement to poorly differentiated, biologically aggressive tumors. Invasion of adjacent organs such as the stomach, colon, or spleen was the second independent imaging predictor, with an odds ratio of 3.84, intuitively reflecting the tumor’s invasive behavior.

The blood test proved equally powerful. Patients whose preoperative serum CA19-9 exceeded 180 units per milliliter had an odds ratio of 7.67 for early recurrence, consistent with a large body of literature showing that elevated levels of this carbohydrate antigen reflect a heavier tumor burden and worse prognosis. Notably, conventional staging measures—tumor size reflecting T stage and lymph node status reflecting N stage—showed associations in the initial univariate analysis but lost their predictive value once the MRI features were taken into account, suggesting that direct radiographic signatures of tumor aggressiveness may outperform traditional anatomic staging when all variables compete within a single multivariate model.

From these three independent predictors, the team constructed a nomogram, a visual scoring tool that assigns weighted points to each risk factor and sums them into a single probability estimate. The model’s discriminatory power, measured by the area under the receiver operating characteristic curve, reached 0.87 in the training cohort and 0.83 in the validation cohort, values considered good to excellent for a clinical prediction tool. Calibration curves confirmed close agreement between predicted and observed recurrence rates, and decision curve analysis showed the model delivered a net clinical benefit across threshold probabilities from 0.0 to 0.9 in both cohorts. Using an optimal cutoff score of 64.32, derived from the maximum Youden index, the model achieved a sensitivity of 80.30 percent and a specificity of 70.77 percent across the entire patient group.

The most clinically consequential result came from survival analysis. When patients were split into high- and low-risk groups by the nomogram score, those in the high-risk category showed dramatically shorter recurrence-free survival in both the training and validation cohorts, with log-rank tests yielding P values below 0.001 in each. Importantly, the model proved robust in subgroup analyses stratified by tumor location, by CA19-9 level using 37 units per milliliter as the cutoff, and by whether patients received postoperative adjuvant chemotherapy. This last check matters because roughly 5 to 10 percent of patients carry a Lewis antigen-negative phenotype and cannot produce detectable CA19-9 at all, a confounder that could otherwise have undermined the model; the subgroup analysis showed reliable performance even among patients with normal marker levels.

The authors are candid about the study’s limitations. It was a single-center, retrospective analysis validated only on an internal split of the same dataset, so external, multicenter validation will be essential before the nomogram can be widely adopted. The 180 units per milliliter CA19-9 threshold, drawn from the team’s own prior work, is not standardized—published cutoffs range from 37 to 200 units per milliliter—and the findings apply only to patients with resectable tumors who undergo complete R0 resection, not to those with borderline resectable disease or positive margins. Even so, the appeal of the approach lies in its simplicity: no artificial intelligence black box, no specialized biomarkers, just two tests already performed in any oncology workup. If prospective validation succeeds, surgeons could soon review a patient’s MRI and blood work on the day of consultation, calculate a recurrence score, and decide whether the wisest course is immediate surgery or neoadjuvant chemotherapy first—potentially sparing the highest-risk patients from an operation their tumor is destined to outlast.

Subject of Research: A preoperative nomogram using MRI features and serum CA19-9 to predict early recurrence of pancreatic ductal adenocarcinoma after curative resection

Article Title: Predicting early recurrence risk in patients with pancreatic ductal adenocarcinoma patients after curative resection based on preoperative MRI features and CA19-9

Article References: Predicting early recurrence risk in patients with pancreatic ductal adenocarcinoma patients after curative resection based on preoperative MRI features and CA19-9. (n.d.). https://doi.org/10.1007/s44178-026-00274-9

Image Credits: AI Generated

DOI: 10.1007/s44178-026-00274-9

Keywords: pancreatic cancer, pancreatic ductal adenocarcinoma, MRI, CA19-9, early recurrence, nomogram, R0 resection, neoadjuvant therapy, rim enhancement, risk stratification, biomarker, predictive model