ABSTRACT
BACKGROUND: The most used pancreatic cancer (PC) resectability criteria are descriptive in nature or based solely on dichotomous degree of involvement (< 180° or > 180°) of vessels, which allows for a high degree of subjectivity and inconsistency. METHODS: Radiographic measurements of the circumferential degree and length of tumor contact with major peripancreatic vessels were retrospectively obtained from pre-treatment multi-detector computed tomography (MDCT) images from PC patients treated between 2001 and 2015 at two large academic institutions. Arterial and venous scores were calculated for each patient, then tested for a correlation with tumor resection and R0 resection. RESULTS: The analysis included 466 patients. Arterial and venous scores were highly predictive of resection and R0 resection in both the training (n = 294) and validation (n = 172) cohorts. A recursive partitioning tree based on arterial and venous score cutoffs developed with the training cohort was able to stratify patients of the validation cohort into discrete groups with distinct resectability probabilities. A refined recursive partitioning tree composed of three resectability groups was generated, with probabilities of resection and R0 resection of respectively 94 and 73% for group A, 61 and 35% for group B, and 4 and 2% for group C. This resectability scoring system (RSS) was highly prognostic, predicting median overall survival times of 27, 18.9, and 13.5 months respectively for patients in RSS groups A, B, and C (p < 0.001). CONCLUSIONS: The proposed RSS was highly predictive of resection, R0 resection, and prognosis for patients with PC when tested against an external dataset.
