Development and Validation of a MicroRNA Signature Predictive of Local-Regional Recurrence and Overall Survival after Resection of Pancreatic Ductal Adenocarcinoma (PDAC)

Abstract

Prognosis for PDAC remains poor despite oncologic resection and adjuvant chemotherapy, with local-regional recurrence (LRR) rates as high as 60-70% at 2 years and 5-year overall survival (OS) of less than 30%. The use of postoperative radiotherapy (RT) is controversial, and identification of novel molecular markers may assist in selecting patients at high risk for LRR who might benefit from RT. We hypothesized we could identify a novel microRNA (miRNA) expression profile associated with LRR and OS in resectable PDAC. We isolated total RNA from formalin-fixed, paraffin-embedded tumor and adjacent normal tissue from patients who underwent surgery and adjuvant chemotherapy at 3 institutions. We created equal-sized training and validation cohorts balanced by institution and LRR events. MiRNA expression was analyzed using a multiplex analysis platform human miRNA v3 platform. Using mixed modeling, we identified miRNAs differentially expressed in tumor and normal tissue in the full dataset and used elastic net modeling to identify miRNAs predictive of LRR and OS in the training dataset. Risk scores were generated using Cox coefficients for each miRNA. For multivariable analysis (MVA), we evaluated the association of risk score with LRR and OS using Cox proportional hazards, accounting for age, pathologic T and N stage, histologic grade, postoperative CA 19-9, and margin status. For Kaplan-Meier analysis, each cohort was dichotomized by its median risk score. There were 90 patients in the training and 93 patients in the validation cohort. Most (96.2%) patients received no RT. Risk score predicted LRR in the training (hazard ratio [HR] = 1.4; 95% confidence interval [CI] 1.1 - 1.8; p = 0.002) and validation cohort (HR = 1.3; 95% CI 1.0 - 1.6; p = 0.046). It did not predict OS in the training cohort (p = 0.4), but was predictive in the validation cohort (HR = 1.2; 95% CI 1.0 - 1.4; p = 0.04). After dichotomization, binary risk score predicted LRR in the training (HR = 3.3; 95% CI 1.5 - 7.0; p = 0.002) and validation cohort (HR = 2.1; 95% CI 1.1 - 4.0; p = 0.03). It trended toward association with OS in the training cohort (HR = 1.6; 95% CI 0.98 - 2.5; p =0.058) and was significant in the validation cohort (HR = 1.6; 95% CI 1.0 - 2.5; p = 0.04). On MVA, binary risk score independently predicted LRR (HR = 3.7; 95% CI 1.2 - 11.2; p = 0.02) and OS (HR = 2.1; 95% CI 1.0 - 4.4; p = 0.039) in the training cohort. In the validation cohort, there was a trend toward association with LRR (HR = 2.2; 95% CI 0.98 - 5.1; p = 0.056) and OS (HR = 1.8; 95% CI 0.9 - 3.4; p = 0.09). Using a multi-institutional dataset, we identified and validated a miRNA signature associated with LRR and OS in patients with resectable PDAC. This signature may be useful in selecting patients at higher risk of LRR who would benefit from adjuvant RT, and should be tested in prospective clinical trials.