Prediction of early recurrence of pancreatic ductal adenocarcinoma after resection

Abstract

Background: Even after curative resection, pancreatic ductal adenocarcinoma (PDAC) patients suffer a high rate of recurrence. There is an unmet need to predict which patients will experience early recurrence after resection in order to adjust treatment strategies.

Methods: Data of patients with resectable PDAC undergoing surgical resection between January 2005 and September 2018 were reviewed to stratify for early recurrence defined as occurring within 6 months of resection. Preoperative data including demographics, tumor markers, blood immune-inflammatory factors and clinicopathological data were examined. We employed Elastic Net, a sparse modeling method, to construct models predicting early recurrence using these multiple preoperative factors. As a result, seven preoperative factors were selected: age, duke pancreatic monoclonal antigen type 2 value, neutrophil:lymphocyte ratio, systemic immune-inflammation index, tumor size, lymph node metastasis and is peripancreatic invasion. Repeated 10-fold cross-validations were performed, and area under the receiver operating characteristic curve (AUC) and decision curve analysis were used to evaluate the usefulness of the models.

Results: A total of 136 patients was included in the final analysis, of which 35 (34%) experienced early recurrence. Using Elastic Net, we found that 7 of 14 preoperative factors were useful for the predictive model. The mean AUC of all models constructed in the repeated validation was superior to the standard marker CA 19-9 (0.718 vs 0.657), whereas the AUC of the model constructed from the entire patient cohort was 0.767. Decision curve analysis showed that the models had a higher mean net benefit across the majority of the range of reasonable threshold probabilities.

Conclusion: A model using multiple preoperative factors can improve prediction of early resectable PDAC recurrence.

Fig 1. Kaplan-Meier analysis of overall survival.

Fig 2

Decision curve analysis for the constructed models and other risk factors in (A) repeated 10-fold cross validation and (B) the entire cohort. The net benefit (y-axis) was calculated by summing the benefits (true-positive results) and subtracting the harms (false-positive results). The constructed models had the highest (mean) net benefit compared with the other assessments.

Fig 3. Kaplan-Meier analysis of relapse-free survival of patients with a high (≥ -0.90) versus low (< -0.90) Elastic Net model score in the whole cohort.