Real-time mortality risk calculator following pancreatoduodenectomy: quantifying the impact of perioperative events.

Abstract

Estimation of mortality risk traditionally has only included preoperative factors. We sought to develop "real-time" mortality risk-calculator for patients who undergo pancreatoduodenectomy (PD) based on preoperative factors, as well as events that occurred during the course of patient's surgery and hospitalization. Patients who underwent PD from 2014 to 2018 were identified in the ACS-NSQIP dataset. Training and validation cohorts were created. Pre-, intra-, and post-operative models to predict 30-day mortality were developed based on perioperative variables selected by stepwise cox regression analyses; model performance was assessed using AUC. Among 17,683 patients who underwent PD, 1.6% died within 30-days. Patient factors and events associated with 30-day mortality were incorporated into a risk calculator (https://ktsahara.shinyapps.io/Real-timePD/). The accuracy of the risk-calculator increased relative to hospital time-course in both the training (AUC, pre-:0.696, intra-:0.724, post-operative:0.871) and validation (AUC, pre-:0.681, intra-:0.702, post-operative:0.850) cohorts. One in 3 patients had a concordant calculated risk of mortality using pre-versus postoperative variables to inform the risk model (kappa=0.474). Risk of mortality fluctuated over the hospital course following PD and preoperative risk assessment was often discordant with risk assessed at other periods. The proposed "real-time" calculator may help better stratify patients with increased risk of 30-day mortality.