Development and Validation of a Survival Prediction Model for Patients With Pancreatic Cancer.

Abstract

Introduction: Patients with pancreatic ductal adenocarcinoma (PDAC) face challenging treatment decisions following their diagnosis. We developed and validated a survival prognostication model using routinely available clinical information, patient-reported symptoms, performance status, and initial cancer-directed treatment.

Methods: This retrospective cohort study included patients with PDAC from 2007 to 2020 using linked administrative databases in Ontario, Canada. Patients were randomly selected for model development (75%) and validation (25%). Using the development cohort, a multivariable Cox proportional hazards regression with backward stepwise variable selection was used to predict the probability of survival. Model performance was assessed on the validation cohort using the concordance index and calibration plots.

Results: There were 17,450 patients (49% female) with a median age of 72 years (interquartile range 63-81) and a mean survival time of 9 months. In the derivation cohort, 1,469 patients (11%) had early stage, 4,202 (32%) had advanced stage disease, and 7,417 (57%) had unknown stage. The following factors were associated with an increased risk of death by more than 10%: tumor in the tail of the pancreas; advanced stage; hospitalization 3 months before diagnosis; congestive heart failure or dementia; low, moderate, or high pain score; moderate or high appetite score; high dyspnea and tiredness score; and a performance status score of 60-70 or lower. The calibration plot indicated good agreement with a C-index of 0.76.

Discussion: This model accurately predicted one-year survival for PDAC using clinical factors, symptoms, and performance status. This model may foster shared decision making for patients and their providers.