External Validation of the Mayo Delirium Prediction Tool in Individuals With Cancer.

Abstract

To validate the Mayo Delirium Prediction (MDP) tool in a population of patients with cancer. This institutional review board-approved, retrospective validation study included 930 hospitalized cancer patients, aged 50 years or older, admitted to the University of Texas MD Anderson Cancer Center from September 1, 2022, to September 1, 2023. Patients with acute substance use disorder were excluded. Recalibrated MDP variables recorded within 24 hours of admission were extracted from the electronic health record. The MDP tool's predictive accuracy was assessed by receiver operating characteristic curves, sensitivity, specificity, positive predictive value, and negative predictive value for patients admitted to medical and surgical services. The cohort comprised 767 medical and 163 surgical patients (median age, 67 years). The MDP tool demonstrated areas under the receiver operating characteristic curve of 0.8047 (95% CI, 0.7585 to 0.8509) for the entire cohort, 0.7867 (95% CI, 0.7371 to 0.8363) for medical patients, and 0.9488 (95% CI, 0.8795 to 1.0000) for surgical patients. At a high risk for delirium score of 0.3 or above, sensitivity was 42.6% (95% CI, 32.6% to 52.5%) and specificity was 94.6% (95% CI, 93.1% to 96.1%). At a low risk for delirium score of 0.05 or lower, sensitivity was 88.3% (95% CI, 81.8% to 94.8%) but specificity was 50.5% (95% CI, 47.1% to 53.9%). The MDP tool demonstrates good capability for predicting delirium risk in hospitalized cancer patients. Prediction was even more robust in surgical patients compared with medical patients. Performance differs by predictive risk, highlighting the need for tailored risk cutoff application in clinical practice.