Background: Delirium is associated with increased morbidity, mortality and health care costs in the hospital setting secondary to increased length of stay and poor long term outcomes (1,2). A systematic review of different delirium prediction models found limitations with these models including derivation from small patient populations, lack of heterogeneity, accuracy and reliability as noted with poor validation studies (3). No one particular delirium prediction tool is used widely in clinical practice. Effective reduction in delirium incidence by 40% has been proven using a multicomponent intervention program in high-risk individuals (4) which emphasizes the need for better prediction of high risk patients.
Methods: A retrospective electronic data extraction was performed to include all Mayo Clinic Hospital, Rochester, MN, admissions. Patients age 50 years and older who have the diagnosis of “Delirium, Acute Encephalopathy, or Toxic/ Metabolic encephalopathy” at hospital admission, during hospitalization or at discharge, or a CAM test being positive during their admission stay as screened by nursing staff between January 2012- December 2017 were included in the study. IRB approval was obtained prior to the data extraction.We analyzed the electronic data for thirty different established known risk factors for delirium as listed in Table 1. The population is randomly divided into two cohorts namely a derivation cohort (2/3 proportion) and validation cohort (1/3 proportion). Data Analysis was performed using SAS software. Predictive modeling was performed utilizing least absolute shrinkage and selection operator (LASSO) penalized logistic regression using 10-fold cross-validation. We assumed a priori that contribution of many risk factors may be different for medical versus surgical admissions. Therefore, estimation of regression coefficients was performed separately for medical and surgical patients. The tool developed from data analysis of derivation cohort was later analyzed to measure performance separately in validation cohort.
Results: Mayo delirium prediction (MDP) tool is derived from a derivation cohort (n=80,000) based on regression coefficients and then internally validated in a separate validation cohort (n=40,764). The area under the ROC curve (AUROC) for MDP tool using derivation cohort was 0.85 (95% C.I. 0.846 to 0.855). For the validation cohort, AUROC was 0.84 (95% C.I. 0.834 to 0.847).For both derivation and validation cohorts, patients were classified into one of the three risk groups based on their predicted probability of delirium: Low (≤5%), Moderate (6-29%) and High (≥30%). About 60% of the patients were classified as low risk and 5.5% were classified as high risk based on prediction probability stratification in both cohorts. In the derivation cohort, observed incidence of delirium was 1.7%, 12.8%, and 44.8% (low, moderate, and high risk respectively) which is similar to the incidence rates in validation cohort of 1.9%, 12.7% and 46.3%.
Conclusions: Mayo delirium prediction tool appears to be very promising in predicting delirium risk. Further prospective internal validation and external validation studies would further analyze and identify the strengths and weakness of this prediction tool.